Integrated Environmental Mapping and Monitoring:

Thesis for the Degree of Philosophiae Doctor

Trondheim, December 2016

Norwegian University of Science and TechnologyFaculty of Natural Sciences and TechnologyDepartment of Biology

Ingunn Nilssen

Integrated EnvironmentalMapping and Monitoring:

A methodological approach for end-users

NTNUNorwegian University of Science and Technology

Thesis for the Degree of Philosophiae Doctor

Faculty of Natural Sciences and TechnologyDepartment of Biology

SBN 978-82-326-1896-5 (printed ver.) ISBN 978-82-326-1897-2 (electronic ver.) ISSN 1503-8181

Doctoral theses at NTNU, 2016:279

Printed by NTNU Grafisk senter

Acknowledgements

Through my whole career, during my studies, while working with the pollution

control authorities and in the oil and gas industry, I have been working with environmental

monitoring and environmental monitoring related issues. During this time I have seen large

technological advances, potentially enabling a much better understanding of the environment

and environmental processes. When moving between academia, the regulating authorities and

industry my impression is that we, in most cases, all see the need for collaboration to obtain

satisfactory results. That said, there are unfortunately still cultures that are primarily

concerned with "feathering one's own nest”. I am convinced that to solve the environmental

challenges we need a holistic approach that utilises the technological opportunities, that

makes data available and that develops analytical methodologies for further exploration of the

inherent knowledge available in the data. To manage this we have to work with an

interdisciplinary mindset. I truly believe that the proposed approach and the different papers

included in this thesis can contribute to show the potential for enhancing the environmental

knowledge through proper planning, data sharing and interdisciplinary collaboration and

communication. Such an approach is a precondition for a successful integrated environmental

mapping and monitoring (IEMM) and thereby reasonable environmental management.

I would like to thank Statoil for given me the opportunity to perform this industrial

PhD. Furthermore I would like to thank my open minded and interdisciplinary supporting

supervisors; Geir Johnsen, Tim W. Nattkemper, Asgeir J. Sørensen, Ståle Johnsen and Vidar

Hepsø. Thanks to Mona Låte and Ingvar Eide for all your help and support, to all co-authors,

colleagues at Statoil Research Centre in Trondheim and staff and fellow students at

Trondheim Biological Station (none mentioned, none forgotten). This has been a great

interdisciplinary journey and I have learned so much from each of you. The last four years

have been hard work and a lot fun; it has been a true honour working with each of you!

I would also like to thank my family for all their support. A special thanks to Helge

who have encouraged and supported me 24/7. You are patient, kind, a brilliant amateur cook

and rather strict when required. I believe you have balanced the performance of these

characteristics pretty well through this 4-years journey.

Ingunn Nilssen

Trondheim 14.09.2016

Preface

“The world has only seen a few geniuses.

The rest of us need to collaborate.”

Asgeir J. Sørensen

Contents Acknowledgements .................................................................................................................................. i

Preface ................................................................................................................................................... iii

List of papers ......................................................................................................................................... vii

Contribution ....................................................................................................................................... ix

Abbreviations and definitions ................................................................................................................ xi

1 Introduction ................................................................................................................................ 1

1.1 Background ......................................................................................................................... 1

1.2 Scientific contribution ......................................................................................................... 5

1.3 Data sources ...................................................................................................................... 8

2 Methodological approach ......................................................................................................... 11

2.1 Selection of sensors, sensor carrying platforms and exposure studies ............................ 11

2.1.1 Temporal and spatial coverage and resolution ............................................................ 12

2.1.2 Sensors and sensor carrying platforms: Technology development and improvement 14

2.1.3 Laboratory experiments to close gaps of knowledge ................................................... 15

3 Data sampling and processing .................................................................................................. 17

3.1 Adaptive sampling ............................................................................................................. 17

3.2 Data processing and knowledge accumulation ................................................................ 18

3.2.1 Interpretation of complex data sets from field measurements ................................... 19

3.2.2 Combining laboratory experiments, field measurements and modelling in environmental risk assessment .................................................................................................... 19

4 Basis for decisions: Tools enabling improved collaboration and communication .................... 23

4.1 Interdisciplinary communication and collaboration ......................................................... 24

4.2 Images in environmental monitoring ............................................................................... 25

5 Conclusions and future perspectives ........................................................................................ 29

5.1 Concluding remarks .......................................................................................................... 29

5.2 Future perspectives........................................................................................................... 30

References ........................................................................................................................................ 33

List of papers

I. Nilssen, I., Ødegård, Ø., Sørensen, A. J., Johnsen, G., Moline, M. A., Berge, J.,

2015a. Integrated environmental mapping and monitoring, a methodological

approach to optimise knowledge gathering and sampling strategy. Marine

Pollution Bulletin Vol 96, pp 374-383,

doi: http://dx.doi.org/10.1016/j.marpolbul.2015.04.045.

II. Berge, J., Johnsen, G., Sørensen, A. J., Nilssen, I., 2015. Enabling technology for

Arctic research. Science and Technology Magazine Vol 15, pp 199-201.

III. Johnsen, S., Nilssen, I., Pinto, A. P. B., Torkelsen, T., 2014. Monitoring of impact

of drilling discharges to a calcareous algae habitat in the Peregrino oil field in

Brazil. SPE International Conference on Health, Safety, and Environment in Oil

and Gas Exploration and Production, Long Beach, California, USA.

Doi: http://dx.doi.org/10.2118/168356-MS

IV. Figueiredo, M. A. O., Eide, I., Reynier, M., Villas-Bôas, A. B., Tâmega, F. T. S.,

Ferreira, C. G., Nilssen, I., Coutinho, R., Johnsen, S., 2015. The effect of

sediment mimicking drill cuttings on deep water rhodoliths in a flow-through

system: Experimental work and modeling. Marine Pollution Bulletin Vol 95,

pp 81-88, doi: http://dx.doi.org/10.1016/j.marpolbul.2015.04.040.

V. Osterloff, J., Nilssen, I., Eide, I., de Oliveira Figueiredo, M. A., de Souza

Tâmega, F. T., Nattkemper, T. W., 2016b. Computational visual stress level

analysis of calcareous algae exposed to sedimentation. PLoS ONE Vol 11,

pp e0157329, doi: http://dx.doi.org/10.1371/journal.pone.0157329.

VI. Eide, I., Westad, F., Nilssen, I., de Freitas, F. S., dos Santos, N. G., dos Santos, F.,

Cabral, M. M., Bicego, M. C., Figueira, R., Johnsen, S., 2016. Integrated

environmental monitoring and multivariate data analysis – A case study.

Integrated Environmental Assessment and Management.

Doi: http://dx.doi.org/10.1002/ieam.1840.

VII. Nilssen, I., dos Santos, F., Coutinho, R., Gomes, N., Cabral, M. M., Eide, I.,

Figueiredo, M. A. O., Johnsen, G., Johnsen, S., 2015b. Assessing the potential

impact of water-based drill cuttings on deep-water calcareous red algae using

species specific impact categories and measured oceanographic and discharge

data. Marine Environmental Research Vol 112, Part A, pp 68-77,

doi: http://dx.doi.org/10.1016/j.marenvres.2015.09.008.

VIII. Osterloff, J., Nilssen, I., Nattkemper, T. W., 2016a. A computer vision approach

for monitoring the spatial and temporal shrimp distribution at the LoVe

observatory. Methods in Oceanography, Vol 15-16, pp 114-128,

doi: http://dx.doi.org/10.1016/j.mio.2016.03.002.

IX. Möller, T., Nilssen, I., Nattkemper, T. W., In press. Change detection in marine

observatory image streams using Bi-Domain Feature Clustering. International

Conference on Pattern Recognition, Cancún, México.

X. Nilssen, I., Hepsø, V., Nattkemper, T. W., Johnsen, G., Images in environmental

monitoring: Enhanced knowledge through interdisciplinary collaboration

-two case studies from marine science and engineering.

Submitted to Marine Pollution Bulletin

Contribution

The principles presented in Paper I were the preconditions for the initiation my thesis.

The paper presents a methodological approach for optimised data sampling and knowledge

gathering. The paper was initiated after a UNIS course on underwater robotics (AB834) at

Svalbard in 2014. IN and ØØ were the main contributors to the paper and IN had the lead of

the writing process.

Paper II was initiated by three of the professors from the UNIS course in underwater

robotics (AB334/834). These professors were also co-authors of Paper I. The paper focuses

on how enabling technologies can contribute to increase knowledge about the Arctic

ecosystem. The paper is published in a journal without a peer-review process. IN was

involved in the final stage of the paper contributing with the methodological principles for

the approach presented in paper I.

Paper III is a conference paper summarising the experiences from the Peregrino

environmental monitoring of calcareous algae (PEMCA) project in Brazil with particular

focus on hardware. The conference does not provide peer-review of the presented papers.

The paper was initiated by SJ. IN took an active part in the planning and execution of the

PEMCA project, as well as in writing the paper.

Paper IV is part of the PEMCA project and presents the results from the long-term

exposure study, establishing threshold levels for particulates on two species of red calcareous

algae. The paper was initiated by SJ. Planning, sampling of organisms, building of the flow

through system and laboratory experiments were carried out by the team of MAOF, IR,

ABV-B, FTST, GCF and RC. IE performed the experimental design, the multivariate data

analyses and led the writing process. In addition to her contribution in the planning and

execution process of the PEMCA project, IN took an active part in writing the paper, with

specific focus on the biological parts.

Paper V is also a PEMCA paper, using automatic image analysis to detect changes in

size and colour from photos of red calcareous algae from long-term exposure to particulates.

JO initiated the work and was the main contributor. IN contributed with biological skills in

discussions and in the writing process, particularly Introduction and Discussion. An

important part of the writing process was to facilitate the text for an interdisciplinary

audience.

Paper VI is another PEMCA publication, using multivariate data analysis on various

field measurements and discharge data to investigate possible influence of water-based drill

cuttings on a calcareous algae habitat. IE and FW did all the data processing and the

multivariate data analysis. IE led the writing process, while IN took an active part in writing

the paper, with specific focus on the biological parts.

Paper VII uses the results from PEMCA, including threshold levels from laboratory

study (Paper IV), measured field data and data on drilling discharges (Paper VI), as input data

to the dispersion and risk modelling to improve the risk assessment. IN initiated the paper and

led the writing process. The modelling was performed by FdS, NG and MMC.

Paper VIII uses photos from the online LoVe Ocean Observatory, Norway, for

development of algorithms to monitor shrimp abundance in a cold-water coral reef

automatically. The work was initiated by JO. JO was the main contributor and he led the

writing process. IN contributed with biological skills in the planning and writing process,

particularly the Introduction and Discussion chapters. An important part of the writing

process was to facilitate the text for an interdisciplinary audience.

Paper IX is a conference paper where photos from the online LoVe Ocean

Observatory, Norway, are used to develop algorithms for simultaneous and automatic

detection and grouping of organisms. The paper was initiated and the main contribution has

been performed by TM. IN contributed with biological skills in the writing process and in the

process of facilitating the text for an interdisciplinary audience.

Paper X describes the potential of images to improve environmental knowledge in

general. Why images have the potential of being good communicative and collaborative tools

and how interdisciplinary communication and collaboration can contribute to enhance the

inherent knowledge available in the images is being elucidated. The work was initiated by

VH and IN. IN was the main contributor and led the writing process.

Abbreviations and definitions

Adaptive sampling – sensors or sensor carrying platforms are operating automatically and/or autonomously. Sensors can be programmed to automatically switch on/off when certain conditions are present or the sampling mission is programmed to autonomously follow agiven phenomenon, parameter or object of interest.

Automatic – The system follows a defined program. In the context of this thesis, the term “automatic” is used to describe data sampling and/or data interpretation

Autonomous – action initiated independent of control from outside, e.g. the system makes choices without human interaction. In the context of this thesis “autonomous” is exclusively used with respect to sensor carrying platforms

CPI – Carbon Preference Index

Drill cuttings – particulates of crushed rock originating from the formation rock a well is drilled through

Flexible environmental mapping and monitoring – the measured parameters and the spatial and temporal coverage and resolution are selected based on (and will vary dependent on) the purpose of the mission, the area of interest and possible anthropogenic influence in the area

Infauna – Aquatic animals that live beneath the surface of a sea or lake floor, such as clams or burrowing worms

IEMM – Integrated environmental mapping and monitoring, is performed as a part of an already ongoing activity, such as the daily operations of an oil field or routine cruises for fish stock estimations. Such an approach can provide cost-efficient monitoring through targeted measurements, multiuse of data, e.g. leak detection and current measurements, or access toadditional sampling for limited costs due to available infrastructure.

Interdisciplinary – collaboration between/activity involving two or more scientific disciplines

IOP – Inherent Optical Properties

Lander – stationary sensor carrying platform

LoVe – Lofoten-Vesterålen

Ocean Observatory – stationary sensor carrying platform

PAH – Polycyclic Aromatic Hydrocarbons

Parameter – a coefficient that might be either constant or time varying, in contrast to a variable that is a time varying state of a dynamic system. So, all variables are parameters but a parameter might not necessarily be a variable. For instance is a manually set white balance in a camera or a threshold value derived from literature a constant (parameter) and not a variable. The term parameter is in the context of this thesis used for both terms.

PCA – Principal Component Analysis

PEC – Predicted Exposure Concentration

PEMCA – Peregrino Environmental Monitoring of Calcareous Algae

Photosynthetic efficiency – in vivo maximum quantum yield of charge separation in photosystem II in dark acclimated cells ( PSII_max)

PLS – Partial Least Squares

PNEC – Predicted No Effect Concentration

RGB – Red, Green, Blue (the three colour channels utilised in ordinary cameras)

ROV – Remotely Operated Vehicle

SSD – Species Sensitivity Distribution

TAR – Terrestrial-Aquatic Ratio

Transect – a path or track investigated by a sensor carrying platform counting and/orrecording occurrences of parameters or species

Variable – In the context of this thesis used related to the multivariate data analysis,describing the characteristic of a unit

Water-based drill cuttings – particulates from the formation rock mixed with residuals of water-based drilling fluids. Drilling fluids are used to lubricate and withstand the pressure from the formation when drilling a well

1 Introduction

The motivation for this thesis is to describe a methodological approach for integrated

environmental mapping and monitoring (IEMM). Furthermore, to demonstrate how IEMM

can benefit from a flexible and interdisciplinary approach by adjusting sampling strategy and

data interpretation to mission and processes, object and/or area of interest. To optimise data

processing and visualisation of the results presented to end-users and decision makers, joint

effort from a suite of scientific disciplines is required. Data with adequate temporal and

spatial coverage and resolution can either be utilised as recorded or as input to applications

such as dispersion models, ecosystem modelling, risk assessment and/or being visualised in

portals for shared situation awareness. Access to such data can improve environmental

management and enhance environmental performance, if provided to the end-users in due

course.

In the following sections the importance of 1) measurements adjusted to purpose,

2) use of enabling technologies 3) awareness of the human capabilities of interpreting objects

and movements and 4) interdisciplinary communication and collaboration will be elaborated.

The latter covering both, utilising the inherent knowledge available in the images themselves

and their communicative capabilities for use in decisions and for management purposes. The

pullet points listed above are all essential aspect of awareness for a flexible environmental

monitoring. The thesis presents the entire concept. The different aspects (steps) will briefly

be described in the main text and the different papers (Paper I – X) will be used to exemplify

and elucidate the individual aspects. The thesis is, to a large extent, organised in accordance

with the layout of Paper I and Paper X, describing the planning, execution and knowledge

accumulation steps, human factors related to data interpretation and visualisation and

interdisciplinary communication and collaboration, respectively.

1.1 Background

Recent advances in enabling technologies are expected to enhance the current

understanding of the environment and its dynamic processes. Some examples from the Arctic

where use of enabling technologies has led to new knowledge about the ecosystem during the

polar night are Berge et al. (2015a), Berge et al. (2015b) and Cohen et al. (2015). Although

the human visual system has an impressive ability in recognising human movements in time

and space, research shows that it performs worse when confronted with objects and

unfamiliar organisms (Shiffrar and Thomas, 2013; Smith and Kosslyn, 2014), such as marine

organisms. According to the technology philosopher Don Ihde, it is the applied sensor

technology used that determines the nature of the captured data and how they are presented

(Ihde, 1979; 1991). From this perspective the rapid technology developments are challenging

the human ability to interpret and understand the variety of novel data provided. Ultimately,

this might even contribute to challenge the implementation of integrated environmental

management systems.

Globally, a variety of approaches for environmental management systems exist and

challenges related to implementation have been described, among others, by Borja and Dauer

(2008), Elliott (2014) and Leslie and McLeod (2007). The European Commission launched in

2010 the Europe 2020 strategy on a “smart, sustainable and inclusive growth” (European

Commission, 2010). The Blue Growth strategy represents the maritime area of the strategy. It

states that “The individual sectors of the blue economy are interdependent. They rely on

common skills and shared infrastructure …” and “They depend on others using the sea

sustainably” (Directorate-General for Maritime Affairs and Fisheries, 2012). Examples of

increased focus from industry on the benefits of an integrated use and monitoring of the

environment are Hepsø et al. (2012) and Kongsberg Maritime (n.d.). Several guidelines, of

instance from CCME (2015) and USEPA (2006) describes planning procedures to ensure a

cost-efficient selection of the data needed (qualitatively and quantitatively) to address

possible impact on resources present when evaluating different alternatives, such as

infrastructure developments, use of chemicals and/or discharge regimes. A similar approach

has been established for risk assessment related to oil and gas related activities in cold-water

coral areas in Norway (DNV, 2013). The principles of environmental risk assessment

implemented among others by the European Commission (ECHA, n.d.) and in the United

States (USEPA, 2016) are based on the PEC/PNEC ratio (Predicted Exposure Concentration

and Predicted No Effect Concentration). The PEC/PNEC approach is commonly used by

offshore operators in Europe (Smit et al., 2011). These principles are accepted by the

Convention for the Protection of the Marine Environment of the North-East Atlantic

(OSPAR) as a basis for environmental management of produced water discharges (OSPAR,

2012a, b). Under such a regime dispersion and risk modelling is used to identify the major

risks related to added chemicals and/or natural occurring substances of planned discharges

from a specific activity or a set of simultaneously ongoing industrial activities. Typical inputs

to the models predicting the concentrations of the discharges in the environment (PEC) are

discharge characteristics (type, amounts, rates or volumes, density, depth, etc.) and

geographical information such as bathymetry, water depth and current speed and direction

(modelled or preferably measured). The risk is then assessed based on the comparison of PEC

and PNEC, where PNEC is commonly derived from species sensitivity distribution (SSD)

curves (de Zwart and Posthuma, 2005; Harbers et al., 2006; Smit et al., 2008) or by

application of an assessment factor to the most sensitive species (ECHA, 2008). The lack of

ecosystem relevance by the use of standard toxicity data is a major criticism against the

PEC/PNEC approach. However, if available, species specific effect data can replace the

standard toxicity data in the risk modelling (Beyer et al., 2012; Larsson et al., 2013; Purser

and Thomsen, 2012).

Systems accepting a risk-based approach require an environmental monitoring that is

flexible in all aspects from selection of parameters and sensor carrying platforms to data

access (online or offline) and methodology for data processing and analyses. The measured

parameters and the spatial and temporal coverage and resolution should in each case be

selected based on the mission, the identified risks, the object of interest and anthropogenic

influence in the area (Godø et al., 2014b; Hepsø et al., 2012). The aim of mission can span

from measuring natural fluctuations to monitoring of point discharges from anthropogenic

activities. The object of interest can either be a phenomenon, (an) organism(s), an area

(habitat) or a piece of infrastructure. In addition to sampling of data with adequate temporal

and spatial coverage and resolution, getting access to already existing data from the area of

interest is of importance for the interpretation. Furthermore, of particular importance for

decision makers, also the timeframe for data deliveries are crucial. For mapping purposes an

offline solution should normally be sufficient, while monitoring in cases where immediate

mitigation measures are crucial, online solutions for data access will be required. Also use of

models should constitute a vital part of IEMM, as modelling as part of the planning process

can be used to optimise both the design of the environmental monitoring programme by

predicting the area of influence (Nilssen and Johnsen, 2008) and/or to extrapolate results

obtained from a limited part of an habitat to the habitat in general or experiences from one

area to similar habitats in other geographical areas (Godø et al., 2014b; Godø et al., 2014c;

Hepsø et al., 2012).

In addition to a flexible IEMM concept, including use of models, an interdisciplinary

effort is also required to enhance the environmental knowledge (Brown et al., 2015; Rylance,

2015; Wuchty et al., 2007) by utilising the inherent knowledge available in the data.

According to Carlile (2004) development, change and extension of knowledge occurs when

actors with different scientific background use their already existing knowledge and

experiences on new tasks. A constructive dialogue between open minded actors, experts

within their own disciplines and interested in others, are essential to succeed in

interdisciplinary collaborations (Brown et al., 2015). Elliott (2014) emphasises the

importance of educating graduates “ … willing and able to link the natural and social

sciences …” to successfully achieve an “… integrated marine management framework …”1.

In addition, Brown et al. (2015) pinpoint institutional support and bridging research, policy

and industry as absolute preconditions to facilitate interdisciplinary collaboration in daily

work. The importance of building interdisciplinary scientific capacities has also been

stressed, among others, by Leslie and McLeod (2007).

Due to the vast amounts of novel and complex data, focus on human factors such as

mindset, interdisciplinary communication and the human capability to interpret and

understand various types of data are all essential when presenting data to end-users and

decision makers. A comprehensive interdisciplinary collaboration therefore requires

development or access to appropriate communicative tools (boundary objects) that are

understood by all actors involved (Carlile, 2004). Based on the human cognitive capacities to

detect and classify objects and movements (Biederman, 1987; Shiffrar and Thomas, 2013;

Smith and Kosslyn, 2014), images could be expected to be well suited for the purpose.

Although use of imaging has a long tradition within disciplines such as geology (Almklov

and Hepsø, 2011) and archaeology (Adams, 2013), and that use of photos and video

recordings in marine environmental mapping and monitoring increases rapidly, the numbers

of well established analytical tools are currently relatively limited, with few exceptions

(Beijbom et al., 2016; Faillettaz et al., 2016; Orenstein et al., 2016; Osterloff et al., In Press;

Prasad et al., 2016; Purser et al., 2009; Schoening et al., 2012).

1 Integrated refers here to cross sectorial management

consideration (Figure 1). Selection of technology with appropriate sensitivity, resolution and

coverage (sensor carrying platforms, software etc.) as well as efficient data gathering and

access, interpretation and visualisation are all essential if end-users such as operators or

decision makers shall be provided with optimised data in due course for mitigation measures

to be made. Furthermore, a methodological approach will identify gaps in knowledge,

methodology and technology, and thereby any needs for further development and

improvements in the different steps of the concept from planning and execution to

interpretation of results.

As illustrated in Figure 2, the individual papers of the thesis are all linked to one or

more of the steps in Figure 1 (where the papers are positioned in Figure 2 indicates their area

of contribution). In paper I and II the methodology is proposed and the different aspects of

the methodological approach are addressed and discussed. The planning step (Step 3 in

Figure 1) is exemplified and discussed more in depth in Paper III and IV, hardware

development and bridging gaps of knowledge with laboratory experiments, respectively. In

Paper III the technological experiences and proposed changes with respect to robustness,

handleability and sensors are given. Establishment of threshold levels for sedimentation of

inorganic particulates to the calcareous red algae Mesophyllum engelhartii (Foslie) Adey and

1.2 Scientific contribution

The flexibility introduced with the IEMM concept enhances the need of a

methodological approach where, in each case, all aspects from planning to interpretation and

visualisation of results should assist users and decision makers to identify elements of

Lithothamnion sp. is described in Paper IV. In Paper V, VIII and IX development of

algorithms for automated and semi-automated data analysis of photos related to biological

questions, such as detection of organisms and measurement of biological behaviour, is

described (Step 5, 6 and 10, Figure 1). Paper VI and Paper VII are examples on data analysis

and different data processing and interpretation strategies. In Paper VI multivariate data

analyses were used to increase environmental knowledge identifying correlations in the

complex datasets. Paper VII uses data from exposure studies (lab) on relevant organisms

(Paper IV), field measurements and discharge data (Paper VI) as input data to the dispersion

and risk modelling to reduce the uncertainties in the risk assessment at the Peregrino oil field

(Step 6 and 10 in Figure 1). Paper X demonstrates why and how images are suitable

communicative tools in communication and collaboration between different research fields.

Furthermore, the contribution of interdisciplinary communication and collaboration to

improve environmental understanding in general through enhanced use of the inherent

knowledge available in images, is illustrated. The advantages of using images, such as

photos, video recordings and visualisations from modelling in future applications for end-

users are also addressed. Since Paper X addresses both current use and future opportunities

provided by images, it is located between Step 10 and 11 in Figure 1.

Figure 2: Illustration of the individual papers’ area of contribution in the proposed

methodological model. The shaded green boxes are placed in accordance with their area of

contribution in the methodology figure (Figure 1) and the corresponding steps are indicated

in each of the boxes.

1.3 Data sources

The data used for this thesis are mainly generated from the Peregrino environmental

monitoring of calcareous algae (PEMCA) project (Salgado et al., 2010) and from the

Lofoten-Vesterålen (LoVe) Ocean Observatory (Godø et al., 2014a; Statoil, n.d.). The

experiences from PEMCA, together with other environmental monitoring programmes using

lander technology, such as Morvin (Godø et al., 2014b), formed the basis for the technology

improvements prior to the LoVe deployment. Technology development was related to

robustness, reliability and logistical issues. Also the experiences with photo collection and

processing for documentation of possible impact from drilling discharges at Peregrino

(unpublished data) and Morvin (Buhl-Mortensen et al., 2015; Godø et al., 2014b) initiated

several changes. For more details, see section 2.1.2 Sensors and sensor carrying platforms

and Paper III.

Figure 3: Illustration of the lander hardware development from the PEMCA-lander (a) to the

frame deployed at LoVe (b). The weight and the height have been reduced and, for instance

the fragile echo sounder system has been much more protected from the surroundings.

Illustrations: Terje Torkelsen, Metas AS.

The three-year PEMCA project was initiated at the Peregrino offshore oil field in

Brazil (23.31° S, 41.31º W)2 in 2010. The purpose of the PEMCA project was to monitor the

possible impacts from discharges of water-based drill cutting to the calcareous algae habitat

present at the Peregrino oil field and to develop and perform an alternative environmental

monitoring programme, aligned to the actual discharges and the resources present in the area

of interest (Paper III - VII). Since the dominant calcareous algae habitat at Peregrino is

present 100 meters water depth, it was anticipated that discharge of additional particulates

from the drilling operations could harm the calcareous algae due to increased sedimentation,

appearing as bleaching. The drill cuttings could potentially also increase the turbidity in the

water column above the calcareous algae habitat, reducing the irradiance (E) and change the

. However, the latter

was for practical reasons not included in the experimental design of the PEMCA project.

Threshold levels for exposure of particulates to red calcareous algae (M. engelhartii and

Lithothamnion sp.) under realistic light conditions were obtained from laboratory studies

2 WGS84 given in decimal degrees

(Paper IV; Reynier et al., 2015). Furthermore, the results from the laboratory experiments

were, together with field measurements and discharge data from the drilling log (Paper VI),

used as input data to the dispersion and risk modelling to improve the environmental risk

assessment at the Peregrino field (Paper VII).

The LoVe Ocean Observatory was deployed in September 2013 at a cold-water coral

area, 20 km off the north western coast of Norway (N 68°54.474’, E 15°23.145’)3 at

approximately 260 meters water depth (Godø et al., 2014a). Time-lapse photos (RGB) of the

coral habitat, dominated by Lophelia pertusa, were taken with a temporal resolution of one

hour. In addition, key environmental parameters such as temperature, salinity, echograms

presenting zooplankton and fish biomass at different depths, current speed and direction at

several depths, sound, concentration of Chlorophyll a presenting phytoplankton biomass,

coloured dissolved organic material (indicating fresh water run-off and different water types)

and total suspended matter (water clarity) were measured (Godø et al., 2014a; Statoil, n.d.).

The LoVe Ocean Observatory serve several purposes: To 1) further develop and test

technology, 2) gather data for improved ecosystem understanding and knowledge of the area

in general and 3) improve understanding and knowledge of the cold-water coral reef habitat

in particular, and 4) develop methodology for automated data analysis and visualisation. The

online-cabled ocean observatory enables observations with high temporal coverage and

resolution of a variety of organisms in their natural environment. So far, two different

approaches for automatic or semi-automatic image analysis of organisms have been

demonstrated at LoVe. For more details see section 4.2 Images in environmental monitoring,

Paper VIII, Paper IX and Osterloff et al. (In Press).

3 WGS84 given in degrees and decimal minutes

2 Methodological approach

The proposed approach is meant to be generic and should therefore be relevant for

authorities, industry, academia and public in general. The individual steps for IEMM are

described in Figure 1 and Step 3-10 will be further elaborated and exemplified under the

sections in Chapter 2 and 3.

Step 1 (number 1 in Figure 1) represents the a priori knowledge and understanding of

the processes, ecosystems, object or area of interest. The mission objectives, formulated

based on existing knowledge (Step 1), are given in Step 2. The operational constraints cover

availability of appropriate technologies and methods, environmental limitations and cost,

detailed planning and replanning/adaption that may take place both autonomously and/or

with human-in-the-loop during execution of the mission (Step 3-6). Criteria for sufficient

data gathering and basis for decision-making (Step 7) will either define a positive result

where mission is completed in accordance with objectives (Step 10), or a negative result

where the research questions and mission objectives may need to be reformulated or divided

into new sub-goals (Step 8). Considering the given operational constraints the mission will

then either be repeated with the new input conditions (Step 3-6) or terminated (Step 11).

Step 11 could either give input to further research and development needs or the conclusion is

noted without further actions.

2.1 Selection of sensors, sensor carrying platforms and exposure studies

Step 3 in the methodological approach (Figure 4) represents the planning phase where

the appropriate parameters (sensors), sensor carrying platforms and exposure (lab) studies (if

needed) are selected according to the mission objectives and data requests defined in Step 2,

5, 6 and/or 9, respectively. Depending on the environment, there will be a minimum number

of parameters that always should be measured, either as essential parameters or as supporting

parameters for interpretation of data gathered. In addition to measuring essential parameters,

the quality of the interpretation is dependent on the temporal and spatial resolution and

coverage of the data. Hence, considering the dynamics both in time and space is crucial in the

mission planning process to enable proper interpretation of the measurements.

Figure 4: Step 3 Selection of sensors and sensor carrying platforms, including exposure (lab)

studies. Exposure studies include both laboratory studies, as in this thesis, and field

experiments.

2.1.1 Temporal and spatial coverage and resolution

Data with sufficient temporal and spatial coverage and resolution is a precondition for

a reasonable interpretation of the data and thereby to gain environmental knowledge. Such

data is crucial for appropriate decisions and hence, for sustainable management of our

ecosystems. For instance, the increasing numbers of fixed ocean observatories will provide

data with high temporal coverage and resolution, data that can provide new insight in the

natural dynamics of hydrographical data, chemistry and species specific distribution. Paper

V, VIII and Paper IX describe different approaches using photos to detect and document

natural species specific behaviour and/or distribution (Paper VIII and IX) and potential

impact from anthropogenic activities, such as increased sedimentation from drilling

discharges (Paper V) with time. Use of photos will be further described under section 4.2

Images in environmental monitoring. When combining various sensor measurements, for

instance by using multivariate data analysis, it is important to note that the sampling

frequency may vary between different sensors and preprocessing of the data is needed to

align them to the same temporal resolution (Paper VI). Taking this into consideration, the

maximum sampling frequency of the individual sensors should be carefully considered as this

might be a limiting factor for the entire analysis. On the other hand, if high temporal

resolution is of less importance, the sampling frequency should be reduced to avoid

unnecessary “noise” and to reduce storage volume.

Figure 5: Spatial and temporal coverage and resolution of different sensor carrying platforms

(Figure 3 in Paper 1). ROV (remotely operated vehicle), AUV (autonomous underwater

vehicle), UAV (unmanned aerial vehicle) and glider (buoyancy driven autonomous

underwater vehicle).

The suitability of the different sensor carrying platforms will vary based on the spatial

and temporal coverage and resolution on the data needed, and their capabilities are illustrated

in Figure 5 (Figure 3 in Paper I). Often, a combination of sensor carrying platforms will be

needed to fulfil the mission requirements. Figure 5 is also illustrating the need for an

integrated approach in general. I addition to the different sensor carrying platforms, the figure

is also applicable for dynamic processes on different biological taxa in time and space or to

sensor technologies and how the different technologies determines the nature of the captured

data and how they are presented, as described in Paper X, Figure 1.

2.1.2 Sensors and sensor carrying platforms: Technology development and improvement

Although sensor carrying platforms with appropriate temporal and spatial coverage

and resolution are used, the platform and/or sensors might be inappropriate for the specific

mission either due to the environmental conditions in the area or, if applicable, the vessel

dedicated for deployment, maintenance and/or recovery (Godø et al., 2014b). One example of

the latter is the stationary sensor carrying platform, or lander technology, used as part of the

PEMCA project (Figure 3 a) (Paper III). The chosen lander was relatively large and heavy

(approximately 2.5 meter high and weighing 3 tonnes). In general, the lander did not

represent a good design with respect to robustness and handleability. This suboptimal design,

combined with the vessel available, led to complicated deployments and recoveries related to

downloading of data and maintenance operations. In addition, unexpected environmental

conditions with respect to the continuous strong currents and the corrosive properties of the

seawater at Peregrino affected the sensor carrying platform (Paper III) and made the field

missions challenging. The experiences from PEMCA lead to a considerable change of the

lander design before the LoVe deployment, both with respect to weight (handleability),

robustness in general and particularly with respect to protection of fragile instruments such as

the echo sounder system. Also the camera settings were improved based on the experiences

from PEMCA. First and foremost the camera was moved from a fixed position on the lander

with fixed settings and flash to a standalone satellite where the camera and corresponding

flash could be adjusted both vertically and horizontally. The angle of the camera, the light

scattering on the photos induced by the unexpected high levels of marine snow, and some

technical problems related to the settings of the flash, resulted in uneven illumination and

overexposure in a majority of the photos from the PEMCA project. Furthermore, at LoVe the

collaboration between biologists, engineers and computer scientists was initiated on an early

stage of the planning process enabling, as far as possible, to meet the needs from the different

disciplines. In contrast to the solution on Peregrino using batteries, the cabled online two-way

communication at LoVe also allowed high temporal resolution of the photos and

opportunities with respect to adjustments of camera settings after deployment.

2.1.3 Laboratory experiments to close gaps of knowledge

Another important aspect, normally not a part of the traditional environmental

monitoring, is to identify lack of key knowledge with respect to the mission objectives. For

instance, being able to evaluate if a specific activity could impact the abundance or dynamics

of species negatively, certain background knowledge about the species itself is required. To

bridge such knowledge gaps, laboratory studies and/or field experiments might be required to

establish threshold levels for the possible impacts. Two examples where such laboratory

studies have been performed to establish threshold levels for exposure of water-based drill

cuttings are the coral risk assessment, monitoring and modelling (CORAMM) project (Purser

and Thomsen, 2012) and the PEMCA project (Paper IV; Salgado et al., 2010). These studies

were initiated to establish general knowledge and thresholds for exposure of water-based drill

cuttings to the cold-water coral L. pertusa and the calcareous red algae species M. engelhartii

and Lithothamnion sp., respectively. Furthermore, controlled experiments can also be

performed to investigate how some parameters, such as the composition of the water inherent

optical properties (IOPs), might affect other data, such as image quality due to the absorption

and/or scattering of light (Johnsen et al., 2013).

In Paper IV the physiological status of the calcareous algae, measured as

photosynthetic efficiency ( PSII_max) as function of sediment coverage, was studied in a flow-

through system as function of the three design variables light intensity, flow rate and added

amount of sediment, in addition to exposure time. Multivariate data analysis, using Partial

Least Squares (PLS) regression, showed that added amount of sediment was the major factor

reducing photosynthetic efficiency, while water flow rate, light intensity and exposure time

had only minor impact within the experimental domain. Exposure–response curves for

photosynthetic efficiency as function of sediment coverage were created based on results

from the PLS-model predicting photosynthetic efficiency and corresponding effect of

sediment coverage by varying the added amount of sediment within the experimental domain

(within the maximum and minimum levels). Using the exposure-response curve, the

photosynthetic efficiency was reduced by 50% after 1-2 weeks with 70% sediment coverage.

These results were further used to establish impact categories with the purpose of reducing

the uncertainties in the field specific environmental risk assessment and to enable common

acceptance of stress criteria adjusted to the present ambient conditions, as described in

Paper VII.

3 Data sampling and processing

In Step 4–6 data sampling and processing is performed (Figure 6), based on the

sampling regime specified in Step 3. If data collected in Step 4 is not in accordance with the

data request, a new adjusted request can be made either autonomously/automatically (Step 5)

or manually (Step 6) before going back to Step 3. Types of requests could typically be:

Restrict the sampling area to where the object of interest is present (mobile sensor carrying

platform or movable echo sounder), adjust sampling frequencies (increase sampling when

object of interest and/or parameters important for objective of mission is present), adjust

range (if mobile sensor carrying platform), etc. until request information has been obtained.

Figure 6: Step 4-6 covering data sampling, autonomous/automatic and manual analyses, interpretation of data and knowledge accumulation.

In addition to optimisation of sampling with sensor carrying platforms, Step 4-6 also

cover issues such as automatic or semi-automatic image analysis. Image analyses are further

described under section 4.2 Images in environmental monitoring.

3.1 Adaptive sampling

Adaptive sampling can increase the value of the measurements. In addition to an

approach being adjusted to purpose as part of the planning process (Step 3), adaptive

sampling enables a selective sampling also during a mission. Sensors could be programmed

to switch on and off based on predefined settings, such as concentrations or presence of a

certain parameter could trigger other sensors to be switched on. Furthermore, the sensor

carrying platform itself could be preprogrammed to follow a desired target (a phenomenon or

given organism) of interest instead of a pre-defined transect, which is the common situation.

Due to a more targeted data sampling, providing more relevant data and less “noise”,

adaptive sampling can also reduce the mission time for data gathering and thereby reduce the

operational risk. The approach requires access to real-time data deliveries, or at least fast data

processing if the data requests are being made manually.

Adaptive sampling will not be further elaborated as part of this thesis. Power

consumption, software development, signal processing guidance and control of sensor

platforms are currently at a level that allows implementation of autonomous adaptive

sampling (Brito et al., 2012; Cruz and Matos, 2010). However, AUVs for instance still face

operational limitations with respect to shortages of power supply and lack of the autonomy

needed.

3.2 Data processing and knowledge accumulation

While access to data with sufficient temporal and spatial resolution still is a main

limitation when trying to understand the marine environment, the ability to handle, analyse

and interpret the increasing amounts of data has become another bottleneck for knowledge

accumulation (Step 6). When including images and video recordings, new multi-modal data

collections are approaching big data dimensions. Examples of different methodological

approaches of data interpretation are given in Papers V - IX. Paper V, VIII and IX describe

different approaches for analysis of marine organisms from photos, and these are further

described in section 4.2 Images in environmental monitoring. In Paper VI multivariate data

analyses are used on a variety of field measurements to evaluate the possible impact of

discharges from water-based drill cuttings on a calcareous red algae habitat. In Paper VII

species specific toxicity data obtained from laboratory studies are combined with measured

discharge and oceanographic data as input in the risk modelling to reduce the uncertainty in

the environmental risk assessment.

3.2.1 Interpretation of complex data sets from field measurements

Paper VI demonstrates the benefit of multivariate data analysis for optimized data

interpretation in IEMM by enabling integration of a high number of variables. A total number

of 178 variables (initially 81 plus 97 derived), 300 sediment trap samples from three sediment

traps, and more than 20000 observations from the lander were combined and interpreted by

the multivariate data analyses. The lander and sediment traps were deployed at different

distances downstream of the discharge point. Current speed and direction at several depths,

turbidity, temperature and salinity were measured at the lander, while the sediment traps were

used to determine suspended particulate matter which was characterised with respect to a

number of chemical parameters (26 alkanes, 16 polycyclic aromatic hydrocarbons (PAHs),

nitrogen, carbon, calcium carbonate and barium). Data on discharges of drill cuttings and

water based drilling fluid was provided from the drilling log on a daily basis.

Principal component analysis (PCA) was used to evaluate similarities and differences

among samples and correlations between variables. PLS regression was used to identify

correlations between predictor variables (x) and response variables (y). The multivariate data

analysis showed no systematic difference in levels of sedimentation or measured organic and

inorganic components between campaigns or traps. In addition, the analysis showed a strong

co-variation between suspended particulate matter and total nitrogen and organic carbon,

suggesting that the majority of the particles sampled had a natural and biogenic origin.

Furthermore, the PLS showed no correlation between discharges of drill cuttings and

sediment traps data or turbidity data from the lander in periods with prevailing current

directed against these sensors. The different multivariate models supported each other and

they were also supported by chemical indicators, such as the Terrestrial-Aquatic Ratio

(TAR), the carbon preference index (CPI) and the isometric ratios for PAHs, and numerical

modelling of dispersion and deposition.

3.2.2 Combining laboratory experiments, field measurements and modelling in environmental risk assessment

Paper VII describes how the uncertainty in environmental risk assessment can be

reduced by utilising data obtained from exposure studies on organisms relevant for the

particular area of interest in combination with measured discharge data and measured

oceanographic data as input to risk modelling (Figure 7). Dispersion modelling of drill

cuttings was performed for a two-year period using measured oceanographic data and

discharge data with a temporal resolution of 24 h (Paper VII). The same model (Beyer et al.,

2012; Durgut et al., 2015; Johnsen et al., 2000; Rye et al., 2008) was used to assess the risk of

impact on the two algae species M. engelhartii and Lithothamnion sp. from the Peregrino oil

field after establishing four species specific impact categories: No, minor, medium and severe

impact. The corresponding intervals for photosynthetic efficiency ( PSII_max) and sediment

coverage were obtained from exposure-response relationship for photosynthetic efficiency as

function of sediment coverage for the two calcareous red algae species (section 2.1.3

Laboratory experiments to close gaps of knowledge and Paper VI). The temporal resolution

enabled more accurate model predictions as short-term changes in discharges and

environmental conditions could be detected. The assessment shows that there is a patchy risk

for severe impact on the calcareous algae stretching across the transitional zone and into the

calcareous algae habitat at Peregrino.

The impact category approach presented in Paper VII (no impact – severe impact)

provides improved management flexibility as the accepted impact level can be adjusted based

on varying geographical environmental conditions. One can, for instance, expect that areas

with low current velocity will tolerate less discharge of particulates than areas with stronger

currents since stronger currents should be expected to give both less sedimentation due to a

broader dispersion and lead to increased resuspension. For industrial developments this

knowledge may allow a site specific evaluation of the discharge strategies, adjusting the

acceptance of stress criteria expressed by the impact categories to the health of the present

habitat and ambient conditions.

Figure 7: Schematic overview of the data sources used to assess the risk of impact on the two

red calcareous algae species M. engelhartii and Lithothamnion sp. The different data sources

include dispersion modelling, using input data from field (oceanographic data), discharge

data from discharge log and the effect data, derived from the exposure studies. The latter was

used to establish impact categories (Paper VII). The sedimentation experiment enabled

sediment coverage from the exposure studies to be converted to sediment deposition, data

required for the modelling (Figure 2 in Paper VII).

4 Basis for decisions: Tools enabling improved collaboration and communication

Step 10 represents the enhanced knowledge acquired from Step 4-6 and is considered

sufficient as basis for end-users and decision makers (Figure 8). Results considered not

adequate for decisions or to give the sufficient understanding needed should be further

specified or divided into “sub questions” (Step 8) before directed via Step 9 either back to

Step 3 (replanning of mission) or Step 11 which covers input to further research or “leave as

is” (“no further actions”).

In Paper X Don Ihde’s (1979, 1991) theory on instrumental realism is used to explain

why images in general are easy to understand and discussed by humans independent of

background and experiences and therefore have the potential of acting as a boundary object in

interdisciplinary communication and collaboration. Furthermore, Carlile’s (2004) model on

innovation and collaboration across disciplines is used to argue for how interdisciplinary

collaboration, utilising such boundary objects, can create new knowledge and understanding

of the environment.

Figure 8: Step 10 representing knowledge which in each case is acquired during the mission and considered as sufficient for end-users and decision makers.

4.1 Interdisciplinary communication and collaboration

In addition to optimised data sampling and aggregation (Step 3-6), a sufficient unified

understanding among the actors involved in the task(s) in question is crucial to generate new

knowledge. As described by, among others, Elliott (2014), Leslie and McLeod (2007) and

Rylance (2015) interdisciplinary collaboration is a precondition to solve complex

environmental problems and to perform sound environmental management. To succeed when

working across interests and disciplines, both domain specific and common knowledge must

be acknowledged by the actors (Carlile, 2004). The inverted triangle presented in Figure 9 a)

illustrates the different levels of complexity that categorises a task, from syntactic to semantic

and pragmatic level, with the latter as the most complex. The complexity of a task

corresponds to the scientific deviation and experiences between the actors involved; the more

complex, the more deviated. Furthermore, Carlile (2004) claim that innovation has occurred

and new knowledge is established and visible only after new tasks (Semantic and Pragmatic

level in Figure 9) have been institutionalised in new methods, tools and representations

(Syntactic level in Figure 9).

Figure 9b) exemplifies the different levels of complexity with use of photos, from a

standard digital red-green-blue (RGB) photo of a coral reef (syntactic level), via the semantic

level where the photos are converted to numerical values, here presented as a heat map where

different colours visualise the varying densities of shrimp. Due to the transformation of the

RGB photo (syntactic level) to numerical values (semantic level), quantifying object of

interest and segmenting regions of interest, photos can be incorporated in multivariate data

analyses, here represented by a PCA plot (pragmatic level).

Figure 9: A general model describing collaboration across boundaries on tasks of varying novelty, or

complexity. a) The syntactic, semantic and pragmatic levels represent data of increasing novelty and

uncertainty with a corresponding decrease in confirmation of the data. Discipline A and discipline B

represents different research fields or scientific disciplines. b) The three levels (syntactic, semantic

and pragmatic) are exemplified with different methods for detection of biological behaviour and

multisensory data from an ordinary photo, the shrimp abundance represented as numerical values

(heat map) and incorporation of photos in multivariate data analysis, represented by a principal

component analysis (PCA) plot, respectively. The more transformed the data is from how humans

visually perceive the world, the more specialised knowledge is required to understand the data

(Figure 3 in Paper X).

4.2 Images in environmental monitoring

Due to human ability to understand images the rapid increase in use of photos and

video recordings in environmental mapping and monitoring should contribute to enhanced

environmental understanding and knowledge. The increasing numbers of long-term deployed

cameras should therefore enable a much better understanding of the natural dynamics of

different species. The technology is non-destructive, it measures the visual fauna in its natural

environment and it enables increased temporal coverage and resolution compared to e.g. ship

campaigns. Understanding the natural dynamic is a precondition to distinguish between

anthropogenic impacts and natural variations. However, use of underwater photos is not

always applicable. For instance, in cases where the mission is focused on the infauna, other

imaging techniques penetrating into the sediments would be more suitable.

The vast amounts of underwater photos and video recordings with increasing

temporal and/or spatial coverage and resolution are challenging our capability to interpret

these data. The obstacle can be solved with an interdisciplinary effort, developing and

applying algorithms for (semi)automatic image analyses. In addition to enhance

environmental understanding and knowledge as such, our general perception of photos can

also be utilised in communication to explain more abstract and complex data, such as the

results of multisensory data and multivariate data analysis.

Different biological questions require different algorithmic approaches to image

analysis. As part of this thesis one approach has focused on identifying behaviour of specific

species of interest, information that can be used to measure health status and habitat function

in environmental monitoring (Paper V and Paper VIII). In these cases the species are known

and manual annotations form the basis for the algorithm development. The other approach

has focused on use of unsupervised computer learning to detect and group interesting events,

such as presence and behaviour of different mobile species, independent of the level of

a priori knowledge of the area or habitat (Paper IX). Due to the lack of knowledge about the

present species the algorithms are designed to detect any kind of change patterns from regular

events.

In Paper V changes in size and colour is monitored in single individuals of red

calcareous algae with time. A machine learning based approach was used on photos collected

from laboratory experiments to assess the potential impact of water-based drill cuttings on

deep-water rhodolith-forming calcareous algae. In this pilot study (Paper V), imaging

technology was used to quantify and monitor the stress levels of calcareous algae caused by

various degrees of light exposure, flow intensity and amount of sediment, further described in

Paper IV. Due to the selected colour normalisation strategy combined with a new flexible

classifier design, the algorithm enable to measure change in size and colour of the calcareous

algae over time. With this approach the photos were converted to numerical values, enabling

to perform multivariate data analysis of the results from the image analysis and the

experimental variables. The study showed correlation between calcareous algae size as a

function of time, as well as correlation between colour and fluorescence. These results are

consistent with the findings in Paper IV.

In Paper VIII the change in abundance of organisms of one species is quantified (by

colour) in time and space. An automatic detection of shrimp in photos from the fixed camera

at the LoVe Ocean Observatory with high temporal resolution (one hour) over a time period

of six weeks is presented. Also in this approach the photos are converted to numerical values,

providing unique opportunities to study shrimp behaviour (number and abundance at

different locations within the camera frame) in their natural environment over time. This is

the first time a feature representation based on local temporal colour contrast has been

applied to detect the semi-transparent shrimp from the challenging background of the coral

In a IEMM context the species composition of a habitat might not be known

beforehand. In such situations algorithms are required to search incoming image streams for

any kind of events or patterns that could be of interest, mark these events in the images and

build up links between similar events/patterns, such as passing species or changes in colour,

size or shape of fixed objects. In contrast to Paper V and Paper VIII, Paper IX proposes an

approach applying algorithms that automatically categorise and group mobile organisms

simultaneously. As described above, this approach based on categorising events based on

features and change detection pattern does not require a priori knowledge of the specific

habitat. The approach detects a large number of local changes, such as different species that

passes the photo frame in a given temporal sequence. In addition to the automatic grouping,

an applied relevance index presents the most relevant representatives for the group first. The

observer can then quickly check the preferred groups and only focus on the relevant findings

of each group. With this method the observer can bypass the time consuming inspection of

the whole data set.

5 Conclusions and future perspectives

5.1 Concluding remarks

The proposed concept for IEMM is a flexible system adjusted to purpose and object

of interest. As a planning tool, the methodological approach proposed assists to, in each case,

optimise selection of parameters and the most suitable sensor carrying platforms providing

data with adequate temporal and spatial coverage and resolution. In addition to improving the

quality and relevance of the data, the concept also helps to secure a mutual understanding

between the actors involved with respect to what should be done and why before, under and

after the mission. It can also help to reveal gaps of knowledge related to the objects of interest

or enabling initiation of experiments in due course before a field study. The approach is cost-

efficient as more targeted data and less “noise” is gathered in a time efficient way.

Access to data of sufficient spatial and temporal coverage and resolution is a

precondition for environmental knowledge in general. The rapid advances in enabling

technology permit access to technology for adequate data sampling.

Technology development challenges the human ability to interpret new data due to the

limitations in detecting and interpreting unfamiliar features. How the different measurements

are perceived is determined by the sensor technology. The more transformed the data is from

how humans visually perceive the world, the more specialised knowledge is required to

understand the data.

Images are in general understood by humans independent of background and

experience. They should therefore be suitable boundary objects for interdisciplinary

communication and collaboration. To exploit the inherent knowledge available in the vast

amounts of photos and video recordings of high temporal and/or spatial coverage and

resolution, interdisciplinary collaboration is a precondition. The collaboration between

biologists and computer scientists on use and interpretation of photos has resulted in new

insights in the dynamic responses of several marine species over time. By converting photos

to numerical values, as demonstrated in this thesis, they can be incorporated with other data,

such as current speed, current direction and temperature, for instance in multivariate data

analysis and modelling.

5.2 Future perspectives

Established methods for image analysis of marine species are a present relatively

limited. However, the rapid increase in use of underwater images for environmental mapping

and monitoring purposes has lately resulted in high focus on further development of suitable

analytical methodology. In addition to the work presented in this thesis, several workshops in

connection with this year’s 23rd International Conference on Pattern Recognition (ICPR)4 are

dedicated to underwater imaging. Also a workshop on marine imaging5 is planned in Kiel

early 2017. Commercially available analytical tools for image analysis are a precondition for

future cost-efficient use of environmental data.

The potential of images to act as a boundary object in interdisciplinary

communication and collaboration should in a future perspective be utilised to ensure better

decisions in operational settings and for management purposes.

From an industrial perspective, the most important drivers for introducing an

approach such as the IEMM concept are improved environmental performance and cost

reduction. Both these aspects are within reach with the present methodology and technology,

as described in the present thesis. Thus, a successful implementation will require a significant

practical and political effort to create a common interest in sharing and assessing data.

Institutional acceptance and focus on interdisciplinary collaboration will be critical success

factors in this process.

By pinpointing important aspects and opportunities that can improve how IEMM is

being performed, this thesis can hopefully contribute to the early stage discussion within the

offshore oil and gas industry on how to improve environmental management with introducing

more flexibility in future environmental monitoring and by strengthening the link between

risk assessment and monitoring. Flexible environmental monitoring also challenges the

regularity needed for the authorities monitoring of trends, both with respect to impact

between different sites and/or areas and the development within the area with time. In

Norway, the discussion between authorities and the offshore oil and gas industry has been

initiated. More experiences with the IEMM concept, open discussions and communication

will enhance environmental knowledge through implementation of new technologies and

4 http://www.icpr2016.org/site/ 5 http://marine-imaging-workshop.com/

furthermore, that the IEMM concept will be an advantage for all end-users of the data

provided.

References

Adams, J., 2013. Experiencing shipwrecks and the primacy of vision. In: Adams, J., Rönnby, J. (Eds.), Interpreting Shipwrecks: Maritime Archaeological Approaches. Highfield press, pp 85-96.

Almklov, P. G., Hepsø, V., 2011. Between and beyond data: How analogue field experience informs the interpretation of remote data sources in petroleum reservoir geology. Social Studies of Science Vol 41, pp 539-561,doi: http://dx.doi.org/10.1177/0306312711403825.

Beijbom, O., Treibitz, T., Kline, D. I., Eyal, G., Khen, A., Neal, B., Loya, Y., Mitchell, B. G., Kriegman, D., 2016. Improving automated annotation of benthic survey images using wide-band fluorescence. Scientific Reports Vol 6, pp 23166, doi: http://dx.doi.org/10.1038/srep23166.

Berge, J., Renaud, P. E., Darnis, G., Cottier, F., Last, K., Gabrielsen, T. M., Johnsen, G., Seuthe, L., Weslawski, J. M., Leu, E., Moline, M., Nahrgang, J., Søreide, J. E., Varpe, Ø., Lønne, O. J., Daase, M., Falk-Petersen, S., 2015a. In the dark: A review of ecosystem processes during the Arctic polar night. Progress in Oceanography Vol 139, pp 258-271, doi: http://dx.doi.org/10.1016/j.pocean.2015.08.005.

Berge, J., Daase, M., Renaud, Paul E., Ambrose Jr, William G., Darnis, G., Last, Kim S., Leu, E., Cohen, Jonathan H., Johnsen, G., Moline, Mark A., Cottier, F., Varpe, Ø.,

sabuau, J.-C., Falk-Petersen, S., Kosobokova, K., Hoppe, Clara

-Kowalczuk, M., Vogedes, D., Camus, L., Tran, D., Michaud, E., Gabrielsen, Tove M., Granovitch, A., Gonchar, A., Krapp, R., Callesen, Trine A., 2015b. Unexpected levels of biological activity during the polar night offer new perspectives on a warming Arctic. Current Biology Vol 25, pp 2555-2561, doi: http://dx.doi.org/10.1016/j.cub.2015.08.024.

Beyer, J., Myhre, L. P., Sundt, R. C., Meier, S., Tollefsen, K.-E., Vabø, R., Klungsøyr, J., Sanni, S., 2012. Environmental risk assessment of alkylphenols from offshore produced water on fish reproduction. Marine Environmental Research Vol 75, pp 2-9,doi: http://dx.doi.org/10.1016/j.marenvres.2011.11.011.

Biederman, I., 1987. Recognition-by-components: A theory of human image understanding. Psychological review Vol 94, pp 115-147, doi: http://dx.doi.org/10.1037/0033-295X.94.2.115.

Borja, A., Dauer, D. M., 2008. Assessing the environmental quality status in estuarine andcoastal systems: Comparing methodologies and indices. Ecological Indicators Vol 8, pp 331-337, doi: http://dx.doi.org/10.1016/j.ecolind.2007.05.004.

Brito, M. P., Bose, N., Lewis, R., Alexander, P., Griffiths, G., Ferguson, J., 2012. The role of adaptive mission planning and control in persistent autonomous underwater vehicles presence. 2012 IEEE/OES Autonomous Underwater Vehicles (AUV). Southampton. pp 1-9, doi: http://dx.doi.org/10.1109/AUV.2012.6380748.

Brown, R. R., Deletic, A., Wong, T. H. F., 2015. Interdisciplinarity: How to catalyse collaboration. Nature Vol 525, pp 315–317, doi: http://dx.doi.org/10.1038/525315a.

Buhl-Mortensen, P., Tenningen, E., Tysseland, A. B. S., 2015. Effects of water flow and drilling waste exposure on polyp behaviour in Lophelia pertusa. Marine Biology Research Vol 11, pp 725-737, doi: http://dx.doi.org/10.1080/17451000.2014.993651.

Carlile, P. R., 2004. Transferring, translating, and transforming: An integrative framework for managing knowledge across boundaries. Organization Science Vol 15, pp 555-568,doi: http://dx.doi.org/10.1287/orsc.1040.0094.

CCME (Canadian Council of Ministers of the Environment), 2015. Guidance manual for optimizing water quality monitoring program design. Document no: PN 1543.

Cohen, J. H., Berge, J., Moline, M. A., Sørensen, A. J., Last, K., Falk-Petersen, S., Renaud, P. E., Leu, E. S., Grenvald, J., Cottier, F., Cronin, H., Menze, S., Norgren, P., Varpe, Ø., Daase, M., Darnis, G., Johnsen, G., 2015. Is ambient light during the High Arctic polar night sufficient to act as a visual cue for zooplankton? PLoS ONE Vol 10, pp e0126247, doi: http://dx.doi.org/10.1371/journal.pone.0126247.

Cruz, N. A., Matos, A. C., 2010. Adaptive sampling of thermoclines with autonomous underwater vehicles. OCEANS 2010.MTS/IEEE SEATTLE. Seattle, WA. pp 1-6,doi: http://dx.doi.org/10.1109/OCEANS.2010.5663903.

de Zwart, D., Posthuma, L., 2005. Complex mixture toxicity for single and multiple species: Proposed methodologies. Environmental Toxicology and Chemistry Vol 24, pp 2665-2676, doi: http://dx.doi.org/10.1897/04-639R.1.

Directorate-General for Maritime Affairs and Fisheries, 2012. Blue growth: Opportunities for marine and maritime sustainable growth. Catalog no: KL-32-12-321-EN-C, doi: http://dx.doi.org/10.2771/43949.

DNV (Det Norske Veritas), 2013. Guideline: Monitoring of drilling activities in areas with presence of cold water corals. Document no: 2012-1691/12NCQKD-2. http://www.norskoljeoggass.no/Global/2013%20Dokumenter/Publikasjoner/Monitoring%20of%20drilling%20activities%20-%20Areas%20with%20Cold%20Water%20Corals.pdf.

, Rye, H., Reed, M., Smit, M. G. D., Ditlevsen, M. K., 2015. Dynamic modeling of environmental risk associated with drilling discharges to marine sediments. Marine Pollution Bulletin Vol 99, pp 240-249,doi: http://dx.doi.org/10.1016/j.marpolbul.2015.07.019.

ECHA (European Chemical Agency), 2008. Guidance on information requirements and chemical safety assessment Chapter R.10: Characterisation of dose (concentration)-response for environment. https://echa.europa.eu/documents/10162/13632/information_requirements_r10_en.pdf/bb902be7-a503-4ab7-9036-d866b8ddce69

ECHA (European Chemical Agency), n.d. https://echa.europa.eu/support/guidance(7 September 2016)

Elliott, M., 2014. Integrated marine science and management: Wading through the morass. Marine Pollution Bulletin Vol 86, pp 1-4,doi: http://dx.doi.org/10.1016/j.marpolbul.2014.07.026.

European Commission, 2010. Europe 2020: A strategy for smart, sustainable andinclusive growth. Document no: COM(2010) 2020 final. http://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52010DC2020.

Faillettaz, R., Picheral, M., Luo, J. Y., Guigand, C., Cowen, R. K., Irisson, J.-O., 2016. Imperfect automatic image classification successfully describes plankton distribution patterns. Methods in Oceanography Vol 15–16, pp 60-77,doi: http://dx.doi.org/10.1016/j.mio.2016.04.003.

Godø, O. R., Johnsen, S., Torkelsen, T., 2014a. The LoVe Ocean Observatory is in operation. Marine Technology Society Journal Vol 48, pp 24-30,doi: http://dx.doi.org/10.4031/MTSJ.48.2.2.

Godø, O. R., Klungsøyr, J., Meier, S., Tenningen, E., Purser, A., Thomsen, L., 2014b. Real time observation system for monitoring environmental impact on marine ecosystems from oil drilling operations. Marine Pollution Bulletin Vol 84, pp 236-250,doi: http://dx.doi.org/10.1016/j.marpolbul.2014.05.007.

Godø, O. R., Handegard, N. O., Browman, H. I., Macaulay, G. J., Kaartvedt, S., Giske, J., Ona, E., Huse, G., Johnsen, E., 2014c. Marine ecosystem acoustics (MEA): Quantifying processes in the sea at the spatio-temporal scales on which they occur. ICES Journal of Marine Science Vol 71, pp 2357-2369,doi: http://dx.doi.org/10.1093/icesjms/fsu116.

Harbers, J. V., Huijbregts, M. A. J., Posthuma, L., van de Meent, D., 2006. Estimating the impact of high-production-volume chemicals on remote ecosystems by toxic pressure calculation. Environmental Science & Technology Vol 40, pp 1573-1580,doi: http://dx.doi.org/10.1021/es051633m.

Hepsø, V., Låte, M., Gramvik, G., Johnsen, S., Nilssen, I., Wesenberg, H., 2012. Integrated environmental monitoring in daily operations. SPE Intelligent Energy International. Utrecht, The Netherlands. Doi: http://dx.doi.org/10.2118/150054-ms.

Ihde, D., 1979. Technics and praxis. D. Reidel Publishing company.

Ihde, D., 1991. Instrumental realism: The interface between philosophy of science and philosophy of technology. Indiana University Press.

Johnsen, G., Volent, Z., Dierssen, H., Pettersen, R., Ardelan, M. V., Søreide, F., Fearns, P., Ludvigsen, M., Moline, M., 2013. Underwater hyperspectral imagery to create biogeochemical maps of seafloor properties. In: Watson, J., Zielinski, O. (Eds.), Subsea Optics and Imaging. Woodhead Publishing, pp. 508-540,doi: http://dx.doi.org/10.1533/9780857093523.3.508.

Johnsen, S., Frost, T. K., Hjelsvold, M., Utvik, T. R., 2000. The Environmental Impact Factor - a proposed tool for produced water impact reduction, management and regulation. SPE International Conference on Health, Safety, and Environment in Oil and Gas Exploration and Production. Stavanger, Norway.Doi: http://dx.doi.org/10.2118/61178-MS.

Kongsberg Maritime, n.d. https://www.km.kongsberg.com/ks/web/nokbg0237.nsf/AllWeb/0875C 8CE982B5EFCC1257BD500388283?OpenDocument (22 August 2016)

Larsson, A. I., van Oevelen, D., Purser, A., Thomsen, L., 2013. Tolerance to long-term exposure of suspended benthic sediments and drill cuttings in the cold-water coral Lophelia pertusa. Marine Pollution Bulletin Vol 70, pp 176-188,doi: http://dx.doi.org/10.1016/j.marpolbul.2013.02.033.

Leslie, H. M., McLeod, K. L., 2007. Confronting the challenges of implementing marine ecosystem-based management. Frontiers in Ecology and the Environment Vol 5, pp 540-548, doi: http://dx.doi.org/10.1890/060093.

Nilssen, I., Johnsen, S., 2008. Holistic environmental management of discharges from the oil and gas industry - Combining quantitative risk assessment and environmental monitoring. SPE International Conference on Health, Safety, and Environment in Oil and Gas Exploration and Production. Nice, France. Doi: http://dx.doi.org/10.2118/111586-MS.

Orenstein, E. C., Haag, J. M., Gagnon, Y. L., Jaffe, J. S., 2016. Automated classification of camouflaging cuttlefish. Methods in Oceanography Vol 15–16, pp 21-34,doi: http://dx.doi.org/10.1016/j.mio.2016.04.005.

OSPAR, 2012a. OSPAR Recommendation 2012/5 for a risk-based approach to the management of produced water discharges from offshore installations. http://www.ospar.org/documents/dbase/decrecs/recommendations/12-05e_RBA%20Recommendation.doc.

OSPAR, 2012b. OSPAR Guidelines in support of Recommendation 2012/5 for a risk-based approach to the management of produced water discharges from offshore installations, (OSPAR Agreement: 2012-7). http://www.ospar.org/documents/dbase/decrecs/agreements/12- 07e_rba%20guidelines.doc.

Osterloff, J., Nilssen, I., Nattkemper, T. W., In Press. Computational coral feature monitoring for the fixed underwater observatory LoVe. IEEE OCEANS. Monterey (CA), USA.

Prasad, L., Singh, H., Gallager, S., 2016. Edge-based cuing for detection of benthic camouflage. Methods in Oceanography Vol 15–16, pp 35-48,doi: http://dx.doi.org/10.1016/j.mio.2016.05.005.

Purser, A., Thomsen, L., 2012. Monitoring strategies for drill cutting discharge in the vicinity of cold-water coral ecosystems. Marine Pollution Bulletin Vol 64, pp 2309-2316,doi: http://dx.doi.org/10.1016/j.marpolbul.2012.08.003.

Purser, A., Bergmann, M., Lundälv, T., Ontrup, J., Nattkemper, T. W., 2009. Use of machine-learning algorithms for the automated detection of cold-water coral habitats:a pilot study. Marine Ecology Progress Series Vol 397, pp 241-251,doi: http://dx.doi.org/10.3354/meps08154.

Reynier, M. V., Tâmega, F. T. S., Daflon, S. D. A., Santos, M. A. B., Coutinho, R., Figueiredo, M. A. O., 2015. Long- and short-term effects of smothering and burial by drill cuttings on calcareous algae in a static-renewal test. Environmental Toxicology and Chemistry Vol 34, pp, 1572-1577, doi: http://dx.doi.org/10.1002/etc.2938.

Rye, H., Reed, M., Frost, T. K., Smit, M. G. D., Durgut, I., Johansen, Ø., Ditlevsen, M. K., 2008. Development of a numerical model for calculating exposure to toxic and nontoxic stressors in the water column and sediment from drilling discharges.

Integrated Environmental Assessment and Management Vol 4, pp 194-203, doi: http://dx.doi.org/10.1897/IEAM_2007-039.1.

Rylance, R., 2015. Grant giving: Global funders to focus on interdisciplinarity. Nature Vol 525, pp 313–315, doi: http://dx.doi.org/10.1038/525313a.

Salgado, C., Johnsen, S., Gomes, S., Tokelsen, T., 2010. Real-time environmental monitoring of discharges of drill cuttings in the Peregrino field (Campos Basin, Brazil). SPE International Conference on Health, Safety, and Environment in Oil and Gas Exploration and Production. Rio de Janeiro, Brazil. Doi: http://dx.doi.org/10.2118/127175-MS.

Schoening, T., Bergmann, M., Ontrup, J., Taylor, J., Dannheim, J., Gutt, J., Purser, A., Nattkemper, T. W., 2012. Semi-automated image analysis for the assessment of megafaunal densities at the Arctic deep-sea observatory HAUSGARTEN. PLoS ONE Vol 7, pp e38179, doi: http://dx.doi.org/10.1371/journal.pone.0038179.

Shiffrar, M., Thomas, J. P., 2013. Beyond the scientific objectification of the human body: Differentiated analyses of human motion and object motion. In: Rutherford, M. D., Kuhlmeier, V. A. (Eds.), Social perception : Detection and interpretation of animacy, agency, and intention. Cambridge, Mass. : MIT Press., pp 83-108,doi: http://dx.doi.org/10.7551/mitpress/9780262019279.003.0005.

Smit, M. G. D., Frost, T. K., Johnsen, S., 2011. Achievements of risk-based produced water management on the Norwegian continental shelf (2002–2008). Integrated Environmental Assessment and Management Vol 7, pp 668-677,doi: http://dx.doi.org/10.1002/ieam.215.

Smit, M. G. D., Holthaus, K. I. E., Trannum, H. C., Neff, J. M., Kjeilen-Eilertsen, G., Jak, R. G., Singsaas, I., Huijbregts, M. A. J., Hendriks, A. J., 2008. Species sensitivity distributions for suspended clays, sediment burial, and grain size change in the marine environment. Environmental Toxicology and Chemistry Vol 27, pp 1006-1012, doi: http://dx.doi.org/10.1897/07-339.1.

Smith, E. E., Kosslyn, S. M., 2014. Cognitive psychology : Mind and brain. Edinburgh: Pearson.

Statoil, n.d. http://love.statoil.com/ (13 february 2015)

USEPA (United States Environmental Protection Agency), 2006. Guidance on systematic planning using the data quality objectives process. Document no: EPA QA/G-4. https://www.epa.gov/quality/guidance-systematic-planning-using-data-quality-objectives-process-epa-qag-4.

USEPA (United States Environmental Protection Agency), 2016. https://www.epa.gov/risk/risk-assessment-guidelines#tab-2 (7 September 2016)

Wuchty, S., Jones, B. F., Uzzi, B., 2007. The increasing dominance of teams in production of knowledge. Science Vol 316, pp 1036-1039,doi: http://dx.doi.org/10.1126/science.1136099.

Paper I

Nilssen, I., Ødegård, Ø., Sørensen, A. J., Johnsen, G., Moline, M. A., Berge, J., 2015a. Integrated environmental mapping and monitoring, a methodological approach to

optimise knowledge gathering and sampling strategy. Marine Pollution Bulletin Vol 96, pp 374-383, doi: http://dx.doi.org/10.1016/j.marpolbul.2015.04.045

Integrated environmental mapping and monitoring, a methodologicalapproach to optimise knowledge gathering and sampling strategy

Ingunn Nilssen a,b,1,⇑, Øyvind Ødegård c,d,1, Asgeir J. Sørensen d, Geir Johnsen b,d,g, Mark A. Moline e,Jørgen Berge f,g

a Statoil ASA, Research, Development and Innovation, N-7005 Trondheim, Norwayb Trondhjem Biological Station, Department of Biology, Norwegian University of Science and Technology, N-7491 Trondheim, NorwaycDepartment of Archaeology and Cultural History, University Museum, Norwegian University of Science and Technology, N-7491 Trondheim, NorwaydCentre for Autonomous Marine Operations and Systems (AMOS), Department of Marine Technology, Norwegian University of Science and Technology, N-7491 Trondheim, Norwaye School of Marine Science and Policy, University of Delaware College of Earth, Ocean, and Environment, Newark, DE 19716, USAfBFE Faculty, Institute for Arctic and Marine Biology, University of Tromsø, 9037 Tromsø, NorwaygUniversity Centre in Svalbard (UNIS), P.O. Box 156, N-9171 Longyearbyen, Norway

a r t i c l e i n f o

Article history:Received 6 March 2015Revised 24 April 2015Accepted 25 April 2015Available online 5 May 2015

Keywords:Integrated environmental mapping andmonitoringSensor platformDecision makingAdaptive samplingMultidisciplinary approach

a b s t r a c t

New technology has led to new opportunities for a holistic environmental monitoring approach adjustedto purpose and object of interest. The proposed integrated environmental mapping and monitoring(IEMM) concept, presented in this paper, describes the different steps in such a system from missionof survey to selection of parameters, sensors, sensor platforms, data collection, data storage, analysisand to data interpretation for reliable decision making. The system is generic; it can be used by author-ities, industry and academia and is useful for planning- and operational phases. In the planning processthe systematic approach is also ideal to identify areas with gap of knowledge. The critical stages of theconcept is discussed and exemplified by two case studies, one environmental mapping and one monitor-ing case. As an operational system, the IEMM concept can contribute to an optimised integrated environ-mental mapping and monitoring for knowledge generation as basis for decision making.

1. Introduction

Better understanding of ecosystems and how natural processes(abiotic and biotic) and anthropogenic factors might influencethem has during the last decade attained increased attention fromauthorities, scientific community and industry. The complexity ofintegrated environmental management is described by Elliott(2014). Borja and Dauer (2008) describe adaptive management,as one of several management approaches, and that the results ofsuch an approach may lead to corresponding changes in dataneeds, analytical procedures and interpretation. In this paper weargue that integrated environmental mapping and monitoring(IEMM), using different instrument (sensor) carrying platformscombined with proper analytical methods, are essential in creatingsuch an understanding (Fig. 1).

Exploration and mapping of unknown areas requires that a var-ied and wide scope for data gathering must be applied. Mapping, orexploration, in a research context is often aiming to discover orreveal hitherto unknown objects, organisms, processes and phe-nomena. Although there are usually one or several features orqualities that one are particularly looking for, a general under-standing of an area requires mapping of basic environmental char-acteristics such as topography, temperature and oxygenconcentration to get an overview of an unknown environment. Inan environmental management perspective, understanding of thearea as part of the larger eco-system is considered important.Hence, parameters to measure, choice of sensors and sensor plat-forms subject to operational constraints are decisions to be madeand remade in an environmental mapping process. This will supplyrelevant data for better accommodation of diverse and complexmission purposes with potentially multiple stakeholders, repre-senting different needs and requirements.

Environmental monitoring is focusing on current status and anychange detection of behaviour and characteristics of key parame-ters and objects of interest (OOI). The selection of parameters tomonitor is based on the best available understanding of the

⇑ Corresponding author at: Statoil ASA, Research, Environmental Technology,N-7005 Trondheim, Norway.

E-mail address: innil@statoil.com (I. Nilssen).1 The main contributors to the manuscript were Nilssen and Ødegård.

Marine Pollution Bulletin 96 (2015) 374–383

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier .com/locate /marpolbul

resources, processes and other relevant qualities present in thearea of interest. Efficient monitoring depends on the ability toacquire relevant data at relevant temporal and spatial resolutionto optimise information for knowledge-based decisions. Forauthorities and industry mapping is often conducted to establisha baseline and monitoring related to follow-up and repeat mea-surements of key parameters relative to the baseline data pro-vided. This means that also parameters essential in a monitoringperspective must, as far as possible, be taken into considerationduring the baseline study.

By integrating mapping and monitoring into the same opera-tional framework in terms of tools for data acquisition and knowl-edge production, the relevance of data for different purposesshould increase. As gathering of data is time-consuming and oftenalso expensive, it is essential to spend the time and money avail-able in a cost-effective manner according to relevant data gather-ing and expectations. Easy access to already existing data fromthe area of interest is also an important aspect in this context.

The spatial and temporal coverage and resolution needs willvary dependent on mission purpose (e.g. processes, organisms ofdifferent sizes) and the different decision-makers may have indi-vidual needs and requirements. In general, the requirements fromauthorities are often related to status and trends; e.g. has the areaof influence from a particular activity increased or decreased sincelast sampling? For industry this knowledge may have less value aslong as the information may be reported months or even yearsafter an occurrence. To make monitoring data useful in industry’sdaily operations, data sampling must be adjusted to the activityof interest (Hepsø et al., 2012).

While surveys have traditionally been performed as ship cam-paigns with temporal and spatial resolution limited to the capabil-ities of the ship with its installed sensors, a variety of fixed andmobile sensor platforms now enables gathering of vast amountsof, more or less, continuous or online real-time multisensory data.This is reflected through several initiatives where authorities e.g.MAREANO (MAREANO, n.d.), industry e.g. the PEMCA project(Johnsen et al., 2014; Salgado et al., 2010) and academia e.g.AMOS (AMOS, n.d.) and NORUS (NORUS, 2011) have supportedmulti sensor and multiplatform approaches by the use of a varietyof sensors mounted on different types of instrument platforms tomeasure parameters of interest (Hepsø et al., 2012; Schofieldet al., 2010). Sensor platforms can typically be fixed e.g. buoys,moorings (Berge et al., 2009; Johnsen et al., 1997) and landers(Ocean Network Canada, 2014; Statoil, n.d.), or mobile e.g.Remotely Operated Vehicles (ROVs) (SERPENT, n.d.), AutonomousUnderwater Vehicle (AUVs) (Kukulya et al., 2010; Maxson et al.,2011; Moline et al., 2005) and gliders (MBARI, 2014).

The problem of undersampling (Munk, 2000), have to somedegree been alleviated by advances in technology (Godø et al.,2014), but still huge scientific efforts are in general necessary to fillthe knowledge gaps and understanding of marine ecosystems.Many recent developments in platforms and sensors have beenguided by needs in the marine sciences, seeing end-users withspecific applications often involved in the engineering process(Bellingham, 2014; Glenn et al., 2005; Moline et al., 2005). Asnew technologies helps to fill knowledge gaps in the understand-ing of the environment, the level of complexity regarding datamanagement and processing also increases. When the amount ofdata increases, the work necessary to transform these data to use-ful information increases accordingly. The scientific community,authorities and industry are at present struggling with optimisingthe use of these data and to utilise the opportunities of enhancedknowledge that can be obtained from combining multi sensor data(Fagerås, 2011; Hepsø et al., 2012; Johnsen et al., 2011). The timeaspect is also of great importance as data must be transformed intoinformation in due time for decisions or as input for dependentoperations and processes. The increased volume and diversity ingathered data, including on-line data, requires better analyticaltools to meet requirements for rapid data analysis and immediateaction.

This paper presents an IEMM concept for data acquisition, pro-cessing and knowledge production. The challenges related to avail-ability of analytical tools and data processing that must be solvedto provide users or decision makers with relevant data in due timewill be discussed. The use and functionality of the concept will beillustrated by a generic model showing steps and sequences, andtwo case studies using new technologies relevant for marine envi-ronmental mapping and monitoring purposes. In Case study 1 theIEMM concept is exemplified by scientific knowledge productionfrom an arctic research project. Case study 2 shows the IEMM con-cept for a monitoring project initiated by the oil and gas industry.

2. Methodological approach

The methodological approach in IEMM is proposed to be genericand therefore relevant for authorities, industry and academia.Furthermore, it may be applicable independent of the operationalenvironment, e.g. terrestrial and marine environments in situ,including laboratory experiments.

Using the systematic approach described in this section will aidusers and decision makers to identify essential elements for con-sideration when planning a mission, help in selecting enablingtechnology (sensor, sensor platforms, software, etc.) for sufficient

Fig. 1. Sensor platforms used in IEMM. Satellites for remote sensing (1), research vessel (2), lander (3), ROV (4), AUV (5), glider (6) with sites such as deep water coral reefs,wreck for cultural heritage, under ice studies, oil and gas infrastructure (7), water column and seabed. Image: NTNU CeO AMOS/B. Stenberg.

I. Nilssen et al. /Marine Pollution Bulletin 96 (2015) 374–383 375

mapping and monitoring and to optimise data gathering and inter-pretation to provide end-users such as operators or decision mak-ers with appropriate data in due time. A methodological approachwill also in a systematic way identify gaps in knowledge, method-ology and technology, and thereby any needs for further develop-ment and improvements in the different steps of the concept fromplanning to execution of e.g. a survey.

The proposed steps in the IEMM concept are illustrated in Fig. 2and described below. Step 1 is the a-priori knowledge and under-standing of e.g. ecosystems, any object or area of interest. Subjectto the defined mission objectives (Step 2) and operational con-straints such as availability of appropriate technology and meth-ods, environmental limitations and cost, detailed planning andre-planning may take place both autonomously and with human-in-the-loop during the execution of the mission (Steps 3–6).Criteria for sufficient data gathering in addition to metrics for deci-sion-making (Step 7) will define either a positive result where mis-sion is completed in accordance with objectives (Step 10), or anegative result where the research questions and mission objec-tives may need to be reformulated or divided into new sub-goals(Step 8). Considering the given operational constraints the missioneither will be repeated with the new input conditions (Steps 3–6)or terminated (Step 11). Step 11 could either give input to furtherresearch and development needs or the conclusion is noted with-out further actions. We will describe Steps 3–6 in more detail. Inthe following the term ‘‘data request’’ is used to describe a set ofdata attributes or qualities such as resolution, sampling rate andquality required for current purpose (i.e. information needed toanswer mission objective).

2.1. Selection of parameters, sensors and sensor platforms (Step 3)

Step 3 represents the planning phase, comprising of a toolboxand sequences of activities where the appropriate sensors and plat-forms (including laboratory studies, if needed) are selected,adjusted and applied according to the mission objectives andrevised data requests defined in Steps 2, 5, 6 or 9, respectively.The wide range of available sensors and platforms allows therequests to be formulated in both broad and specific manners.Based on the data request, selections of sensors, platforms and

parameters are made within appropriate temporal and spatialranges, coverage and resolutions. A decision could also be madeto enable other sensors and sensor platforms additional to the onesnecessary to meet data request. The rationale behind such anapproach could be multiple; access to the area of interest, capacityof data sampling and an awareness of knowledge gaps or informa-tion needs beyond the scope of the present mission or, most likely,a combination of these.

2.1.1. Experimental studiesA certain basic knowledge of the object and/or area of interest

related to the mission objective in question (Step 2) is essential.If this basic knowledge is missing, initial experimental studies bothin laboratory or in situ may be performed in order to establish e.g.the biological basis to guide the mission planning (Step 3) and toenable interpretation of data gathered (Steps 4–6).

2.1.2. Measurements adjusted to purposeSelection of parameters to measure should fit the environment

in question and the aim of mission. Dependent on the environ-ment, there will be a minimum set of parameters that alwaysshould be measured, either as an essential or supporting parameterfor interpretation of other data gathered. When working withorganisms the measurements of oxygen concentration may e.g.be an essential parameter since it reflects bacterial, photosyntheticand respiration activity. Since oxygen solubility is dependent ontemperature in water, this is an essential supporting parameterinterpreting the oxygen data. Presence of anthropogenic activitiesor possible anthropogenic influence in the area of question shouldalso be considered when planning the sampling design. In the arc-tic, the latter could typically be long-range transported pollutants.

2.1.3. Sensor platforms covering different spatial and temporal scalesIn addition to measurements of essential parameters, the qual-

ity of the data interpretation is dependent on the spatial and tem-poral resolution and coverage of the data. Hence, the dynamics inboth space and time have to be considered in the mission planningprocess. As illustrated in Fig. 3, the temporal and spatial resolutionand coverage capabilities varies between the different platformsand corresponding sensors.

Fig. 2. The steps in IEMM from planning of a mission, data gathering and storage to interpretation of data as basis for decision making.

376 I. Nilssen et al. /Marine Pollution Bulletin 96 (2015) 374–383

Remote sensing platforms have high spatial coverage withvarying resolution. For satellites in particular, improvements inoptics, navigation and communications have made it possible toproduce geo-localised images with 1–1000 m spatial resolutionper image pixel (Johnsen et al., 2011) at high speed providing verylarge areal coverage. However, for marine applications, satelliteshave their limitations as they, dependent on illumination fromthe sun light, cloud cover, sun glint and waves can characterisethe sea surface and upper water column only. Aircrafts providessimilar capabilities as satellites and are useful to map and monitorshallow water areas and coastlines in an effective manner.However, being closer to the sea surface the spatial resolutionincreases, while the areal coverage decreases. As for the satellites,sensors mounted on aircrafts will be able to analyse the uppermostpart of the water column only due to water transparency. Lately,research on unmanned aerial vehicles (UAV) and autonomy hasincreased the interest to apply low-cost UAV as sensors platformsand communication hubs between sensor platforms in the surfaceor the air and e.g. a mother vessel supporting AUV operations withsome distance to the launching vessel (AMOS, n.d.).

The payload capacity and spatial coverage of a ship can be high.The ship serves as both a sensor platform as well as a mother plat-form launching and supporting other sensor platforms such asROVs and AUVs. The ship can provide data from the sea surface,the water column and the seabed. The ship may be an excellentplatform as the laboratory with its scientist is brought out on thesea.

For fixed platforms, such as landers, moorings and oceanobservatories, the temporal resolution can be high provided suffi-cient energy supply and data storage capacity. However, the spatialresolution will be limited to the coverage and range of themounted sensors. Most sensors are point samplers, while others,such as active acoustics can cover a wide area. The range of activeacoustics is dependent of the frequencies used, varying frommetres to several kilometres (Godø et al., 2014).

Mobile sensor platforms operating in the water column arenormally deployed from a ship/floater. The ROV can be equippedwith different sensors and manipulators for sampling and, depen-dent of type and size of the ROV, the depth range varies from 100to 6000 m. Recent ROV motion control systems (Dukan and

Sørensen, 2014; Sørensen et al., 2012) provides manoeuvring capa-bilities with station keeping/hovering (dynamic positioning) andtarget and bottom tracking. High-resolution data for targeted areacan be provided with detailed seabed mapping and sampling withdown to mm spatial resolution. The umbilical gives unlimited elec-trical power and high bandwidth communication. AUVs may bedivided into small AUVs (0–100 m depth) and large AUVs (0–6000 m). Small AUVs may be operated from small boats and fromshoreline. Torpedo shaped survey AUVs are mostly used in map-ping providing good hydrodynamic and manoeuvring capabilitiesfor tracking and path following (Moline et al., 2005). So far, thereis limited access to AUVs with station keeping/hovering capabili-ties. This is at present also the situation for AUVs with manipulatorcapabilities doing light intervention and sampling. The AUVs mainstrength is the longitude, latitude and depth mapping capabilities.AUVs can provide seabed and water column mapping with highspatial resolution data over large areas. The survey area coverageper time is significantly higher compared to ROV as the ROV haslimited spatial range due to the exposed current loads/drag forceson the umbilical (Johnsen et al., 2011). Gliders’ operational rangeand spatial coverage are high compared to AUVs and ROVs. Thespeed is rather low, following ocean current systems at a minimumof energy, using principles of buoyancy driven propulsion. Sincethe operation may go on for weeks, the spatial range is high. Formeasurements in the water column, the glider is good. However,the accuracy in navigation and control is limited. The ability forbenthic surveys is also limited (Testor et al., 2010). More detailson underwater sensor platforms, including ships are found inAppendix A.

From signal reconstruction theory it is well established that atime-periodic signal at least should be sampled twice each period(Nyquist frequency), or ideally 20–40 times each period in order toreconstruct the evolution including mean, maximum and mini-mum values (Åstrøm and Wittenmark, 1997). The same approachmay also be considered for reconstruction of spatial variations ofthe area of interest. For a homogenous system the geographicallocation of the measurements is not essential. However, if horizon-tal and vertical spatial variations are of interest or concern, properarea planning and selection of technology platform(s) are critical.For instance, a lander (Appendix A) may provide both high

Fig. 3. Spatial and temporal resolution and coverage of different platforms shown in magnitudes. Figure based on (Haury et al., 1978).

temporal and even high spatial resolution, however, on a limitedspatial coverage. This motivates use of mobile platforms such asAUV and ships. On the contrary, data from mobile platforms maybe limited by the temporal resolution on a given location depen-dent on how often that point is re-visited. Hence, providing simul-taneously high spatial and temporal resolution and coveragemight, in addition to proper planning of mission, require a combi-nation of several platforms. Other constraints such as payloadcapacity, energy endurance and level of autonomy related toenergy and control system intelligence including communicationcapabilities will impact the sensor platforms’ spatial and temporalresolution. The accessibility of the data depends on the solution forpower and data transfer from the sensor platform and, if notcabled, the duration of the deployment and any communicationnetwork and methodology set up using e.g. acoustics, optics, satel-lites or radio waves.

2.2. Data sampling (Steps 4–6)

In Steps 4–6 the sampling/data acquisition is in progress.Adaptive sampling may occur if specified in Step 3, depending ondata request, or if sensor data triggers predefined interdependentsequences. If data collected in Step 4 is not in accordance withthe data request, new adjusted data request can be made eitherautomatically (Step 5) or manually (Step 6) and fed back to Step3 e.g. adjust sampling area (if mobile platform or active acousticswith movable echo sounder), adjust sampling frequencies, adjustrange (if mobile platform), etc. until request is satisfied.

In Steps 5 and 6 data are processed, analysed and transformedinto information contributing to e.g. answering research questionsand mission objectives. After assessment of information and com-parison with prior knowledge, it will be accumulated knowledge.An assessment of the acquired information with regard to the orig-inal research question (Step 2) is made (Step 7). If acquired data isinsufficient in terms of scope, resolution, etc., then either repro-cessing in Step 6 or the data request is re-specified (Step 8). Ifinformation from acquired data is irrelevant to the original ques-tion (Step 2), or the question turns out to be erroneously formu-lated or imprecise, a new request must be formulated based onaccumulated knowledge (Step 10), or else Step 11.

2.3. Data processing (Steps 5 and 6)

While the understanding of the oceans in the past has been lim-ited due to sampling constrains such as lack of technology and dis-tributed observations, it could now be stated that the ability tohandle, analyse and interpret the vast amount of data, includingdata noise, received from an increasing variety of sensors hasbecome the main bottleneck for transforming data to knowledge.For monitoring purposes, multisensory data can be processed toproduce early warning information on a few key parameters criti-cal to whatever is to be monitored (Ølmheim et al., 2010). The keyparameters are chosen based on the understanding of the processor environment to be observed. Hence, unless there is an exhaus-tive understanding of what to be mapped or monitored, the selec-tion of key parameters is liable to change in time with change inlevel of knowledge.

Allowing development of new data analyses and interpretationof knowledge over time requires appropriate long time storage,accessibility and sharing of data. First and foremost is storage ofraw data essential to secure that the data can be used independentof software development and available analytical tools. Secondly,easy access to the stored data is essential. Data availability isdependent on several issues from selected software solution, tothe semantic structure of the data and organisational attitude orstrategy for data sharing. Further, the present bottleneck of data

processing of the vast amount of incoming data requires develop-ment of automated and/or semi-automated systems helpinghumans to pinpoint the object and/or event of interest.

3. Case studies

Two case studies are selected to demonstrate the applicabilityof the IEMM concept for mapping (Case study 1) and monitoring(Case study 2) purposes. Detailed results from the case studies willnot be presented as such, as these are reported in separatepublications.

Case study 1 is from a survey carried out in Ny-Ålesund,Svalbard, Norway in January 2014. The study was performed aspart of a Master/PhD course at the University of Svalbard (UNIS)in underwater robotics, comprising laboratory studies and use ofmobile platforms for in situ data gathering. The case study showshow a widely formulated research question was approached byuse of different sensors deployed on different platforms, dependingon the changing need for specific data to produce relevant informa-tion in a challenging environment. The study was performed in aremote area (78�5503000N, 11�5502000E) during the polar night,which entails low temperatures down to �30 �C and darkness(nautical polar night). The main aim of the Svalbard survey wasto gather knowledge about how organisms are adapted and reactto the present ambient light climate such as moon, star and north-ern light during the polar night when the sun is under the horizonfor months in general, and to identify the environmental trigger forzooplankton diurnal vertical migrations (DVM) in particular.

The second case (Case study 2) is from the PeregrinoEnvironmental Monitoring of Calcareous Algae (PEMCA) projectcarried out related to the Peregrino offshore oil field in Brazil (atabout 23.31�S, 41.31�W). This case study shows how new andnon-destructive technology can be applied to obtain more relevantknowledge about the environment in question. The main chal-lenges related to this case study were the limited backgroundknowledge about the organisms of interest (calcareous algae), thecorrosive environment, the strong currents and the robustness ofthe monitoring platform. The purpose of the PEMCA project wasto monitor whether discharges of drill cuttings had an impact ona calcareous algae bed located on around 100 m water depth andapproximately 1.5 km away from the discharge point.

3.1. Case study 1: importance of light on marine life in the polar night

Several recent studies have indicated that the polar night is animportant period for key process such as reproduction, e.g. forpolar cod (Boreogadus saida) (Graham and Hop, 1995) and amphi-pod Onisimus caricus (Nygård et al., 2009). Low photon fluxes of dif-fuse light from the sun which is below the horizon, moonlight andthe northern light (Aurora borealis) are important environmentalparameters utilised by arctic organisms (Berge et al., in press).Berge et al. (2009) and Båtnes et al. (2013) documented that zoo-plankton continue to perform DVM also during the polar night.Furthermore, overlapping presence of phytoplankton and zoo-plankton during night was documented by Berge et al. (2014).

To understand these mechanisms and to gain knowledge abouthow to measure these mechanisms in situ (Step 2 in model), dedi-cated experimental work with key marine organisms were per-formed in the laboratory (Step 3) on the two krill speciesThysanoessa longicaudata and Thysanoessa inermis to identify theirthreshold levels for light (spectral sensitivity and intensity) (Step4). Furthermore, for further understanding of the importance oflight in predator–prey interactions in the arctic ecosystem, thepresence of bioluminescent zooplankton and reflectance in severalmacroalgae (Laminaria digitata, Saccharina latissima, Alaria

esculenta, Palmaria palmata and Ulva sp.) were also measured, bothin the laboratory and in situ (Step 4) (Unpublished results).

Several sensors and sensor platforms were selected (Step 3) todetect and measure presence/activity of pelagic organisms in thewater column, including DVM (Step 4). An AUV equipped with asuite of sensors to map aggregations of zooplankton and phyto-plankton in the water column was pre-programmed to run thesame 1000 m long transect repeatedly at different depths. Theacoustic backscattering data acquired by the Acoustic DopplerCurrent Profiler (ADCP) on the AUV indicated, in addition to pres-ence, a vertical movement of zooplankton in the water column atdifferent depths and times. ADCP data in combination with othersensor data on the AUV indicated zooplankton grazing on phyto-plankton in the upper layers of the water column (Step 6).Increased concentration of chlorophyll a [Chl a] indicating biomassof phytoplankton, while decreased oxygen concentration [O2] andincreased concentrations of coloured dissolved organic matter[cDOM] indicated presence of organisms, grazing and decay(release and decomposition of organic material), respectively. Theaddition of conductivity (to obtain salinity), temperature and pres-sure (to obtain depth) by CTD measurements complemented thebiological datasets with physical characteristics of the watermasses. In shallow waters (4–30 m) two ROVs were used for visualmapping of benthic species distribution according to depth. Todocument the temporal resolution of the biological activity(macrofauna) over a period of some days, a stationary time-lapsecamera took images every 30 min. In addition, divers collectedand documented macroalgae and fauna in the shallower parts ofthe study area twice a day (day and night).

In the Svalbard case study, adaptive sampling was consideredon daily basis during the campaign by the course supervisors eval-uating the data recorded (Steps 5 and 6). Operational constrainssuch as low temperatures, strong wind, snow showers and wavesdid influence the operations on hourly and daily basis. This clearlyindicated the need to plan for sufficient days of operation. Thescope of the mission was well defined prior to the course, sensorsneeded were deployed, no discharges from point sources were pre-sent and, if other sensor needs should have been identified, thesensors could not reach the remote area in due time (Step 9).Challenges related to data processing were experienced. While dis-crete values of [O2], [cDOM], [TSM], salinity and temperature easilycould be presented in a joined figure for a visual comparison of thespatial presence of the parameters, the data processing from e.g.the side scan sonar was much more time consuming and difficultto integrate with the former data. The Svalbard Study 2014 gavenew insight into the importance of polar night light regime andhow it affects marine organisms and parts of the physiologicalmechanisms behind this (Step 7) which will be brought along tillfuture studies (Steps 10 and 11). However, during the study alsotechnical restrictions on the sensor platforms were made clearwhen planning for overlapping sampling areas (Step 9). Umbilicalon the ROV was too short to reach the location of the fixed time-lapse camera and the region of interest was also too shallow forthe AUV (Step 9).

3.2. Case study 2: Peregrino Environmental Monitoring of CalcareousAlgae

Another project where the IEMM approach is used is the PEMCAproject (Johnsen et al., 2014; Salgado et al., 2010). The objective(Step 2) of the PEMCA project was to monitor whether dischargesof drill cuttings, which is grained rock generated when drilling aborehole, had an impact on a calcareous algae bed located onaround 100 m water depth and approximately 1.5 km away fromthe discharge point. The drill cuttings discharged were not toxic,but due to the limited light available there was a fear that the

particulates could lead to calcareous algae death due to reducedlight levels and/or gas exchange cause by increased sedimentation.

A monitoring programme using non-destructive methods wasplanned (Step 3). Due to lack of knowledge about the impact ofincreased sedimentation on calcareous algae in a light limitedenvironment in general, there was also a need to perform labora-tory studies to measure the calcareous algae’s threshold levelsfor effect of sediment coverage (Step 3). A set of laboratory exper-iments, some lasting up till 9 weeks were exposing calcareousalgae for particulates (drill cuttings and sediments mimicking thegrain size of drill cutting) (Step 4). In the field, the monitoringequipment were deployed and retrieved several times during asampling period of approximately 2.5 years (Step 4). The fixedplatform (lander) was equipped with a surface buoy for online datatransfer. The data transfer lasted only for a short time before theweak link mounted on the cable between the fixed platform andthe surface buoy broke. The weak link was designed to break at acertain force and the exact cause is unknown (Johnsen et al.,2014). Due to this, the lander, in addition to the sediment trapsused in the field measurements, had continuous offline sampling.Hence no on-line analysis or options for adaptive sampling (Step5) were available. Knowledge accumulation after each retrievalcampaign (Step 6) did not identify needs to broaden the parametermeasurements.

However, when working with the time-lapse images from thefield, we see that there is a high degree of movements on theseabed. The rhodoliths (calcareous porous structures on whichthe calcareous algae is growing) change location and occasionallyalso the seabed topography changed. The temporal resolution ofthe images (two hours intervals) gives the opportunity to identifythe movement of rhodoliths in different areas of the images(Unpublished results), but the temporal resolution is too low totrack individual rhodolith movements. The knowledge gained fromthe PEMCA study (Step 10) has been used as basis for a new pro-posed monitoring programme at the Peregrino field (Johnsenet al., 2014). Further, as described in Johnsen et al. (2014), theexperiences with the lander lead to a redesign. It was made morerobust as the sensors now are more protected by the platformframe and the weight has been reduced to improve deploymentand retrieval. As the discharges of water-based drill cuttings atthe Peregrino field are determined to have a low risk for impactingthe calcareous algae seabed (Unpublished results), the proposedmonitoring programme does not include further on-line measure-ments. Furthermore, the outcome from the laboratory studies (Step10) has been used to optimise the environmental risk assessmenton the Peregrino field (Unpublished results). Results also show thatthere might be a temporal aspect related to the impact on the cal-careous algae. The 9 weeks experiment showed that, within theexperimental domain, the impact was related to sediment coverand that irradiance was of minor importance (Figueiredo et al.,2015). However, the photosynthetic efficiency (maximum quan-tum yield of charge separation in photosystem II: /PSIImax) of theexperimental controls that were kept in darkness without addedsediments, decreased slightly at the end of the experiment. Thismight indicate an incipient light limited impact on the calcareousalgae (Figueiredo et al., 2015). This finding is input to furtherresearch (Step 11). The experiences with the temporal resolutionwith the images have initiated further research activities (Step 11).

4. Discussion

The IEMM concept has through the case studies, demonstratedthat the model (Fig. 2) enables holistic and flexible approachesadjusted to the environment(s) in question and aim of missionregardless if it is initiated by authorities, academia or industry.Using existing data and/or identifying areas where knowledge

gathering is needed, this methodological approach will in a plan-ning process help to pinpoint critical aspect to consider at an earlystage of a process which makes the IEMM concept a tool to iden-tify, map and monitor objects of interest in areas with differentdegrees of a-priori knowledge. Such an approach enables an opti-mised and cost efficient knowledge and data gathering.

If used in a regime allowing a risk based approach, the IEMMconcept can, as used in Case study 2, encourage knowledge gather-ing essential to reduce assessment factors used in the risk assess-ment. Figueiredo et al. (2015) describes the use of laboratoryexperiments to establish exposure threshold levels for drilling dis-charge on calcareous algae. These results were used as input to therisk modelling to improve the risk assessment for drilling dis-charges at the Peregrino oil field (Unpublished results). Anotherexample where laboratory studies have been successfully per-formed in order to gain knowledge on key species with limitedknowledge is the coral risk assessment, monitoring and modelling(CORAMM) project (Purser and Thomsen, 2012) where thresholdvalues for water-based drill cuttings were established in the labo-ratory for the cold water coral Lophelia pertusa.

In contrast to the past, recent sensor developments andincreased use of mobile sensor platforms enables mapping andmonitoring technology that are non-destructive to the environ-ment as many of the measurements are based on optical measure-ments which is not depending on physical sampling of theenvironment. As described in Section 2.1.3 and Appendix A, differ-ent sensor platforms have their pros and cons and, if used in anoptimised manner, alone or in combination, they can enable sam-pling with the right temporal and spatial resolution and coverageadjusted to purpose of mission. However, besides of some ‘‘lessonslearned experiences’’, such as with the time-lapse cameradescribed for calcareous movements in Case study 2, operationalexperiences with combined use of sensor platforms in an IEMMconcept are still limited.

The need to optimise data sampling has increased the focus onthe ability for real-time observations of what is being sampled.Being able to alter a real-timemission plan could increase the valueof the gathered data. Reduced mission time, and thereby reducedrisk related to the operation, is an additional important advantageof such an adaptive sampling approach. The potential gain of chang-ing the mission plan in real-time is well known (Brito et al., 2012).For Case study 1 it is obvious that an improved autonomous adap-tive sampling capability using AUV to document both DVM andzooplankton grazing on phytoplankton would have optimised thetime and energy consumption of the AUV operation. Since the zoo-plankton migration will be at different depths throughout the day,the AUV could utilise the ADCP data to alter its desired depth, whilethe other sensors measuring [Chl a], [cDOM], [TSM] and oxygenconsumption only were activated when the desired target wasfound. Such an approach would provide more valuable scientificdata, in addition to higher temporal resolution and lower powerconsumption. As there were several parallel operations ongoing,the AUV could also have been used for other missions at an earlierstage. Another example describing the use of autonomous adaptivesampling is the use of an AUV to track and sample the thermoclineoff the Portuguese coast, as reported in Cruz and Matos (2010). TheAUVwas programmed to descend to a certain temperature gradientand then stay within the depth interval containing the desired tem-perature. Other relevant examples can be found in Berge et al.(2012), Ferri et al. (2010), Giguere et al. (2009), Moline et al.(2005) and Zhang et al. (2012).

Adaptive control based on sensor readings does, however,require the data to be processed in real, or near real-time, whichplaces strict requirements on the computer processing capacityand power on the sensor platform. Only recently the computa-tional power versus electrical power consumption, as well as the

development of software architecture, rapid signal processingand multivariate data analysis, guidance and control of sensor plat-forms has reached a point that allows practical implementationsfor autonomously adaptive sampling (Brito et al., 2012; Cruz andMatos, 2010). One challenge in data analysis will be to adapt modelreduction schemes in order to improve real-time performance. Themain operational limitations for e.g. an AUV are the shortage inendurance due to limited power supply and lack of needed auton-omy. Research attention and efforts on battery and autonomycapabilities, including real time data analysis, guidance, navigationand control systems, will increase the abilities providing smarter,safer and more efficient operations in an unstructured environ-ment with on-board planning and re-planning capabilities suchas collision avoidance for handling of unexpected events and opti-mised data gathering near real-time. However, it is essential tobear in mind that even though sensor platform developmentsenable adaptive sampling, the quality of the data gathered isdependent on the sensors mounted on these platforms. In this per-spective it is important to be aware of the individual sensor’s con-strains when it comes to signal quality and degradation due toresponse time and any sensor faults, including platform generatednoise.

Despite improved coverage of mapped and monitored areas dueto extended use of mobile platforms, there are still huge demandsrelated to data gathering. Modelling might be used to predict sim-ilar occurrences in similar habitats in a broader geographical scale.Furthermore, modelling should also be used to optimise the envi-ronmental monitoring programme. Data from monitoring shouldalso be of such quality (type of parameter(s) with sufficient tempo-ral and spatial resolution) that it can be used to improve futuremodel predictions. Modelling can help optimising the environmen-tal monitoring programme by focusing on the most critical aspectsand sampling strategy (Nilssen and Johnsen, 2008). The impor-tance of linking modelling and environmental monitoring isdescribed, among others by Godø et al. (2014), Nilssen andJohnsen (2008) and OGP (2012). In a real-time monitoring conceptalso systems enabling online risk modelling should be developed.

Optimising knowledge gathering requires more than just meth-ods for autonomous adaptive sampling. Optimised use and under-standing are also to a large extent linked to development ofanalytical tools combining a variety of data types from various datasources. Results both from the laboratory and from the field part ofCase study 2 have been processed using multivariate data analysis.However, none of the case studies involved ‘‘on the fly’’ analysis. Ifdata is to be used for decision support as an integrated part of theindustry’s daily operations, the time aspect is specifically criticalfor real-time monitoring and control. Ideally, information enablingactions before impact should be available. Use of multivariate anal-yses is starting to become more commonly used for early warningof changes (Johnsen et al., 1994; Robbins et al., 2006) and there areexamples where existing software on real-time multivariate anal-ysis have been applied on environmental data (Statoil, n.d.).However, suitable software and analytical tools processing partic-ularly big data, such as sound, optical images, hyperspectral data,acoustic images and others, are with few exceptions (Purseret al., 2009; Schoening et al., 2012a,b), lacking. Hence, importantdata is at present not included in existing multivariate data analy-sis. Effort to overcome this bottleneck is essential if these data is tobe used with other sensor data.

Despite all the advantages with the IEMM’s holistic approachfor mapping and monitoring and all the recent development insensor and sensor platform technology, there are several externalfactors that have to be fulfilled if the IEMM concept shall be a suc-cess. National regulations are diverse (Borja and Dauer, 2008; Borjaet al., 2010), and an implementation of the IEMM concept requirese.g. that the regulatory bodies are willing to re-think the basis for

their regulatory framework and accept the flexibility needed. Ingeneral the need for an integrated management system is acceptedbut there are still challenges with regards to implementation(Elliott, 2014; Leslie and McLeod, 2007). To ensure a positiveimplementation of IEMM in industries’ daily operations, recognis-ing the value of multidisciplinary use of already existing data aswell as mutual needs for new data is vital, e.g. is leak detectionboth a maintenance and an environmental related issue. Webelieve that data sharing and availability of data e.g. from industryto academia can drive research and innovation, and thereby lead tofurther improvement of all aspects from sensors and sensor tech-nology to analytical tools and use of knowledge.

In addition to available technology and opportunities providedby amultidisciplinary team, it is also crucial to be aware of the needfor mutual understanding and the need for behavioural changebetween the actors involved to drive innovation (Carlile, 2004).Carlile (2004) describes the need to acknowledge both domain-specific and common knowledge when working across interestsand disciplines. In addition he presents a model describing howthe need for communication increases with the degree of complex-ity. According to Carlile, successful implementation of a complexmodel such as IEMMwill require a significant practical and politicaleffort to create a common interest in sharing and assessing data.This is also in line with Elliott (2014), claiming that the only wayto obtain an integrated marine management framework is to havesectorial managers willing to think across both vertical and hori-zontal levels of integration. Establishing common knowledge andunderstanding the complexity of each other’s needs is necessaryfor mutual trust, to understand the individual data needs and fora successful data sharing and interpretation. Seeing these contexts,using a system facilitating data flow and access and having systemsprocessing the data on an appropriate temporal scale, authoritiescan utilise them for regulatory purposes, while industry can takethe knowledge into their day-to-day operations.

5. Conclusion

The proposed IEMM concept is a flexible system which enablesa case-by-case approach adjusted to purpose and object of interest.The purpose can span from measuring natural variations to moni-toring of point discharges from anthropogenic activities, and theobject of interest can either be (an) organism(s), an area or athing/an infrastructure. With this flexibility the concept can facili-tate easier cross disciplinary cooperation, utilisation of commoninfrastructure and sharing of data among different stakeholders.As a planning tool the IEMM concept aids to optimise selectionof parameters to measure and sensor platforms to use to get thebest temporal and spatial resolution fitted for purpose in each case.

The concept will also help to reveal gap of knowledge of the objectof interest, which could enable initiation of laboratory experimentsin due time before a field study, if deemed needed. In an opera-tional phase an adaptive sampling, either fully automated or withhuman interactions, can optimise the data gathering eitherthrough switching sensors on/off or through an adjusted samplingmission following the object of interest instead of following a pre-programmed sampling route in dynamic and changing environ-ments. As for data processing and analytical tools, there is still alack of knowledge/available systems to perform a cost-efficientadaptive sampling. However, we believe that the IEMM conceptwill contribute to optimise knowledge gathering and samplingstrategies through data and information feedback loops. In addi-tion it addresses areas with potential for further developmentand optimisation for mapping and monitoring purposes.

6. Abbreviations and definitions

ADCP acoustic doppler current profilerAUV autonomous underwater vehicleBig data huge amount of complex data collected with

high velocity, cauterised by non-linearity andhigh intrinsic dimensions

cDOM coloured dissolved organic matterCTD conductivity, temperature, depthDVM dial vertical migrationIEMM integrated environmental mapping and

monitoringParameter a coefficient that may be either constant or time

varying, in contrast to a variable that is a time-varying state of a dynamic system. As bothparameters and variables may be time varyingthe term parameter is used for both.

ROV remotely operated vehicleSensor

platformfixed or mobile instrument carrier

Acknowledgements

This work has been carried out as part of Ingunn Nilssen’sindustrial PhD, enabled by Statoil ASA and the Research Councilof Norway through the Industrial PhD programme, ProjectNumber 222718.

The work was also partly supported by the Research Council ofNorway through the Centres of Excellence funding scheme –AMOS, Project Number 223254 and the EWMA project (NRC195160).

Appendix A. Underwater sensor platforms

Platform Characteristics Pros Cons Arctic challenges

� Landers andmoorings

� Instrument platform either standingon the seabed (lander), or suspendedin the water column moored to theseabed or tethered to floating deviceat the surface

� Offline or accessible via cable or satel-lite/radio communication

� Limited risk for collisionswhen installed

� High payload capacity� High temporal resolutionof data sampling (timeseries)

� Low cost operation wheninstalled

� When online, access toreal-time data

� Possibility for synopticmeasurements along ver-tical profile

� Samplings limited to location, i.e.low spatial coverage

� Bio fouling and corrosion compli-cate calibration and degrade thesensors and data quality over time

� For online systems high cost forinstallation of power and commu-nication systems

� Lander and cables exposed fordamage due to e.g. trawling anddredging

� Increased risk and costduring deployment

� Access to instru-ments/data limited byharsh environment andice

� On shallow water icebergs may damageequipment

(continued on next page)

References

AMOS. n.d. <http://www.ntnu.edu/amos> (retrieved 12.02.15).Åstrøm, K.J., Wittenmark, B., 1997. Computer-Controlled Systems: Theory and

Design. Prentice Hall Inc, Upper Saddle River, New Jersey.Båtnes, A.S., Miljeteig, C., Berge, J., Greenacre, M., Johnsen, G., 2013. Quantifying the

light sensitivity of Calanus spp. during the polar night: potential fororchestrated migrations conducted by ambient light from the sun, moon, oraurora borealis? Polar Biol., 1–15. http://dx.doi.org/10.1007/s00300-013-1415-4.

Bellingham, J.G., 2014. Have robot, will travel. Methods Oceanography 10, 5–20.http://dx.doi.org/10.1016/j.mio.2014.10.001.

Berge, J., Cottier, F., Last, K.S., Varpe, Ø., Leu, E., Søreide, J., Eiane, K., Falk-Petersen, S.,Willis, K., Nygård, H., Vogedes, D., Griffiths, C., Johnsen, G., Lorentzen, D.,Brierley, A.S., 2009. Diel vertical migration of Arctic zooplankton during thepolar night. Biol. Lett. 5, 69–72. http://dx.doi.org/10.1098/rsbl.2008.0484.

Berge, J., Båtnes, A.S., Johnsen, G., Blackwell, S.M., Moline, M.A., 2012.Bioluminescence in the high Arctic during the polar night. Mar. Biol. 159,231–237. http://dx.doi.org/10.1007/s00227-011-1798-0.

Berge, J., Cottier, F., Varpe, Ø., Renaud, P.E., Falk-Petersen, S., Kwasniewski, S.,Griffiths, C., Søreide, J.E., Johnsen, G., Aubert, A., Bjærke, O., Hovinen, J., Jung-Madsen, S., Tveit, M., Majaneva, S., 2014. Arctic complexity: a case study on dielvertical migration of zooplankton. J. Plankton Res. 36, 1279–1297. http://dx.doi.org/10.1093/plankt/fbu059.

Appendix A (continued)

Platform Characteristics Pros Cons Arctic challenges

� Ships � Important components: hull, deckspace, crane, power, thruster andpropulsion system, DP system, sensors

� Mother ship and control centre forpayloads such as ROV, AUV, UAV

� Very largepayloadcapacity� Very large temporal andspatial coverage andresolution

� Conduct and handle sam-ples from any gears andsensor platforms

� Ship-based sensors will have lim-ited resolutions for increasingwater depth

� Costly operation

� Ice and icebergs limitsthe operability

� Icing on sensitiveequipment and sensors

� RemotelyOperatedVehicle(ROV)

� Main components: vehicle, umbilicaland control stations

� Delivered in different sizes, depth andthrust ratings, functionality, manipu-lator and sensors

� Power supply and communication byumbilical

� Navigation by means of acoustics, INS,DVL

� ROV is normally launched from craneon ship with station keeping capabil-ity by anchors or dynamic positioning(DP) system

� High payload capacity� High temporal and spatialresolution data for tar-geted area providingdetailed seabed mappingand sampling

� Umbilical gives almostunlimited electricalpower and high band-width communication

� Manipulator arms forsampling and intervention

� Collection units (watermasses and seabed)

� Limited spatial range and cover-age: usually < 1 km transect lines,< 100 m2

� Umbilical is exposed to currentloads/drag forces limiting spatialcoverage even more on deeperwaters

� Expensive operation due to dayhigh rates on ships

� Data quality could be degraded bypoor ROV guidance, navigationand control

� Increased risk and costduring operations

� Ice and icebergs reducespatial coverage andincrease risk for collisions

� Ship may be unable tomaintain position dueto ice drift

� AutonomousUnderwaterVehicle(AUV)

� Main components: vehicle, controlstation, acoustic navigation and forbigger AUVs launch and recoverysystem

� Delivered in different sizes, depth andthrust ratings, functionality, andsensors

� Carries its own power supply� Operates supervised or autonomouslywith limited communication ability

� High payload capacity(however less capacitycompared to ROV)

� 3D (long, lat and depth)mapping capabilities areunique

� High spatial resolutiondata for large area provid-ing detailed seabed andwater column mappingand monitoring

� High spatial coverage pertime

� Operates untethered andautonomously

� New research on auton-omy improves AUV intel-ligence and ability tooperate in an unstruc-tured environment

� Allows operations inareas that have limitedor no accessibility withother platforms

� Risk of operation – loss of data andvehicle

� Limited power supply on-board� Today: need for competence onAUV crew for launch and recovery,planning of operation and trou-bleshooting during different oper-ational scenarios

� Possible limitations in operationdue to ship traffic and risk forcollision

� Increased risk and costduring operations

� Ice and icebergs reducespatial coverage andincrease risk for collisions

� Low temperatures reducebattery efficiency

� Ice coverage preventsdirect ascent to surfacefor GPS fix for naviga-tion and increase riskfor loss of AUV

� Layers of fresh, muddyand salt waters chal-lenge navigation andcontrol

� Gliders � Components: vehicle and controlstation

� Delivered in different sizes, depth andpower ratings, and sensors

� Carries its own power supply� Operates normally autonomously� Operation typical 15–30 days covering600–1500 km range

� Deployed from boats and quayside� Buoyancy driven propulsion by vari-able ballasting of water

� Very large water columncoverage over long distances

� Few personnel involvedduring operation

� Follow large ocean cur-rent systems

� Low payload capacity� Slow speed operation – 0.4 m/saverage.

� Due to control concept useless forbenthic (seabed) investigations

� Limited power, payload, controland navigation accuracy

� Risk of operation – loss of data andvehicle

� Limited on-line control� Possible limitations in operationdue to ship traffic and risk forcollision

� Ice and icebergs reducespatial coverage andincrease risk for colli-sions and loss of vehicle

� Low temperatures reducebattery efficiency

� Ice coverage preventsdirect ascent to surfacefor GPS fix for navigation

Berge, J., Renaud, P.E., Darnis, G., Cottier, F., Last, K., Gabrielsen, T., Johnsen, G.,Seuthe, L., Weslawski, J.M., Leu, E., Moline, M., Nahrgang, J., Søreide, J., Varpe, Ø.,Lønne, O.L., Falk-Petersen, S., in press. In the dark: paradigms of Arcticecosystems during polar night challenged by new understanding. Progress inOceanography.

Borja, A., Dauer, D.M., 2008. Assessing the environmental quality status in estuarineand coastal systems: comparing methodologies and indices. Ecol. Ind. 8, 331–337. http://dx.doi.org/10.1016/j.ecolind.2007.05.004.

Borja, Á., Elliott, M., Carstensen, J., Heiskanen, A.-S., van de Bund, W., 2010. Marinemanagement – towards an integrated implementation of the European MarineStrategy Framework and the Water Framework Directives. Mar. Pollut. Bull. 60,2175–2186. http://dx.doi.org/10.1016/j.marpolbul.2010.09.026.

Brito, M.P., Bose, N., Lewis, R., Alexander, P., Griffiths, G., Ferguson, J., 2012. The roleof adaptive mission planning and control in persistent autonomous underwatervehicles presence. Autonomous Underwater Vehicles (AUV), 2012 IEEE/OES, 1–9.

Carlile, P.R., 2004. Transferring, translating, and transforming: an integrativeframework for managing knowledge across boundaries. Organ. Sci. 15, 555–568. http://dx.doi.org/10.1287/orsc.1040.0094.

Cruz, N.A., Matos, A.C., 2010. Adaptive sampling of thermoclines with autonomousunderwater vehicles. Oceans 2010, 1–6. http://dx.doi.org/10.1109/OCEANS.2010.5663903.

Dukan, F., Sørensen, A.J., 2014. Sea floor geometry approximation and altitudecontrol of ROVs. Control Eng. Pract. 29, 135–146. http://dx.doi.org/10.1016/j.conengprac.2014.04.006.

Elliott, M., 2014. Integrated marine science and management: wading through themorass. Mar. Pollut. Bull. 86, 1–4. http://dx.doi.org/10.1016/j.marpolbul.2014.07.026.

Fagerås, B., 2011. Time to retool the future of seabed monitoring and surveillance.First Break, 29.

Ferri, G., Jakuba, M.V., Yoerger, D.R., 2010. A novel trigger-based method forhydrothermal vents prospecting using an autonomous underwater robot.Autonomous Robots 29, 67–83. http://dx.doi.org/10.1007/s10514-010-9187-y.

Figueiredo, M.A.O., Eide, I., Reynier, M., Villas-Bôas, A.B., Tâmega, F.T.S., Ferreira,C.G., Nilssen, I., Coutinho, R., Johnsen, S., 2015. The effect of sediment mimickingdrill cuttings on deep water rhodoliths in a flow-through system: experimentalwork and modeling. Mar. Pollut. Bull. 95, 81–88. http://dx.doi.org/10.1016/j.marpolbul.2015.04.040.

Giguere, P., Dudek, G., Prahacs, C., Plamondon, N., Turgeon, K., 2009. Unsupervisedlearning of terrain appearance for automated coral reef exploration. Can. Conf.Comput. Robot Vision 2009, 268–275. http://dx.doi.org/10.1109/CRV.2009.41.

Glenn, S.M., Schofield, O.M.L., Chant, R., Kohut, J., McDonnell, J., 2005. Educationalneeds in the changing field of operational oceanography: training the peoplethat will sustain Munk’s 1+1 = 3 scenario. In: OCEANS, 2005. Proceedings ofMTS/IEEE, vol. 3. pp. 2549–2556. http://dx.doi.org/10.1109/OCEANS.2005.1640154.

Godø, O.R., Handegard, N.O., Browman, H.I., Macaulay, G.J., Kaartvedt, S., Giske, J.,Ona, E., Huse, G., Johnsen, E., 2014. Marine ecosystem acoustics (MEA):quantifying processes in the sea at the spatio-temporal scales on which theyoccur. ICES J. Mar. Sci. 71, 2357–2369. http://dx.doi.org/10.1093/icesjms/fsu116.

Graham, M., Hop, H., 1995. Aspects of Reproduction and Larval Biology of Arctic Cod(Boreogadus saida). Arctic 48, 130–135. http://dx.doi.org/10.14430/arctic1234.

Haury, L.R., McGowan, J.A., Wiebe, P.H., 1978. Patterns and processes in the time-space scales of plankton distributions. In: Steele, J. (Ed.), Spatial Pattern inPlankton Communities, vol. 3. Springer, US, pp. 277–327. http://dx.doi.org/10.1007/978-1-4899-2195-6_12.

Hepsø, V., Låte, M., Gramvik, G., Johnsen, S., Nilssen, I., Wesenberg, H., 2012.Integrated Environmental Monitoring in Daily Operations. SPE IntelligentEnergy International. http://dx.doi.org/10.2118/150054-ms.

Johnsen, G., Samset, O., Granskog, L., Sakshaug, E., 1994. In vivo absorptioncharacteristics in 10 classes of phytoplankton: taxonomic characteristics andresponses to photoadaptation by means of discriminant and HPLC analysis. Mar.Ecol. Prog. Ser. 105, 149–157.

Johnsen, G., Volent, Z., Tangen, K., Sakshaug, E., 1997. Time series of harmful andbenign phytoplankton blooms in northwest European waters using theSeawatch buoy system. In: Kahru, M., Brown, C. (Eds.), Monitoring AlgalBlooms: New techniques for Detecting Large-scale Environmental Change, vols.115–143. Springer, New York.

Johnsen, G., Moline, M.A., Pettersson, L.H., Pinckney, J., Pozdnyakov, D., Egeland, E.S.,Schofiled, O.M., 2011. Optical monitoring of phytoplankton bloom pigmentsignatures. In: Roy, S. et al. (Eds.), Phytoplankton Pigments: Characterization,Chemotaxonomy and Applications in Oceanography, vol. 784. CambridgeUniversity Press.

Johnsen, S., Nilssen, I., Pinto, A.P.B., Torkelsen, T., 2014. Monitoring of impact ofdrilling discharges to a calcareous algae habitat in the Peregrino oil field inBrazil. In: SPE International Conference on Health, Safety, and Environment inOil and Gas Exploration and Production. http://dx.doi.org/10.2118/168356-MS.

Kukulya, A., Plueddemann, A., Austin, T., Stokey, R., Purcell, M., Allen, B., Littlefield,R., Freitag, L., Koski, P., Gallimore, E., Kemp, J., Newhall, K., Pietro, J., 2010.Under-ice operations with a REMUS-100 AUV in the Arctic. AutonomousUnderwater Vehicles (AUV), 2010 IEEE/OES, 1–8. http://dx.doi.org/10.1109/AUV.2010.5779661.

Leslie, H.M., McLeod, K.L., 2007. Confronting the challenges of implementing marineecosystem-based management. Front. Ecol. Environ. 5, 540–548. http://dx.doi.org/10.1890/060093.

MAREANO. n.d. <http://www.mareano.no/en/start> (retrieved 12.02.15).Maxson, M., Moline, M., Layton, S., Johnsen, G., Berge, J., Clark, C.M., 2011. Ice

detection for under ice AUV navigation. In: The Twenty-first InternationalOffshore and Polar Engineering Conference.

MBARI, 2014. <http://www.mbari.org/> (retrieved 12.02.15).Moline, M.A., Blackwell, S.M., Von Alt, C., Allen, B., Austin, T., Case, J., Forrester, N.,

Goldsborough, R., Purcell, M., Stokey, R., 2005. Remote environmentalmonitoring units: an autonomous vehicle for characterizing coastalenvironments. J. Atmos. Oceanic Technol. 22, 1797–1808.

Munk, W., 2000. Chapter 1 oceanography before, and after, the advent of satellites.In: David, H. (Ed.), Elsevier Oceanography Series, vol. 63. Elsevier, pp. 1–4(Chapter 1).

Nilssen, I., Johnsen, S., 2008. Holistic environmental management of dischargesfrom the oil and gas industry – combining quantitative risk assessment andenvironmental monitoring. In: SPE International Conference on Health, Safety,and Environment in Oil and Gas Exploration and Production. http://dx.doi.org/10.2118/111586-MS.

NORUS, 2011. <http://norus-science.com/> (retrieved 12.02.15).Nygård, H., Vihtakari, M., Berge, J., 2009. Life history of Onisimus caricus

(Amphipoda: Lysianassoidea) in a high Arctic fjord. Aquat. Biol. 5, 63–74.http://dx.doi.org/10.3354/ab00142.

Ocean Network Canada, 2014. <http://www.oceannetworks.ca/> (retrieved13.02.15).

OGP, 2012. Offshore environmental monitoring for the oil & gas industry.International Association of Oil & Gas Producers (OGP), Report no. 457.<http://www.ogp.org.uk/pubs/457.pdf>.

Ølmheim, J., Landre, E., Spillum, Ø., Hepsø, V., 2010. Decision support andmonitoring using autonomous systems. In: SPE Intelligent Energy Conferenceand Exhibition. http://dx.doi.org/10.2118/128202-MS.

Purser, A., Thomsen, L., 2012. Monitoring strategies for drill cutting discharge in thevicinity of cold-water coral ecosystems. Mar. Pollut. Bull. 64, 2309–2316. http://dx.doi.org/10.1016/j.marpolbul.2012.08.003.

Purser, A., Bergmann, M., Lundälv, T., Ontrup, J., Nattkemper, T.W., 2009. Use ofmachine-learning algorithms for the automated detection of cold-water coralhabitats: a pilot study. Mar. Ecol. Prog. Ser. 397, 241–251. http://dx.doi.org/10.3354/meps08154.

Robbins, I.C., Kirkpatrick, G., Blackwell, S.M., Hillier, J., Knight, C.A., Moline, M.A.,2006. Improved monitoring of HABs using autonomous underwater vehicles(AUV). Harmful Algae 5, 749–761. http://dx.doi.org/10.1016/j.hal.2006.03.005.

Salgado, C., Johnsen, S., Gomes, S., Tokelsen, T., 2010. Real-time environmentalmonitoring of discharges of drill cuttings in the Peregrino field (Campos Basin,Brazil). In: SPE International Conference on Health, Safety, and Environment inOil and Gas Exploration and Production. http://dx.doi.org/10.2118/127175-MS.

Schoening, T., Bergmann, M., Nattkemper, T.W., 2012a. Investigation of hiddenparameters influencing the automated object detection in images from the deepseafloor of the HAUSGARTEN observatory. Oceans 2012, 1–5. http://dx.doi.org/10.1109/OCEANS.2012.6405040.

Schoening, T., Bergmann, M., Ontrup, J., Taylor, J., Dannheim, J., Gutt, J., Purser, A.,Nattkemper, T.W., 2012b. Semi-automated image analysis for the assessment ofmegafaunal densities at the arctic deep-sea observatory HAUSGARTEN. PLoSOne 7, e38179. http://dx.doi.org/10.1371/journal.pone.0038179.

Schofield, O., Ducklow, H.W., Martinson, D.G., Meredith, M.P., Moline, M.A., Fraser,W.R., 2010. How do polar marine ecosystems respond to rapid climate change?Science 328, 1520–1523. http://dx.doi.org/10.1126/science.1185779.

SERPENT. n.d. <http://www.serpentproject.com/> (retrieved 12.02.15).Sørensen, A.J., Dukan, F., Ludvigsen, M., Fernandes, D.A., Candeloro, M., 2012.

Development of dynamic positioning and tracking system for the ROV Minerva.Further Advances in Unmanned Marine Vehicles. Institution of Engineering andTechnology, pp. 113–128. http://dx.doi.org/10.1049/PBCE077E_ch6.

Statoil. n.d. <http://love.statoil.com> (retrieved 13.02.15).Testor, P., Meyers, G., Pattiaratchi, C., Bachmayer, R., Hayes, D., Pouliquen, S., Petit de

la Villeon, L., Carval, T., Ganachaud, A., Gourdeau, L., 2010. Gliders as acomponent of future observing systems. OceanObs’09: Sustained OceanObservations and Information for Society, vol. 2.

Zhang, Y., Godin, M.A., Bellingham, J.G., Ryan, J.P., 2012. Using an autonomousunderwater vehicle to track a coastal upwelling front. IEEE J. Oceanic Eng. 37,338–347. http://dx.doi.org/10.1109/JOE.2012.2197272.

Paper II

Berge, J., Johnsen, G., Sørensen, A. J., Nilssen, I., 2015. Enabling technology for Arctic research. Science and Technology Magazine Vol 15, pp 199-201

that marine zooplankton have a (circadian)biological clock entrained by the day/night cycleand that this clock is adaptive in initiating amigratory response even when external lightcues are limited at depth. However, during thepolar night, when the strongest source ofillumination is no longer coming from the Sun,moonlight probably plays an important role inentraining migratory behaviour. Which speciesare implicated, and what the ultimate drivers areat this time, remains to be discovered.

As shipping routes open up through theNorthwest and Northeast passages and theexpanding oil, gas, fishery and mineralindustries become increasingly active, the riskof environmental damage in the Arcticincreases. There is a growing demand forsound management, decision making andgovernance guidelines that rely on a thoroughresearch-based understanding of theecosystem, not just snapshots from the brightpart of the year.

For example, we now know that many speciesacross most phyla and trophic levels utilise thepolar night for reproduction, hence increasingtheir potential vulnerability, particularly duringthis part of the year. Knowledge of the patternsand processes that characterise the entiremarine habitat during the polar night is,therefore, one of the most important gaps inknowledge, preventing the informedmanagement and sound decisions of the Arctic.

The unknownThe exploration, mapping and monitoring ofunknown areas are not trivial tasks. In a researchcontext, and particularly so when related to theArctic Ocean during the polar night, these areoften aimed at discovering or revealing hithertounknown objects, organisms, processes andphenomena. Although there are usually one orseveral features or qualities that one isspecifically looking for, a general understandingof an area requires the identification and

Enabling technology for Arctic research

PROFESSORS JØRGEN BERGE, GEIR JOHNSEN AND ASGEIR J SØRENSEN,AND PHD STUDENT INGUNN NILSSEN ON ENABLING TECHNOLOGIES FOR

ARCTIC MARINE ECOSYSTEM RESEARCH

The Arctic winter and polar night are emerging as key periodsduring which many reproductive and other ecologically importantprocesses occur. Recent studies have demonstrated

unexpectedly high levels of activity in the pelagic zone during the polarnight. These include the diel vertical migration of zooplankton (Berge etal. 2009) and nekton (Webster et al. 2013), increased understanding ofthe microbial community structure and function (Rokkan Iversen andSeuthe 2011), patterns of bioluminescence by different pelagicorganisms (Berge et al. 2012, Johnsen et al. 2014), foraging bypredators believed to rely at least in part on a visual search (Kraft et al.2012), and the ability to detect extreme low light levels in certain keyzooplankton species (Båtnes et al. 2013).

These recent discoveries, in addition to the pioneering studies by Polishand Russian researchers in the 1990s, are currently representing adramatic shift in our understanding of the marine system. It is no longerpossible to ignore the processes occurring during the polar night in orderto achieve a thorough and comprehensive understanding of the Arcticmarine ecosystem.

Synchronicity By way of example, there are many types of cyclic activity in marineorganisms that are highly synchronised within populations. It seems likely

Pan European Networks: Science & Technology 15 www.paneuropeannetworks.com1

Fig. 1 Platformsused and developedthrough AMOS

Cutting-edge interdisciplinary research involving technology and marine

science fields such as marine biology and archaeology will provide the

needed bridge to make high levels of autonomy a reality towards

autonomous underwater operations.

Enabling technologyPlatforms included in the AMOS landscape (Fig. 1) are remotely

operated vehicles (ROV), autonomous underwater vehicles (AUV),

gliders, landers/moorings, floating buoys, benthic landers, autonomous

sensor arrays operating in sea ice, and unmanned aerial vehicles (UAV).

The work within Mare Incognitum – the umbrella for several research

projects exploring the Arctic – is tightly integrated with this on-going

research at AMOS and the Applied Underwater Robotics Laboratory

(AUR-Lab) at NTNU.

An example on how such technological development provides new insight

into ecological processes under the ice comes from a recent work, in

which enabling technology allowed for sampling on broader spatial scales

than previously possible. It had been suggested in the scientific literature

that massive pelagic under-ice blooms occurred in the Arctic and that

these were of vital importance to understanding and reliably predicting

the fate of production in ice-covered waters of a warming Arctic.

Recently, however, Professor Johnsen and his colleagues have

demonstrated, based upon an unprecedented campaign using an AUV

under the Arctic pack ice (Fig. 2), that such pelagic under-ice blooms

are most likely a result of advection of phytoplankton from open water

areas, at least in the Eurasian sector of the Arctic.

Analogous studies could be designed to sample across temporal scales

or extended periods in great detail to investigate, for example, the

phenology of ice-algal bloom development. Autonomous measurements

and observations could then guide experimental studies to test for

mechanisms responsible and could provide important data for numerical

models assessing ecosystem sensitivity.

corresponding mapping of basic environmentalcharacteristics such as topography, temperatureand oxygen concentration in order to get anoverview of an unknown environment.

From an environmental managementperspective, an understanding of the area aspart of the larger ecosystem is consideredimportant. Hence parameters to measure,choice of sensors and sensor platforms subjectto operational constraints are decisions to bemade and remade in an environmentalmapping process. This will supply relevant datafor the better accommodation of diverse andcomplex mission purposes with potentiallymultiple stakeholders representing differentneeds and requirements.

Complex environmentsOperations in cold and distant areas with limitedcommunication possibilities motivate researchon autonomous, multifunctional, driftingobservatory platforms with reliable energysupplies, adaptive functionality and optimisationof technology and modes of operation.

The ability to sample in complex environmentscreates a need for onboard, in situ andautonomous data collection, quality control andsignal processing. Through the Mare Incognitumprojects (www.mare-incognitum.no), we areworking closely with the Centre for AutonomousMarine Operations and Systems (AMOS) atNTNU (Norway).

AMOSAMOS is, as a ten-year research programme anda centre of excellence (2013-2022), addressingresearch challenges related to autonomousmarine operations and systems applied in, forexample, maritime transportation, oil and gasexploration and exploitation, fisheries andaquaculture, oceans science, offshore renewableenergy, and marine mining.

Fundamental knowledge is created throughmultidisciplinary theoretical, numerical andexperimental research within the knowledgefields of hydrodynamics, structural mechanics,guidance, navigation and control. AMOS isengaged in the research challenges,achievements and experiences of selected fieldtrials related to integrated autonomousunderwater operations for mapping andmonitoring purposes in coastal waters inNorway and Arctic operations outside Svalbard.

www.paneuropeannetworks.com Pan European Networks: Science & Technology 15 2

Fig. 2 A combinedeffort involving

multiple sensor-carrying platforms

such as AUV under seaice, remote sensing

from a polar-orbitingsatellite and a research

vessel allows for amore holistic view on

productivity in ice-covered waters thanpreviously possible.

Each platform isequipped with an array

of sensors giving bio-geo-chemical-physicalinformation over larger

areas and depths

Pan European Networks: Science & Technology 15 www.paneuropeannetworks.com3

10.1007/s00300-013-1415-4, Electronic supplementarymaterial doi:10.1007/s00300-013-1415-4

Johnsen G, Candeloro M, Berge J, Moline M (2014):Glowing in the dark: discriminating patterns ofbioluminescence from different taxa during the Arctic polarnight. Polar Biology, 37, 707-713

Kraft A, Berge J, Varpe Ø, Falk-Petersen S (2012): Feedingin Arctic darkness: mid-winter diet of the pelagic amphipodsThemisto abyssorum and T. libellula. Marine Biology, 160(1),241–248. doi:10.1007/s00227-012-2065-8

Rokkan Iversen K, Seuthe L (2011): Seasonal microbialprocesses in a high-latitude fjord (Kongsfjorden, Svalbard): I.Heterotrophic bacteria, picoplankton and nanoflagellates.Polar Biology, 34, 731-749

Webster C, Varpe Ø, Falk-Petersen S, Berge J, Stübner E,Brierley A (2013): Moonlit swimming: vertical distributions ofmacrozooplankton and nekton during the polar night. PolarBiology, 1-11

Winds of changeChanges in the Arctic ocean-sea ice-atmosphere interface are leadingto rapid shifts in the structure, resilience and function of Arcticecosystems. Rapid decline in sea ice extent and thickness, increased airand ocean temperatures, increased water-column stratification, andmultiple dynamic physical and chemical changes significantly alter thepatterns of productivity at the base of marine food webs. Such changesare also anticipated to affect ecosystem structure and productivity higherin the food web.

Ultimately, Arctic marine ecosystem structure and productivity within thenext decades may be substantially different from what we observe today.Predictions as to how Arctic marine ecosystems may change arehindered by our inability to understand the year-round response of theArctic system.

The implementation of new enabling technologies in marine research istherefore likely to drastically increase our ability to understand the Arcticmarine ecosystem as one interlinked and connected system and,moreover, will provide the essential tools needed to explore one of theleast known realms of our planet – the Arctic Ocean during the polar night.

ReferencesBerge J, Cottier F, Last K S, Varpe O, Leu E, Søreide J, Eiane K, Falk-Petersen S, Willis K,Nygard H, Vogedes D, Griffiths C, Johnsen G, Lorentzen D, Brierley A S (2009): Diel verticalmigration of Arctic zooplankton during the polar night. Biology Letters, 5, 69-72

Berge J, Varpe O, Moline M A, Wold A, Renaud P E, Daase M, Falk-Petersen S (2012):Retention of ice-associated amphipods: possible consequences for an ice-free Arctic Ocean.Biology Letters, 8, 1012-1015

Berge J, et al.: In the dark: paradigms of Arctic ecosystems during the polar night challengedby new understanding. Progress in Oceanography. In press

Båtnes A, Miljeteig C, Berge J, Greenacre M, Johnsen G (2013): Quantifying the lightsensitivity of Calanus spp. during the polar night: potential for orchestrated migrationsconducted by ambient light from the sun, moon, or aurora borealis? Polar Biology, DOI

Professor Jørgen BergeUiT The Arctic University of Norway

browse www.arctosresearch.net

Professor Geir Johnsen Norwegian University of Science andTechnology (NTNU)

browse www.ntnu.no

Professor Asgeir J Sørensen Director AMOS, NTNU

browse www.ntnu.edu/amos

PhD student Ingunn NilssenStatoil ASA and NTNU

Changes in the Arcticocean-sea ice-atmosphere interfaceare leading to rapidshifts in the structure,resilience and functionof Arctic ecosystems

Paper III

Johnsen, S., Nilssen, I., Pinto, A. P. B., Torkelsen, T., 2014. Monitoring of

impact of drilling discharges to a calcareous algae habitat in the

Peregrino oil field in Brazil. SPE International Conference on Health,

Safety, and Environment in Oil and Gas Exploration and Production.

Long Beach, California, USA.

Doi: http://dx.doi.org/10.2118/168356-MS

Please note that the following manuscripts were under preparation at the time the paper was

written and consequently the references are incorrect. The correct citations are as follows:

Eide et al. (2013):

Eide, I., Westad, F., Nilssen, I., de Freitas, F. S., dos Santos, N. G., dos Santos, F.,

Cabral, M. M., Bicego, M. C., Figueira, R., Johnsen, S., 2016. Integrated

environmental monitoring and multivariate data analysis – A case study.

Integrated Environmental Assessment and Management,

doi: http://dx.doi.org/10.1002/ieam.1840

Elbers et al. (2013) has not been published

Figueiredo et al. (2013):

Figueiredo, M. A. O., Eide, I., Reynier, M., Villas-Bôas, A. B., Tâmega, F. T. S.,

Ferreira, C. G., Nilssen, I., Coutinho, R., Johnsen, S., 2015. The effect of sediment

mimicking drill cuttings on deep water rhodoliths in a flow-through system:

Experimental work and modeling. Marine Pollution Bulletin Vol 95, pp 81-88,

doi: http://dx.doi.org/10.1016/j.marpolbul.2015.04.040

Godø et al. (2013):

Godø, O. R., Klungsøyr, J., Meier, S., Tenningen, E., Purser, A., Thomsen, L., 2014.

Real time observation system for monitoring environmental impact on marine

ecosystems from oil drilling operations. Marine Pollution Bulletin Vol 84, pp 236-

250, doi: http://dx.doi.org/10.1016/j.marpolbul.2014.05.007.

Nilssen et al. (2013):

Nilssen, I., dos Santos, F., Coutinho, R., Gomes, N., Cabral, M. M., Eide, I.,

Figueiredo, M. A. O., Johnsen, G., Johnsen, S., 2015b. Assessing the potential impact

of water-based drill cuttings on deep-water calcareous red algae using species specific

impact categories and measured oceanographic and discharge data. Marine

Environmental Research Vol 112, Part A, pp 68-77,

doi: http://dx.doi.org/10.1016/j.marenvres.2015.09.008

SPE 168356

Monitoring of Impact of Drilling Discharges to a Calcareous Algae Habitat in the Peregrino Oil Field in Brazil Ståle Johnsen, Ingunn Nilssen, and Ana Paula Brandão Pinto, Statoil ASA, and Terje Torkelsen, METAS

This paper was prepared for presentation at the SPE International Conference on Health, Safety, and Environment held in Long Beach, California, USA, 17–19 March 2014.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract

The Peregrino field, operated by Statoil ASA, is located 80km south of Cabo Frio, in the Campos Basin area, Brazil. The field is operated with two well head platforms drilling production and injection wells and a FPSO located between these. The seabed hosts a calcareous algae (CA) habitat, an area with relatively rich biodiversity. Conventional sediment monitoring has been found not to give sufficient information about potential impact of drill cuttings.

A tailor suited, in-situ sensor based monitoring approach was developed for the field including testing and qualification of a number of sensor systems for visual observation, oceanographic parameters, light and turbidity, all placed on a seabed observatory frame. 4, 6-months sampling campaigns were carried out with this system. In addition, three sediment traps were placed at different locations, over totally 7, 4-months campaigns. Data were combined with discharge information and laboratory studies of drill cuttings effects to CA and associated species, to identify potential exposure and impact in the field. Discharge and environmental risk modeling were performed to identify which data gave significant and vital information for the purpose of environmental monitoring of the CA habitat.

Combining discharge information with in-situ observations and modeling gives a strongly improved basis for identifying impact of drill cuttings discharges in a field like Peregrino. Technology for this has been developed to a readiness level available for multiple use. A new, cost beneficial monitoring program is proposed as an alternative to traditional sediment monitoring.

Introduction

The Peregrino oil field is geographically located 80 km off the coast of Brazil, south of Cabo Frio, in the Campos Basin area. The field consists of two fixed well head platforms for drilling of production and water injection wells, and one floating production storage and offloading unit (FPSO). The two well head platforms (WHPA and WHPB) are located 10 km apart with the FPSO in between (Figure 1a). Drilling at Peregrino commenced in November 2010, and until May 2013, 19 production and 3 injection wells had been completed. The field came in production in May 2011. The well streams are transported to the FPSO through subsea pipelines, while produced water (water that is produced together with the hydrocarbons from the reservoir) is routed back to the well head platforms and re-injected to the reservoir. Drill cuttings with residuals of water based drilling fluid are discharged to the sea, while cuttings from the hydrocarbon containing reservoir section are sent to shore for disposal.

Water depth in the area varies from 90 – 120 m and seabed sediments consist in general of sand and silt with a hard surface. The more shallow part of the Peregrino seabed (<110m) hosts a calcareous algae (CA) community. At the Peregrino field the CA forms calcified structures up to a few cm diameters (rhodoliths) built of dead CA and other calcifying organisms such as Bryozoan and different species of Polychaeta. The rhodoliths act as substrate for a large number of marine fauna species. The living CA grows on the surface of the rhodoliths at a very low growth rate. The density of rhodoliths on the seabed varies, but may for some areas be very high, including several layers where CA lives on top of dead rhodoliths. Figure 1outlines the

2 SPE 168356

infrastructure of the Peregrino field and shows an example of a typical CA rhodolite.

Figure 1a) Illustration of the Peregrino field with the FPSO and the WHPs, and b) rhodolith with live calcareous algae (pink area)

The CA habitat of Peregrino is regarded as a deep water rhodolith bank, the species probably living on the very limit of physical confitions sufficient for their survival, an assumption is supported by the fact that CA communities are not observed at water depths larger than 110 – 115m in the area. While shallow water communities have been subject to several scientific studies over the past decades (Steller et al., 2007; Steller et al. 2009), less knowledge exists on the function, the sensitivity and the importance of the deep water communities of these rhodolith beds.

When Statoil acquired the operation of the Peregrino field in 2008, it was therefore decided to launch a R&D project to increase the knowledge of deep water CA communities, with the purpose of enabling sustainable environmental management of the Peregrino CA habitat, under the regime of discharges of water based drilling fluids and cuttings. The Peregrino Environmental Monitoring and Calcareous Algae (PEMCA) project was initiated in 2010 and completed in 2013, under the Brazilian National Petroleum and Gas Agency (ANP) Federal Participation Program (FPE). The project included four main activities; i) taxonomy studies of the habitat, ii) exposure and effect studies of drill cuttings, iii) environmental risk assessment and iv) development of environmental monitoring technology.

A comprehensive taxonomy and classification study of the Peregrino seabed was carried out by sampling CA and associated species through four field campaigns, followed by analyses and characterisation of the samples by the Federal University of Rio de Janeiro. In general, the Peregrino seabed is characterized as a rich biodiverse habitat. More than 200 species were identified through this activity a comprehensive overview over these have been given byTâmega et al. (2013).

Figueiredo et al (2013) performed controlled laboratory studies on exposure on impact of light variation, sediment (representing drill cuttings) coverage and water flow rates on CA and associated species (Bryozoa and Polychaeta), The exposed CA species proved to be relatively robust towards fluctuations in these parameters compared to other species reported in the literature, with highest sensitivity towards sediment coverage. Dose-response relationships were established for all variables and utilised as input to the environmental risk assessment studies performed in the project.

Nilssen et al (2013) performed environmental risk assessment of discharges of drill cuttings to the CA habitat by applying the Dose-related Risk and Effect Assessment Model (DREAM). DREAM has been developed of E&P operators in the North Sea (Johnsen et al. 2000, Singsås et al 2008, Nilssen and Johnsen 2008, Smith 2009) and is frequently used for discharges to the marine environment from the E&P industry. By utilising the results from the exposure and effect studies of the PEMCA project, in-situ oceanographic data and the discharge logs from the field, Nilssen et al. (2013) found that after two years of drilling, a significant risk of impacting the seabed environment close to the two well head platforms was present. This result would be expected for such an operation. However, since the well head platforms are located in deeper waters, outside the CA bank, the environmental risk for the CA habitat was less evident. Nevertheless, areas in the CA bank closest to the discharge points were found to be exposed to a significant environmental risk. Nilssen et al. (2013) underlines that even realistic and representative input data were used for the modelling, the output still represents a conservative approach to the real situation.

Only by designing an environmental monitoring program able to validate the environmental risk estimated from the above studies, can the true impact of drilling discharges be documented. In general, Statoil’s ambition is to utilise environmental risk assessment to identify the vital environmental variables to include in the environmental monitoring program for a specific field. The traditional approach to environmental monitoring of drilling discharges implies sampling and analysis of the upper layer of the sediment in the vicinity of the discharge point. Physical, chemical and biological parameters are normally included in this approach. Components like barium, alkanes and PAH are frequently used as markers for drill cuttings discharges, and are normally analysed along with grain size distribution, biodiversity and species (International Association of Oil & Gas

SPE 168356 3

Producers (2012), Climate and Pollution Agency (2011)). A similar approach to sediment monitoring in the vicinity of discharges of drill cuttings is also applied by the regulatory authorities in Brazil (IBAMA). In the operational licence for Peregrino sampling of sediment by grab/box corer is required for a transect system around each of the two well head platforms, covering an area out to 1600 m in the direction of the prevailing current (IBAMA 2010). By 2013, two surveys have been performed, showing, as expected, increased levels of barium in the area close to the platforms. However, no information concerning potential impact on the CA community can be extracted from these surveys. For the Peregrino seabed, it was evident from the baseline environmental survey that conventional sediment sampling by grab system and box corer would not give satisfying results. Due to the hard sediment surface and the presence of rhodoliths these systems were not able to bring an undisturbed sample to the surface, which indeed is the core of the traditional monitoring approach.

The PEMCA project was partly designed to compensate for this and to develop, install and qualify an alternative monitoring approach, using in-situ observations of environmental variables to validate and document potential impact of the discharges to the CA habitat. The ambition and the approach of this development were described by Salgado et al. (2010). In short, the approach included the building and deployment of a seabed platform with different sensor systems to observe presence, exposure and effects of drill cuttings with drilling fluid residues in a location at the CA bank in the Peregrino field. The use of such systems, known as ‘landers” or ‘ocean observatories’ has over the past decade found a number of applications within oceanographic and marine environmental research. The NEPTUNE (NEPTUNE, 2013) and ESONET/EMSO (ESONET, 2013) programs are probably the most recognized within the international scientific community, representing US/Canadian and EU initiatives, respectively. For the E&P industry, use of this type of technology represents a relatively new approach. Statoil has used sensor based systems, mounted on landers or floating devices for the monitoring of drilling discharges in areas with cold water coral reefs in the Norwegian Sea, at the Morvin and Hyme fields (Godø et al. 2013, Frost et al, 2013). These operations have, however, lasted for relatively limited time windows, over a few weeks. For the Peregrino field the horizon of the drilling activity represents minimum 5 – 7 years, implying more stringent requirements for system reliability and robustness.

The present paper outlines the results of this development along with a cost-benefit analysis of the technology and a potential implementation approach in the Peregrino field.

System design and functionality

Original system design and functionality

The design and functionality of the Peregrino ocean observatory described by Salgado et al. (2010), was based on the use of real-time observations through different sensor systems. The idea was to install the system in the vicinity of well head platform B, approximately 1.5 km southwest of the platform, in the direction of the prevailing current. This area represents the outer boundary of the CA bank, since the water depth increases closer to the platform. Figure 2 shows a schematic drawing of the overall system featuring the surface buoy, the anchor, the lander or sensor platform and a satellite holding the time lap camera. The figure also includes pictures of the buoy and the lander with the mounted sensor systems.

Figure 2: Schematic drawing of the Peregrino ocean observatory system and pictures of yhe sirface buoy and the main instrument frame (the lander)

4 SPE 168356

The surface buoy was designed to serve as a communication unit, enabling communication through satellite, a wireless router for short range communication (vessel based) and wireless internet through a base station installed on Peregrino well head platform B.

In the original design, the buoy was moored to an anchor with further connection to the lander which in turn was connected to a stainless steel frame holding a time lap video camera. The anchor mooring from the buoy was a steel armored cable with 2 pcs of single modus fiber and 4 x 1.5 mm2 power leads, working loaf 3 tons. A ‘weaklink’ with a breaking strength of 1.5 tons was mounted on the mooring cable. Amore detailed description of materials, cables, connections and power supply setup is given by Salgado et al. (2010).

In addition to the ocean observatory system, three sediment traps were included in the monitoring development project. These were deployed at different positions in and outside the expected influence area to verify the settling of natural particles and cuttings. The sediment traps were deployed with moorings placing the collection funnel 10 m above the seabed. A detailed outline of the applied sensor systems is given in Table 1.

Table 1: Sensor systems applied in the Peregrino ocean observatory (Salgado et al. 2010)

Sensor Producer Type Spec: Echo sounder 1 Simrad

Simrad/Metas Simrad

Simrad

Echo sounder EK60 X-GPT transceiver, 38 kHz Transducer: ES-38DD

ER60 Software

Power: 2.000 kW Frequency: 38 kHz Sensitivity: -202,5 dB re 1V per µPa Transducer beam: 7º

Echo sounder 2 Simrad Simrad/Metas Simrad

Simrad

Echo sounder EK60 X-GPT transceiver, 120 kHz Transducer: ES-120 CD

ER-60 Software

Power 500 W Frequency: 120 kHz Sensitivity: -187 dB re 1V per µPa Transducer bream: 7º

ADCP AADI -Aanderaa RDCP600 Frequency; 600 kHz Beams: 4 Depth Capacity; 300m

Light sensor Biospherical Instruments Inc

QCP-2000 Cosine PAR sensor 1,4x10-5 µE/cm2 sec to 0,5 µE/cm2 sec

Depth/Temp sensor

AADI -Aanderaa Accuracy: ± 0,005 S/m Resolution: 0,0002 S/m

Oxygen sensor: AADI -Aanderaa Accuracy: < 8 µM Resolution: < 1 µM

Turbidity sensor AADI -Aanderaa Accuracy: 2% of full scale Resolution: 0,1% of f.s

Salinity sensor AADI -Aanderaa Accuracy: ± 0,005 S/m Resolution: 0,0002 S/m

Compass, tilt and roll sensor

METAS MC 1500 Compass accuracy: < 0,5º Tilt: ± 90º Roll: ± 180º

Tilt sensor METAS MT 1500 Range: ± 180º

Sediment traps McLane LTD, USA

Parflux Mark 78H-21 Equipped with 21 collection inits fpr sequencial sampling

SPE 168356 5

Pre-deployment adjustments

The technical approach applied for the PEMCA ocean observatory implied the use of relatively large and heavy equipment, leading to relatively comprehensive requirements to the vessel used for deployment and retrieval of the system. A vessel with sufficient lifting capacity and heave compensated cranes was necessary to deploy and retrieve the whole system. A work-class ROV was also needed. In practice, these requirements were not met by Statoil supply vessels, and some adjustments to the plan were made to simplify the operations.

The plan of a separate camera satellite platform was abandoned and the camera was mounted directly on the Lander frame. This led to an adjustment of the angle of the camera that was mounted in a 45 degree angle, pointing downwards. It was assumed that this would not affect the quality of the images.

To reduce the weight of the lander, the battery containers were redesigned. The weight was reduced with approximately 10%.

The lander was originally equipped with an Aanderaa RDCP (ADCP) unit for ocean current and conductivity, temperature and pressure measurements. The RDCP unit was only able to measure the current in the lower 60 m of the water column, but this was regarded as sufficient since the cuttings discharge point is at 50 m water depth. However, Statoil internal Met Ocean group agreed with Peregrino asset that full water column current measurements were needed and an additional ADCP unit, a Nortek Continental was installed on the lander frame. This unit was not connected to computer and power system in the lander, but operated on separate batteries and stored data on an internal hard disc.

Field campaigns Totally, 4 field campaigns were planned during the three year PEMCA project, each lasting approximately six months. Table 2 presents an overview over these, including the main outcome of the campaigns with respect to data collection.

Table 2: Overview over field campaigns

Period Location Overall results May – Oct 2010 250 m S of WHPB Data collection failed after 3 weeks due to leak in GPT

container (-included main computer)

Jan – Jun 2011 1.4 km SW of WHPB Data collection for only 4 months due to high power consumption. Poor images, light sensor failed. Leaks and corrosion. Major service and upgrading performed

Nov 2011 – Apr 2012

1.4 km SW of WHPB Communication buoy deployed but weak link broke. Concept abandoned. Data collection successful, except light sensor failure and overexposure of parts of the images

Jul 2012 – Jan 2013 1.4 km SW of WHPB Data collection successful except light sensor

Field campaign 1

The first campaign was designed to start before drilling discharges commenced. The lander was deployed May 1st 2010, 250 meters south of well head platform B with the sonar system oriented to observe discharged drilling particulates. This was done to clarify to what extent the sonar system was able to track the discharge plume from the platform. For this campaign the lander was deployed as a standalone system without the communication buoy, since it was in a temporary position. Deployment was carried out by the vessel “7 Oceans”, which at the time was participating in the installation work at Peregrino.

Two major problems occurred during the first field campaign. Drilling operatioms, and thereby cuttings discharges, were delayed, and did not start until after the lander was retrieved for battery shift in October 2010. As a result no cuttings discharges were present in the field during the deployment period. The second problem occurred party because the lander was damaged during handling onboard the vessel. The cable and connector to the power supply unit was damaged during lifting. Repair was attempted prior to deployment and appeared successful, but the system probably short-circuited and drained the

6 SPE 168356

batteries for power after a few weeks. When the lander was retrieved, sea water was found both in the battery containers and the GPT container (computer container). Data was only gathered for about 4 weeks.

Field campaign 2

After repair and battery replacement the lander was again deployed in January 2011, this time from “Maersk Attender”, a crane vessel participating in the installation work at Peregrino. A position 1.4 km southwest of well head platvorm B, at the edge of the CA bank was chosen. The plan was to also deploy the communication buoy during this campaign, but this was postponed because the fiber cable between the buoy and the lander broke during the preparations for deployment. An unsuccessful attempt to repair was made onboard the vessel.

This time, the system worked better, but the batteries were drained after 4 months. The systems still had problems with corrosion of stainless steel connectors, and this led to higher energy consume than expected. The light sensor did not work during the campaign and the images were of very poor quality, mainly due to overexposure (high flash intensity). A large number of the acoustic files were ‘corrupted’ for reasons unknown, indicating a certain degree of instability in the set up and systems.

Some major changes were necessary based on the experience so far. A technical audit and risk assessment was carried out between Statoil subsea installation experts and METAS, where all aspects of the equipment and operation were reviewed. The lander was therefore sent to Sao Paulo for repair and service, the major adjustment was that all stainless steel connectors were replaced by titanium to avoid corrosion. A comprehensive set of tools and spare part was purchase to always be available during the field campaigns.

For the retrieval of the lander in campaign 2, Statoil supply vessel “Skandi Peregrino” was upgraded to host a work class ROV for the purpose of participating in the PEMCA project. The vessel and the ROV was hired from Fugro, responsible for the remaining campaigns.

Field campaign 3

During field campaign 3, from November 2011 – April 2012, the buoy and anchor were deployed. Because of the periodically very strong current in the field, the buoy came under very strong pressure and almost lay flat in the sea surface. This made the communication even from very close range unstable. After less than 24 hours the weak link between the buoy and the anchor broke, and the buoy had to be rescued by “Skandi Peregrino”. A possible reason for this can be interaction with fishing vessels using the buoy as mooring. There is, however, no clear evidence for this. Since a real-time solution was not essenial for the project and due to a cost/benefit consideration the buoy concept was abandoned for the PEMCA project after this.

Data collection on the 3rd campaign was more successful, all sensor systems except the light meter worked. The turbidity sensor failed after about 3 months and the image quality was still reduced due to overexposure or too strong flash. The camera house was also exposed to biofouling. After three months this started to impact the quality of the images. Except for these minor to moderate issues, all systems operated satisfactory.

Field campaign 4

Also the 4th and final campaign, July 2012 – January 2013, was successful, but the light sensor failed also this time. This in spite it had been returned to the manufacturer after the previous cruise, repaired and tested. Apart from this all systems worked well and collected data for the whole period. The images were this time of much better quality and biofouling did not seem to be a significant problem, probably since the sampling took place during the winter and spring season.

General learning from the field campaigns

One of the most challenging issues for the PEMCA project was the logistics and solutions related to vessel capacity and quality. When the project was initiated, “Skandi Peregrino” was planned to carry a work class ROV or at least work class ROV platform, but this was not installed. A ROV platform was installed on “Skandi Peregrino” in 2011 and thereafter used for deployment and retrieval of the equipment.

In general, the initial PEMCA ocean observatory system did not represent a good design with respect to robustness and handle-ability. Only after major improvements did the lander work properly and in combination with anchor, buoy and satellite it represents a system which is very challenging to handle in the field. As a result of the experiences from PEMCA, and other campaigns, has the lander design been changed considerabelly.

Utilizing seabed based observatory or a stand alone landers for environmental observations is not a new concept. For the oil industry, however, limited experience in this area is present, meaning that the suppliers of this technology are not familiar with

SPE 168356 7

working with offshore requirement’s and level of quality control needed. The PEMCA project has from this viewpoint contributed significantly to develop this experience for all involved parties. Sediment traps campaigns

Three sediment traps were purchased for the PEMCA project in order to monitor natural sedimentation and potential input from discharges of drill cuttings in the area. In general, this equipment is reliable and easy to handle. It can be deployed and retrieved without use of ROV or heavy lifting equipment. The traps are simply dropped from the surface attached to a weight; they sink calmly to the seabed and are retrieved by releasing the trap from the weight by an acoustic signal. Figure 3 shows a photo of a sediment trap under deployment.

Figure 3: Deployment of sediment trap Table 3 gives an overview over the sediment trap sampling campaigns. The sediment traps were deployed and retrieved by the supply vessels “Anita”, “Carolina” and “Skandi Peregrino”. Campaigns 1 and 2 were complete, while campaign 3 only included 1 trap. A technical problem led to that one trap could not be redeployed after campaign 2. In addition failure of the acoustic release system made one trap impossible to retrieve and deeply again for campaign 3. This could have been avoided by installing a double acoustic release system. This trap was later retrieved by an ROV. In addition, one trap was destroyed by the vessel during retrieval after campaign 4, so from the 5th campaign only two traps were available. Table 3: Overview over sediment trap campaigns

Campaign Period Trap 1 Trap 2 Trap 3 1 19.06.2010 - 21.08.2010 X X X 2 28.10.2010 - 16.01.2011 X X X 3 07.04.2011 - 26.06.2011 X 4 16.08.2011 - 17.11.2011 X X 5 26.11.2011 - 21.02.2012 X X 6 22.03.2012 - 17.06.2012 X X 7 01.07.2012 - 01.10.2012 X X

The funnels for sample collection on the sediment traps were located 10 m above the seabed. Ideally, a location closer to the seabed would be preferable since the results are to be compared with sedimentation on the seabed. Since the sediment trap has positive buoyancy in the water column it is sensitive to strong currents. This may influence its ability to sample realistically since the angle of the funnel opening may tilt away from the current direction. The present solution is therefore not optimal for an area with strong currents such as the Peregrino field. Sampling locations

As described above, the lander was during the first campaign located 250 m south of the Peregrino well head platform B, while it for the following three campaigns was located approximately 1.5 km southwest of the platform. The sediment traps were placed in three different positions, 500 m southwest of well head platform B, close to the lander and approximately 4 km southwest of platform B. The latter position was expected to represent an area outside the potential influence area of cuttings

8 SPE 168356

discharges. When the number of traps was reduced to two, as described above, the two latter positions were maintained. Figure 4 shows the positions of the sampling equipment in an infrastructure and seabed structural map of the Peregrino field.

Figure 4: The Peregrino infrastructure and location of the lander and the three sediment traps and a map showing the area with high density of calcareous algae rhodoliths (pink dotted area) Overall results of the monitoring campaigns

The present paper focuses on the development, implementation and qualification of monitoring technology for the Peregrino seabed environment. However, to complete the picture reflecting the outcome of the PEMCA project, this section addresses the results of the observations and measurements made through the sampling program of the project. Eide et al (2013) and Nilssen et al (2013) have described the analysis of the sampling campaigns and the risk assessment, respectively. Eide et al. (2013) performed correlation studies and multivariate analysis of the sampling results from the lander and sediment trap campaigns, the discharge logs from the two well head platforms and the dispersion of drill cuttings in the Peregrino seabed modelled by the DREAM model. The intention with the corelation study was to document the distribution of drill cuttings in the area and to identify any potential impact of this on the CA community. Also, the goal of these analyses was to identify which of the sensor and sampling systems contributes significantly to the understanding and documentation of fate and effects of drill cuttings discharges in the area. Eide et al. (2013) showed that no evidence of drill cuttings exposure was present for the sampling sites in the area of the fields hosting the CA habitat, and thus, no effects were observed. The results from the modelling and sampling were in agreement on this point. However, the study by Nilssen et al. indicated the presence of a significant environmental risk of impact to the CA in the outer boundaries of the CA bank, also in the area where the lander was located, The reason for this disagreement lies in the conservative nature of the DREAM model, only reflecting the build-up of drill cuttings in the sediment over the whole discharge period. The sampling campaigns, on the other hand, reflect a more realistic dynamic situation where natural processes of sediment turbation are evident. It was suggested by Nilssen et al. (2013) that the model should be developed further to include this aspect by introducing a time variable sediment reworking function. From the image analysis of photos collected by the lander, Elbe et al. (2013) has determined an average movement frequency of CA rhodiliths. This may be utilised in the risk assessment model to account for the self-cleaning ability of the CA. A new monitoring approach for the CA community

Technology approach

Based on the experience from use of ocean observatory technology in several environmental monitoring programs over the past years, with Peregrino as the major contributor, a new technical solution for the purpose of seabed based monitoring platforms, or landers, has been developed. The solution attempts to meet and solve the challenges met in the PEMCA project, related to robustness, reliability and logistics. A major priority was to decrease the weight and the size of the sensor platform to increase flexibility and ease the hanfling during field operations.Another major objective was to enable deployment and retrieval without the need for work class ROV. Figure 5 shows an ilustration of a stand-alone lander designed and built by METAS, meeting these criteria. The unit has a total weight of 800 kg and is tailor suited for environmental monitoring in the Peregrino field, with a built-in sediment trap and the sensor package described below. This 2nd generation Peregrino lander is designed as a drop unit with inflectable flotation cells that can be activated by a remote acoustic signal. Alternatively, it can be built with positive buoyancy and connected to drop weights that can be released acoustically to make it surface again. It can also be deployed and retrieved by a special carrier frame with small electrical trusters for positioning. This stand-alone unit can also be integrated in an ocean observatory system and has been tested and qualified by Statoil. It has already found several applications in environmental monitoring programs on the Norwegian shelf.

SPE 168356 9

Figure 5. The ‘X-frame’, a new technical design and sollutiom for astand-alone lander or for integration in an ocean observatory Alternative monitoring approach for the CA habitat in Peregrino

Based on the results from the PEMCA project and the development of a new lander structure, an alternative approach to environmental monitoring of drill cuttings discharged in the Peregrino field has been designed. The approach has a special focus on the potential impact of the discharges to calcareous algae habitat, but is also aimed to replace the traditional sediment monitoring regime for the field. The program includes two main elements;

• Field sampling campaigns • Cuttings dispersion and risk assessment modeling

By combining these elements the program aims to improve understanding, management and documentation of the fate and effect of drill cuttings discharges to the Peregrino seabed. The potential for improvement is significant based on the results from PEMCA and the first two years of traditional environmental monitoring. Field sampling will be carried out by use of two separate stand-alone landers, designed as described above, so they can be operated from any of Statoil’s supply vessels at Peregrino. The landers will hold the following sensor and sampling systems, based on the results from the PEMCA field campaigns, as summarized by Eide et al. (2013) and the observations from the image analyses described by Elbers and Sumida (2013).

• Sediment trap for analysis of particle flux, organic carbon, hydrocarbons, metals and isotopes These measurements will enable documentation of sediment fluxes and contribution from natural sources vs cuttings discharges. The results can also be used directly for validation of dispersion modeling

• Turbidity sensor Will be used for correlation of particles in the near sediment water column with the discharge log, sediment trap results, current and dispersion model results

• Bottom current profiler Will be used as input to the modeling and for correlation with sediment traps results, turbidity and discharge log

• Water column current profiler (ADCP) if not obtained from WHB As above

• 2 time lap cameras Based on the laboratory studies and image analyses from PEMCA, field images will be used to identify impact of sedimentation (natural and drill cuttings) to the CA. Further, the images will be used to valisate the risk and dispersion modeling results by applying colour change and movement as variables.

• Battery power and data storage system The ambition of the proposed monitoring program is to combine field campaigns with dispersion and risk/impact modeling, and thereby present an overview over the environmental status of the Peregrino seabed environment. The results from the model will be visualized by a map over the field where areas are colored related to the risk of impact from the discharges as shown Nilssen et al. (2013). Distribution of drill cuttings will also be visualized by the model. The field sampling campaigns will partly provide input to the model, partly serve to validate the model and also stand alone to document the environmental status of selected locations in the field. In addition, the monitoring program will make it possible to distinguish between natural sedimentation and contamination due to drilling activities. The principle is to run 3 – 9 weeks sampling campaigns where the landers are placed on different locations of the field to give a best possible fundament for modeling and

10 SPE 168356

documentation. As the quality of the modeled results improves and are better documented through field validation, the frequency of the sampling campaigns will be reduced. Table 3 outlines how the alternative monitoring program can be carried out for the remaining drilling period of Peregrino (updated plan from May 2013) with respect to campaign frequency and modeling. To also cover the need for monitoring the close area of the platforms, conventional grab/box corer sampling campaugns are included at the end of the drilling activities for each platform. Table 4: Proposed campaign frequency for the Peregrino monitoring program, based on present (May 2013) drilling program

Activity / Year Frequency 2014 2015 2016 2017 2018 2019 2020

Field sampling campaigns 2 2 1 1 1 1 1 Modelling 1 1 1 1 1 1 1 Sediment sampling campaigns 1 1

Cost benefit evaluation

To generate a basis for evaluation of cost for the proposed monitoring program, a request was presented to four potential suppliers of a modified lander system. The suppliers were asked to independently propose a design and estimate production and operational costs for building two lander systems as described above, and make these available for Statoil Brazil. Both local Brazilian and international companies were included in the survey. Re-use of the sensor systems from the PEMCA project was not considered in the request, but it was assumed that the PEMCA sediment traps can be reused as a part of the lander system. The estimates from the four potential suppliers varied significantly, but an average of the three lowest estimates was applied for the overall program cost estimation. Costs related to field work, sample analyses, modelling and reporting were estimate based on experience from the PEMCA project and other environmental monitoring programs conducted in Brazil by local suppliers. Vessel costs assumed the use of Statoil hired supply vessels, using daily rates as basis for the estimate. By summarizing the investment costs and the costs related to field campaigns, modeling, sample analyses and reporting for the planned drilling period in the Peregrino field (Table 4), a total expected cost of the alternative monitoring program can be established. This figure can be directly compared to the expected cost of the traditional monitoring program required by Brazilian pollution control authorities (IBAMA) and presently implemented in the Peregrino field. The expected costs were estimated to 2.2 and 2.5 million USD, respectively. These figures are of the same order of magnitude, in fact the difference is relatively low. However, over the remaining planned period of drilling the alternative monitoring approach will accumulate to a lower total cost than the on-going conventional program. Because of the up-front investment in equipment the alternative approach will be more expensive than the on-going for the initial four years, but after this period the total cost will be lower for the PEMCA based alternative. If the drilling program is extended, which is not unlikely, this difference will increase. In conclusion, the results indicate that a program using the technology developed can be more cost efficient than the current sediment sampling methodology demanded by Brazilian authorities. In addition to the expected cost reduction, the alternative program represents a clear improvement in quality of the information obtained through the environmental monitoring campaigns, with respect to documenting the presence or absence of impact of drill cuttings discharges to the CA community. This is simply not possible with the on-going conventional monitoring program. This, in turn, will enable improved environmental management and ability to document the environmental performance of the drilling operation. The alternative approach may also represent a significant step in the direction of better control and environmental management for other areas where special focus is needed on sensitive seabed habitats. Further use and thereby increased experience with this type of environmental monitoring may also contribute to cost reductions in the equipment applied. A general criticism towards environmental monitoring by the traditional grab sampling approach is that the methodology itself poses certain harm to the seabed environment, by disturbing the habitat when removing samples of this for further analysis. As shown during the baseline environmental survey, the only efficient way to sample CA rodoliths is to dredge the seabed surface. This approach to monitoring will pose significant impact on the habitat, since relatively large areas will be affected. The monitoring approach presented in the present paper eliminates this problem, since it is based on passive observation of the seabed environment, with minimum collection of physical samples, e.g. insignificant disturbance on the ambient environment

SPE 168356 11

Aknowledgements

The authors wish to thank Agência Nacional de Petróleo Gás Natural e Biocombustíveis (ANP) in Rio de Janeiro, Brazil, for accepting the PEMCA Project under the Federal Participation Agreement (FPE). Further, the Institute of Oceanography of the University of Sao Paulo, Brazil and the Institute of Marine Research, Bergen, Norway, played an active and important role in the field work, sample analyses and data interpretation. Finally, this project was made possible by the efforts and support by the Statoil Brazil Field Development department, the Peregrino Operational Department and the Statoil Brazil Logistics Department. References Climate and Pollution Agency (2011). The petroleum sector on the Norwegian Continental Shelf; Guidelines for offshore environmental monitoring, TA 2849 (http://www.miljodirektoratet.no/old/klif/publikasjoner/2849/ta2849.pdf) Eide, I., Nilssen, I., Westad, F. and Johnsen, S. (2013)- Multivariate data analysis in environmental monitoring of the Peregrino oilfield. In prep Elbers, K.L. and Sumida, P.Y.G. (2013). Long-term monitoring through imaging: crustose coralline algae mobility at Peregrino oil field observatory. In prep. ESONET (2013) - European Seas Observatory NETwork. Available at http://www.esonet-emso.org/ Figueiredo, M.A.O., Eide, I., Coutinho, R., Reynier, M., Villas-Bôas, A.B., Tâmega, F.T.S., Ferreira,, C.G.W, Nilssen, I. and Johnsen, S. (2013) The effect of drilling particulates on a deep water rhodolith bed; Experimental work and modelling using a mesocosm flow through system. In prep Frost, T.K, Myrhaug, J.L., Ditlevsen, M.K. and Rye, H.(2014). Environmental Monitoring and Modeling of Drilling Discharges at a Location with Vulnerable Seabed Fauna: Comparison between Field Measurements and Model Simulations. SPE Error! Reference source not found., PE international Conference on Health, Safety and Environment in Oil and Gas Exploration and Production at Lobg Beach, XA, USA. Godø, O.R., Klumgsøyr, J. and Meyer, S. (2013). Real time observation system for monitoring environmental impact on marine ecosystems from oil drilling operations. In prep. IBAMA, 2010, Drilling Permit Nº962/201 and for the Production Permit º1016/2011”. International Association of Oil & Gas Producers (2012) Offshore environmental monitoring for the oil & gas industry Report no: 457 http://www.ogp.org.uk/pubs/457.pdf) Johnsen, S., Frost, T.K., Hjelsvold M., Utvik T.R. (2000). The Environmental Impact Factor—A proposed tool for produced water impact reduction, management and regulation. SPE paper 61178. In: SPE International Conference on Health, Safety and Environment in Oil and Gas Exploration and Production; 26–28 June 2000; Stavanger, Norway. Richardson (TX): Society of Petroleum Engineers. NEPTUNE (2013) - North-East Pacific Time-Series Underwater Networked Experiments. Available at http://www.neptunecanada.ca/ Nilssen, I. and Johnsen, S. (2008) Holistic Environmental Management of Discharges From the Oil and Gas Industry Combining Quantitative Risk Assessment and Environmental Monitoring. SPE-111586-MS, SPE International Conference on Health, Safety, and Environment in Oil and Gas Exploration and Production, Nice France ProOceano (2013) Drilling Discharge Modeling; Peregrino Field, Campos Basin. Report /PRO_STA_1204 Salgado, C.S, Johnsen, S., Gomes, S., Torkelsen, T. (2010) Real-time environmental monitoring of discharges of drill cuttings in the peregrine field (Campos basin, Brazil). SPE 127175. SPE international Conference on Health, Safety and Environment in Oil and Gas Exploration and Production, Rio de Janeiro, Brazil Singsaas, I., Rye, H., Frost, T.K., Smit, M.G.D., Garpestad, E., Skare, I., Bakke, K., Veiga, L.F., Buffagne, M., Follum, O-A., Johnsen, S., Moltu, U-E. and Reed, M. (2008.) “Development of a Risk-Based Environmental Management Tool for Drilling Discharges. Summary of a Four-Year Project”. Integrated Environmental Assessment and Management 4 (2), pp. 171-176 Smit, M.G.D (2009). Quantifying the response of marine species to physiochemical stressors at different levels of biological organisation. PhD thesis Steller, D.L.; Riosmena-Rodriguez, R.; Foster, M.S. & Roberts, C. (2003). Rhodolith bed diversity in the Gulf of California: the importance of rhodolith structure and consequences of anthropogenic disturbances. Aquatic Conservation Marine Freshwater Ecosystems, 13:5-20.

12 SPE 168356

Steller, D.L.; Foster, M.S. & Riosmena-Rodriguez, R. (2009). Living rhodolith bed ecosystems in the Gulf of California. In: Jhonson, M.E. & Ledesma-Vázquez, J. (Eds) Atlas of Coastal ecosystems in the Gulf of California: Past and Present. University of Arizona Press, chapter 6.

Tâmega, F.T.S., Spotorno-Oliveira, P. Figueiredo, M.A.O. (2013). Catalogue of the benthic marine life from Peregrino oil field, Campos Basin, Brazil. In prep.

Paper IV

Figueiredo, M. A. O., Eide, I., Reynier, M., Villas-Bôas, A. B., Tâmega, F. T. S., Ferreira, C. G., Nilssen, I., Coutinho, R., Johnsen, S., 2015. The effect of sediment mimicking drill

cuttings on deep water rhodoliths in a flow-through system: Experimental work and modeling. Marine Pollution Bulletin Vol 95, pp 81-88,

doi: http://dx.doi.org/10.1016/j.marpolbul.2015.04.040

The effect of sediment mimicking drill cuttings on deep water rhodolithsin a flow-through system: Experimental work and modeling

Marcia A.O. Figueiredo a,b,c, Ingvar Eide d,⇑, Marcia Reynier e, Alexandre B. Villas-Bôas a,c,Frederico T.S. Tâmega b,c, Carlos Gustavo Ferreira c, Ingunn Nilssen d,f, Ricardo Coutinho c, Ståle Johnsen d

a Instituto de Pesquisa Jardim Botânico do Rio de Janeiro, Rua Pacheco Leão 915, Jardim Botânico 22460-030, Rio de Janeiro, RJ, Brazilb Instituto Biodiversidade Marinha, Avenida Ayrton Senna 250, Sala 208, Barra da Tijuca, 22793-000, Rio de Janeiro, RJ, Brazilc Instituto de Estudos do Mar Almirante Paulo Moreira, Departamento de Oceanografia, Divisão de Biotecnologia Marinha, Rua Kioto 253, 28930-000, Arraial do Cabo, RJ, Brazild Statoil ASA, Research, Development and Innovation, N-7005 Trondheim, Norwaye LABTOX – Laboratório de Análise Ambiental Ltda., Avenida Carlos Chagas Filho, 791, Cidade Universitária, Ilha do Fundão, Rio de Janeiro, RJ 21941-904, Brazilf Trondhjem Biological Station, Department of Biology, Norwegian University of Science and Technology, N-7491 Trondheim, Norway

Article history:Received 4 February 2015Revised 17 April 2015Accepted 18 April 2015Available online 29 April 2015

Keywords:Environmental impactSediment loadPhotosynthesisCorrelation analysisCalcareous algaeMultivariate data analysis

a b s t r a c t

The impact of sediment coverage on two rhodolith-forming calcareous algae species collected at 100 mwater depth off the coast of Brazil was studied in an experimental flow-through system. Natural sedi-ment mimicking drill cuttings with respect to size distribution was used. Sediment coverage and photo-synthetic efficiency (maximum quantum yield of charge separation in photosystem II, /PSIImax) weremeasured as functions of light intensity, flow rate and added amount of sediment once a week for nineweeks. Statistical experimental design and multivariate data analysis provided statistically significantregression models which subsequently were used to establish exposure–response relationship forphotosynthetic efficiency as function of sediment coverage. For example, at 70% sediment coverage thephotosynthetic efficiency was reduced 50% after 1–2 weeks of exposure, most likely due to reducedgas exchange. The exposure–response relationship can be used to establish threshold levels and impactcategories for environmental monitoring.

1. Introduction

Rhodolith beds are benthic communities dominated by calcare-ous red algae (Rhodophyta: Corallinales and Sporolithales) whichbuild calcified nodules (rhodoliths) (Bosence, 1983; Foster, 2001).Rhodoliths with multi spherical structures are classified within themorphological group ‘‘boxwork’’ (Basso, 1998). Due to the structure,the rhodoliths are inhabited by other organisms increasing the bio-diversityof soft-bottomcommunities (Bordehore et al., 2003; Stelleret al., 2003; Figueiredo et al., 2007; Harvey and Bird, 2008; Sciberraset al., 2009) and therefore they play an important ecological role incoastal areas (Hall-Spencer, 1998; Ávila and Riosmena-Rodriguez,2009; Steller et al., 2009; Riosmena-Rodriguez et al., 2010).

The largest occurrence of rhodolith beds in the world arefound in the southwest Atlantic along most of the Brazilian con-tinental shelf (Kempf, 1970; Foster, 2001; Lavrado, 2006; Amado-

Filho et al., 2012). Calcareous algae are found down to approxi-mately 250 m depth (Littler et al., 1986, 1991). These communi-ties may be disturbed and have a potential to get buried due tonatural sedimentation and anthropogenic activities such as fish-trawling and mining (Nelson, 2009). Rhodolith beds are increas-ingly exposed to discharges of drill cuttings from oil and gasactivities, for instance in the Gulf of Mexico and on theBrazilian shelf (Davies et al., 2007). It has been demonstrated thatfine sediments (<250 lm grain size) may reduce the photosyn-thetic activity of coralline algae to a larger extent than coarse cal-careous sediments from shallow nutrient-rich estuarine andcoastal environment (Wilson et al., 2004; Harrington et al.,2005; Riul et al., 2008) as well as deep rhodolith soft-bottoms(Villas-Bôas et al., 2014). The ability to withstand sedimentationvaries greatly among calcareous algae species (Harrington et al.,2005; Villas-Bôas et al., 2014). Different species may have differ-ent survival strategy towards sedimentation, for example slowgrowth and low metabolic demand, shooting branches abovethallus surface, translocation of photosynthesis through cell-fu-sions from healthy to damaged area of thallus, or tilting(Steneck, 1986; Dethier and Steneck, 2001).

Abbreviations: PLS, partial least squares; /PSIImax, photosynthetic efficiency(maximum quantum yield of charge separation in photosystem II).⇑ Corresponding author. Mobile: +47 90997296.

E-mail address: ieide@statoil.com (I. Eide).

Marine Pollution Bulletin 95 (2015) 81–88

Marine Pollution Bulletin

journal homepage: www.elsevier .com/locate /marpolbul

The purpose of the present study was to measure the possibleimpact of sediment coverage on two rhodolith-forming calcareousalgae species collected at approximately 100 m water depth at thePeregrino oil field off the coast of Brazil.

Due to the slow growth of the calcareous algae (Adey andMacintyre, 1973; Foster, 2001), photosynthetic efficiency (maxi-mum quantum yield of charge separation in photosystem II,/PSIImax) was selected as endpoint. A decrease in photosyntheticefficiency is generally considered to be a response to environmen-tal stress (Genty et al., 1989; Wilson et al., 2004; Evertsen andJohnsen, 2009).

Natural sediment mimicking drill cuttings with respect to sizedistribution was used in the study. Although water based drillingfluids consist of water soluble and non- or low-toxic components(Bakke et al., 2013) some laboratory studies on single species(Larsson and Purser, 2011; Larsson et al., 2013) and soft bottomseabed communities (Trannum et al., 2010; Bakke et al., 2013)have shown larger effects of drill cuttings (water based) comparedto natural sediments. However, initial short- and long-term toxic-ity testing with the algae species used in the present work verifiedthat there was no significant difference in photosyntheticefficiency when comparing sediment and drill cuttings fromPeregrino (Reynier et al., in press). It was therefore assumed thatthis sediment was considered relevant as a surrogate for the drillcuttings from Peregrino. Besides, discharge of drill cuttings fromthe laboratory experiments would not be permitted.

Sediment coverage and photosynthetic efficiency were studiedas functions of light intensity, flow rate and added amount of sed-iment for nine weeks in an experimental flow-through systemespecially built for the experiments. Statistical experimentaldesign and multivariate data analysis and statistics have been use-ful in previous toxicological and environmental studies (Eide andJohnsen, 1998; Søfteland et al., 2009) and were used to obtain mul-tivariate regression models. These models were subsequently usedto establish exposure–response relationship for photosyntheticefficiency as function of sediment coverage. In addition, colorchanges indicating stress or mortality of the calcareous algae wererecorded.

2. Material and methods

2.1. Sampling site

The organisms used in the present study were collected at thePeregrino oil field 80 km off the coast of Brazil, in the CamposBasin area. Drilling started in November 2010 and the field camein production in May 2011. The water depth in the area is approx-imately 100 m. The seabed sediments consist in general of sandand silt with a hard surface (Salgado et al., 2010). In order to avoidimpact of drilling activities to the calcareous algae habitat, the dril-ling platform and the discharge were located approximately1.5 km away from the habitat.

2.2. Species and their environment

A field assessment was carried out at the Peregrino oil field toidentify rhodolith-forming species (Tâmega et al., 2013).Rhodoliths and the associated fauna were collected by dredging22 sampling sites at 94–103 m depth in June and November2010 and April 2011. For specimen identification, a selected num-ber of the calcareous algae samples were preserved in 10% formalinsolution, and associated fauna samples were preserved with mag-nesium chloride (8%) for 30 min and then fixed in P70% alcohol.The main organisms identified were the long lived encrustingcalcareous algae and bryozoans.

The choice of species for the experiments was based on theirecological importance and abundance at the Peregrino oil field,their resistance to sampling, and survival under laboratory condi-tions. For species identification the calcareous algae weresectioned using histological techniques for optical microscopyaccording to Moura et al. (1997) and identification followeddescriptions of Woelkerling (1988). Two of the most abundantspecies of encrusting calcareous algae covering rhodoliths werechosen. Organisms for the exposure studies were kept alive in aflow-through system under controlled levels of temperature andlight intensity (15 �C and 15 lmol m�2 s�1, respectively) similarto natural conditions.

2.3. The calcareous algae

The two dominant species of calcareous algae, Mesophyllumengelhartii (Foslie) Adey and Lithothamnion sp. (Figueiredo et al.,2012), were chosen for the exposure studies. Rhodoliths free of epi-bionts and with the most common size (40–60 mm in averagediameter) and shape (compact-bladed to bladed) were selectedand tagged with numbers glued by epoxy putty (TUBOLIT MEN)to their surface. The calcareous algae used in the exposure studieswere healthy specimens, purple to pinkish in surface color withbleached spots.

2.4. Sediment

Natural sediment mimicking drill cuttings with respect to sizedistribution was used in the study. Composition of the sedimentgrain size was based on sieving analysis of drill cuttings fromone of the wells at the Peregrino oil field (well A-18).

The dominant drill cutting particle size is medium to fine(63–250 lm). Natural sediment was sieved and the fine and coarsefractions were mixed at a ratio of 3:1 in order to mimic drillingparticles. Initial toxicity testing verified that this natural sedimentand the drilling particles from Peregrino had comparable effects onthe two calcareous algae species (Reynier et al., in press).

2.5. The flow-through system

A novel large-scale flow-through system was designed to testthe effects of particles on selected species as a function of addedsediment amount, light intensity, water flow rate and exposuretime. The entire flow-through system consisted of four units witheight loops in each unit placed in a rack system. Each loop con-sisted of an exposure chamber where the organisms were posi-tioned. The water movement was driven by a propeller asillustrated in Fig. 1. Water flow could be varied within each ofthe four units. Light intensity and added amount of sediment couldbe varied independently within each loop. Each exposure chamberwas optically shielded from its neighbors and was illuminated bythree blue diodes (peak wavelength 465 nm) of the Lambertiantype (Seoul Semiconductor type B42182). The entire experimentalsystem of four units, each with eight loops, was placed inside atemperature regulated chamber with the same temperature asthe inlet water to avoid temperature fluctuations. A cooling systemwith the capacity to cool 2 L min�1 of sea water to 15 �C was estab-lished outside the cooled room. In addition to a biological filter thewater was mechanically filtered using 25 lm and 5 lm filters inseries. A 20 L seawater reservoir for the chilled water was placedinside the cooled room.

2.6. Statistical experimental design

Photosynthetic efficiency and sediment coverage were studiedas function of the three design variables light intensity, flow rate

82 M.A.O. Figueiredo et al. /Marine Pollution Bulletin 95 (2015) 81–88

and added amount of sediment in addition to exposure time. Inorder to perform the experiments as efficient as possible, and toenable regression modeling to describe possible non-linear rela-tionships and interactions, the predictor variables were combinedas a full factorial design (Central Composite Face, CCF) with centerpoints as shown in Table 1 (Box et al., 1978). Replicates (same con-ditions in parallel loops) are indicated in the table and the legend.

The light levels were 3, 6.6, and 10 lmol m�2 s�1, the latter cor-responding to light intensity at 100 mwater depth at the Peregrinofield where the calcareous algae were collected. The different flowrates were 0.04, 0.07, and 0.09 m s�1, higher flow rates caused toomuch turbulence in the flow-through system. The added amountsof sediment were 600, 900 and 1200 g per chamber. Introductorystudies had shown that the range 600–1200 g resulted in 0–100%coverage.

The full factorial design gave 15 combinations of light intensity,flow rate and sediment amount (Table 1). In addition there were anumber of replicates (same conditions in parallel loops) and paral-lel rhodoliths within each loop (exposure chamber). In the firstexperimental series there were six parallels (rhodoliths) withineach loop, one loop for each of the 15 combinations plus six repli-cates, giving a total of 21 loops and 126 individual rhodoliths withthe calcareous algae M. engelhartii plus nine controls. In the secondexperimental series there were four parallels of each rhodolithwith either M. engelhartii or Lithothamnion sp. in each loop, one

loop for each of the 15 combinations plus 13 replicates, giving atotal of 28 loops and 112 individual rhodoliths with each of thealgae species. In addition there were 16 controls of each species.Three rack units were used in the first trial, and all four were usedin the second trial; the intermediate flow rate was used in two ofthe rack units resulting in a larger number of replicates in thesecond trial.

The controls were placed in loops distributed between thedifferent units (and hence different flow rates). The controls wereneither exposed to light nor sediment and were used to evaluatehow long the algae could live in darkness without deleteriouseffects (Wilson et al., 2004).

Photosynthetic efficiency was measured prior to the addition ofsediment, and both photosynthetic efficiency and sediment cover-age were measured each week for nine weeks implying that alsoexposure time was a predictor variable in addition to flow, lightand sediment amount.

2.7. Photosynthetic efficiency measurement and color evaluation ofcalcareous algae

Once a week during nine weeks (T0–T9), except the first week ofthe first trial due to technical problems, all rhodoliths were takenout of the flow-through system to measure photosynthetic effi-ciency as maximum quantum yield of charge separation in photo-system II (/PSIImax) in dark acclimated algae in vivo. Eachrhodolith was placed in sterilized sea water at 15 �C and carefullybrushed to remove biofilm and sediments. Prior to the measure-ments the rhodoliths were transferred to a box for dark acclimationfor 20 min. Three readings of photosynthetic efficiency were takenrandomly from each rhodolith in complete darkness. The measure-ments were done using a Diving-PAM fluorometer andWin-Controlsoftware (Walz GmbH, Effeltrich. Germany). A blue measuring lightwith a wavelength of 470 nm was used. The saturating pulse was0.6 s and saturation intensity 150–200 lmol m�2 s�1, gain3 lmol m�2 s�1 and dumping 3 lmol m�2 s�1. To ensure that thereadings always were taken at a distance of 3 mm to the algae sur-face, a plastic spacer were attached to the fiber optic sensor (2 mmdiameter) (Harrington et al., 2005). The average of the three mea-surements made on each rhodolith was used in the data analysis.After the photosynthetic efficiency readings all rhodoliths werereturned to the exposure chambers.

F0, the ground fluorescence, and Fm, the maximum fluorescencewere measured. The variable fluorescence was calculated asFv = (Fm � F0). The photosynthetic efficiency was calculated accord-ing to the following equation: /PSIImax = Fv/Fm = (Fm � F0)/Fm.

Fig. 1. The loop design.

Table 1Matrix for the central composite face design and the predictor variables flow rate(m s�1), light intensity (lmol m�2 s�1), and added sediment amount (g). Time(0–9 weeks) is the fourth predictor variable. Samples 1, 6 (centerpoint), 10 and 15were duplicated or triplicated (bolded). Controls were distributed among the differentflow rates (no light or sediment).

Flow Light Sediment

1 0.04 6.6 9002 0.04 3 12003 0.04 3 6004 0.04 10 12005 0.04 10 6006 0.07 6.6 9007 0.07 10 9008 0.07 6.6 12009 0.07 6.6 600

10 0.07 3 90011 0.09 10 60012 0.09 10 120013 0.09 3 60014 0.09 3 120015 0.09 6.6 900

M.A.O. Figueiredo et al. /Marine Pollution Bulletin 95 (2015) 81–88 83

In addition to the photosynthetic efficiency measurement,changes in color tone and intensity of the calcareous algae wereevaluated by visual inspection according to a standard color chart(Figueiredo et al., 2000; Villas-Bôas et al., 2014). Furthermore,health status was evaluated according to a classification scheme(Figueiredo et al., 2000). Data on photosynthetic efficiency corre-sponding to color tone and intensity were organized in a table inorder to relate these parameters to health status and stress levels(Table 4).

2.8. Sediment coverage measurement

Sediment coverage was calculated from photos taken of eachrhodolith at the beginning of the experiment and each week fornine weeks. To avoid disturbing the sediment, seawater was slowlydrained off before taking the rhodoliths out for measurements.Photos were taken at a standard distance with a measuring scaleclose to each rhodolith. To ensure a constant distance from the rho-doliths, a digital camera (Canon G12) was fixed on a frame.Sediment coverage (%) was calculated by outlining the border ofeach rhodolith image and estimating the covered area by usingthe software ImageJ (National Institutes of Health, Bethesda,Maryland). After the measurements were completed, the rhodo-liths were brought back to their original position in the exposurechambers and sediment was added to replace the amount thatwas lost when draining off the water.

2.9. Multivariate data analysis

Multivariate regression was performed with partial leastsquares (PLS) (Martens and Næs, 1993; Wold et al., 1983) usingSimca P+ 11.5 (Umetrics, Umeå, Sweden). The purpose was to cor-relate the predictor variables (X matrix) to the measured responses(Y matrix). Prior to the PLS-regression, the data were scaled to unitvariance and mean centered. The PLS models were validated withrespect to explained variance, goodness of fit (R2) and prediction(shown as Q2), the latter obtained after cross-validation (Wold,1978).

Introductory, the PLS-regression was carried out for each trialand each algal species separately and with all individual observa-tions in order to verify repeatability in the measurements and toidentify possible outliers (not shown). Subsequently, averages ofparallels (six in each loop in the first trial, and four for each algalspecies in the second trial) were calculated. However, still therewere replicates (same conditions in parallel loops) as describedin Section 2.6 and Table 1. The averages were used in the nextPLS analyses which were carried out for each trial and each algalspecies separately and also on the combined data from both trialsand both algal species. The combined data matrix with averageshad 21 rows (observations) from the first experimental series withMesophyllum, and from the second experimental series, 28 rowswith Mesophyllum and 28 with Lithothamnion resulting in a totalof 77 rows for each sampling time. The X-matrix had four columns,one per predictor variable (light, flow, sediment amount and time),and the Y-matrix had one column per response variable (photosyn-thetic efficiency and sediment coverage).

The PLS-model was subsequently used to predict photosyn-thetic efficiency and sediment coverage at fixed levels of lightintensity, flow rate and time in order to provide data for expo-sure–response curves for photosynthetic efficiency as function ofsediment coverage at different scenarios.

The controls were treated separately since they were neitherexposed to light nor sediment; these data were analyzed in twodifferent ways: (1) PLS regression of photosynthetic efficiency ofcontrols as function of flow and time (light and sediment amountwere constant at zero). (2) Linear regression of average values for

photosynthetic efficiency, one for each algae, trial and flow, asfunction of time.

3. Results

3.1. Controls

Fig. 2 shows the result of the linear regression using averagevalues for photosynthetic efficiency in controls as function of timefor each algae, trial and flow rate. The regression line has a corre-lation coefficient of 0.40. There is an indication of larger variabilityat T9. Performing liner regression from T0 to T8 gives a regressionline slightly less steeper with a correlation coefficient of 0.56.Although the correlation is not strong, the data indicate a slightdecrease in photosynthetic efficiency in controls from 0.49 to 0.43.

Creating nine regression lines (not shown), one for each trial,species and flow rate, gives correlation coefficients of 0.1–0.6.There is no systematic variation of photosynthetic efficiency ofcontrols in relation to flow rate, species or trial. This is also con-firmed by PLS-regression (not shown) of photosynthetic efficiencyof controls as function of flow rate and time (light and sedimentconstant at zero). The PLS-models had goodness of fit and predic-tion of only 0.28 and 0.23, respectively, implying weak correlationand low predictive property.

3.2. PLS modeling of photosynthetic efficiency and sediment coverageas function of predictor variables

Table 2 summarizes the results of the PLS-regression of photo-synthetic efficiency and sediment coverage as function of the four

-1 1 3 5 7 9Time (weeks)

Fig. 2. Photosynthetic efficiency (/PSIImax) in controls from T0 to T9 for the first trialwith Mesophyllum (diamonds) and for the second trial with Mesophyllum (dots) andLithothamnion (squares). Controls were distributed among the different flow rates0.04 (blue), 0.07 (red) and 0.09 m s�1 (green). Lower regression line T0–T9(R2 = 0.40). Upper regression line T0–T8 (R2 = 0.56).

Table 2Goodness of fit (R2) and goodness of prediction (Q2) for photosynthetic efficiency andsediment coverage with flow, light, added sediment amount and time as variables.Data from the first trial with Mesophyllum and the second trial with Mesophyllum andLithothamnion were analyzed separately and also merged into one matrix andanalyzed combined.

Photosyntheticefficiency

Sedimentcoverage

R2 Q2 R2 Q2

Mesophyllum 1st trial 0.95 0.89 0.93 0.88Mesophyllum 2nd trial 0.91 0.86 0.91 0.87Lithothamnion 2nd trial 0.91 0.90 0.89 0.88Combined 0.89 0.89 0.90 0.90

predictor variables. Data from the first trial with Mesophyllum andthe second trial with Mesophyllum and Lithothamnion were ana-lyzed separately and also merged into one matrix and analyzedcombined. The regression models are very good in terms of good-ness of fit and prediction. The correlation coefficients (R2) are gen-erally slightly better when analyzing each trial and speciesseparately. However, the prediction properties (Q2) are generallybetter when analyzing the data combined. Using all data in thesame regression model makes the model more generic and robust.

The best model was achieved with one square term in thepolynome that describes photosynthetic efficiency as a non-linearfunction of added amount of sediment, light intensity, flow rateand time. Table 3 shows the PLS regression coefficients and corre-sponding 95% confidence limits for the four different models. Thetable illustrate that the polynomes describing photosynthetic effi-ciency and sediment coverage as functions of the predictor vari-ables are very similar, confirming that the combined modelsummarizes all data from both trials and species. Fig. 3 illustratesthe regression coefficients for the combined model (same data asin the five last rows in Table 3) and clearly show that addedamount of sediment is the major factor reducing photosynthetic

efficiency and increasing sediment coverage. Flow rate, light inten-sity and time had minor impact in the flow-through system withinthe experimental domain described by the chosen levels of the pre-dictor variables.

3.3. PLS modeling of photosynthetic efficiency as function of sedimentcoverage

The combined PLS-model was subsequently used to predictphotosynthetic efficiency and sediment coverage by varying theadded amount of sediment at fixed intermediate levels of lightintensity, flow rate and time. These three parameters were keptconstant since they had minor impact on the responses. The pre-dicted values for photosynthetic efficiency and sediment coveragewere then used to create exposure–response curves for photosyn-thetic efficiency as function of sediment coverage. Fig. 4 shows theexposure–response curve obtained from the regression model. Forexample, a 50% reduction in photosynthetic efficiency is observedat 70% sediment coverage already after 1–2 weeks of exposureThe upper and lower dotted curves represent the uncertainties inthe modeling (/PSIImax ± 0.06) taking into account the minorimpacts, although largely insignificant, of flow rate, light intensityand time.

3.4. Photosynthetic efficiency measurements and color evaluation

Throughout the experiment the two calcareous algae specieschanged from a healthy purple to pinkish color with bleached spots

Table 3Partial least squares regression coefficients and 95% confidence interval for the fourPLS models. The major and significant regression coefficients bolded.

Photosyntheticefficiency

Sediment coverage

Regr.coeff.

Conf.int.

Regr.coeff.

Conf.int.

Mesophyllum Flow 0.07 0.39 0.08 0.371st trial Light �0.08 0.40 0.03 0.42

Sediment �0.82 0.19 0.89 0.17Time 0.13 0.11 �0.01 0.13Sediment2 �0.50 0.31 0.37 0.34

Mesophyllum Flow 0.03 0.32 0.00 0.272nd trial Light �0.05 0.27 �0.01 0.29

Sediment �0.77 0.20 0.90 0.16Time 0.07 0.03 0.04 0.03Sediment2 �0.56 0.38 0.29 0.37

Lithothamnion Flow 0.10 0.13 �0.02 0.092nd trial Light �0.03 0.11 �0.05 0.07

Sediment �0.74 0.06 0.86 0.06Time 0.06 0.22 0.05 0.20Sediment2 �0.57 0.13 0.34 0.09

Combined Flow 0.07 0.04 0.02 0.05Light �0.05 0.10 �0.01 0.09Sediment �0.76 0.07 0.88 0.05Time 0.08 0.17 0.03 0.15Sediment2 �0.54 0.10 0.33 0.08

(b) Sediment coverage (a) Photosynthetic efficiency

Fig. 3. Partial least squares regression coefficients with confidence limits for photosynthetic efficiency (a) and sediment coverage (b) obtained from PLS regression ofphotosynthetic efficiency and sediment coverage as functions of light intensity, flow rate, added sediment amount and time.

Fig. 4. Exposure–response curve based on predicted photosynthetic efficiency(/PSIImax) and sediment coverage from partial least squares regression model.Dashed lines illustrate uncertainty. Based on the combined data from both trialsand both algae species.

at the beginning of the experiment (T0), to a stressed pale colorafter one week when completely buried (1200 g sediment) andlater also when partially buried (900 g sediment). By the end ofthe experiments (T9) there were larger areas of white calcareousalgae thallus judged already dead under complete burial andincreased patchy white areas due to partial burial. High stresslevels were observed by an orange color of thallus and low read-ings of photosynthetic efficiency (/PSIImax = 0.16–0.25). In the mostsevere cases dead white areas of thallus were observed, consistentwith low readings (/PSIImax 6 0.15) of photosynthetic efficiency.Less frequently, whitish patches and areas with pale color wereobserved in the treatment with the lowest amount of sediment(600 g). No bleaching or color change was observed in the controlskept in darkness without sedimentation during the experimentalperiod of nine weeks. Photosynthetic efficiency of the controlswas 0.49 at T0 classifying the algae in the ‘‘no stress’’ category(Table 4). Photosynthetic efficiency decreased slightly to 0.43 atT9 classifying the controls in the ‘‘low stress’’ category (Table 4and Fig. 2).

A computational approach to quantitatively extract size andcolor features from images is in preparation and will be publishedin a separate paper.

4. Discussion

A major finding from the present study is the statistically signif-icant relationship between the exposure of two species of calcare-ous algae to particulate matter and the decrease in photosyntheticefficiency which is regarded as a parameter for environmentalstress. An exposure–response relationship describing the observedeffect on the calcareous algae as function of sediment coverage hasbeen established. For example, at 70% sediment coverage the pho-tosynthetic efficiency was reduced 50% already after 1–2 weeks ofexposure. The new insight in the impact of sedimentation on twospecies of calcareous algae can be used to establish threshold levelsand impact categories in environmental monitoring of calcareousalgae. Establishment of impact categories will be presented in aseparate paper.

The multivariate data analysis showed that the major contribu-tor to sediment coverage and reduction in photosynthetic effi-ciency was the added amount of sediment (positive and negativecorrelations, respectively), while flow rate, light intensity and timehad minor and mostly insignificant impact on the response param-eters within the experimental domain described by the chosenlevels of the predictor variables. Although flow rate did not influ-ence sediment coverage in these experiments, flow rate is animportant factor to take into consideration because it also affectsthe supply of oxygen and nutrients. The fact that flow rate and alsolight had minor impact on photosynthetic efficiency implies thatsediment coverage was the major contributor, probably due toreduced gas exchange. This assumption is supported by the con-trols kept in darkness without sedimentation where no bleachingor color change was observed. Furthermore, both algae speciesused in the present study were incrusting and unbranched

growth-forms which also support the assumption that the mea-sured stress was due to oxygen depletion and accumulation ofmetabolic products like carbon dioxide caused by the sedimentcoverage. Reduced gas exchange due to burial of fine sedimentsis consistent with previous findings (Wilson et al., 2004;Harrington et al., 2005).

No visual changes of the controls were observed during the nineweeks experiments. However, the photosynthetic efficiency mea-surements decreased slightly from 0.49 at T0 to 0.43 at T9 andmay be interpreted as an incipient light limited impact on the cal-careous algae.

The handling procedures of the algae are not expected to havesignificant effect on the photosynthetic efficiency. The removalfrom the exposure chamber and light source was too short to influ-ence on photosynthetic efficiency. This is confirmed by the controlskept in darkness where only a minor effect was seen even afternine weeks. The 20 min dark acclimation ensured that all reactioncenters were open for the determination of ground fluorescence(F0). The brushing and replacing of sediment were carried out verycarefully in a standardized way. Although calcareous algae are gen-erally considered to be rather robust (Steneck, 1986), initial mea-surements of photosynthetic efficiency were carried out beforeand after brushing to ensure that the brushing did not influencethe results.

Although the impact of light intensity is shown to be insignifi-cant during the nine week period, the regression coefficient inFig. 3 and the exposure–response curve (Fig. 4) indicate that lowlevels of sediment coverage on the calcareous algae may producea negative effect on photosynthetic efficiency. Photoinhibition ofphotosystem II has been demonstrated previously (Aro et al.,1993). This indicates that the calcareous algae are well adapted tolow light levels and also, to some extent, prefer some light reduc-tion. Similar responses have been found for color patterns of theoutermost pigmented layer of the calcareous algae (Villas-Bôaset al., 2014). The assumptions are also supported by environmentalmonitoring carried out on the Peregrino field demonstrating highdegree of natural sedimentation (unpublished data).

The present study has been performed with calcareous algae onan individual level. Possible impact of drilling discharges on popu-lation level has not been investigated. However, using all data fromboth trials and both species in the same regression model makesthe model more generic and robust. On a habitat level, large pro-portions of white dead patches have been attributed to a reductionof rhodolith bed vitality (Harvey and Bird, 2008). In addition toreduced survival as response to smothering and burial, alsoreduced recruitment of calcareous algae on coral reefs has beendocumented (Steneck, 1997; Figueiredo and Steneck, 2002).

As already described, one of the main intentions with the pre-sent study has been to use the new insight in the calcareous algaetolerance levels to sedimentation to develop threshold levels forcalcareous algae for use in risk assessment. Risk models incorpo-rating physical disturbances such as burial (Johnsen et al., 2000;Rye et al., 2008; Smit et al., 2008) can utilize these species specificthreshold levels instead of the generic set of threshold levelsderived from the literature. A similar approach has previously beenused for the cold water coral Lophelia pertusa (Purser and Thomsen,2012; Larsson et al., 2013).

It is emphasized that although sediment coverage can be mod-eled in the experimental flow-through system, these data are notexpected to be used to model sediment coverage at sea floor. Theenvironmental relevance of the data is first of all related to thereduction in photosynthetic efficiency as function of sedimentcoverage. Using the experimental results in risk assessment mayrequire a conversion to sediment thickness although sediment cov-erage (in %) was considered to be more suitable in the present worksince the experiments were carried out with calcareous algae living

Table 4Photosynthetic efficiency (/PSIImax) and color (tone and intensity) referring to stresslevels of calcareous algae thallus.

/PSIImax Color (tone) Color (intensity) Stress level

0–0.15 0–1 0–2.25 Dead0.16–0.25 1.1–2.0 2.26–3.25 High stress0.26–0.35 2.1–3.0 3.26–5 Intermediate stress0.36–0.45 3.1–3.75 5.1–6 Low stress0.46–0.55 3.76–4.25 6.1–7 No stress

on rhodoliths with varying surface structure and porosity. Thestrong correlation (critically validated) between sediment cover-age and photosynthetic efficiency implies that sediment coverageis a valid parameter, and also a parameter that can be measuredin the field e.g. by imaging. As a consequence, the results are rele-vant for environmental monitoring and risk assessment and theresults are valid for two important calcareous algae species.

5. Conclusions

The results from the present study are unique with respect todemonstrating the relationship between exposure of deep watercalcareous algae to particulate matter and the photosyntheticresponse. The statistical approach for the experimental designand interpretation of the sampled data secures the high reliabilityand significance of the results. This enables establishment of expo-sure–response relationship describing photosynthetic efficiency ofthe calcareous algae as function of sediment coverage. The expo-sure-response relationship can be used to establish thresholdlevels and impact categories for environmental risk assessmentof drill cuttings discharges to the Peregrino and other calcareousalgae habitats. Although the present results are of a conservativenature due to the design of the experiments, they provide a morerealistic basis for risk assessment than generic effect data whichare usually applied.

Acknowledgments

We are grateful to Mauricio Andrade, Sarah Dario, JonasOsterloff, Fernanda Siviero and Kristin Collier Valle for valuablesupport. Financial support was given by STATOIL Brasil Óleo eGás Ltda (STATOIL Brazil Oil and Gas LLC) and Agência Nacionalde Petróleo, Gás Natural e Biocombustíveis – ANP (NationalAgency of Petroleum, Natural Gas and Biofuels – ANP). SISBIOlicenses n� 20820-2 and 20826-1.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.marpolbul.2015.04.040.

References

Adey, W.H., Macintyre, I.G., 1973. Crustose coralline algae: a re-evaluation in thegeological sciences. Geol. Soc. Am. Bull. 84, 883–904. http://dx.doi.org/10.1130/0016-7606(1973) 84<883:ccaari>2.0.co;2.

Amado-Filho, G.M., Moura, R.L., Bastos, A.C., Salgado, L.T., Sumida, P.Y., Guth, A.Z.,Francini-Filho, R.B., Pereira-Filho, G.H., Abrantes, D.P., Brasileiro, P.S., Bahia, R.G.,Leal, R.N., Kaufman, L., Kleypas, J.A., Farina, M., Thompson, F.L., 2012. Rhodolithbeds are major CaCO3 bio-factories in the tropical south West Atlantic. PLoSONE 7, 35171. http://dx.doi.org/10.1371/journal.pone.0035171.

Aro, E.-M., Virgin, I., Andersson, B., 1993. Photoinhibition of photosystem II.Inactivation, protein damage and turnover. Biochim. Biophys. Acta – Bioenerg.1143, 113–134. http://dx.doi.org/10.1016/0005-2728(93)90134-2.

Ávila, E., Riosmena-Rodriguez, R., 2009. Rhodolith beds as critical habitat formonitoring areas in the Gulf of California. In: Polisciano, G., Farina, O. (Eds.).National Parks, Vegetation, Wildlife, pp. 2–8.

Bakke, T., Klungsøyr, J., Sanni, S., 2013. Environmental impacts of produced waterand drilling waste discharges from the Norwegian offshore petroleum industry.Mar. Environ. Res. 92, 154–169. http://dx.doi.org/10.1016/j.marenvres.2013.09.012.

Basso, D., 1998. Deep rhodolith distribution in the Pontian islands, Italy: a model forthe paleoecology of a temperate sea. Palaeogeogr. Palaeoclimatol. Palaeoecol.137, 173–187. http://dx.doi.org/10.1016/S0031-0182(97)00099-0.

Bordehore, C., Ramos-Esplá, A.A., Riosmena-Rodríguez, R., 2003. Comparative studyof two maerl beds with different otter trawling history, southeast IberianPeninsula. Aquat. Conservation. Mar. Freshw. Ecosyst. 13, S43–S54. http://dx.doi.org/10.1002/aqc.567.

Bosence, D.W.J., 1983. Description and classification of rhodoliths (Rhodoids,Rhodolites). In: Peryt, T.M. (Ed.), Coated Grains. Springer – Verlag Press,Berlin, Germany, pp. 217–224.

Box, G.E.P., Hunter, W.G., Hunter, J.S., 1978. Statistics for Experimenters: AnIntroduction to Design, Data Analysis, and Model Building. JohnWiley and Sons,New York, p. 650.

Davies, A.J., Roberts, J.M., Hall-Spencer, J., 2007. Preserving deep-sea naturalheritage: emerging issues in offshore conservation and management. Biol.Conserv. 138, 299–312. http://dx.doi.org/10.1016/j.biocon.2007.05.011.

Dethier, M.N., Steneck, R.S., 2001. Growth and persistence of diverse intertidalcrusts: survival of the slow in a fast-paced world. Mar. Ecol. Prog. Ser. 223, 89–100. http://dx.doi.org/10.3354/meps223089.

Eide, I., Johnsen, H.G., 1998. Mixture design and multivariate analysis in mixtureresearch. Environ. Health Perspect. 106, 1373–1376.

Evertsen, J., Johnsen, G., 2009. In vivo and in vitro differences in chloroplastfunctionality in the two north Atlantic sacoglossans (Gastropoda,Opisthobranchia) Placida dendritica and Elysia viridis. Mar. Biol. 156, 847–859.http://dx.doi.org/10.1007/s00227-009-1128-y.

Figueiredo, M.A.O., Steneck, R.S., 2002. Floristic and ecological studies of crustosecoralline algae on Brazil’s Abrolhos reefs. Proceedings of the 9th InternationalCoral Reef Symposium. Bali, pp. 493–498.

Figueiredo, M.A.O., Kain, J.M., Norton, T.A., 2000. Responses of crustose corallines toepiphyte and canopy cover. J. Phycol. 36, 17–24.

Figueiredo, M.A.O., Menezes, K.S., Costa-Paiva, E.M., Paiva, P.C., Ventura, C.R.R.,2007. Experimental evaluation of rhodoliths as living substrata for infauna atthe Abrolhos Bank, Brazil. Ciencias Mar. 33, 427–440.

Figueiredo, M.A.O., Coutinho, R., Villas-Boas, A.B., Tâmega, F.T.S., Mariath, R., 2012.Deep-water rhodolith productivity and growth in the southwestern Atlantic. J.Appl. Phycol. 24, 487–493. http://dx.doi.org/10.1007/s10811-012-9802-8.

Foster, M.S., 2001. Rhodoliths: between rocks and soft places. J. Phycol. 37, 659–667. http://dx.doi.org/10.1046/j.1529-8817.2001.00195.x.

Genty, B., Briantais, J.M., Baker, N.R., 1989. The relationships between the quantumyield of photosynthetic electron transport and quenching of chlorophyllfluorescence. Biochim. Biophys. Acta 990, 87–92. http://dx.doi.org/10.1016/S0304-4165(89)80016-9.

Hall-Spencer, J.M., 1998. Conservation issues concerning the molluscan fauna ofmaërl beds. J. Conchol. Special Publ. 2, 271–286.

Harrington, L., Fabricius, K., Eaglesham, G.C., Negri, A., 2005. Synergistic effects ofdiuron and sedimentation on photosynthesis and survival of crustose corallinealgae. Mar. Pollut. Bull. 51, 415–427. http://dx.doi.org/10.1016/j.marpolbul.2004.10.042.

Harvey, A.S., Bird, F.L., 2008. Community structure of a rhodolith bed from cold–temperate waters (Southern Australia). Australian J. Bot. 56, 437–450. http://dx.doi.org/10.1071/BT07186.

Johnsen, S., Frost, T.K., Hjelsvold, M., Utvik, T. R., 2000. The Environmental ImpactFactor – a proposed tool for produced water impact reduction, management andregulation. Society of Petroleum Engineers, International Conference on Health,Safety, and Environment in Oil and Gas Exploration and Production. Stavanger,Norway, 26–28 June 2000, SPE-61178-MS. doi: http://dx.doi.org/10.2118/61178-MS.

Kempf, M., 1970. Notes on the benthic bionomy of the N-NE Brazilian shelf. Mar.Biol. 5, 213–224. http://dx.doi.org/10.1007/BF00346909.

Larsson, A.I., Purser, A., 2011. Sedimentation on the cold-water coral Lopheliapertusa: cleaning efficiency from natural sediments and drill cuttings. Mar.Pollut. Bull. 62, 1159–1168. http://dx.doi.org/10.1016/j.marpolbul.2011.03.041.

Larsson, A.I., van Oevelen, D., Purser, A., Thomsen, L., 2013. Tolerance to long-termexposure of suspended benthic sediments and drill cuttings in the cold-watercoral Lophelia pertusa. Mar. Pollut. Bull. 70, 176–188. http://dx.doi.org/10.1016/j.marpolbul.2013.02.033.

Lavrado, H.P., 2006. Capítulo 1. Caracterização do ambiente e da comunidadebentônica. In: Lavrado, H.P., Ignácio, B.L. (Eds.), Biodiversidade Bentônica daRegião central da Zona Econômica Exclusiva Brasileira. Museu Nacional, Rio deJaneiro, Brazil, pp. 19–64.

Littler, M.M., Littler, D.S., Blair, S.M., Norris, J.N., 1986. Deep-water plantcommunities from an uncharted seamount off San Salvador Island, Bahamas:distribution, abundance, and primary productivity. Deep Sea Res. Part A.Oceanographic Res. Papers 33, 881–892. http://dx.doi.org/10.1016/0198-0149(86)90003-8.

Littler, M.M., Littler, D.S., Dennis Hanisak, M., 1991. Deep-water rhodolithdistribution, productivity, and growth history at sites of formation andsubsequent degradation. J. Exp. Mar. Biol. Ecol. 150, 163–182. http://dx.doi.org/10.1016/0022-0981(91)90066-6.

Martens, H., Næs, T., 1993. Multivariate Calibration. John Wiley and Sons,Chichester, p. 438.

Moura, C.W.N., Kraus, J.E., Cordeiro-Marino, M., 1997. Metodologia para obtençãode cortes histológicos com historesina e coloração com azul de toluidina o paraalgas coralináceas (Rhodophyta, Corallinales). Hoehnea 24, 17–27.

Nelson, W.A., 2009. Calcified macroalgae – critical to coastal ecosystems andvulnerable to change: a review. Mar. Freshw. Res. 60, 787–801. http://dx.doi.org/10.1071/MF08335.

Purser, A., Thomsen, L., 2012. Monitoring strategies for drill cutting discharge in thevicinity of cold-water coral ecosystems. Mar. Pollut. Bull. 64, 2309–2316. http://dx.doi.org/10.1016/j.marpolbul.2012.08.003.

Reynier, M.V., Tâmega, F.T.S., Daflon, S.D.A., Santos, M.A.B., Coutinho, R., Figueiredo,M.A.O., 2015. Long-and short-term effects of smothering and burial by drillcuttings on calcareous algae in a static-renewal test. Environ. Toxicol. Chem. (inpress). doi: http://dx.doi.org/10.1002/etc.2938.

Riosmena-Rodriguez, R., Steller, D.L., Hinojosa-Arango, G., Foster, M.S., 2010. Reefsthat rock and roll: biology and conservation of rhodolith beds in the Gulf of

California. In: Brusca, R.C. (Ed.), The Gulf of California: Biodiversity andConservation. University of Arizona and Arizona-Sonora Desert MuseumPress, Arizona, pp. 49–71.

Riul, P., Targino, C.H., Farias, J.N.B., Visscher, P.T., Horta, P.A., 2008. Decrease inLithothamnion sp. (Rhodophyta) primary production due to the deposition of athin sediment layer. J. Mar. Biol. Ass. UK 88, 17–19. http://dx.doi.org/10.1017/S0025315408000258.

Rye, H., Reed, M., Frost, T.K., Smit, M.G.D., Durgut, I., Johansen, Ø., Ditlevsen, M.K.,2008. Development of a numerical model for calculating exposure to toxic andnontoxic stressors in the water column and sediment from drilling discharges.Integr. Environ. Assess. Manag. 4, 194–203. http://dx.doi.org/10.1897/IEAM_2007-039.1.

Salgado, C.S., Johnsen, S., Gomes, S., Torkelsen, T., 2010. Real-time environmentalmonitoring of discharges of drill cuttings in the Peregrino field (Campos Basin,Brazil). SPE 127175, Rio de Janeiro, Brazil, pp. 2–9. doi: http://dx.doi.org/10.2118/127175-MS.

Sciberras, M., Rizzo, M., Mifsud, J.J., Camilleri, K., 2009. Habitat structure andbiological characteristics of a maërl bed off the northeastern coast of theMaltese Islands (central Mediterranean). Mar. Biodivers. 39, 251–264. http://dx.doi.org/10.1007/s12526-009-0017-4.

Smit, M.G.D., Holthaus, K.I.E., Trannum, H.C., Neff, J.M., Kjeilen-Eilertsen, G., Jak,R.G., Singsaas, I., Huijbregts, M.A.J., Hendriks, A.J., 2008. Species sensitivitydistributions for suspended clays, sediment burial, and grain size change in themarine environment. Environ. Toxicol. Chem. 27, 1006–1012. http://dx.doi.org/10.1897/07-339.1.

Søfteland, L., Eide, I., Olsvik, P.A., 2009. Factorial design applied for multipleendpoint toxicity evaluation in Atlantic salmon (Salmo salar L.) hepatocytes.Toxicol. In Vitro 23, 1455–1464. http://dx.doi.org/10.1016/j.tiv.2009.07.014.

Steller, D.L., Riosmena-Rodríguez, R., Foster, M.S., Roberts, C., 2003. Rhodolith beddiversity in the Gulf of California: the importance of rhodolith structure andconsequences of anthropogenic disturbances. Aquat. Conserv. 13, S5–S20.http://dx.doi.org/10.1002/aqc.564.

Steller, D.L., Riosmena-Rodriguez, R., Foster, M.S., 2009. Living rhodolith bedecosystems in the Gulf of California. In: Johnson, M.E., Ledesma-Vázquez, J.

(Eds.), Atlas of Coastal Ecosystems in the Western Gulf of California. TrackingLimestone Deposits on the Margin of a Young Sea. The University of ArizonaPress, Arizona, pp. 72–82, <http://www.uapress.arizona.edu/Books/bid2141.htm>.

Steneck, R.S., 1986. The ecology of coralline algal crusts: convergent patterns andadaptive strategies. Ann. Rev. Ecol. Systemat. 17, 273–303, <http://www.jstor.org/discover/10.2307/2096997>.

Steneck, R.S., 1997. Crustose corallines, other algal functional groups, herbivoresand sediments: complex interactions along reef productivity gradients.Proceedings of the 8th International Coral Reef Symposium. Panama, pp. 695–700.

Tâmega, F.T.S., Spotorno-Oliveira, P., Figueiredo, M.A.O., 2013. Catalogue of thebenthic marine life from Peregrino oil Field, Campos Basin. InstitutoBiodiversidade Marinha, Rio de Janeiro, Brazil, Brazil.

Trannum, H.C., Nilsson, H.C., Schaanning, M.T., Øxnevad, S., 2010. Effects ofsedimentation from water-based drill cuttings and natural sediment onbenthic macrofaunal community structure and ecosystem processes. J. Exp.Mar. Biol. Ecol. 383, 111–121. http://dx.doi.org/10.1016/j.jembe.2009.12.004.

Villas-Bôas, A.B., Tâmega, F.T.S., Andrade, M., Coutinho, R., Figueiredo, M.A.O., 2014.Experimental effects of sediment burial and light attenuation on two corallinealgae of a deep water rhodolith bed in Rio de Janeiro, Brazil. CryptogamieAlgologie 35, 67–76.

Wilson, S., Blake, C., Berges, J.A., Maggs, C.A., 2004. Environmental tolerances offree-living coralline algae (maërl): implications for European marineconservation. Biol. Conserv. 120, 279–289. http://dx.doi.org/10.1016/j.biocon.2004.03.001.

Woelkerling, W.J., 1988. The Red Coralline Algae: An Analysis of the Genera andSubfamilies of Nongeniculate Corallinaceae. Oxford University Press, New York.

Wold, S., 1978. Cross-validatory estimation of the number of components in factorand principal components models. Technometrics 20, 397–405. http://dx.doi.org/10.2307/1267639.

Wold, S., Martens, H., Wold, H., 1983. The multivariate calibration problem inchemistry solved by the PLS method. Matrix Pensils 973, 286, 286–29.

Paper V

Osterloff, J., Nilssen, I., Eide, I., de Oliveira Figueiredo, M. A., de Souza Tâmega, F. T., Nattkemper, T. W., 2016b. Computational visual stress level analysis of calcareous

algae exposed to sedimentation. PLoS ONE Vol 11, pp e0157329, doi: http://dx.doi.org/10.1371/journal.pone.0157329

RESEARCH ARTICLE

Computational Visual Stress Level Analysis ofCalcareous Algae Exposed to SedimentationJonas Osterloff1*, Ingunn Nilssen2,3, Ingvar Eide2, Marcia Abreu de Oliveira Figueiredo4,5,Frederico Tapajós de Souza Tâmega5,6, TimW. Nattkemper1

1 Biodata Mining Group, Faculty of Technology, Bielefeld University, Bielefeld, Germany, 2 Statoil ASA,Research and Technology, N-7005 Trondheim, Norway, 3 Trondhjem Biological Station, Department ofBiology, Norwegian University of Science and Technology, N-7491 Trondheim, Norway, 4 Instituto dePesquisa Jardim Botanico do Rio de Janeiro, CEP 22.460-030, Rio de Janeiro, RJ, Brazil, 5 InstitutoBiodiversidade Marinha, 22.793-000, Rio de Janeiro, RJ, Brazil, 6 Instituto de Estudos do Mar AlmirantePaulo Moreira, 28.930-000, Arraial do Cabo, RJ, Brazil

* josterlo@cebitec.uni-bielefeld.de

AbstractThis paper presents a machine learning based approach for analyses of photos collected

from laboratory experiments conducted to assess the potential impact of water-based drill

cuttings on deep-water rhodolith-forming calcareous algae. This pilot study uses imaging

technology to quantify and monitor the stress levels of the calcareous algaeMesophyllumengelhartii (Foslie) Adey caused by various degrees of light exposure, flow intensity and

amount of sediment. A machine learning based algorithm was applied to assess the tempo-

ral variation of the calcareous algae size (*mass) and color automatically. Measured size

and color were correlated to the photosynthetic efficiency (maximum quantum yield of

charge separation in photosystem II, FPSIImax) and degree of sediment coverage using multi-

variate regression. The multivariate regression showed correlations between time and cal-

careous algae sizes, as well as correlations between fluorescence and calcareous algae

colors.

IntroductionIncreasing anthropogenic activities in marine areas call for a holistic environmental monitor-ing approach using new and more sophisticated models and sensor systems that enable predic-tions and measurement of impact on organisms of interest [1]. One important input tomodeling is knowledge generated through laboratory experiments, determining various levelsof stress on the organisms of interest. A common way to measure stress is through visual docu-mentation of behavioral changes, change in abundance [2–4], change of the spectral response[5] or change of the visual appearance [6, 7] of organisms.

The investigated calcareous algae species,Mesophyllum engelhartii (Foslie) Adey, belongs tothe Phylum Rhodophyta under the Order Hapalidiales. Calcareous algae play an important eco-logical role in coastal habitats [8–10] contributing to the formation of rhodoliths. Rhodolithsare made from dead calcareous algae and other calcifying organisms [11]. These multi-

PLOSONE | DOI:10.1371/journal.pone.0157329 June 10, 2016 1 / 22

a11111

OPEN ACCESS

Citation: Osterloff J, Nilssen I, Eide I, de OliveiraFigueiredo MA, de Souza Tâmega FT, NattkemperTW (2016) Computational Visual Stress LevelAnalysis of Calcareous Algae Exposed toSedimentation. PLoS ONE 11(6): e0157329.doi:10.1371/journal.pone.0157329

Editor: James P. Meador, Northwest FisheriesScience Center, NOAA Fisheries, UNITED STATES

Received: April 15, 2015

Accepted: May 27, 2016

Published: June 10, 2016

Copyright: © 2016 Osterloff et al. This is an openaccess article distributed under the terms of theCreative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in anymedium, provided the original author and source arecredited.

Data Availability Statement: All images areavailable under doi:10.4119/unibi/2775316 (http://doi.org/10.4119/unibi/2775316).

Funding: Financial support was given by STATOILBrasil Óleo e Gás Ltda (STATOIL Brazil Oil and GasLLC) (http://www.statoil.com/brazil/). AgénciaNacional de Petróleo, Gás Natural e Biocombustíveis- ANP (National Agency of Petroleum, Natural Gasand Biofuels - ANP) (http://www.anp.gov.br/), Rio deJaneiro, Brazil, has accepted the Project under theFederal Participation Agreement (FPE). The decisionto publish was not influenced by the funders.

spherical structures are creating a habitat for other organisms living on, between and in thestructures [12]. Calcareous algae are found down to about 250 meters water depth [13–15] andthe largest occurrence of rhodolith beds are found in the southwest Atlantic on the Braziliancontinental shelf [16–18]. These algae communities may be disturbed by natural sedimentationand/or sedimentation from anthropogenic activities such as fish-trawling, mining [19] and dis-charges of drill cuttings from oil and gas drilling activities [20].

In a previous study the impact from sedimentation of drill cuttings on live calcareous algaewas studied by measuring sediment coverage (SC) and photosynthetic efficiency (P) [6]. Thestudy was part of the Peregrino Environmental Monitoring Calcareous Algae Project(PEMCA). Rhodoliths, partly covered with calcareous algae, were collected from the Peregrinooil field off the coast of Brazil. The impact of sedimentation was investigated in a laboratoryflow-through system, varying light exposure (L), flow rate (F ), amount of sediment (S) andtime (T ). Photosynthetic efficiency (P) was measured as maximum quantum yield of chargeseparation in photosystem II, FPSIImax

. These results have been published elsewhere [6].

The present paper describes a new method to extend the observed impact variables (SC, P)by measurements of size and color of calcareous algae in digital photos recorded during theexperiments.

Using photos for environmental monitoring generates a large number of images implyingthat a manual evaluation is extremely demanding regarding time and resources and requires acareful and professional execution, as for instance outlined in [21].

To support the evaluation and labeling of the photos performed by marine biology expertsdifferent software tools have been proposed such as Coral Point Count with Excel extensions(CPCe) [22]. Recently, web-based image annotation and labeling systems have been proposedto support web-based sharing of photos and collaboration, such as BIIGLE (Benthic ImageIndexing, Graphical Labeling and Exploration) [23], which was successfully applied in [4, 24–28], and CoralNet [29], the latter one providing for instance an automated point classificationfor corals as well.

Since humans have a limited capability to quantify visual features, such as color change, inan objective way, we propose a machine learning based approach to compute size and color ofthe calcareous algae from photos recorded at different time points. A machine learning algo-rithm is trained with a number of manual image annotations. The trained classifier is appliedfor the full automatic segmentation of the calcareous algae, enabling the quantification of thecalcareous algae size and color over time. To learn the classification function a H2SOM (Hier-archical Hyperbolic Self-Organizing Map) [30] algorithm was applied. This unsupervisedlearning algorithm has previously been used for cold-water coral segmentation in ROV videoframes [24] and for poly-metallic nodule segmentation [31, 32]. Nevertheless, the integrationof the basic algorithm into an image analysis pipeline needed to be modified substantiallyregarding pre-processing, feature computation, pixel classification function and post-processing.

A simplified overview of the whole impact quantification process is given in Fig 1.

MaterialSamples of live calcareous algaeMesophyllum engelhartii (Foslie) Adey were collected from94–103m water depth at the Peregrino oil production field (23°13028.3400S, 41°50055.7300W)located off the Brazilian Atlantic coast [11]. Light intensity (L), flow rate (F ) and sedimentamount (S) could be varied independently as described in [6]. In order to optimize the experi-ment, the three predictor variables (L, F , S) were combined according to a Central CompositeFace (CCF) design with center point [33]. This enables regression modeling to describe linear

Computational Visual Analysis of Calcareous Algae

PLOS ONE | DOI:10.1371/journal.pone.0157329 June 10, 2016 2 / 22

Competing Interests: The authors of this manuscripthave read the journal’s policy and have the followingcompeting interests: Two of the authors, IngunnNilssen and Ingvar Eide are employees in Statoil,Research and Technology, Trondheim, Norway. Theyhave contributed scientifically to the manuscript andhave been involved in study design and dataanalyses. The decision to publish was not influencedby the funders; on the contrary, the intention and aprerequisite have been to publish the work in peerreviewed scientific journals and release the data.

and non-linear relationships and interactions between the variables. The full factorial designgives 15 combinations of L, F , and S. Two of these combinations were duplicated and twowere triplicated (i.e. technical replicates). Using the 15 combinations and technical replicatesresulted in a total of 21 experiments which were carried out in 21 chambers (Table 1). In addi-tion controls were kept in complete darkness and without sediment. The area of the chamberbottom was approximately 0.06m2 and the height of was 0.1 m resulting in a volume of approx-imately 6 L. The light levels (L) used in the exposure studies were 3, 6.6, or 10 μmol m−2 s−1,the latter value corresponding to measured light conditions at the sea floor on the Peregrinofield. The flow rates (F ) used were 0.04, 0.07, or 0.09 m s−1. The amounts of sediment (S),applied at the beginning of the experiment (T0), were 600, 900, or 1200 g per chamber. Thechosen amounts resulted in a sediment coverage (SC) of the calcareous algae ranging fromuncovered to completely covered, implying that also non- and very low sediment covered cal-careous algae are included in the analysis. Six parallels (i.e. biological replicates) of rhodoliths

Fig 1. A simplified illustration of the impact quantification process. The process is starting on the upper left with the laboratory experiment, continuingwith the image recording, the labeling by the experts, the application of the unsupervised machine learning algorithm (see Machine Learning section fordetails) and ending with the image segmentation and the measuring of size and color of the segmented area.

doi:10.1371/journal.pone.0157329.g001

covered with healthy calcareous algae were placed in each chamber, so each of the 21 experi-ments were conducted with six samples.

In addition to the measurements reported in [6], photos of the calcareous algae sampleswere taken for purpose of the present paper during the weekly examination. For image-acquisi-tion, the calcareous algae rodoliths were removed from their chambers and manually placed ina transparent, water-filled cylinder together with a color reference plate (Figs 2 and 3A). Thereference plate, essential for a comparative study of the color over time, can be seen in theupper part of the images (Fig 3 upper left). The photos were taken with a Canon PowerShotG10, which was placed on a tripod to control the distance (15 cm) between camera and sam-ples. The camera was in automatic mode with macro function and internal flash on. To elimi-nate light refractions and reflections samples and camera, the latter protected by a waterproofhousing, were placed underwater inside the cylinder (Fig 2). The digital images had a resolutionof 3456 × 2592 pixels. Each week 63 images (21 chambers × 3 images, 2 samples per photo),were taken summing up to N = 630 images for the whole experiment (T0, . . ., T9).

In order to collect a set of image annotations for training and testing the machine learningalgorithm, the recorded images were uploaded to BIIGLE. With this online image annotationtool, biological experts from the Instituto Biodiversidade Marinha were manually labelingphoto regions (pixel) as “live” calcareous algae (2404 labels), “stressed” calcareous algae (2834labels), “dead” calcareous algae (4358 labels) and “bare substratum” (43 labels) on all N = 630images. For the annotation, the experts were allowed to select single point label or customizableframe labels (Fig 4). These labels provided examples for the later machine learning training

Table 1. Chamber configurations for the CCF design.

Chamber L mmol

� �F m

� �S½g�

1 3.0 0.04 600

2 10.0 0.09 600

3 3.0 0.09 600

4 10.0 0.04 600

5 3.0 0.04 1200

6 10.0 0.04 1200

7 3.0 0.09 1200

8 10.0 0.09 1200

9 3.0 0.07 900

10 3.0 0.07 900

11 10.0 0.07 900

12 6.6 0.04 900

13 6.6 0.04 900

14 6.6 0.04 900

15 6.6 0.09 900

16 6.6 0.09 900

17 6.6 0.09 900

18 6.6 0.07 600

19 6.6 0.07 1200

20 6.6 0.07 900

21 6.6 0.07 900

The configurations for the 21 chambers for light intensity (L), flow rate (F ), and sediment amount (S) are

displayed. Replicated configurations are grouped.

doi:10.1371/journal.pone.0157329.t001

step (see Machine Learning section below). The labeling strategy was chosen to reduce the timeneeded for the labeling and simultaneously cover the whole variation between and within eachlabel category over time. The “live” and “stressed” calcareous algae labels represent the so called“positive” class in the segmentation and the “dead”- and “bare substratum”-labels representthe so called “negative” class. All images and labels can be accessed via BIIGLE (https://ani.cebitec.uni-bielefeld.de/biigle/). To browse the data, a login with username pemca and pass-word pemca is required. The data is stored in the project Pemca3rd, with the transects T0, . . .,T9. Image material is also published under DOI 10.4119/unibi/2775316.

Fig 2. Illustration of color reference plate and camera setup.On the left the original color reference plate, used to obtain color constancy within theexperiment, is displayed. Colors are referring to the RGB triplets (top-down): (0,0,255), (255,0,0), (255,0,255), (0,255,0), (0,255,255), (255,255,0). On theright a schematic layout of the underwater camera setup is presented, consisting of a water filled cylinder containing the sample, the color reference plateand the camera in the waterproof housing fixed on a tripod.

MethodsThe permission to collect the calcareous algae samples for the experimental study described inthe Material section was granted by the Chico Mendes Institute for Biodiversity ConservationSISBIO license n 20820-2 and 20826-1. The experimental study did not involve endangered orprotected species.

The basic idea of the computational calcareous algae segmentation is to classify each pixelin each image to be i)“live” or “stressed” calcareous algae or ii) other, including “dead” calcare-ous algae. A pixel classifier is trained using data collected from the annotations provided by theexperts (see Material section). From the pixel-wise classification, the regions of “live”/“stressed”calcareous algae are determined for each sample at each time point. The results are analyzed in

Fig 3. Example of an image recorded during the experiment completion and its different processing results of the pre-processing. Theoriginal image (A), the result of the illumination correction (B), the result of the color correction (C) and the difference image (D) are displayed. In theoriginal image (A) the color reference plate is located in the upper part, providing input for the pre-processing. The two calcareous algae samples arelocated below the reference plate. Comparing the result image from the illumination correction (see Illumination Correction section for details) with theoriginal image a slight decrease in brightness can be noticed. Due to the stable illumination conditions during image recording only little differencesbetween the images can be recognized visually. The color correction (see Reference Plate Detection, Color and Zoom Correction section for details)is changing the colors very little as only minor color shifts appear. Most of the visible differences in the difference image are caused by the illuminationcorrection.

combination with the previously measured parameters photosynthetic efficiency (P) and sedi-ment coverage (SC) using multivariate analysis. The individual steps for image processing,pixel classification and final data analysis are described in detail in the following sections.

Pre-ProcessingBefore the feature extraction and the machine learning can be applied, the recorded imagesneed to be pre-processed.

Illumination Correction. In order to compensate illumination fluctuations in the imagesthe local space average color scaling [34] is used on each image individually. The local spaceaverage color image is computed using a Gaussian blurring with σ = 0.093 �max(Iheight, Iwidth).

Fig 4. A screenshot of the BIIGLE system in a web-browser. The images of the calcareous algae were examined and labeled using theBIIGLE system. Images can also be zoomed to examine more details. The experts were allowed to select single point label or customizableframe labels. The single point labels are represented as filled colored circles, the customizable frame labels are represented as whiteoutlined rectangles with a filled colored circle in the upper left corner of the individual rectangle. The colors of the filled circles indicate theclass of the individual label. Red is representing “live” calcareous algae, yellow “stressed” calcareous algae, green “dead” calcareous algaeand Pink “bare substratum”.

As the images were captured in sRGB [35], which assumes a gamma correction of 1/2.2, theimages are gamma corrected and then divided pixel- and channel-wise (c) by the correspond-ing Gaussian filtered image G(n):

I ðnÞx;y;c ¼ðIðnÞx;y;cÞ2:2

t � GðnÞx;y;c

!1=2:2

τ is a scaling factor and set to τ = 2 in the whole study. An example for an illumination cor-rected image is displayed in Fig 3 (upper right).

Reference Plate Detection, Color and Zoom Correction. The first step in the color cor-rection is the detection of the color circles in the reference plate using a Hough circle transfor-mation [36]. The Hough transform computes a list of circle candidates and those that show alow color variance inside are selected taking the spatial layout of the color plate into account as

well. The identified six color circles Cl (l = 1, . . ., 6) of each I ðnÞ are then used to calculate one

RGB color triplet cðnÞl for each circle over all its pixels pðnÞðx;yÞ;c.

cðnÞl ¼

medianðx;yÞ2circlel

pðnÞðx;yÞ;red

� �medianðx;yÞ2circlel

pðnÞðx;yÞ;green

� �medianðx;yÞ2circlel

pðnÞðx;yÞ;blue

� �

0BBBBBB@

1CCCCCCA ð2Þ

In addition, overall median color references are computed for each of the six color circles (Eq3) of all images (Fig 5).

�cl ¼

mediann

cðnÞl;red

� �median

ncðnÞl;green

� �median

ncðnÞl;blue

� �

0BBBBB@

1CCCCCA ð3Þ

For each image I ðnÞ and color channel c 2 {red, green, blue}, the individual, channel-wise

gamma correction is calculated using the color RGB triplet cðnÞl (Eq 2) of the individual image

Fig 5. The average color reference plate. This reference is computed by averaging over the detected reference plates from all imagesused in this study (see text for details). Colors are referring the RGB triplets (left to right) (181,185,100), (147,179,200), (146,173,106),(173,101,147), (183,67,80), and (69,88,154).

and the overall median color references (Eq 3).

gðnÞc ¼ medianl

logcðnÞl;red

� �log

cl;c255

� �0BB@

1CCA ð4Þ

This image specific gamma correction is applied channel-wise to all the images resulting in anew set of color corrected images (Fig 3 lower left)

J ðnÞðx;yÞ;c ¼ I ðnÞðx;yÞ;c

� �gðnÞc

: ð5Þ

A difference image (ðIðnÞx;y;cÞ�ðJðnÞx;y;cÞ=2 þ 128) is presented in Fig 3 (lower right) to demonstrate the

impact of the described illumination and color correction.The layout of the reference plate is also used to correct zoom variations, which occurred for

a significant number of images. The average distance �d between the neighboring color circles(e.g. green and magenta) on each plate is evaluated from the five Euclidian circle distancesdi, i+1

�d ¼ 1

dðnÞði;iþ1Þ: ð6Þ

The difference between the average distance and the observed individual distances in image{J(n)} is used to rescale each image individually.

SegmentationUnsupervised learning is applied to learn a vector quantization of the image feature space (seeMachine Learning section). Vector quantization fits a set of so called prototype vectors itera-tively to a data set of feature vectors from the images describing pixel features. After the train-ing step is finished, each prototype represents a group of pixels sharing similar features, i.e. adense cloud in the feature space. The prototypes are assigned to the classes “live / stressed cal-careous algae” and “other”, including “dead calcareous algae” and “bare substratum”, and areapplied to classify all image pixel (see Prototype Label Mapping section). For the final segmen-tation result the classified image is post-processed (see Post Processing section).

Feature Extraction. One important step in vector quantization-based image segmentationis the selection of appropriate features and the collection of a training set. To collect a trainingset for the vector quantization learning, a selection of image pixels {pi} from a subset of all Nimages are chosen and their feature vectors {x(i)} are computed. Here, Median RGB values

extracted on a 11 × 11 pixel neighborhood ðZðiÞ11Þ for each color channel, are used as features

since we observed such a representation to be sufficient for the initial task of calcareous algaesegmentation by pixel classification.

xðiÞ ¼

medianðx;yÞ2ZðiÞ

pðx;yÞ;red

� �medianðx;yÞ2ZðiÞ

pðx;yÞ;green

� �medianðx;yÞ2ZðiÞ

pðx;yÞ;blue

� �

0BBBBBBBB@

1CCCCCCCCAð7Þ

To reduce the computation time, features are only extracted for pixels in a 2 × 2 grid resolutioncollecting a training set of about 2.5 million feature vectors {x(i)}.

Machine Learning. To perform vector quantization on the training data the unsupervisedmachine learning algorithm H2SOM [30] is applied.

H2SOMs with different parameter settings, like the number of prototypes, were trained andthe best parameter set was determined using a customized criterion described in (S1 Text). Atraining set of feature vectors generated from 12.5% of the pixels chosen from a small subset(2%) of the entire image collection is used to construct the H2SOM with the best segmentationresults. The H2SOM is trained using 30 × |{x(i)}| iterations, an exponentially decreasing learn-ing rate of α = [0.99, 0.1], an eight-neighbor topology and three rings and therefore is consist-ing of 161 prototypes.

Prototype Label Mapping. The next step is to build a pixel classifier out of the trainedH2SOM, using the feature vectors x(p) of all images J(n). We distinguish between feature vectorsx(i) used for the H2SOM training and feature vectors x(p) from all images, but of course bothare computed as described in Eq 7. The expert labels “live” and “stressed” from BIIGLE areused to build the classifier as follows. In general, we have observed that a differentiationbetween the categories “live” and “stressed” is problematic regarding the reproducibility of theresults of the visual assessment. As a consequence a discrimination between these two catego-ries has been neglected. The fused category is referred to as “live calcareous algae”.

First, a label yp is assigned to each feature vector x(p)

yp ¼1; if the pixel was labeled as “live”=“stressed” calcareous algae

0; else

(ð8Þ

Second, each feature vector x(p) is assigned to one of the H2SOM prototypes {u(j)} (j = 1..161)that represents the features of x(p) in the best possible way. This prototype is selected by the socalled Best Matching Unit (BMU) criterion, that selects the prototype with the minimal Euclid-ian distance to x(p) in the feature space. Next, all cluster prototypes are investigated in order toclassify them. The prototypes constitute a Voronoi tessellation of the feature space

Vj ¼ fxðpÞjj ¼ argminj 0

feuclidðuðj 0Þ;xðpÞÞgg ð9Þ

with qj = |Vj| as the number of feature vectors that have been assigned to u(j) and therefore arelocated in Vj. In each Voronoi cell Vj we now analyze how many feature vectors stem frompixel, labelled as “live” or “stressed”:

Vþj ¼ fxðpÞjj ¼ argmin

j 0feuclidðuðj 0Þ;xðpÞÞg ^ yp ¼ 1g; ð10Þ

with qþj ¼ jVþ

j j as the number of feature vectors in Vj that are labeled as “live”/“stressed” cal-

careous algae. The relation sj ¼ qþj =qj quantifies how well a prototype u(j) resembles the fea-

tures of live calcareous algae. Since we do not expect all x(p) with yp = 1 to be perfectrepresentatives, only prototypes u(j) with a high sj will represent the positive class (presence oflive calcareous algae).

Next the relation values {sj} are sorted in descending order:

sj ! sl; with sl > sl�18l: ð11Þ

From this list, the first L prototypes u(l) are chosen as a model for typical live calcareous algaecolor features. L is chosen so at least 80% of the entire positive training data (i.e. yi = 1) is cov-ered. This set of the first 16 “most live calcareous algae-like” prototypes is used for the pixel

classification based segmentation in all images. In each image, each pixel p is mapped to eachfeature vector x(p). The BMU of the feature vector is determined and if this unit is a live calcar-eous algae-like prototype, as described above, the pixel is labeled as live calcareous algae(L(x(p)) = 1) or otherwise as background (L(x(p)) = 0).

Post Processing. By classifying all feature vectors x(p), binary images are created showingwhite pixels on live calcareous algae classified positions and black pixels otherwise. Thesebinary images are post-processed with an opening filter mask [37] of a size 10 × 10 pixels toremove small areas of false positive classifications. The post-processed binary images representthe final segmentation result.

Multivariate Data AnalysisTo analyze the possible biological impact of sedimentation on calcareous algae, the size and thecolor of the segmented areas in the images are computed. (Fig 6). The size (A) of the live calcar-eous algae area is measured as the amount of segmented pixels per image J(n). It is weighted by

the ratio of the average color circle distance of circle centers �d and the average circle center dis-

tance dn of the color reference plate from the Image J(n).

A ¼ size��d

ð12Þ

The relative size A ¼ AA0, with A0 being the live calcareous algae size at the first measurement, is

finally used as one variable. The HSV (hue, saturation, value) color space is used for measuringthe color. Note that we use a different color space here than the one we used for the segmenta-

tion. Averages of hue (averaging is applied with respect that it is a circular quantity) (�h), satura-tion (�s) and value (�v) of all segmented pixels per image J(n) are computed of the segmentedpixels. These four variables are combined to one measurement for the image n:

f ðnÞ ¼ A; �h;�s; �v� T ð13Þ

All calcareous algae samples from the same chamber are exposed to the same treatment param-eters. On each image two samples were captured. To avoid the need of a colocation of the sam-ples as well as taking into account that all rhodoliths are different, averages of size and colorwithin each chamber are used. The dataset can be found in S1 Table.

Multivariate regression is performed with partial least squares (PLS) [38, 39] on the live cal-careous algae area measurements f(n), using Unscrambler X 10.3 (CAMO Software, Oslo, Nor-way) to correlate the predictor variables (X matrix) to the measured responses (Y matrix). Thepreviously published data (first experimental series withM. engelhartii in [6]) on P and SC areincluded in the data matrix for the regression analysis. To match f(n) also data for P and SCwere averaged chamber wise (six parallels). The final data matrix has 168 rows (21 chambers × 8time-points) (observations), four columns for the four predictor variables (L, F , S and T ),

and six columns for the response variables (A; �h;�s; �v; P; SC). Prior to the multivariate dataanalysis, the data are mean centered and scaled to unit variance. The models were evaluatedwith respect to explained variance, and goodness of fit and prediction, the latter obtained aftercross validation [40, 41].

In addition to the PLS regression, a traditional statistical analysis was conducted by per-forming ANOVA, MANOVA and pairwise correlation analysis. These results are presented inthe supplementary (Table A, Table B, Table C in S2 Text and Fig A, B, C in S3 Text). In the fol-lowing we will concentrate on the results of the PLS regression as it takes all variables into

Fig 6. Example of one pair of calcareous algae samples. In the top row the original images recorded at T2, T5 and T9 are shown. In the middlerow the segmentation results are highlighted and in the lower row the live calcareous algae relative size (A) and hue (�h), starting from 10 (slightlyorange red) to 0 (red), are presented. The samples were exposed to S ¼ 900 g (a partially covered sample), L ¼ 6:6mmol m�2 s�1 andF ¼ 0:07m s�1.

consideration simultaneously and gives a better understanding of the relative importance andthe significance of all variables.

Results

Pre-ProcessingFluctuations in sample illumination are a well-known problem in computational image analy-sis and the majority of the N = 630 images I(n) (n = 1, . . ., N) suffered at least partially from it.The procedures described above compensated this effect (Fig 3(B)).

In 15% of the images used in the analysis, the reference color circles could not be detectedautomatically. In these images the circles were marked manually or only a subset of detectedcircles was used for the color correction. Although reflections were avoided on the calcareousalgae samples, serious reflection problems were observed on the color reference plates in someof the images. The images from T0 and T1 had to be excluded from further analysis because ofthis problem.

Visual inspection showed that the channel-wise gamma correction produced robust results.The pre-processing stages described above could be computed all together in about three min-utes per image.

SegmentationThe segmentation results were reviewed by comparing the human expert labels with the seg-mented areas of all images using standard measures for image analysis accuracy assessmentlike recall (or sensitivity), precision (or positive predictive value) and false positive rate (FPR).The recall is the percentage of all pixels labeled “live” or “stressed” p(+) that have been seg-mented correctly as live calcareous algae by the algorithm.

recall ¼ jfpðþÞg\fpðsÞgjjfpðþÞgj ð14Þ

The precision is the percentage of all pixels segmented as live calcareous algae (p(s)) that havebeen segmented correctly by the algorithm.

precision ¼ jfpðþÞg\fpðsÞgjjfpðsÞgj ð15Þ

From the segmentation results we computed a recall value of 0.89 and a precision of 0.03.Since the manually labeling did not attempt to measure the precise live calcareous algae

extent in each image, also the “dead” and “bare substratum” labels were used to review the seg-mentation results. The false positive rate (FPR) was used to measure the number of pixels thatwere segmented to be “live” or “stressed” p(s) but were labeled “dead” or “bare substratum” p(−).

FPR ¼ jfpð�Þg\fpðsÞgjjfpð�Þgj ð16Þ

The FPR was very low compared to the recall. Looking at the false positive rates in detail itshould be noted that the FPR for the “bare substratum” is higher than the one of the “dead”labels (Table 2). Segmentation results are also displayed in Fig 7.

Multivariate Data Analysis

The best PLS regression model was obtained after expansion with one square term (S2) in thepolynomial describing the relationship between predictors (L, F , S and T ) and responses

(A; �h;�s; �v; P; SC). One significant outlier was identified and removed from the data set, as itappeared that the segmentation failed in the corresponding image (no “live” or “stressed” cal-careous algae were detected). Fig 8 shows the correlation loading plot obtained from the PLS-regression. The outer circle represents 100% explained variance and the inner circle 50%explained variance for the different variables.

SC and P, were both positively and negatively (respectively) correlated with amount of sedi-

ment (S) with excellent goodness of fit and prediction. Also the HSV-variables (�h;�s; �v)extracted with our approach, were correlated with S. Goodness of fit and prediction, the latterobtained after cross validation, are summarized in Table 3. Furthermore, there were inter-cor-relations between the response variables: the HSV-variables were inter-correlated with SC, andinversely correlated with P. The predictor parameters F and L had low influence on the mea-sured responses. Time (T) showed no correlation with P, SC, or HSV, but was well explained

and inversely correlated with relative size (A). This implies that there are two major directionsin the data reflected in the loading plot. The first (horizontal) PLS-component describes S andmeasured P, SC and the HSV-variables. The second PLS-component (vertical) describes T and

A, which are not correlated with the other variables shown in the loading plot.The apparent inverse correlation between photosynthetic efficiency (P) and the HSV-vari-

ables suggests an alternative PLS-regression with HSV-variables as predictors and P asresponse. Fig 9 shows the resulting loading plot obtained after PLS-regression with linear rela-tions. Goodness of fit and prediction were both 0.82, enabling prediction of photosynthetic effi-ciency from HSV-values.

DiscussionThe use of BIIGLE enabled a flexible and swift exchange of data between the researchersinvolved in this interdisciplinary study, independent of their geographical location.

The selected strategy for the manual labeling in this study was to apply an “example based”labeling approach, rather than to fully outline the different calcareous algae regions. The ratio-nale behind the chosen strategy was that color changes were expected to be time and/or treat-ment related and that the use of a “example based” annotation approach (rectangles andpoints) would balance the time spent on labeling and the amount of training data needed. Dueto this approach, no representative ground truth for the negative class, i.e. pixel regions thatshow no calcareous algae, existed. In principle the negative class could have been constructedby a random selection from the unlabeled regions of the images. However this random selec-tion might have introduced some false-negatives to the negative class. This motivated us toapply an unsupervised learning approach. The prototype label mapping in the H2SOM fol-lowed a conservative strategy to classify only 80% of the positive training data as “live”/“stressed” calcareous algae. Despite of this strategy, our segmentation result showed a high

Table 2. Results for the evaluation of the segmentation.

Recall Prec. FPR

“dead“[ “bare” “dead” “bare”

0.89 0.03 0.02 0.02 0.05

The very low values for precision were expected, since a realistic quantification of the FP is not possible in

our dataset (see text for details). The low FPR though indicates that the segmentation generates low FP in

the “dead” or “bare substratum” labeled regions.

Fig 7. Examples of manually labeled rhodoliths (left) and the corresponding segmented rhodoliths(right).White rectangles with a green dot mark “dead” calcareous algae labeled regions and rectangles witha yellow dot mark “stressed” calcareous algae labeled regions. The pixel regions of “live” or “stressed”calcareous algae are marked red.

recall of 0.89. Based on the selected “example based”manual labeling strategy the precision waslow, as we were expecting. The rationale behind the chosen strategy was that color changeswere expected to be time and/or treatment related and that the use of a “example based” anno-tation approach (rectangles and points) would balance the time spent on labeling and theamount of training data needed. A posterior visual inspection of the results showed that theselected approach produced many false positives that actually turned out to be true positives.Computing the precision as defined in Eq 15 was therefore not a reasonable assessment for the

Fig 8. The correlation loading plot obtained after PLS-regression. It was computed with light (L), flow rate (F ), amount of sediment (S, S2),and time (T ) as predictors (blue dots), and HSV-variables (�h; �s; �v), relative size (A), photosynthetic efficiency (P) and sediment coverage (SC) asresponses (red dots). The outer circle represents 100% explained variance and the inner circle 50% explained variance for the differentvariables.

accuracy of our segmentation system (Fig 7). The FPR scores should be low, as an indication ofthe effect, that almost none of the pixels labeled as “dead” or “bare substratum” were seg-mented into “live” or “stressed” calcareous algae regions. The FPR for the “bare substratum” isnoticeable, but was interpreted as still low and does not seem to have a significant effect on theevaluated colors.

The PLS regression showed that the photosynthetic efficiency is correlated (R2 = 0.82) to theHSV-values, implying that it is possible to predict the photosynthetic efficiency from the color(Fig 9). The segmented images indicate that the area of live calcareous algae decreases with

time during the exposure study. However, the correlation of P, SC and the color (�h;�s; �v) withtime was low, as goodness of fit and prediction are 0.39 and 0.35, respectively (Fig 8). Only the

relative size (A) seemed to be associated with time although the correlation was relatively weak.

According to Table 3 goodness of fit and prediction were 0.5 for A as a function of all predictor

variables, including time. Performing linear regression with A as a function of time alone gavea correlation coefficient of R2 = 0.4. The latter low correlation could have been caused by rota-tions of rhodoliths around an axis not perpendicular to the image plane between the differenttime points of the images (Fig 10).

The results of the pre-processing steps show that the illumination fluctuations could bereduced significantly. Furthermore, the reference plate enabled us to achieve color constancyand to correct camera zooming for all images. Although we were able to design a pre-process-ing pipeline to enhance the image quality and to compensate feature shifts caused by the exper-imental set up, some experimental steps can still be improved. For instance, in most of theexcluded images, the color reference plate showed strong reflections due to a suboptimal mate-rial surface and the angle between camera and reference plate. Hence, the circle detection failedor the gamma correction produced wrong corrections if the reflection influenced too many cir-cles. The strong reflections that appeared in the color reference plate may be avoided by select-ing a material that gives minimal reflections. Also using an imaging setup where the locationand angle of the reference plate are fixed and optimized to avoid strong reflections will improvethe image quality. Furthermore using a camera with manual settings enables the use of a fixedwhite balance that reduces possible fluctuations in color. The necessity to apply illuminationcorrection could maybe be avoided using an optimized stable light setup system with artificiallight sources only. Reference plate and sample should be illuminated evenly throughout thewhole image. By experience even in laboratory studies these conditions are often hard toachieve throughout the whole experiment. The presented semi-automated software solutioncan be used to evaluate the images even if not recorded under the optimized conditions, asshown here.

At this state the whole automated segmentation progress takes about 4 minutes for eachimage, where most of the time is spent on the pre-processing stage. As shown by [31] a signifi-cant saving of processing time is achievable by code optimizations and the use general-purpose

Table 3. Goodness of fit and prediction of the PLS-regression.

R2 h s v bA P SC

fit 0.73 0.64 0.76 0.52 0.93 0.93

pred. 0.72 0.62 0.74 0.50 0.93 0.93

Goodness of fit and prediction for the response variables as function of the four predictor variables L, F , S

and T .

computing on graphics processing units (GPGPU). A similar acceleration can be expected forour approach, as no code optimization and no GPGPU has been applied for this study.

In comparison to this laboratory study, the analysis of underwater images recorded in thefield is even more challenging for the issues discussed above as the levels of the inherent opticalproperties (IOP) will affect the image quality. Various concentrations of phytoplankton (Chla), colored dissolved organic matter (cDOM) and total suspended matter (TSM) will alter the

Fig 9. The correlation loading plot for the photosynthetic efficiency (P). It was obtained after PLS-regression with the HSV-variables(�h; �s; �v) as predictors (blue dots), and photosynthetic efficiency (P) as response (red dot). The outer circle represents 100% explainedvariance and the inner circle 50% explained variance for the different variables. The plot showed that the predictor variables are highlycorrelated to the response variable. A prediction of the photosynthetic efficiency using the HSV-values is therefore possible for ourexperiment.

contrast, sharpness and colors of objects due to absorption and/or scattering of light [42, 43].In shallow waters, (0–20 m depth) reflection from the seafloor and varying light conditionsrelated to the cloudiness and flickering effects (focusing/defocusing due to wave effects) wouldadd even more challenges that should be considered [44]. Therefore, it must be expected thatimage (pre-)processing will be needed when operating under field conditions. As both the cal-careous algae and sediments at the Peregrino field moves quite extensively (unpublishedresults), the monitoring should focus on measurements of possible color change of the seg-mented (visual) calcareous algae present in the image over time.

Furthermore, we believe that this methodological approach has the potential to be appliedto studies with other species where color change is a stress indicator, e.g. tropical corals [7].However, as individual species have species specific features (color, structure, shape, etc.) andtheir surroundings differ, species specific adjustment for computational analysis may berequired.

ConclusionThis pilot study has shown that impact of sedimentation on calcareous algae samples can bedetected by the presented computational approach for automatic size and color measurementfrom photos.

In addition to the previously published correlation between photosynthetic efficiency andsediment coverage [6], the present multivariate regression showed correlations between timeand calcareous algae size, as well as correlations between fluorescence and calcareous algaecolor. Furthermore, we have shown that the use of an unsupervised learning algorithm is apromising attempt to create a classifier out of a sparsely labeled dataset. The chosen approachof an automatic assignment of the H2SOM prototypes was successful. The results of this pilot

Fig 10. The relative position of the sample can have an effect on the segmentation result. Both images are displaying the same calcareous algaesample, but were recorded at different time points during the experiment. The calcareous algae sample was rotated around an axis not perpendicular to theimage plane.

study will enable researchers to conduct more studies in the future with different parameteriza-tions and samplings to improve calcareous algae stress models.

The observed correlation between photosynthetic efficiency and the HSV-values is highlyinteresting for field studies, since photosynthetic efficiency measurements are impractical toperform manually in deeper waters. Color measurement in images has therefore the potentialto play an important role in future field calcareous algae monitoring systems.

Supporting InformationS1 Text.(PDF)

S2 Text.(PDF)

S3 Text.(PDF)

S1 Table.(CSV)

AcknowledgmentsWe are grateful to the Chico Mendes Institute for Biodiversity Conservation for permitting thecollection of the samples. The Instituto de Pesquisas Almirante Paulo Moreira offered their lab-oratory facilities to run the mesocosm trial with the support of Ricardo Coutinho, Carlos Gus-tavo Ferreira and Fernanda Siviero.

Author ContributionsConceived and designed the experiments: JO IN IE MAOF FTST TWN. Performed the experi-ments: JO IN IE MAOF FTST TWN. Analyzed the data: JO TWN IE IN. Contributed reagents/materials/analysis tools: JO IN IE MAOF FTST TWN. Wrote the paper: JO IN IE MAOF FTSTTWN.

References1. Nilssen I,Ø degård Oy, Sø rensen AJ, Johnsen G, Moline MA, Berge Jr. Integrated environmental map-

ping and monitoring, a methodological approach to optimise knowledge gathering and sampling strat-egy. Marine Pollution Bulletin. 2015; 96(1–2):374–383. Available from: http://www.sciencedirect.com/science/article/pii/S0025326X15002477. doi: 10.1016/j.marpolbul.2015.04.045 PMID: 25956441

2. Buhl-Mortensen Pl, Tenningen E, Tysseland ABS. Effects of water flow and drilling waste exposure onpolyp behaviour in Lophelia pertusa. Marine Biology Research. 2015; 11(7):1–13. Available from:http://www.tandfonline.com/doi/abs/10.1080/17451000.2014.993651. doi: 10.1080/17451000.2014.993651

3. Purser A. A Time Series Study of Lophelia pertusa and Reef Megafauna Responses to Drill CuttingsExposure on the Norwegian Margin. Plos One. 2015; 10(7):e0134076. Available from: http://dx.plos.org/10.1371/journal.pone.0134076. doi: 10.1371/journal.pone.0134076 PMID: 26218658

4. Purser a, Ontrup J, Schoening T, Thomsen L, Tong R, Unnithan V, et al. Microhabitat and shrimp abun-dance within a Norwegian cold-water coral ecosystem. Biogeosciences. 2013; 10(9):5779–5791. doi:10.5194/bg-10-5779-2013

5. Villas-Bôas AB, Tâmega FTS, Andrade M, Coutinho R, Figueiredo MAO. Experimental Effects of Sedi-ment Burial and Light Attenuation on Two Coralline Algae of a DeepWater Rhodolith Bed in Rio deJaneiro, Brazil. Cryptogamie, Algologie. 2014; 35(1):67–76. Available from: http://www.bioone.org/doi/abs/10.7872/crya.v35.iss1.2014.67. doi: 10.7872/crya.v35.iss1.2014.67

6. Figueiredo MAO, Eide I, Reynier M, Villas-Bôas AB, Tâmega FTS, Ferreira CG, et al. The effect of sedi-ment mimicking drill cuttings on deep water rhodoliths in a flow-through system: Experimental work andmodeling. Marine Pollution Bulletin. 2015; 95(1):81–88. Available from: http://www.sciencedirect.com/science/article/pii/S0025326X15002428. doi: 10.1016/j.marpolbul.2015.04.040 PMID: 25935812

7. Siebeck UE, Marshall NJ, Klüter a, Hoegh-Guldberg O. Monitoring coral bleaching using a colour refer-ence card. Coral Reefs. 2006; 25(3):453–460. doi: 10.1007/s00338-006-0123-8

8. Hall-Spencer JM. Conservation issues relating to maerl beds as habitats for molluscs. Journal of Con-chology. 1998; 2(2):271–286.

9. Steller DL, Riosmena-rodríguez R, Foster MS. 6 Living Rhodolith Bed Ecosystems in the Gulf of Califor-nia. Ecosystems. 2001;(1944: ):72–82.

10. Riosmena-Rodríguez R, Steller D, Hinojosa-Arango G, Foster M. Reefs that rock and roll: Biology andconservation of rhodolith beds in the Gulf of California. The Gulf of California: Biodiversity and Conser-vation. 2010;p. 49–71.

11. Tâmega FTS, Oliveira PS, Figueiredo MAO. Catalogue of the Benthic Marine Life for Peregrino OilField. Campos Basin, Brazil Instituto Biodiversidade Marinha, Rio de Janeiro. 2013;p. 140.

12. Basso D. Deep rhodolith distribution in the Pontian Islands, Italy: A model for the paleoecology of a tem-perate sea. Palaeogeography, Palaeoclimatology, Palaeoecology. 1998; 137(1–2):173–187. doi: 10.1016/S0031-0182(97)00099-0

13. Henriques MC, Riosmena-Rodríguez R, Coutinho L, Figueiredo MAO Lithophylloideae and Mastophor-oideae (Corallinales, Rhodophyta) from the Brazilian continental shelf. Phytotaxa. 2014; 190(1):112–129. Available from: http://biotaxa.org/Phytotaxa/article/view/phytotaxa.190.1.9. doi: 10.11646/phytotaxa.190.1.9

14. Littler MM, Littler DS, Hanisak MD. Deep-water rhodolith distribution, productivity, and growth history atsites of formation and subsequent degradation. Journal of Experimental Marine Biology and Ecology.1991; 150(2):163–182. Available from: http://www.sciencedirect.com/science/article/pii/0022098191900666. doi: 10.1016/0022-0981(91)90066-6

15. Littler MM, Littler DS, Blair SM, Norris JN. Deep-water plant communities from an uncharted seamountoff San Salvador Island, Bahamas: distribution, abundance, and primary productivity. Deep SeaResearch Part A Oceanographic Research Papers. 1986; 33(7):881–892. Available from: http://www.sciencedirect.com/science/article/pii/0198014986900038. doi: 10.1016/0198-0149(86)90003-8

16. Kempf M. Notes on the benthic bionomy of the N-NE Brazilian shelf. Marine Biology. 1970; 5(3):213–224. doi: 10.1007/BF00346909

17. Foster MS. Rhodoliths: Between rocks and soft places. Journal of Phycology. 2001; 37(5):659–667.doi: 10.1046/j.1529-8817.2001.00195.x

18. Amado-Filho GM, Moura RL, Bastos AC, Salgado LT, Sumida PY, Guth AZ, et al. Rhodolith beds aremajor CaCO 3 BIO-factories in the tropical southWest Atlantic. PLoS ONE. 2012; 7(4):e35171. doi: 10.1371/journal.pone.0035171 PMID: 22536356

19. NelsonW. Calcified macroalgae–critical to coastal ecosystems and vulnerable to change: a review.Marine and Freshwater Research. 2009; 60(8):787–801. Available from: http://www.publish.csiro.au/?paper=MF08335. doi: 10.1071/MF08335

20. Davies AJ, Roberts JM, Hall-Spencer J. Preserving deep-sea natural heritage: Emerging issues in off-shore conservation and management. Biological Conservation. 2007; 138(3–4):299–312. doi: 10.1016/j.biocon.2007.05.011

21. Schoening T, Osterloff J, Nattkemper TW. RecoMIA—Recommendations for Marine Image Annotation:Lessons Learned and Future Directions. Frontiers in Marine Science. 2016; 3(59). Available from:http://www.frontiersin.org/deep-sea_environments_and_ecology/10.3389/fmars.2016.00059/abstract.

22. Kohler KE, Gill SM. Coral Point Count with Excel extensions (CPCe): A Visual Basic program for thedetermination of coral and substrate coverage using random point count methodology. Computers andGeosciences. 2006; 32(9):1259–1269. doi: 10.1016/j.cageo.2005.11.009

23. Ontrup J, Ehnert N, BergmannM, Nattkemper TW. Biigle—Web 2.0 enabled labelling and exploring ofimages from the Arctic deep-sea observatory HAUSGARTEN. In: OCEANS’09 IEEE Bremen: Balanc-ing Technology with Future Needs. IEEE; 2009. p. 1–7.

24. Purser A, Bergmann M, Lundälv T, Ontrup J, Nattkemper TW. Use of machine-learning algorithms forthe automated detection of cold-water coral habitats: A pilot study. Marine Ecology Progress Series.2009; 397:241–251. doi: 10.3354/meps08154

25. BergmannM, Langwald N, Ontrup J, Soltwedel T, Schewe I, Klages M, et al. Megafaunal assemblagesfrom two shelf stations west of Svalbard. Marine Biology Research. 2011; 7(6):525–539. doi: 10.1080/17451000.2010.535834

26. Schoening T, BergmannM, Ontrup J, Taylor J, Dannheim J, Gutt J, et al. Semi-automated image analy-sis for the assessment of megafaunal densities at the Artic deep-sea observatory HAUSGARTEN.PLoS ONE. 2012; 7(6):e38179. doi: 10.1371/journal.pone.0038179 PMID: 22719868

27. BergmannM, Klages M. Increase of litter at the Arctic deep-sea observatory HAUSGARTEN. MarinePollution Bulletin. 2012; 64(12):2734–2741. doi: 10.1016/j.marpolbul.2012.09.018 PMID: 23083926

28. Osterloff J, Nilssen I, Nattkemper TW. A computer vision approach for monitoring the spatial and tem-poral shrimp distribution at the LoVe observatory. Methods in Oceanography. 2016;p. –. Availablefrom: http://www.sciencedirect.com/science/article/pii/S2211122015300062. doi: 10.1016/j.mio.2016.03.002

29. BeijbomO, Edmunds PJ, Roelfsema C, Smith J, Kline DI, Neal BP, et al. Towards Automated Annota-tion of Benthic Survey Images: Variability of Human Experts and Operational Modes of Automation.PLoS ONE. 2015; 10(7):e0130312. doi: 10.1371/journal.pone.0130312 PMID: 26154157

30. Ontrup J, Ritter H. Hyperbolic Self-Organizing Maps for Semantic Navigation. Advances in Neural Infor-mation Processing Systems 14. 2002; 14(14):1417–1424.

31. Schoening T, Langenkämper D, Steinbrink B, Brün D, Nattkemper TW. Rapid image processing andclassification in underwater exploration using advanced high performance computing. In: OCEANS2015—MTS/IEEE Washington; 2015. p. 1–5.

32. Schoening T, Kuhn T, Jones DOB, Simon-Lledo E, Nattkemper TW. Fully automated image segmenta-tion for benthic resource assessment of poly-metallic nodules. Methods in Oceanography. 2016;p. –.Available from: http://www.sciencedirect.com/science/article/pii/S2211122015300104. doi: 10.1016/j.mio.2016.04.002

33. Box GEP, Hunter WG, Hunter JS. Statistics for Experimenters: An Introduction to Design, Data Analy-sis, and Model Building. Wiley Series in Probability and Mathematical Statistics. 1978; 1st:653.

34. Foster DH. Color constancy. vol. 51 of TheWiley-IS&T Series in Imaging Science and Technology.Wiley; 2010.

35. Stokes, M, Anderson, M, Chandrasekar, S, Motta, R. A standard default color space for the internet-srgb. In: Microsoft and Hewlett-Packard Joint Report. Society for Imaging Science and Technology;1996. p. 1–17. Available from: http://fp.optics.arizona.edu/opti588/reading/Stokes_sRGB_colorspace_detail.pdf.

36. Yuen H, Princen J, Illingworth J, Kittler J. Comparative study of Hough Transformmethods for circlefinding. Image and Vision Computing. 1990; 8(1):71–77. doi: 10.1016/0262-8856(90)90059-E

37. Gonzalez RC, Woods RE. Digital Image Processing. 3rd ed. Upper Saddle River, NJ: Pearson Pren-tice Hall; 2007.

38. Wold S, Martens H, Wold H. The multivariate calibration problem in chemistry solved by the PLSmethod. In: Matrix Pencils. vol. 973. Springer; 2012. p. 286–293. Available from: http://dx.doi.org/10.1007%2fBFb0062108.

39. Forina M, Lanteri S, Casale M. Multivariate calibration. vol. 1158. JohnWiley & Sons; 2007.

40. Wold S. Cross-Validatory Estimation of the Number of in Factor and Principal Components Compo-nents Models. Technometrics. 1978; 20(4):397–405. doi: 10.1080/00401706.1978.10489693

41. Esbensen K. Multivariate Data Analysis—In Practice: An Introduction to Multivariate Data Analysis andExperimental Design. 5th ed. Camo Process AS; 2002.

42. Ioccg. Remote Sensing of Ocean Colour in Coastal, and Other Optically-ComplexWaters. In: Reportsof the International Ocean Colour Coordinating Group 3; 2000. p. 144.

43. Johnsen G, Volent Z, Sakshaug E, Sigemes F, Pettersson LH. 6. Remote sensing in the Barents Sea.In: Sakshaug E, Johnsen GH, Kovacs KM, editors. Ecosystem Barents Sea. Tapir Academic Press;2009. p. 139–166.

44. Shihavuddin A, Gracias N, Garcia R. Online Sunflicker Removal using Dynamic Texture Prediction. In:VISAPP (1); 2012. p. 161–167.

Paper VI

Eide, I., Westad, F., Nilssen, I., de Freitas, F. S., dos Santos, N. G., dos Santos, F., Cabral, M. M., Bicego, M. C., Figueira, R., Johnsen, S., 2016. Integrated environmental

monitoring and multivariate data analysis – A case study. Integrated Environmental Assessment and Management. Doi: http://dx.doi.org/10.1002/ieam.1840

Integrated Environmental Monitoring and MultivariateData Analysis—A Case StudyIngvar Eide,*y Frank Westad,z Ingunn Nilssen,y§ Felipe Sales de Freitas,k Natalia Gomes dos Santos,#Francisco dos Santos,# Marcelo Montenegro Cabral,# Marcia Caruso Bicego,k Rubens Figueira,kand Ståle Johnseny

yStatoil ASA, Research Centre, Trondheim, NorwayzCAMO Software AS, Oslo, Norway§Department of Biology, Trondhjem Biological Station, Norwegian University of Science and Technology, Trondheim, NorwaykOceanographic Institute, University of S~ao Paulo, PraSca do Oceanogr�afico, S~ao Paulo, Brazil#PROOCEANO ServiSco Oceanogr�afico, Rio de Janeiro, Brazil

(Submitted 5 January 2016; Returned for Revision 11 March 2016; Accepted 28 July 2016)

ABSTRACTThe present article describes integration of environmental monitoring and discharge data and interpretation using

multivariate statistics, principal component analysis (PCA), and partial least squares (PLS) regression. The monitoring wascarried out at the Peregrino oil field off the coast of Brazil. One sensor platform and 3 sediment traps were placed on theseabed. The sensors measured current speed and direction, turbidity, temperature, and conductivity. The sediment trapsamples were used to determine suspended particulate matter that was characterized with respect to a number of chemicalparameters (26 alkanes, 16 PAHs, N, C, calcium carbonate, and Ba). Data on discharges of drill cuttings and water-baseddrilling fluid were provided on a daily basis. The monitoring was carried out during 7 campaigns from June 2010 toOctober 2012, each lasting 2 to 3 months due to the capacity of the sediment traps. The data from the campaigns werepreprocessed, combined, and interpreted using multivariate statistics. No systematic difference could be observed betweencampaigns or traps despite the fact that the first campaign was carried out before drilling, and 1 of 3 sediment traps waslocated in an area not expected to be influenced by the discharges. There was a strong covariation between suspendedparticulatematter and total N and organic C suggesting that themajority of the sediment samples had a natural and biogenicorigin. Furthermore, the multivariate regression showed no correlation between discharges of drill cuttings and sedimenttrap or turbidity data taking current speed and direction into consideration. Because of this lack of correlation withdischarges from the drilling location, a more detailed evaluation of chemical indicators providing information aboutorigin was carried out in addition to numerical modeling of dispersion and deposition. The chemical indicators and themodeling of dispersion and deposition support the conclusions from themultivariate statistics. Integr Environ Assess Manag2016;X:000–000. © 2016 SETAC

Keywords: Drill cuttings Principal component analysis (PCA) Peregrino Partial least squares regression Sediment traps

INTRODUCTIONNew technology has led to new opportunities for a holistic

and integrated environmental monitoring approach in offshorepetroleumexploration andproductionoperations. It has also ledto new challenges associated with the interpretation ofvast amounts data (Nilssen, Ødegård et al. 2015). Sensor-based environmental monitoring has during the past years beenused as alternative or supplement to conventional sampling-based monitoring campaigns (Godø et al. 2014; Bagley et al.2015). An integrated environmental monitoring approachcomprises selection of parameters, sensors, sensor platforms,data collection, storage and interpretation, the latter requiresadvanced analytical methodology (Nilssen, Ødegård et al.2015). Multivariate data analysis has successfully been used inseveral marine environmental studies for the interpretation of

large data sets, for example on heavy metals and biomarkerresponse (Rodr�ıguez-Ortega et al. 2009), on spatial andtemporal distributionof heavymetals in seawater and sediments(Lin et al. 2013), on element release from sediments (Mu~nozet al. 2015), on PAH in marine and freshwater sediments (Neffet al. 2005; Eide et al. 2011), and to predict toxicity fromsediment chemistry (Alvarez-Guerra et al. 2010).

The present work demonstrates the use of multivariate dataanalysis within the integrated environmental mapping andmonitoring concept using data from the Peregrino oil field inBrazil as a case study. Multivariate data analysis was used tointerpret the large data sets comprising oceanographic data,sediment trap data, chemical and physical environmentalparameters, and also discharge data from offshore drillingobtained during long-term environmental monitoring. Becauseof the variability of the sampling frequency and the number ofvariables per sensor system, routines for preprocessing andcombining data were established. Principal component analysis(PCA) was used as an exploratory approach to evaluatesimilarities between campaigns and samples (Jackson 1991).Partial least squares (PLS) regression was used to revealcorrelation between predictors and responses (Wold et al.

This article includes online-only Supplemental Data.

* Address correspondence to ieide@statoil.com

Published online 8 August 2016 in Wiley Online Library

(wileyonlinelibrary.com).

DOI: 10.1002/ieam.1840

Enviro

talMan

1983; Martens and Næs 1993). The results from the multivari-ate data analysis were critically validated using statisticalmethods. As an additional evaluation of the results, chemicalindicators were interpreted and dispersion and depositionmodeling were carried out.

MATERIALS AND METHODS

The Peregrino oil field

The Peregrino oil field is located 80 kmoff the coast of Brazil,southeast of Cabo Frio, in the Campos Basin area. The fieldconsists of 2 fixedwell head platforms namedA and B (WHPAand B) for drilling of production and water injection wells, and1 floating production storage and offloading unit (FPSO). Amap and a table with positions are available in theSupplemental Data (Figure 6 gives an overview). Drilling atPeregrino began November 9, 2010 and the field came inproduction April 9, 2011. The water depth in the area variesfrom 90 to 120m, and seabed sediments consist, in general, ofsand and silt with a relatively hard surface (Salgado et al. 2010).The shallower part of the Peregrino seabed hosts a calcareous

algae community (Salgado et al. 2010; Figueiredo et al. 2015).Calcareous algae are regarded as important for the ecosystem,as they create a habitat for other species living between, on, andin the calcareous algae structures (rhodoliths) (Basso 1998).These communities can be disturbed by natural sedimentationand anthropogenic activities, for example discharges of drillcuttings (Villas-Boas et al. 2014; Figueiredo et al. 2015). Thedrilling fluid used at Peregrino is water-based, which generallyis non- or low-toxic (Bakke et al. 2013). The concern related tothe discharge of drill cuttings is related to the physical effect ofburial and reduced light and gas exchange (Figueiredo et al.2015). Because of the prevailing currents fromWHP B towardthe calcareous algae bank (Nilssen, Santos et al. 2015), 7 fieldcampaigns during 2010 to 2012 were carried out to monitor ifdrill cuttings could be detected in the vicinity of the Peregrinocalcareous algae community.

Sediment traps and lander campaigns

One sensor platform (lander) and 3 sediment traps wereplaced in predefined positions on the seabed (Figure 6, detailsin the Supplemental Data). The first of the 7 campaigns wascarried out before drilling and thus represents baseline(Table 1). The prevailing current direction in the area istoward the southwest from WHP B toward traps 1 and 3 andthe lander (Nilssen, Santos et al. 2015). Trap 2 was located inan area not expected to be influenced by the discharges. Themajor area with calcareous algae is located approximately1.3 km south-west of WHP B (see the map in the Supplemen-tal Data).The sediment traps and the lander aredescribed inmoredetail

by Salgado et al. (2010) and Johnsen et al. (2014). The funnel ofthe sediment traps was 10m above the seabed. Each trapcaptured 20 samples of sediment during periods of 3 to 4.5d(fixed time intervals within each campaign) implying that eachsediment trap campaign lasted for approximately 60 to 90d.After completing the 20 samples, the traps were retrieved, andthe sediment samples were analyzed in the laboratory.In addition to the sediment traps, data were also available for

campaigns 5 and 7 from the lander located roughly betweentraps 1 and 3. The lander was equipped with sensors providingdata on current speed (CS) and current direction (CD) at 10depths (every 10m) measured every 10min. Temperature,

pressure, conductivity, and turbidity were measured every5min. Details on sensors are available in the SupplementalData. Data on discharges of water-based drilling fluid andcuttings were provided on a daily basis for WHP B for the 5thand the 7th campaign. The discharges were directed verticallydownward 50m below sea surface.

Analyses of sediment trap samples

The sediment trap samples were used to determinesuspended particulate matter (SuspPM) and a number ofchemical compounds as briefly described below. The detailsare available in the Supplemental Data.Analyses of 26 individual alkanes (n-C12 to n-C35 and

pristane and phytane) and 16 PAHs (US EnvironmentalProtection Agency [USEPA] priority list) were carried out inaccordance with UNEP (1992) with minor modifications.PAH data were not available for the first campaign; themethodology was not sufficiently sensitive but was improvedbefore the subsequent campaigns.Terrestrial–aquatic ratio (TAR) was determined according

to Bourbonniere andMeyers (1996). Carbon preference index(CPI) was calculated to evaluate the possible petroleum sourceof n-alkanes (Aboul-Kassim and Simoneit 1996). Isomericratios for different PAH were calculated according to Yunkeret al. (2002).Carbon (12C and 13C) and N (14N and 15N) analyses were

carried out as described byThornton andMcManus (1994) andZhou et al. (2006) with a few modifications. The analysesprovided data on organic C (Corg), total N (Ntot), organic Cand N as percentages of SuspPM (CORG% andNTOT%), andbulk C to N (C:N). Nitrogen (14N and 15N) and organic C(12C and 13C) were used to calculate the bulk organic proxiesd15N and d13C (Meyers 2003).CaCO3 were determined gravimetrically (campaign 2-7).

Barium analysis was based on the procedure described in theSW 846 (USEPA 1996).

Preprocessing

As the variables were recorded at different time intervals,the data were preprocessed in various ways to match thevariables with the lowest temporal resolution before themultivariate analysis. For multivariate data analysis includingsediment trap data, the averages for the corresponding timeintervals (3–4.5 d) were also calculated for the other param-eters. Models with turbidity, temperature, conductivity, andcurrentwere based on common interpolation of data to 10-minintervals. The data for CS and CD at 10 different depths wereused in different ways in the analyses

� As CD (angle relative to north) and CS (correspondinghorizontal speed)

� As “effective current” (CD_CS) calculated from CS andCD according to the following procedure: calculation ofthe angle (in degrees) between CD and the direction fromWHP B to the lander. This angle will be referred to as“relative current angle.” The “effective current” is obtainedbymultiplying CSwith cosine of the relative current angle.A histogram for the frequency of effective current wascalculated from all the individual observations for the timeintervals corresponding to the sediment trap data.

� As score vectors obtained from PCA based on CS, CD, andeffective current histograms (PCA described in nextsection)

2 Integr Environ Assess Manag 9999, 2016—I Eide et al.

The rationale behind “effective current” is that dischargedparticles are assumed to be transported in the direction of thelander and sediment traps 1 and 3. Using the average currentspeed, it takes approximately 6 h for discharged drillingparticles to travel 1436m from WHP B to the lander, andthis delay was used when matching the discharge data to thelander data before the data analysis. This delaywas also used forsediment trap 3 and was considered sufficiently accuratebecause sampling periods were 3 to 4.5 d. With this approach,any perpendicular current direction will not contribute to theeffective current. The direction from WHP B to the lander is221° (north is 0°). The dominating current direction wasbetween 180° and 240°, confirming that the lander was locateddownstream of WHP B.

Multivariate data analysis

PCA (Jackson 1991) was used to evaluate similarities anddifferences between samples and for tracking changes, forexample between campaigns, traps, and samples. This isachieved by evaluating all variables simultaneously, obtainingthe structured information in the data, and visualizing theresult in score and loading plots. Multivariate regression wascarried out with PLS (Wold et al. 1983; Martens and Næs1993) because it overcomes the problem of intercorrelatedpredictor variables and also takes into account relationshipsbetween response variables.

The sensors for turbidity and current were placed on thelander, and PLS-regression was done in an attempt to relateturbidity to CS and CD. As the lowest sampling frequency forthese variables was 10min, turbidity was also calculated every10min after interpolation. A total of 8267 observations forcampaign 5 were included in the model.

PLS-regression was also carried out for the 5th and the 7thcampaigns using discharge data and score vectors fromindividual PCA models for CS, CD, and effective currenthistograms, CD_CS (described in previous section) aspredictors, and SuspPM, Ba, and turbidity as responsevariables. The PCA models were calculated independentlyfor campaigns 5 and 7 implying 6 individual PCA models (notshown) providing score vectors for the PLS model. The scorevector approach was chosen because the high number ofvariables would make the interpretation difficult and the plotsoverloaded. Furthermore, the score vectors represent trends incurrent speed as function of depth. All data were adjusted tothe sediment trap sampling intervals. The score vectors were

combined with individual parameters in a multiblock model.All score values are available in the Supplemental Data.

PCA and PLSwere carried out with the Unscrambler X 10.3(CAMO Software, Oslo, Norway). Before the multivariatedata analysis, the data were mean-centered and scaled to unitvariance. The data were not divided into training and test setbecause the objective for the analyses was to visualize anydifferences due to trap or campaign; therefore, all sampleswere used for the analyses. The models were validated withrespect to explained variance and goodness of fit andprediction, the latter obtained after cross-validation (Wold1978) carried out with 20 random segments.

Dispersion and deposition modeling

The numerical simulations of the horizontal and verticaldispersion in the water column and the deposition of thedrilling discharges on the seabed during campaigns 5 and 7were carried out using the dose-related risk and effectassessment model (DREAM) (Rye et al. 2008; Singsaas et al.2008; Durgut et al. 2015). DREAM uses discharge character-istics and environmental data such as current, wind, etc. topredict the behavior of the discharge plumes in the watercolumn and the deposition on the seabed. Particle sizedistribution of drill cuttings from the Peregrino oil field wasobtained from sieving analyses. Volumes of discharged drillcuttings and drilling fluids with a temporal resolution of 24 hwere obtained from the platform drilling logs. The totalamount of solids in the water-based drilling fluid wascalculated in accordance with the bulk density of the drillingfluid through the given drilling operation period at Peregrino.Because of the temporal 24-h resolution of the discharge log,themodelwas run in daily sequences, enabling a daily overviewof the accumulation of the deposition. More details on thenumericalmodeling is described inNilssen, Santos et al. (2015)who modeled deposition from WHP B over a 2-y period.

RESULTS AND DISCUSSION

PCA on data on alkanes and PAHs from all sediment trapsamples (all campaigns)

An initial PCA was carried out to evaluate grouping andcovariance of the individual hydrocarbons. PAH data were notavailable for the first campaign and the PCAwas carried out forcampaigns 2 to 7. The correlation loading plot in Figure 1shows 2 groups of alkanes, and all alkanes except pristane and

Table 1. Overview over sediment traps and lander campaigns

Campaign Period Trap 1 Trap 2 Trap 3 Lander

1 19.06.2010–21.08.2010 X X X

2 28.10.2010–16.01.2011 X X X

3 07.04.2011–26.06.2011 X

4 16.08.2011–17.11.2011 X X

5 26.11.2011–21.02.2012 X X X

6 22.03.2012–17.06.2012 X X

7 01.07.2012–01.10.2012 X X X

Drilling began November 9, 2010, and the field came in production April 9, 2011.

IEM and Multivariate Data Analysis—Integr Environ Assess Manag 9999, 2016 3

n-C35 vary systematically and arewell explained. PC-1 and PC-2 explain 31% and 20% of the variation, respectively. PC-3 andPC-4 explain 11% and 6%, respectively. For PC-5 and onward,no individual variable is explained to more than 15%. Thevariance in PC-3 is, in essence, due to 1 sample from sedimenttrap 2, campaign 6, and the corresponding correlation loadingplot shows that some of the heavier PAH span this dimension.The Hotelling’s T2 statistic indicated this sample as an outlier.PC-4 is explained by fluorene and phenanthrene due to highvalues in a few samples in sediment trap 2.The alkanes can be grouped into low- and high-molecular

weight alkanes (LMWHC andHMWHC). The low-molecularcomprises the n-alkanes C12 to C22 plus C24, pristane (C19),and phytane (C20). The high-molecular comprises the n-alkanes C23 and C25 to C35. Although C24 has a highermolecular weight than C23, it is grouped with LMWHC andC23 is groupedwith theHMWHC.Wehave no explanation forthis. However, the 2 groups of alkanes are summarizedaccording to their grouping in the loading plot before thesecond PCA that also included the other sediment trap data.The loadings for the PAHs are located close to the middle of

the plot that means there is little systematic variation amongthe samples. According to molecular weight, there is nosystematic grouping (Figure 1). Hence, the sum of all PAHswas used for the second PCA. Descriptive statistics of thehydrocarbons are presented in the Supplemental Data.

PCA on data from all sediment trap samples, all campaigns

The second PCA which was carried out on data from the3 sediment traps (all 7 campaigns), did not include dataon CaCO3 and PAH because these data were not availablefor campaign 1. However, a third PCA was carried out forcampaigns 2–7 including CaCO3 and sum PAH. The majorconclusion from the score plot (Figure 2A) from the secondPCA is that most of the 298 samples from all trapsand campaigns are clustered within the Hotelling’s T2 ellipsedescribing the distance to the model center as spanned by theprincipal components. This implies that most of the samplesare relatively similar both qualitatively and quantitatively, andthere are no systematic differences between the differentcampaigns or traps. However, there are a number of individual

samples that are significantly different from the majoritywithin theHotelling’s T2 ellipse, themost extreme samples areST3-7-14 and ST3-7-21 (Figure 2A).

By superimposing the loading plot on the score plot, it ispossible to identify the variables that are responsible fordifferences between samples. ST3-7-14, ST3-7-21, and a fewother samples have a very high content of SuspPM, Ntot, andCorg. The fact thatNtot andCorg are so closely intercorrelatedwith SuspPM strongly suggests sedimentation of a naturaland biogenic origin (Premuzic et al. 1982; Deyme et al. 2011).Furthermore, NTOT % and CORG% do not vary much, or atleast unsystematically, and are not correlated with SuspPM,Ntot, or Corg. d13C, d15N, and C:N as well as TAR and CPIshow low or at least no systematic variability, and they are notcorrelated with SuspPM. PC-1 and PC-2 explain 23% and 15%of the variation in the sediment data, respectively (Figure 2).The reason for the relatively low total explained variation isthat only 3 of the parameters are well explained (SuspPM,Ntot, and Corg).

A PCA without the 2 extreme samples (ST3-7-14 andST3-7-21) gives essentially the same conclusions, i.e. the sameexplained variance and correlation between variables (scoreand loading plots not shown). However, Ba and HMWHC arebetter explained but are not correlated with SuspPM. Barium,

Figure 1. Correlation loading plot obtained after PCA on all data on 26alkanes and 16 PAHs from the sediment traps, campaigns 2–7. The outer circlerepresents 100% explained variance and the inner circle 50% explainedvariance. The full names of Benz[g,h,i and Inden[1,2 are Benz[g,h,i]peryleneand Inden[1,2,3-c,d]pyrene, respectively.

Figure 2. Score plot (A) with Hotelling's T2 ellipse and correlation loading plot(B) obtained after PCA on all data from the 3 sediment traps, all 7 campaigns(data on CaCO3 and PAH not included because they were not available for thefirst campaign). The alkanes were summarized into LMWHC and HMWHCbased on the initial PCA (Figure 1). The notation is ST1–ST3 for the 3 sedimenttraps, followed by 1–7 for each of the 7 campaigns, and the last digits aresample number (20 samples in each trap each campaign; some samples had tobe combined to provide sufficient sample material for the analyses). Legendshows symbols and colors for each of the 7 campaigns. The intercorrelationsbetween Ntot, Corg, and SuspPM are illustrated by the blue circle. Theapproximate location of PAH after PCA of data from campaigns 2–7 isindicated.

which is an indicator of drilling activity, is high in a fewsamples, particularly in ST1 and ST3 from the 2nd campaignand also in 1 sample from ST2, 5th campaign. A few samplesfrom ST2 2nd campaign (samples 3–10) have a relatively highcontent of HMWHC. The elevated concentrations cannot beexplained by drilling discharges from WHP B because ST2 islocated 4 km away from WHP B and in a direction notexpected to be influenced by discharges from the platform.

The PCA for campaigns 2 to 7, including CaCO3 and sumPAH (missing for campaign 1), does not change Figure 2 or theconclusions (not shown, however, the approximate location ofPAH after PCA of data from campaigns 2–7 is indicated in theloading plot in Figure 2).

CPI and TAR values support that the sedimentation is of anatural and biogenic origin (Eglinton et al. 1962; Volkmanet al. 1992; Meyers 1997; Hostettler et al. 1999). The TARvalues generally indicate terrestrial organic matter (TAR >1).At some periods, lower TAR values were observed (TAR <1)due to an increase in the relative contribution of nC15 and nC17

(algae-derived n-alkanes). As a result, TAR values decreaseindicating an increased relative contribution of marine organicmatter on SuspPM.

CPI and the general predominance of odd-to-even C-numbered n-alkanes indicate terrestrial sources of n-alkanes,possibly due to riverine input (Eglinton et al. 1962; Biet al. 2005; Zhang et al. 2006); however, the source couldalso be maritime traffic (Deyme et al. 2011). n-Alkanesoriginating frompetroleumusually show awideC chain lengthdistribution but no predominance of odd-to-even C number(Volkman et al. 1992).

PAH isomeric ratios (Yunker et al. 2002) are shown in thecross-plots in Figure 3. PAH sources are most likely related topyrogenic sources, based on BaA/228, IP/IPþBghiP, andFl/FlþPy values. Nevertheless, there is contribution fromdirect input of petrogenic sources, as shown by Ant:178 ratios.Therefore, there are mixed sources of PAH to SuspPM atthe Peregrino oil field. The relative abundance of LWMPAHand HMWPAH, with relative predominance of HMWPAH,confirms the mixed sources of PAH in this area (Neff et al.2005; Deyme et al. 2011).

PLS with current speed and direction as predictors andturbidity as response

PLS with current speed and direction and discharge data aspredictors and turbidity as response was carried out forcampaigns 5 and 7 because lander data were only available forthese 2 campaigns. The loading plot in Figure 4 based on datafrom campaign 5 shows that turbidity is not well explained andis independent onCD andCS and cannot be related to currentsfrom WHP B. The explained variance for turbidity after 2factors was 27% and 15% for calibration and validation,respectively, implying that there is no predictive ability in themodel. However, Figure 4 shows that CS and CD follow verysystematic patterns. CS varies with depth and the correlationchanges accordingly. CD for all depths is correlated showingthat the current is homogeneous for all depths. Turbidity doesnot correlate with CD or CS. On the other hand, CD is wellexplained for all depths. CS is relatively well explained for thedepths 16 to 76m above seabed, whereas the 3 pointsrepresenting CS at 86 to 106m above seabed are inside the50% explained variance circle in Figure 4. It is emphasized thatthe depth values (16–106m) refer to the distance from theseabed, not from the sea surface.

PLS-regression with data from campaign 7 based on 13 236observations (not shown) gives essentially the same result ascampaign 5 (Figure 4).

PLS using score vectors for CS, CD, and effective current

PLS-regression with discharge data, temperature, conduc-tivity, and scores for CS, CD, and CD_CS as predictorvariables, and SuspPM, Ba, and turbidity as response variableswas carried out with data from campaigns 5 and 7, sedimenttrap 3. Sediment trap 3 was used because it was locateddownstream from the discharge (data were not available forsediment trap 1).

The first model for the 40 observations corresponding to thesediment trap intervals for campaigns 5 and 7 indicated 2outliers, ST3-7-14 and ST3-7-21 with very high values forSuspPM. These 2 samples were also extreme in the previousPCA (Figure 2) but were not considered as outliers as thepattern among the variables did not change when these 2samples were excluded from the model. In the PLS-regression,however, they are outliers because they significantly influencethe parameters in the regression model if they were included.This means that they are influential, but the extreme values for

Figure 3. PAH cross-plots for the ratios of (A) Ant/178 versus Fl/Flþ Py, (B)BaA/228 versus Fl/Flþ Py, and (C) IP/IPþBghiP versus Fl/Flþ Py. Ratios werecalculated based on average values of each sampling period for sedimenttraps 2 and 3. The notation is ST2 and ST3 for the sediment traps, followed by2–7 for campaigns number.

SuspPM cannot be explained by the x-variables as the currentderived variables are inside the normal range for these 2samples.The correlation loading plot in Figure 5 shows that SuspPM

does not correlate with any of the other variables. Further-more, although goodness of fit (explained calibration variance)for turbidity is 41% and Ba 30% with 2 PLS factors, the modelis not stable and lacks prediction power (evaluated by cross-validation). This implies that SuspPM, turbidity, andBa cannotbe explained by transport of drilling discharges fromWHPB tothe lander or sediment trap 3.

Dispersion and deposition modeling

For comparison, numerical modeling with DREAM wascarried out to describe dispersion and deposition of thedischarges of drilling fluid and cuttings from WHP B. A

corresponding 2-y modeling has previously been published(Nilssen, Santos et al. 2015); however, the modeling in thepresent workwas carried out for campaigns 5 and 7 specificallybecause lander data were available for these 2 campaigns. Thepurpose was to evaluate whether established numericalmodels would support the observation that the drill cuttingsdid not settle in the sediment traps or increase turbidity at thelander position. Figures 6A and 6B show the deposition ofdischarged drill cuttings and solids in drilling fluid. The lowerlimit of deposition was set to 50 gm�2 corresponding to0.05mm thickness. The deposition figures demonstrate thatthese amounts of drill cuttings and drilling fluid solids hardlyreach sediment trap 3 during campaigns 5 and 7. Obviously,the position for sediment trap 3 was not optimal for capturingdrill cuttings and fluid solids from the discharges duringcampaigns 5 and 7. Furthermore, the deposition is closer tosediment trap 2 than expected.

Table 2 shows the predicted accumulated sedimentationfrom the discharges during the 2 campaigns compared to thelarger measured sedimentation in sediment traps 2 and 3. Thedifference further supports the conclusion that the major partof the sediment collected in the traps had another origin thanWHP B. Furthermore, it is generally assumed that DREAMoverestimates sedimentation (Rye et al. 2012), implying thatthe modeled deposition should probably be even lower thanshown in Table 2.Figure 6C shows a representative snapshot of the horizontal

and vertical dispersion plume in the water column from thedischarge of drill cuttings and drilling fluid solids. The lowerlimit is 0.05 ppm of particles in the water column. Thehorizontal dispersion shows that the plume goes in thedirection of the lander and to some extent to sediment trap3. The vertical cross section shows that there is a significantdeposition close to the platform, whereas the plume continuesat intermediate water layers and will not have any significantimpact on turbidity at the lander position. This may explainthe lack of relationship between turbidity and discharges(Figure 4).

CONCLUSION AND FUTURE PERSPECTIVESThe present article demonstrates the essential benefit of

multivariate data analysis for optimized data interpretation inintegrated environmental monitoring. The multivariate dataanalysis integrates a high number of variables and vast amountsof data. Principal component analysis was used to evaluatesimilarities and differences among samples and correlationsbetween variables. Partial least squares regression was used toobtain correlations between predictors and responses. Becauseof the variability of the sampling frequency and the number ofvariables per sensor system, a multiblock modeling approachwas applied for the final PLS-regression.

The different multivariate models support each other andare also supported by chemical indicators and numericalmodeling of dispersion and deposition using DREAM.

In a future perspective, data from biosensors, images,spectra, etc., may be integrated in the multivariate dataanalysis in addition to the parameters described in the presentwork. A further step in integrated environmental monitoring isthe integration of multiple sensors, online transmission of dataand automated multivariate statistics for real-time monitoringand integration in operational systems. In addition to providinga better basis for rapid mitigation in case of incidences orundesired impact from planned discharges, a monitoring

Figure 4. Correlation loading plot obtained after PLS with turbidity asfunction of the predictors CS and CD (16–106mabove seabed) and dischargesof drill cuttings and fluid. Data are from campaign 5 with temporal resolutionof 24h. A few extreme observations at the very end of the campaign were notused as they were obvious outliers (based on outlier statistics).

Figure 5. Correlation loading plot obtained after PLS-regression withdischarge data, temperature, conductivity, and scores for current speed (CS),direction (CD), and effective current (CD_CS) as predictor variables, andsuspended particulate matter (SuspPM), Ba, and turbidity as responsevariables. Campaigns 5 and 7, sediment trap 3. Samples ST3-7-14 and ST3-7-21 excluded based on outlier statistics. Data were not available for sedimenttrap 1.

system integrated as part of the daily industrial operations will,in most cases, be cost-efficient. The large amount of datagenerated from real-time measurements represents a majorchallenge both with respect to transmission to, and interpreta-tion by the end user. Multivariate data analysis is capable ofhandling both these challenges by reducing the magnitude

before the data are transmitted and through the capability ofidentifying patterns in huge data sets. Software for onlinemultivariate data analysis is already installed at the Lofoten-Vesterålen Ocean Observatory, Norway (http://love.statoil.com). Results and experiences from this online multivariatedata analysis will be published later.

Figure 6. Accumulated deposition of drill cuttings and drilling fluid solids in g m�2 at the Peregrino seabed predicted by DREAM during campaigns 5 (A) and 7(B). Lower limit of sedimentation is set to 50 gm�2 (0.05mm thickness). Snapshot (C) of horizontal and vertical dispersion of drill cuttings and drilling fluid solidsin the water column after discharge from WHP B. Arrow indicates prevailing current direction.

Acknowledgment—Financial support was given by STATOILBrazil �Oleo e G�as Ltda (STATOIL Brazil Oil and Gas LLC).The authors thank Agencia Nacional de Petr�oleoG�as Natural eBiocombust�ıveis (ANP) in Rio de Janeiro, Brazil, for acceptingthe PEMCA Project under the Federal Participation Agree-ment (FPE). EinarNygård, Statoil ASA, EinarOsland,METAS(Marine Ecosystem Technologies AS), Bergen, Norwayand Leandro Cerqueira, Oceanographic Institute, Universityof S~ao Paulo, S~ao Paulo, Brazil are acknowledged for theprovision of data from the lander. Karsten Ofstad, Statoil ASAis acknowledged for the provision of discharge data.Data availability—All raw data will be made available upon

request from the authors.

SUPPLEMENTAL DATATable S1. Table with positions.Table S2. Table with descriptive statistics of hydrocarbons.Table S3. Table with all data used in the multivariate data

analyses.Figure S1. Map.Data S1. Text with methods details and additional results.

REFERENCESAboul-Kassim TAT, Simoneit BRT. 1996. Lipid geochemistry of surficial sediments

from the coastal environment of Egypt: Aliphatic hydrocarbons � characteri-zation and sources. Mar Chem 54:135–158.

Alvarez-Guerra M, Ballabio D, Amigo JM, Viguri JR, Bro R. 2010. A chemometricapproach to the environmental problem of predicting toxicity in contaminatedsediments. J Chemometrics 24:379–386.

Bagley PM, Smith KL, Bett BJ, Priede IG, Rowe G, Clarke J, Wall A, Ruhl HA, BaileyDM, Bazica B. 2015. The DELOS project: Development of a long-termobservatory in an oil field environment in the Tropical Atlantic Ocean. In:Seafloor Observatories. Berlin (DE): Springer. p 325–344.

Bakke T, Klungsøyr J, Sanni S. 2013. Environmental impacts of produced water anddrilling waste discharges from the Norwegian offshore petroleum industry.Mar Environ Res 92:154–169.

Basso D. 1998. Deep rhodolith distribution in the Pontian islands, Italy: Amodel forthe paleoecology of a temperate sea. Palaeogeogr Palaeoclimatol Palaeoecol137:173–187.

Bi X, Sheng G, Liu X, Li C, Fu J. 2005. Molecular and carbon and hydrogenisotopic composition of n-alkanes in leaf waxes. Org Geochem 36:1405–1417.

Bourbonniere RA, Meyers PA. 1996. Sedimentary geolipid records of historicalchanges in the watersheds and productivities of Lake Ontario and Erie. LimnolOceanogr 41:352–359.

Deyme R, Bouloubassi I, Taphanel-Valt MH,Miquel JC, Lorre A, Marty JC, MejanelleL. 2011. Vertical fluxes of aromatic and aliphatic hydrocarbons in theNorthwestern Mediterranean Sea. Environ Pollut 159:3681–3691.

Durgut _I, Rye H, Reed M, Smit MGD, Ditlevsen MK. 2015. Dynamic modeling ofenvironmental risk associated with drilling discharges to marine sediments.Mar Pollut Bull 99:240–249.

Eglinton G, Gonzalez AG, Hamilton RJ, Raphael RA. 1962. Hydrocarbonconstituents of the wax coatings of plant leaves: A taxonomic survey.Phytochemistry 1:89–102.

Eide I, Berg T, Thorvaldsen B, Christensen GN, Savinov V, Larsen J. 2011. Polycyclicaromatic hydrocarbons in dated freshwater and marine sediments along theNorwegian coast. Water Air Soil Pollut 218:387–398.

FigueiredoMAO, Eide I, ReynierM, Villas-Boas AB, Tamega FTS, Ferreira CG, NilssenI, Coutinho R, Johnsen S. 2015. The effect of sediment mimicking drill cuttingson deep water rhodoliths in a flow-through system: Experimental work andmodeling. Mar Pollut Bull 95:81–88.

Godø OR, Klungsøyr J, Meier S, Tenningen E, Purser A, Thomsen L. 2014. Real timeobservation system for monitoring environmental impact on marineecosystems from oil drilling operations. Mar Pollut Bull 84:236–250.

Hostettler FD, PereiraWE, Kvenvolden KA, vanGeen A, Luoma SN, Fuller CC, AnimaR. 1999. A record of hydrocarbon input to San Francisco Bay as traced bybiomarkers profiles in surface sediment and sediment cores. Mar Chem64:115–127.

Jackson JE. 1991. A user's guide to principal components. New York (NY): JohnWiley and Sons. 569 p.

Johnsen S, Nilssen I, Pinto APB, Torkelsen T. 2014. Monitoring of impact of drillingdischarges to a calcareous algae habitat in the Peregrino oil field in Brazil. In:Proceedings SPE International Conference onHealth, Safety, and Environment;2014 March 17-19; Long Beach (CA): SPE 168356.

Lin YC, Chang-Chien GP, Chiang PC, Chen WH, Lin YC. 2013. Multivariate analysisof heavy metal contaminations in seawater and sediments from a heavilyindustrialized harbor in Southern Taiwan. Mar Pollut Bull 76:266–275.

Martens H, Næs T. 1993. Multivariate calibration. Chichester (UK): John Wiley andSons. 438 p.

Meyers PA. 1997. Organic geochemical proxies of paleoceanographic, paleo-limnologic, and paleoclimatic processes. Org Geochem 27:213–250.

Meyers PA. 2003. Applications of organic geochemistry to paleolimnologicalreconstructions: A summary of examples from the Laurentian Great Lakes.OrgGeochem 34:261–289.

Mu~noz I, Mart�ın-TorreMC, Gal�an B, Viguri JR. 2015. Assessment by self-organizingmaps of element release from sediments in contact with acidified seawater inlaboratory leaching test conditions. Environ Monitor Assess 187:748.

Neff JM, Stout SA, Gunster DG. 2005. Ecological risk assessment of polycyclicaromatic hydrocarbons in sediments: Identifying sources and ecologicalhazard. Integr Environ Assess Manag 1:22–33.

Nilssen I, Ødegård Ø, Sørensen AJ, Johnsen G, Moline MA, Berge J. 2015.Integrated environmental mapping and monitoring, a methodologicalapproach to optimise knowledge gathering and sampling strategy.Mar PollutBull 96:374–383.

Nilssen I, Santos F, Coutinho R, Gomes N, Cabral MM, Eide I, Figueiredo MAO,Johnsen G, Johnsen S. 2015. Assessing the potential impact of water-baseddrill cuttings on deep-water calcareous red algae using species specific impactcategories and measured oceanographic and discharge data.Mar Environ Res112:68–77.

Premuzic ET, Benkovitz CM, Gafiney JS,Walsh JJ. 1982. The nature and distributionof organic matter in surface sediments of world oceans and seas. OrgGeochem 4:63–77.

Rodr�ıguez-Ortega MJ, Rodr�ıguez-Ariza A, G�omez-Ariza JL, Mu~noz-Serrano A,L�opez-Barea J. 2009. Multivariate discriminant analysis distinguishes metal-from non-metal-related biomarker responses in the clam Chamaelea gallina.Mar Pollut Bull 58:64–71.

Rye H, Loekken M, Moe J, Ditlevsen MK. 2012. Simulation of concentrations anddepositions of particle matter caused by drilling discharges. Comparisonbetween field measurements and simulation results at coral locations. In:Proceedings SPE International Conference on Health, Safety and Environment;2012 September 11-13; Perth, Australia: SPE 156775.

Rye H, Reed M, Frost TK, Smit MGD, Durgut I, Johansen Ø, Ditlevsen MK. 2008.Development of a numerical model for calculating exposure to toxic andnontoxic stressors in the water column and sediment from drilling discharges.Integr Environ Assess Manag 4:194–203.

Salgado CS, Johnsen S, Gomes S, Torkelsen T. 2010. Real-time environmentalmonitoring of discharges of drill cuttings in the Peregrino field (Campos Basin,Brazil). In: Proceedings SPE International Conference on Health, Safety andEnvironment; 2010 April 12-14; Rio de Janeiro, Brazil: SPE 127175.

Table 2. Predicted (byDREAM) accumulated sedimentation from thedischarges compared to themeasured accumulated sedimentation in

sediment traps 2 and 3 during campaigns 5 and 7 (gm�2)

Measured Predicted

Campaign 5 ST2 134 21

Singsaas I, Rye H, Frost TK, Smit MGD, Garpestad E, Skare I, Bakke K, Veiga LF,Buffagne M, Follum OA, et al. 2008. Development of a risk-basedenvironmental management tool for drilling discharges. Summary of a four-year project. Integr Environ Assess Manag 4:171–176.

Thornton SF, McManus J. 1994. Application of organic carbon and nitrogen stableisotope and C/N ratios as source indicators of organic matter provenance inestuarine systems: Evidence from the Tay Estuary, Scotland. Est Coast Shelf Sci38:219–233.

[UNEP] United Nations Environment Programme. 1992. Determination ofpetroleum hydrocarbons in sediments. Reference methods for marinepollution studies. Vol 20. Nairobi (KE): UNEP. 97 p.

[USEPA] US Environmental Protection Agency. 1996. Microwave assisted aciddigestion of siliceous and organically based matrices. Method 3052. Revision0. Washington DC: US Environmental Protection Agency. 20 p.

Villas-Boas AB, Tamega FTS, Andrade M, Coutinho R, Figueiredo MAO. 2014.Experimental effects of sediment burial and light attenuation on two corallinealgae of a deep water rhodolith bed in Rio de Janeiro, Brazil. CryptogamieAlgologie 35:67–76.

Volkman J.K., Holdsworth DG, Neill GP, Bavor Jr HJ. 1992. Identification of natural,anthropogenic and petroleum hydrocarbons in aquactic sediments. Sci TotEnviron 112:203–219.

Wold S. 1978. Cross-validatory estimation of the number of components in factorand principal components models. Technometrics 20:397–405.

Wold S, Martens H, Wold H. 1983. The multivariate calibration problem inchemistry solved by the PLS method. Matrix Pensils 973:286–293.

Yunker MB, Macdonald RW, Vingarzan R, Mitchell RH, Goyette D, Sylvestre S.2002. PAHs in the Fraser River basin: A critical appraisal of PAH ratiosas indicators of PAH source and composition. Org Geochem 33:489–515.

Zhang Z, Zhao M, Eglinton G, Lu H, Huang CY. 2006. Leaf wax lipids aspaleovegetational and paleoenvironmental proxies for the Chinese LoessPlateau over the last 170 kyr. Quat Sci Rev 25:575–594.

Zhou J, Wu Y, Zhang J, Kang Q, Liu Z. 2006. Carbon and nitrogen compositionand stable isotopes as potencial indicators of source and fate of organicmatter in the salt marsh of the Changjiang Estuary, China. Chemosphere65:310–317.

Paper VII

Nilssen, I., dos Santos, F., Coutinho, R., Gomes, N., Cabral, M. M., Eide, I., Figueiredo, M. A. O., Johnsen, G., Johnsen, S., 2015b. Assessing the potential impact of water-based

drill cuttings on deep-water calcareous red algae using species specific impact categories and measured oceanographic and discharge data. Marine Environmental

Research Vol 112, Part A, pp 68-77, doi: http://dx.doi.org/10.1016/j.marenvres.2015.09.008

Assessing the potential impact of water-based drill cuttings ondeep-water calcareous red algae using species specific impactcategories and measured oceanographic and discharge data

Ingunn Nilssen a, b, *, Francisco dos Santos c, Ricardo Coutinho d, Natalia Gomes c,Marcelo Montenegro Cabral c, Ingvar Eide a, Marcia A.O. Figueiredo d, e, f, Geir Johnsen b,Ståle Johnsen a

a Statoil ASA, Research, Development and Innovation, N-7005 Trondheim, Norwayb Trondhjem Biological Station, Department of Biology, Norwegian University of Science and Technology, N-7491 Trondheim, Norwayc PROOCEANO Serviço Oceanogr�afico, Av. Rio Branco, 311 e sala 1205 Centro, Rio de Janeiro, RJ, Brazild Instituto de Estudos do Mar Almirante Paulo Moreira, Department of Oceanography, Marine Biotechnology Division, Arraial do Cabo, RJ, Brazile Instituto de Pesquisa Jardim Botanico do Rio de Janeiro, Rua Pacheco Le~ao 915, Jardim Botanico 22460e030, Rio de Janeiro, RJ, Brazilf Instituto Biodiversidade Marinha, Avenida Ayrton Senna 250, Sala 208, Barra da Tijuca, 22.793e000, Rio de Janeiro, RJ, Brazil

Article history:Received 15 June 2015Received in revised form3 September 2015Accepted 16 September 2015Available online 24 September 2015

Keywords:Impact assessmentImpact categoriesCalcareous algaeDrill cuttingsModelling

a b s t r a c t

The potential impact of drill cuttings on the two deep water calcareous red algae Mesophyllum engelhartiiand Lithothamnion sp. from the Peregrino oil field was assessed. Dispersion modelling of drill cuttingswas performed for a two year period using measured oceanographic and discharge data with 24 hresolution. The model was also used to assess the impact on the two algae species using four speciesspecific impact categories: No, minor, medium and severe impact. The corresponding intervals forphotosynthetic efficiency (FPSIImax) and sediment coverage were obtained from exposureeresponserelationship for photosynthetic efficiency as function of sediment coverage for the two algae species. Thetemporal resolution enabled more accurate model predictions as short-term changes in discharges andenvironmental conditions could be detected. The assessment shows that there is a patchy risk for severeimpact on the calcareous algae stretching across the transitional zone and into the calcareous algae bedat Peregrino.

1. Introduction

The calcareous red algae, Mesophyllum engelhartii (Foslie) Adeyand Lithothamnion sp. are encrusting species growing on differenthard substrates such as rhodoliths. Rhodoliths are calcified multi-spherical structures of different sizes made of dead calcareousalgae and other calcifying organisms such as bryozoans anddifferent species of polychaetes (Tamega et al., 2013) (Fig. 1). Thesemulti-spherical structures are regarded as important for theecosystem, creating a habitat for other taxa living between, on andin the rhodolith structures (Basso, 1998). Rhodoliths are founddown to 250m (Henriques et al., 2014; Littler et al., 1991, 1986), and

the largest occurrence is discovered along the Brazilian ContinentalShelf (Amado-Filho et al., 2012; Foster, 2001; Kempf, 1970). Theknowledge about growth, sensitivity to environmental stressorsand ecological importance of the deep water rhodolith commu-nities is rather sparse (Henriques et al., 2014; Steller et al., 2009,2007). However, these communities may be disturbed and burieddue to natural sedimentation and anthropogenic activities such asfishetrawling and mining (Nelson, 2009). Burial may result inreduced photosynthetic activity due to reduced gas exchange(Figueiredo et al., 2015; Harrington et al., 2005;Wilson et al., 2004).It is also demonstrated that fine particles (<63 mm grain size) mayreduce the photosynthetic efficiency of coralline algae to a largerextent than coarse calcareous and nutrient-rich shallow estuarinesediments (Figueiredo et al., 2015; Harrington et al., 2005; Riulet al., 2008; Wilson et al., 2004).

Furthermore, rhodolith beds are increasingly exposed to dis-charges of drill cuttings from oil and gas activities, for instance in

* Corresponding author. Statoil ASA, Research, Development and Innovation,N-7005 Trondheim, Norway.

E-mail address: innil@statoil.com (I. Nilssen).

Marine Environmental Research

journal homepage: www.elsevier .com/locate /marenvrev

Marine Environmental Research 112 (2015) 68e77

the Gulf of Mexico and on the Brazilian shelf (Davies et al., 2007).The Peregrino oil field, located 80 km off the coast of Brazil, south ofCabo Frio, in the Campos Basin area, is located close to a rhodolithbed (Fig. 9). The calcareous algae Mesophyllum engelhartii (Foslie)Adey and Lithothamnion sp. are among the most abundant algaespecies present in the area (Figueiredo et al., 2015; Salgado et al.,2010; Tamega et al., 2013). Drill cuttings with residuals of water-based drilling fluids are discharged from the drilling activities atPeregrino. Drill cuttings are formation rock particulates generatedduring drilling (Neff, 2008). The water-based drilling fluid dis-charges only contain residuals of water soluble and none or lowtoxic components such as barite, clay, salts and calcium carbonate(Bakke et al., 2013; Frost et al., 2006; Neff, 2008). Barite is aweighting agent in drilling fluids and one of the most abundantsolid in drilling fluids (Neff, 2008). Laboratory tests verified that theresiduals of water-based fluid from the Peregrino oil field are nottoxic to Mesophyllum and Lithothamnion (Reynier et al., 2015).However, as calcareous algae have no individual motion and sincethe rhodoliths they are living on are expected to have low motioncapabilities (Fig. 1), these discharges may have a physical impact onthe test organisms through sedimentation of particles. Reducedproduction of oxygen and reduced in vivo photosynthetic efficiency,reported by Reynier et al. (2015) was primarily physical and due tothe burial with drill-cuttings. Also sediment mimicking drill cut-tings from the Peregrino oil field resulted in reduced in vivophotosynthetic efficiency in Mesophyllum and Lithothamnion(Figueiredo et al., 2015; Villas-Boas et al., 2014). Photosyntheticefficiency was measured as in vivo Chlorophyll a fluorescence ki-netics of dark acclimated cells (FPSII_max). A decrease in photosyn-thetic efficiency is regarded as a response to stress (Genty et al.,1989).

Modelling is frequently used to assess potential risk or impact ofdischarges to sea from oil and gas exploration and production.Environmental risk assessment based on the PEC/PNEC approach iscommonly used by offshore operators in Europe, and the principleis accepted by the Oslo Paris Commission (OSPAR) as a basis forenvironmental management of produced water discharges (OSPAR,2012a, 2012b). One example of such a model is the Dose-relatedRisk and Effect Assessment Model (DREAM) (Beyer et al., 2012;Durgut et al., in press; Johnsen et al., 2000; Rye et al., 2008; Smitet al., 2008). DREAM predicts dispersion and potential environ-mental risk based on the PEC/PNEC approach for discharges ofproduced water and drill cuttings to the marine environment. Amajor criticism against this approach has been the lack ofecosystem relevance as the PNEC levels derived from the literaturemay not be representative for local species. This is especially of

concern when operating in areas considered as particularlyvulnerable or housing species or habitats of particular interest. Thissituation is indeed valid for the Peregrino field, housing a deepwater calcareous algae community. However, DREAM also allowsthe use of species specific effect data as an option to the moregeneric PNEC approach, giving a risk or impact assessment targetedat selected species.

The aim of the present study was to assess the potential impactof drill cuttings discharges on the dominating species of calcareousalgae at the Peregrino oil field. Furthermore, the purpose was toverify that the assessment of the potential impact as performed byDREAM provides valid and reliable information for environmentalmanagement. This was achieved by:

1) establishment of species specific impact categories for thecalcareous algae Mesophyllum engelhartii (Foslie) Adey andLithothamnion sp. based on exposure studies performed in aflow-through system (Figueiredo et al., 2015) and

2) modelling the potential impact on the two calcareous algaespecies using the established impact categories together withmeasured current velocity and direction and actual drill cuttingdischarges with high temporal resolution.

2. Methods

The methodological approach in this study is based on theprinciples outlined by Nilssen et al. (2015) for integrated environ-mental mapping and monitoring, describing a flexible approach foroptimised knowledge gathering through combined use of labora-tory experiments and measured field and discharge data withmodelling. To enable the link between the impact categories ob-tained from the flow-through exposure studies (Figueiredo et al.,2015) and the dispersion modelling, sedimentation experimentswere performed to convert sediment coverage (%) to sedimentdeposition (kg m�2). A flow chart of the different elements used toperform the assessment of potential impact on the two calcareousalgae species is illustrated in Fig. 2.

2.1. The Peregrino area

The Peregrino oil field consists of two fixed well head platforms(WHP) for drilling of production and water injectionwells, and onefloating production storage and offloading unit (FPSO). The WHPsare located 10 km apart with the FPSO in between. Drilling at thePeregrino field commenced in November 2010 and until May 2013,

Fig. 1. a) Rhodolith bed, b) Rhodolith with healthy calcareous red algae (reddish colour).

I. Nilssen et al. / Marine Environmental Research 112 (2015) 68e77 69

30wells were drilled. The field came in production inMay 2011. Thewell streams are transported to the FPSO through subsea pipelines,while produced water (water that is produced together with thehydrocarbons from the reservoir) is routed back to the WHPs andre-injected to the reservoir. Drill cuttings with residuals of water-based drilling fluid are discharged to the sea, while cuttings fromthe hydrocarbon containing reservoir section are sent to shore fordisposal (Johnsen et al., 2014).

The water depth at Peregrino varies from 90 to 120 m andseabed sediments consist in general of sand and silt with a rela-tively hard surface (Johnsen et al., 2014; Salgado et al., 2010). Therhodolith bed is found in the shallow part of the Peregrino seabed(<110 m).

2.2. Dispersion modelling

The numerical simulations of the dispersion and deposition ofthe drilling discharges over a two years period were carried outusing the DREAM model (Beyer et al., 2012; Durgut et al., in press;Johnsen et al., 2000; Rye et al., 2008; Singsaas et al., 2008). DREAMuses discharge characteristics and environmental data such ascurrent, wind, etc. to estimate the behaviour of the dischargeplumes in the water column and the deposition on the seabed, thepredicted exposure concentration (PEC). Due to the temporal 24 hresolution of the discharge log the model was run in daily se-quences, enabling a daily overview of the accumulation of thedeposition.

2.2.1. Field dataThe environmental variables used were bathymetry, sea tem-

perature, salinity, speed and direction of wind and currents, thelatter being the most complex and crucial for reliable outputs frommodelling of cuttings deposition. Despite the 4-D behaviour (3spatial dimensions and time) of a cuttings discharge plume, fromnear the sea surface to settling on the seabed, the horizontal scale ofthe plume is expected to be much smaller than the horizontalvariability of the currents. It was therefore presumed that thevertical profile of the currents in the vicinity of the operation isrepresentative for the currents for the whole area of interest.

The vertical profile of the currents was obtained from anAcoustic Doppler Current Profiler (ADCP) moored in the vicinity of

WHPB. Current velocity and direction were measured every 10 mfrom 14 m below the sea surface to 104 m water depth. Thedispersion modelling was carried out for the time period 10thNovember 2010 to 29th November 2012. Measured data werelacking in the period of 10th November to 26th January 2011, 16thAugust to 21st November 2011 and 25th April 2012 to 2nd July2012. These gaps were filled by model outputs from the MyOceanproject (Dombrowsky et al., 2013). This project uses the Nucleus forEuropean Models of the Ocean (NEMO) to assimilate current data,sea surface height and sea temperature measurements derivedfrom satellites. The complete current time series used in the studyis presented in Fig. 3. Current velocities varied between 0 and1.01 m s�1 at about 10 m above the seabed.

Both modelled and measured current velocity and directionshowed a barotropic behaviour from surface to bottom, mainlyflowing south-westwards. Inversion on the current direction wasobserved most frequently in the autumn/winter period (from Mayto August). This behaviour is explained by the Brazil Current, aweek western boundary current flowing towards high latitudesalong the continental shelf break (Evans and Signorini, 1985). Itsintensity and direction may be modified by cold front passages(Stech and Lorenzzetti, 1992).

2.2.2. Discharge dataParticle size distribution of drill cuttings from the Peregrino oil

field (well A-18) was obtained from sieving analyses. Volumes ofdischarged drill cuttings and drilling fluids with a temporal reso-lution of 24 h were obtained from the platform drilling logs. Thetotal amount of solids in the water-based drilling fluid was calcu-lated in accordance with the bulk density of the drilling-fluidthrough the given drilling operation period at Peregrino.

2.3. Effect data and impact categories

Stress on calcareous algae is traditionally categorised by visualinspection, using charts for colour tone and intensity (De O.Figueiredo et al., 2000; Villas-Boas et al., 2014) (Table 1).Figueiredo et al. (2015) related the categorised stress levels tocorresponding photosynthetic efficiency in their study with sedi-ment coverage on Mesophyllum and Lithothamnion. The photosyn-thetic efficiency was measured as in vivo maximum quantum yield

Fig. 2. Schematic overview of elements and input data used in the assessment of impact on the two calcareous algae species. This includes dispersion modelling, using input datafrom field (oceanographic data) and discharge log, described in chapter 2.2, 2.2.1 and 2.2.2, respectively. The effect data, derived from the exposure studies used to establish impactcategories and the sedimentation experiment are described in chapter 2.3 and 2.3.1, respectively.

I. Nilssen et al. / Marine Environmental Research 112 (2015) 68e7770

of charge separation in photosystem II in dark acclimated cells(Genty et al., 1989). The measurements were performed with aDiving-PAM fluorometer, using a blue measuring light with awavelength of 470 nm. The saturating pulse was 0.6 s and satura-tion intensity 150e200 mmol m�2 s�1, gain 3 mmol m�2 s�1 anddumping 3 mmol m�2 s�1. The light levels used were 3, 6.6, and10 mmolme2 se1, the latter corresponding to light intensity at 100mwater depth at the Peregrino field where the calcareous algae werecollected.

Sediment coverage (%) was calculated by the use of the softwareImageJ (National Institutes of Health, Bethesda, Maryland)(Figueiredo et al., 2015). Photos were taken of each rhodolith eachweek through the experimental period (T0eT9). The borders of theeach rhodolith were outlined before and after removal of sedimentfor the photosynthetic efficiency measurements. Sedimentcoverage was calculated from these images.

These semi-quantitative data were used as the basis for estab-lishing the following four species specific impact categories forMesophyllum and Lithothamnion: no, minor, medium and severeimpact. The corresponding FPSIImax values are 0.46e0.55,0.36e0.45, 0.26e0.35 and <0.26, respectively, implying that thetwo most severe stress levels were merged into one category. Thecorresponding sediment coverage intervals were obtained from thecombined exposureeresponse relationship for photosynthetic

efficiency as function of sediment coverage for the two algae spe-cies as described by Figueiredo et al. (2015). Sediment coverage andphotosynthetic efficiency were measured in a flow-through systemvarying light intensity, flow rate and added amount of sedimentonce a week for nine weeks. Statistical experimental design andmultivariate data analysis provided statistically significant regres-sion models which subsequently were used to establish the expo-sureeresponse relationship.

All calcareous algae used in the experimental work were clas-sified as healthy, “no stress” (FPSII_max ¼ 0.46e0.55) at the begin-ning of the experiment (T0). This is a conservative approach and it isexpected that lesser healthy individuals to a lesser extent wouldtolerate the sediment exposure during the experiment.

2.3.1. Sedimentation experimentsThe DREAM dispersionmodelling provides sedimentation of the

discharges in kg m�2. In order to relate modelled sedimentation tosediment coverage (%), as provided by the exposure study(Figueiredo et al., 2015), sedimentation experiments were per-formed (Fig. 2). The flow-through system used by Figueiredo et al.(2015) for the exposure studies was also used in the sedimentationexperiments, however, without rhodoliths in the chambers.

Three series of experiments were performed with five differentsediment amounts added to the chambers under three differentflow conditions. The flow rates were 0.04, 0.07, and 0.09 m s�1,identical to those used in Figueiredo et al. (2015) and the amount ofsediment added were 600, 750, 900, 1050 and 1200 g.

A Plexiglas tray with the size of 80.34 cm2 was placed hori-zontally at the bottom of each exposure chamber (approximately6 L) in the flow-through system to collect the deposited sediments.Each experiment lasted 60 min, which was appropriate for theparticles to settle. After settlement, the flow was stopped and theremaining seawater was slowly drained off the exposure chambersto prevent loss of deposited sediments, to minimise the amount ofsalt in the sediment and to reduce the drying process. The sediment

Fig. 3. Time series of current velocity and direction at Peregrino used in the DREAM simulations. The ADCP data (measured) is represented by black arrows and MyOcean data(modelled) by red arrows. The timeline indicated on the x-axes equals the aggregated deposition modelled after 6, 12, 18 and 24 months presented in Fig. 4a and b (24 months),respectively. Prevailing current direction is south-westwards. Measured current velocities were 0e1.56 m s�1 (all depths).

Table 1Photosynthetic efficiency (FPSIImax) and corresponding colour tone and intensity andstress levels of calcareous algae. (Table 4 from Figueiredo et al. (2015).)

FPSIImax Colour (Tone) Colour (Intensity) Stress level

0e0.15 0e1 0e2.25 Dead0.16e0.25 1.1e2.0 2.26e3.25 High stress0.26e0.35 2.1e3.0 3.26e5 Intermediate stress0.36e0.45 3.1e3.75 5.1e6 Low stress0.46e0.55 3.76e4.25 6.1e7 No stress

covered Plexiglas trays were removed from the exposure chambersand dried in air for 24 h before the sediments were transferred tometal trays and oven dried for another 24 h before dry weightmeasurements.

The amounts of sediment deposited in the three sedimentationexperiments were averaged and converted to kg m�2. Corre-sponding data for sediment coverage (%) were obtained from theexposure study (Figueiredo et al., 2015). The relationship betweensediment coverage (%) and deposition (kg m�2) was obtained byregression modelling.

In addition, sediment deposition (kg m�2) was converted tosediment thickness (mm) using an initial density of drill cuttings of2.65 kg L�1 which was reduced to 1.06 kg L�1 using a porosity indexof 60% for the deposited cuttings. The porosity index is commonlyapplied for deposition at this water depth (Bryant et al., 1981).

2.4. Impact assessment

In addition to dispersion modelling, the DREAM model wasused for the assessment of the potential impact of the drilling

Fig. 4. Deposition of drill cuttings at the Peregrino seabed predicted by DREAM. a) Aggregated discharges fromWHPB after 6, 12 and 18 months and b) total discharges from WPHAand WPHB aggregated after two years. Prevailing current direction is south-westwards. Lower limit of sedimentation is set to 1000 g m�2 (~1 mm thickness).

discharges on the two species of calcareous algae, Mesophyllumengelhartii and Lithothamnion sp. One of the modules in DREAMcovers thresholds for physical disturbances on organisms,including burial, oxygen depletion and change in grain size fordischarges from drilling activities. DREAM uses, as a default, ageneric set of effect threshold levels for impacts derived fromliterature studies (PNECs) (Rye et al., 2008; Smit et al., 2008). ThePNEC for the nontoxic stress from burial have been derived by useof a Species Sensitivity Distribution (SSD) approach from availablechronic data (NOEC) (Smit et al., 2008). SSDs can be applied in riskmodelling by predicting the potentially affected fraction of speciesat a certain level of exposure. The 5% affected fraction is in generalregarded as representative for the PNEC (Harbers et al., 2006) andthis is also used in the DREAM model. However, when speciesspecific knowledge is regarded as crucial, specific values for effectsderived from laboratory studies can be used instead of genericPNECs as input to the model.

In the present study, the species specific impact categoriesdescribed in section 2.3 were used as input to the DREAM simu-lations. This provides an impact assessment analogous to theassessment of physical disturbances of drill cuttings on the reefbuilding cold and deep water coral Lophelia pertusa (Det NorskeVeritas, 2013).

3. Results

3.1. Dispersion modelling

The results from the dispersion modelling of drilling dischargesfrom the first two years of drilling at the Peregrino filed are pre-sented as deposition maps of different colours as function of time,represented by the deposition after 6 months (May 2011), 12months (November 2011) and 18 months (May 2012) (Fig. 4a) andthe total aggregation over the two years period (Fig. 4b), respec-tively. As expected, deposition is following the ocean currentpattern and is concentrated in the south-west/north-eastern di-rection. However, periods with divergent current directions occur,as shown after 18 months in Fig. 4a (grey lines).

Examples of aggregation of sedimentation at two differentgeographical positions over time are presented in Fig. 5. Fig. 5ashows the aggregation in a position near the discharge point(WHPB), while Fig. 5b presents the deposition for a position furthersouthwest, close to the calcareous algae rhodolith bed. As expected,deposition of discharges are higher close to the discharge point,while areas further away are expected to receive less amounts andthe deposition therefore increases more gradually that in the im-mediate vicinity of the discharges.

3.2. Effect data and impact categories

Four species specific impact categories for Mesophyllum andLithothamnion were established in section 2.3: No, minor, mediumand severe impact. The corresponding FPSIImax values are0.46e0.55, 0.36e0.45, 0.26e0.35 and <0.26, respectively. Thecorresponding sediment coverage intervals were obtained fromthe combined exposureeresponse relationship for photosyntheticefficiency as function of sediment coverage for the two algaespecies as described by Figueiredo et al. (2015). The result isshown in Fig. 6.

The variability in the exposureeresponse curve (FPSIImax ±0.06)is illustrated by the dotted lines (Figueiredo et al., 2015). Thesetake into account the contribution from the other factors thatwere varied in the exposure studies, namely flow rate, light in-tensity and time. However, the contribution of each of these fac-tors is minor and the dotted curves represent the extreme upper

and lower limits when all three factors contribute simultaneouslyin the same direction, each on its maximum level. The solidexposureeresponse curve has flow rate, light intensity and time atintermediate levels. Due to the fact that other factors maycontribute, the sediment coverage intervals corresponding to theestablished FPSII_max levels are rounded off to the nearest 10%value. The following categories were established for possibleimpact of sediment coverage (%) on the two species of calcareousred algae at the Peregrino field:

� No impact: 0e30% sediment coverage� Minor impact: 30e50% sediment coverage� Medium impact: 50e70% sediment coverage� Severe impact: >70% sediment coverage

3.2.1. Sedimentation experimentsThe results from the sedimentation experiments are presented

in Table 2. The amounts of sediment deposited in the three sedi-mentation experiments were averaged and subsequently convertedto kg m�2. Corresponding data on sediment coverage (%) obtainedfrom the exposure studies (Figueiredo et al., 2015) are presented inthe last column of Table 1.

Average sediment deposition as a function of added sedimentamount and flow rates are presented in Fig. 7. The data points forthe two highest current velocities (0.07 and 0.09 m s�1) are over-lapping, indicating that sedimentation will probably not increasebeyond these velocities.

In order to establish the correlation between sediment deposi-tion (kg m�2) and coverage (%) corresponding data from Table 2were used. The best regression model was obtained with anexponential curve (R2 ¼ 0.89) presented in Fig. 8.

Fig. 5. Aggregated deposition of drill cuttings over the two years drilling period(Fig. 4b) at two selected geographical positions in the model grid; a) near the dischargepoint at WHPB and b) close to the calcareous algae rhodolith bed. Note the differencein scaling on the y-axes.

3.3. Impact assessment

The established impact categories and corresponding photo-synthetic efficiency, sediment coverage, sediment deposition andsediment thickness are shown in Table 3. The impact categories andsediment deposition were, together with the oceanographic data(Fig. 3), used as input data to the DREAM simulations presented inFig. 9.

The predicted impact from the first two years of drilling at thePeregrino field is presented in Fig. 9. The potential of severe impactis mainly found within a radius less than 250 m from WHPA andWHPB. However, as shown in Fig. 9, only WHPB experience anoverlap between the area of predicted impact and the calcareousalgae occurrence (mapped through field campaigns) as scatteredareas along the prevailing current through the rhodolith transi-tional zone and into the rhodolith bed are presented (Fig. 9b).

4. Discussion

The novelty of the present work is the combined use of speciesspecific impact categories, established from laboratory experiments

and measured oceanographic and drilling discharge data in themodelling, enabling access to and use of the best informationavailable as input to decision making. The DREAM modellingdemonstrates a build-up of drill cuttings and thus, potential envi-ronmental impact on the two calcareous algae species, mainlywithin a radius of 250 m of the two platforms over the modelledtwo years period. The levels of sedimentation from the dischargesrepresenting minor, medium and severe impact outside this radiusare mainly observed as scattered areas along the prevailing currentthrough the rhodolith transitional zone and into the rhodolith bed.The study has also demonstrated the importance of the availabilityof input data with sufficient temporal resolution.

By experience, the DREAM model is conservative and has beenshown to overestimate concentrations of drilling particulates (PEC)(Rye et al., 2012). Due to the availability of detailed discharge andfield data the DREAM model was run with a higher temporal res-olution than the traditional long time-averaged approach. This re-duces the uncertainty in the modelling at any time over themodelled period by enabling e.g. identification of worst case sce-narios, such as the raise in deposition due to the increased drillingfluid discharges fromWHPB during February 2012 shown in Fig. 5a,or in situations with unexpected field conditions, such as theabnormal current velocity period shown in Fig. 4a.

Some uncertainty related to the sedimentation experiments isevident since they were carried out without rhodoliths in theexposure chambers. This may probably have led to a change inmicro currents and thereby differences in sediment deposition.However, the calcareous algae species used in the experiment(Figueiredo et al., 2015) lives like a coating on the rough surface ofthe rhodoliths. Since the rhodoliths are calcified multi-sphericalstructures it can be expected that a good portion of the addedsediments would, with the help from increased micro currentscreated by the rhodolith structures, deposit inside the rhodolithsinstead of on top of the calcareous algae. The challenge regardingmeasurement of burial is also discussed in Smit et al. (2008). Due tothe expected rhodolith dependent variations the most conservativeapproach was selected, to run a separate sedimentation experi-ment. This also supports a precautionary approach of being con-servative in case of limited availability of knowledge.

The output from DREAM is a time variable, 3-dimensional riskmaps (Rye et al., 2012) with different colour codes for the differentlevels of potential impact to the calcareous algae species. In com-parison, the generic PEC/PNEC approach would provide a risk mapwere the potential impact to the whole sediment community

Fig. 6. Established impact categories based on exposureeresponse curve for in vivophotosynthetic efficiency (FPSII_max) as function of sediment coverage for M. engelhartiiand Lithothamnion sp. Green: no impact; yellow: minor impact; orange: mediumimpact; red: severe impact. The curve is based on regression modelling of data fromlaboratory experiments (Figure modified from Figueiredo et al. (2015)).

Table 2Sediment deposited in the exposure chambers with different amounts of sediment added and with different flow rates. Sediment deposited (g per 80.34 cm2) are shown asaverages ± stdev and converted to kg m�2. Data on sediment coverage (%) was obtained from the exposure studies (Figueiredo et al., 2015).

Flow(m s�1)

Sediment added(g)

Sediment deposited (average ± stdev)(g per 80.34 cm2)

Sediment deposited(kg m�2)

Sediment coveragea (%)

0.04 600 69.2 ± 7.5 8.6 7.30.04 750 89.0 ± 2.3 11.10.04 900 99.7 ± 7.9 12.4 21.40.04 1050 102.4 ± 6.7 12.70.04 1200 134.5 ± 3.2 16.7 100.00.07 600 62.1 ± 10.4 7.7 9.00.07 750 79.1 ± 8.8 9.80.07 900 83.5 ± 11.0 10.4 29.70.07 1050 96.2 ± 22.2 12.00.07 1200 112.6 ± 9.4 14.0 97.70.09 600 63.8 ± 6.0 7.9 8.90.09 750 80.0 ± 9.7 10.00.09 900 92.2 ± 6.7 11.5 35.30.09 1050 94.2 ± 8.3 11.70.09 1200 116.0 ± 11.3 14.4 96.3

a Numbers from Figueiredo et al. (2015).

would be presented. The PEC/PNEC approach of DREAM includes ageneric PNEC for sediment species related to the effect of physicalcoverage of drill cuttings. The value is derived from SSDs from acomprehensive data set obtained from the literature (Smit et al.,2008). The basis for the SSD approach is NOEC (No Observed Ef-fect Concentration) values of burial of different species, and wherethe PNEC represents 5% of the potentially affected fraction of thespecies. In the present study this approach is replaced by the use ofimpact categories, expressing the potential effect of drill cuttings or

sediment burial on the species of interest. The generic PNEC forburial in DREAM equals 6.3 mm sediment thickness (Smit et al.,2008), while the “no impact” interval (<30%) for the impact cate-gories equals a sediment thickness of <10.7 mm (Table 3). Theseresults show that the sensitivity of calcareous algae towards sedi-ment or drill cuttings burial is well within the range of the speciesrepresented in the SSD curve presented by Smit et al. (2008) andare not among the most sensitive species (NOEC for calcareousalgae > generic PNEC). In the present case with the calcareousalgae, the PEC/PNEC approach would have been applicable to assessthe possible impact.

However, applying the impact categories reduces the un-certainties of the potential impact of drilling discharges to the twospecies of calcareous algae. In addition it provides flexibility to theimpact assessment. While the generic PEC/PNEC approach inDREAM does not distinguish between levels of risk by using onecut-off PNEC value, the impact categories approach introduces arange of severity (no impactesevere impact) in the risk picture. Aspecies specific impact category approach based on degrees ofexposure provides improved management ability adjusted tovarying geographical environmental conditions and the species ofinterest. For instance, field sampling at Peregrino have shown highnatural sedimentation (0.4e2.5 g m�2 d�1) (unpublished results)and Figueiredo et al. (2015) indicated that the calcareous algae atPeregrino is adapted to some degree of natural sedimentation, asthe best light conditions for the photosynthetic efficiency werefound within the interval 12e14% particle coverage (max lightregime used in experiment equals 100% light intensity measured atthe Peregrino rhodolith bed). Reduced fluorescence emission fromalgae due to too high irradiance is also well documented in theliterature (Aro et al., 1993; Johnsen and Sakshaug, 2007). In fieldsituations, variations of particle loads in suspension and deposited(natural and anthropogenic), current conditions and water depthwill determine the potential impact of drilling particulates dis-chargesmight have on the calcareous algae. It is expected that areaswith low current velocity tolerate less discharges of particulatesthan areas with stronger currents since stronger currents will givereduced sedimentation by wider dispersion of a discharge and/orincreased re-suspension Due to this considerable influence ofenvironmental factors, the category or interval approach applied ina planning process on a field may open up for a case by case eval-uation, adjusting the acceptance intervals to the presentgeographical environmental conditions. Such an interval approachcombining health status or environmental conditions with pre-dicted sedimentation, is aligned with the approach described in theguidelines for cold water corals where an individual classification ofhealth status for each reef is evaluated against the predicted sedi-mentation (Det Norske Veritas, 2013).

The established impact categories are based on laboratorystudies and may be adjusted after experience from environmentalmonitoring and assessment in the field has been obtained. An

Fig. 7. Sediment deposited as function of amount of sediment added and flow rates.The data points for the current velocities of 0.07 m s�1 and 0.09 m s�1 are indicatingthat sedimentation above these velocities will probably not increase.

Fig. 8. Correlation between sediment coverage (%) from Figueiredo et al. (2015) asfunction of sediment deposition (kg m�2). Y ¼ 0.0044x3.64 and R2 ¼ 0.89. Note thatonly the three sediment loads corresponding to the sediment loads from Figueiredoet al. (2015) (600, 900 and 1200 g) are used in the calculation.

Table 3Corresponding impact categories, calcareous algae photosynthetic efficiency (FPSII_max), sediment coverage (%), amount of sediment deposited per unit area (kg m�2) andsediment thickness (mm).

Impact categories for calcareousalgae

Physiological status in calcareous algae(FPSII_max)a

Sediment coverage(%)b

Sediment amount/area(kg m�2)c

Sediment thickness(mm)d

No impact 0.46e0.55 0e30 0e11.3 0e10.7Minor impact 0.36e0.45 30e50 11.3e13 10.7e12.3Medium impact 0.26e0.35 50e70 13e14.3 12.3e13.5Severe impact 0e0.25 >70 >14.3 >13.5

a Measured in exposure studies (Figueiredo et al., 2015).b Obtained from measured physiological status in calcareous algae and exposureeresponse curve presented in Fig. 6, obtained from Figueiredo et al. (2015).c Calculated from sedimentation experiments carried out to convert % coverage to kg m�2.d Calculated from kg m�2 using a density of 1.06 kg L�1 (reduced from 2.65 due to 60% porosity as described in 2.3.1).

alternative monitoring program, in accordance with the principlesof a flexible monitoring program aligned with type of discharges,environmental conditions and organisms of interest (Nilssen et al.,2015) has been proposed by the Peregrino Environmental Moni-toring and Calcareous Algae (PEMCA) project (Johnsen et al., 2014).The environmental monitoring focuses on the calcareous algae andpossible impact of discharges from drilling activities. It is proposedto combine sediment traps measurements, current velocity andimages, in agreement with the integrated approach to verify andimprove both the environmental modelling and monitoring(Nilssen and Johnsen, 2008).

Acknowledgements

This work has been carried out as part of Ingunn Nilssen's in-dustrial PhD, enabled by Statoil ASA and the Research Council ofNorway through the Industrial PhD programme, NRC projectnumber 222718. The work was also partly supported by theResearch Council of Norway through the Centres of Excellencefunding scheme e AMOS, NRC project number 223254.

Further, we are grateful to the PEMCA team for valuable support.Financial support was given by STATOIL Brasil �Oleo e G�as Ltda((STATOIL Brazil Oil and Gas LLC) and Agencia Nacional de Petr�oleo,G�as Natural e Biocombustíveis e ANP (National Agency of

Petroleum, Natural Gas and Biofuels e ANP). SISBIO license n�

20820-2 and 20826-1.Thanks to Tone Karin Frost for contributing in fruitful discus-

sions and valid input to the manuscript.

Abbreviations and definitions

DREAM Dose-related Risk and Exposure Assessment ModelFPSO Floating production storage and offloading unitNOEC No Observed Effect ConcentrationPAM Pulse Amplitude Modulated fluorometerPEC Predicted Exposure ConcentrationPhotosynthetic efficiency In vivo maximum quantum yield of

charge separation in photosystem II indark acclimated cells (FPSII_max)

PNEC Predicted No Effect ConcentrationPEMCA Peregrino Environmental Monitoring and Calcareous

AlgaePLS Partial least squaresPSII Photosystem II (oxygen evolving site)SSD Species Sensitivity DistributionWHP Wellhead platformWHPA Wellhead platform AWHPB Wellhead platform B

Fig. 9. Illustration of the potential impact after two years of drilling discharges fromWHPA and WHPB, expressed by the established impact categories. Prevailing current directionis south-westwards. Note: The colour coding is not identical to the colour coding used of the impact categories in Fig. 6.

References

Amado-Filho, G.M., Moura, R.L., Bastos, A.C., Salgado, L.T., Sumida, P.Y., Guth, A.Z.,Francini-Filho, R.B., Pereira-Filho, G.H., Abrantes, D.P., Brasileiro, P.S., Bahia, R.G.,Leal, R.N., Kaufman, L., Kleypas, J.A., Farina, M., Thompson, F.L., 2012. Rhodolithbeds are major CaCO3 bio-factories in the Tropical South West Atlantic. PLoSOne 7, e35171. http://dx.doi.org/10.1371/journal.pone.0035171.

Aro, E.-M., Virgin, I., Andersson, B., 1993. Photoinhibition of photosystem II. Inac-tivation, protein damage and turnover. Biochim. Biophys. Acta (BBA) e Bio-energetics 1143, 113e134. http://dx.doi.org/10.1016/0005-2728(93)90134-2.

Bakke, T., Klungsøyr, J., Sanni, S., 2013. Environmental impacts of produced waterand drilling waste discharges from the Norwegian offshore petroleum industry.Mar. Environ. Res. 92, 154e169. http://dx.doi.org/10.1016/j.marenvres.2013.09.012.

Basso, D., 1998. Deep rhodolith distribution in the Pontian Islands, Italy: a model forthe paleoecology of a temperate sea. Palaeogeogr. Palaeoclimatol. Palaeoecol.137, 173e187. http://dx.doi.org/10.1016/S0031-0182(97)00099-0.

Beyer, J., Myhre, L.P., Sundt, R.C., Meier, S., Tollefsen, K.-E., Vabø, R., Klungsøyr, J.,Sanni, S., 2012. Environmental risk assessment of alkylphenols from offshoreproduced water on fish reproduction. Mar. Environ. Res. 75, 2e9. http://dx.doi.org/10.1016/j.marenvres.2011.11.011.

Bryant, W.R., Bennet, R.H., Katherman, C.E., 1981. Shear strength, consolidation,porosity, and permeability of oceanic sediments. In: Emiliani, C. (Ed.), TheOceanic Lithosphere: the Sea, 7Willey, New York, pp. 1555e1616.

Davies, A.J., Roberts, J.M., Hall-Spencer, J., 2007. Preserving deep-sea natural heri-tage: emerging issues in offshore conservation and management. Biol. Conserv.138, 299e312. http://dx.doi.org/10.1016/j.biocon.2007.05.011.

De O. Figueiredo, M.A., Kain, J.M., Norton, T.A., 2000. Responses of crustose coral-lines to epiphyte and canopy cover. J. Phycol. 36, 17e24. http://dx.doi.org/10.1046/j.1529-8817.2000.98208.x.

Det Norske Veritas, 2013. Guideline; Monitoring of Drilling Activities in Areas withPresence of Cold Water Corals. Det Norske Veritas (Ed.) Report no. 2012-1691 /12NCQKD-2. http://www.norskoljeoggass.no/Global/2013%20Dokumenter/Publikasjoner/Monitoring%20of%20drilling%20activities%20-%20Areas%20with%20Cold%20Water%20Corals.pdf

Dombrowsky, E., Regnier, C., Lellouche, J.-M., Legalloudec, O., Desportes, C.,Drevillon, M., 2013. Quality Information Document for Global Sea PhysicalAnalysis and Forecasting Product. Report no. MYO2-GLO -QUID-001-002.http://catalogue.myocean.eu.org/static/resources/myocean/quid/MYO2-GLO-QUID-001-002-V1.2.pdf.

Durgut, I., Rye, H., Reed, M., Smit, M.G.D., Ditlevsen, M.K., 2015. Dynamic modelingof environmental risk associated with drilling discharges to marine sediments.Mar. Pollut. Bull. http://dx.doi.org/10.1016/j.marpolbul.2015.07.019 (in press).

Evans, D.L., Signorini, S.S., 1985. Vertical structure of the Brazil Current. Nature 315,48e50. http://dx.doi.org/10.1038/315048a0.

Figueiredo, M.A.O., Eide, I., Reynier, M., Villas-Boas, A.B., Tamega, F.T.S.,Ferreira, C.G., Nilssen, I., Coutinho, R., Johnsen, S., 2015. The effect of sedimentmimicking drill cuttings on deep water rhodoliths in a flow-through system:experimental work and modeling. Mar. Pollut. Bull. 95, 81e88. http://dx.doi.org/10.1016/j.marpolbul.2015.04.040.

Foster, M.S., 2001. Rhodoliths: between rocks and soft places. J. Phycol. 37, 659e667.http://dx.doi.org/10.1046/j.1529-8817.2001.00195.x.

Frost, T.K., Nilssen, I., Neff, J.M., Altin, D., Eide Lunde, K., 2006. ERMS Report no. 4;Toxicity of Drilling Discharges. http://www.sintef.no/project/ERMS/Reports/ERMS%20Report%20no%204_Toxicity%20report%20-%20signed%20version.pdf.

Genty, B., Briantais, J.-M., Baker, N.R., 1989. The relationship between the quantumyield of photosynthetic electron transport and quenching of chlorophyll fluo-rescence. Biochim. Biophys. Acta (BBA) e General Subjects 990, 87e92. http://dx.doi.org/10.1016/S0304-4165(89)80016-9.

Harbers, J.V., Huijbregts, M.A.J., Posthuma, L., van de Meent, D., 2006. Estimating theimpact of high-production-volume chemicals on remote ecosystems by toxicpressure calculation. Environ. Sci. Technol. 40, 1573e1580. http://dx.doi.org/10.1021/es051633m.

Harrington, L., Fabricius, K., Eaglesham, G., Negri, A., 2005. Synergistic effects ofdiuron and sedimentation on photosynthesis and survival of crustose corallinealgae. Mar. Pollut. Bull. 51, 415e427. http://dx.doi.org/10.1016/j.marpolbul.2004.10.042.

Henriques, M.C., Riosmena-Rodríguez, R., Coutinho, L.M., Figueiredo, M.A.O., 2014.Lithophylloideae and Mastophoroideae (Corallinales, Rhodophyta) from theBrazilian continental shelf. 2014 190, 18. http://dx.doi.org/10.11646/phytotaxa.190.1.9.

Johnsen, G., Sakshaug, E., 2007. Biooptical characteristics of PSII and PSI in 33species (13 pigment groups) of marine phytoplankton, and the relevance forpulse-amplitude-modulated and fast-repetition-rate fluorometry. J. Phycol. 43,1236e1251. http://dx.doi.org/10.1111/j.1529-8817.2007.00422.x.

Johnsen, S., Frost, T.K., Hjelsvold, M., Utvik, T.R., 2000. The Environmental ImpactFactor e a Proposed Tool for Produced Water Impact Reduction, Managementand Regulation. SPE International Conference on Health, Safety, and Environ-ment in Oil and Gas Exploration and Production. http://dx.doi.org/10.2118/61178-MS.

Johnsen, S., Nilssen, I., Pinto, A.P.B., Torkelsen, T., 2014. Monitoring of Impact ofDrilling Discharges to a Calcareous Algae Habitat in the Peregrino Oil Field inBrazil. SPE International Conference on Health, Safety, and Environment in Oiland Gas Exploration and Production. http://dx.doi.org/10.2118/168356-MS.

Kempf, M., 1970. Notes on the benthic bionomy of the N-NE Brazilian shelf. Mar.Biol. 5, 213e224. http://dx.doi.org/10.1007/BF00346909.

Littler, M.M., Littler, D.S., Dennis Hanisak, M., 1991. Deep-water rhodolith distri-bution, productivity, and growth history at sites of formation and subsequentdegradation. J. Exp. Mar. Biol. Ecol. 150, 163e182. http://dx.doi.org/10.1016/0022-0981(91)90066-6.

Littler, M.M., Littler, D.S., Blair, S.M., Norris, J.N., 1986. Deep-water plant commu-nities from an uncharted seamount off San Salvador Island, Bahamas: distri-bution, abundance, and primary productivity. Deep Sea Res. Part A Oceanogr.Res. Pap. 33, 881e892. http://dx.doi.org/10.1016/0198-0149(86)90003-8.

Neff, J.M., 2008. Estimation of bioavailability of metals from drilling mud barite.Integr. Environ. Assess. Manag. 4, 184e193. http://dx.doi.org/10.1897/IEAM_2007-037.1.

Nelson, W.A., 2009. Calcified macroalgae e critical to coastal ecosystems andvulnerable to change: a review. Mar. Freshw. Res. 60, 787e801. http://dx.doi.org/10.1071/MF08335.

Nilssen, I., Johnsen, S., 2008. Holistic Environmental Management of Dischargesfrom the Oil and Gas Industry e Combining Quantitative Risk Assessment andEnvironmental Monitoring. SPE International Conference on Health, Safety, andEnvironment in Oil and Gas Exploration and Production. http://dx.doi.org/10.2118/111586-MS.

Nilssen, I., Ødegård, Ø., Sørensen, A.J., Johnsen, G., Moline, M.A., Berge, J., 2015.Integrated environmental mapping and monitoring, a methodological approachto optimise knowledge gathering and sampling strategy. Mar. Pollut. Bull. 96,374e383. http://dx.doi.org/10.1016/j.marpolbul.2015.04.045.

OSPAR, 2012a. OSPAR Recommendation 2012/5 for a Risk-based Approach to theManagement of Produced Water Discharges from Offshore Installations. http://www.ospar.org/documents/dbase/decrecs/recommendations/12-05e_RBA%20Recommendation.doc.

OSPAR, 2012b. OSPAR Guidelines in Support of Recommendation 2012/5 for a Risk-based Approach to the Management of Produced Water Discharges fromOffshore Installations, (OSPAR Agreement: 2012-7). http://www.ospar.org/documents/dbase/decrecs/agreements/12-07e_rba%20guidelines.doc.

Reynier, M.V., Tamega, F.T.S., Daflon, S.D.A., Santos, M.A.B., Coutinho, R.,Figueiredo, M.A.O., 2015. Long-and short-term effects of smothering and burialby drill cuttings on calcareous algae in a static-renewal test. Environ. Toxicol.Chem. http://dx.doi.org/10.1002/etc.2938.

Riul, P., Targino, C.H., Farias, J.D.N., Visscher, P.T., Horta, P.A., 2008. Decrease inLithothamnion sp. (Rhodophyta) primary production due to the deposition of athin sediment layer. J. Mar. Biol. Assoc. U. K. 88, 17e19. http://dx.doi.org/10.1017/S0025315408000258.

Rye, H., Loekken, M., Moe, J., Ditlevsen, M.K., 2012. Simulation of concentrations anddepositions of particle matter caused by drilling discharges. Comparison be-tween field measurements and simulation results at coral locations. Int. Conf.Health, Saf. Environ. Oil Gas Explor. Prod. http://dx.doi.org/10.2118/156775-MS.

Rye, H., Reed, M., Frost, T.K., Smit, M.G.D., Durgut, I., Johansen, Ø., Ditlevsen, M.K.,2008. Development of a numerical model for calculating exposure to toxic andnontoxic stressors in the water column and sediment from drilling discharges.Integr. Environ. Assess. Manag. 4, 194e203. http://dx.doi.org/10.1897/IEAM_2007-039.1.

Salgado, C., Johnsen, S., Gomes, S., Tokelsen, T., 2010. Real-time EnvironmentalMonitoring of Discharges of Drill Cuttings in the Peregrino Field (Campos Basin,Brazil). SPE International Conference on Health, Safety, and Environment in Oiland Gas Exploration and Production. http://dx.doi.org/10.2118/127175-MS.

Singsaas, I., Rye, H., Frost, T.K., Smit, M.G.D., Garpestad, E., Skare, I., Bakke, K.,Veiga, L.F., Buffagni, M., Folium, O. a, Johnsen, S., Moltu, U. e, Reed, M., 2008.Development of a risk-based environmental management tool for drilling dis-charges. Summary of a four-year project. Integr. Environ. Assess. Manag. 4,171e176. http://dx.doi.org/10.1897/IEAM_2007-035.1.

Smit, M.G.D., Holthaus, K.I.E., Trannum, H.C., Neff, J.M., Kjeilen-Eilertsen, G.,Jak, R.G., Singsaas, I., Huijbregts, M.A.J., Hendriks, A.J., 2008. Species sensitivitydistributions for suspended clays, sediment burial, and grain size change in themarine environment. Environ. Toxicol. Chem. 27, 1006e1012. http://dx.doi.org/10.1897/07-339.1.

Stech, J.L., Lorenzzetti, J.A., 1992. The response of the South Brazil Bight to thepassage of wintertime cold fronts. J. Geophys. Res. Oceans 97, 9507e9520.http://dx.doi.org/10.1029/92JC00486.

Steller, D.L., Riosmena-Rodríguez, R., Foster, M.S., 2009. Living rhodolith bed eco-systems in the Gulf of California. In: Johnson, M.E., Ledesma, V.,J. (Eds.), Atlas ofCoastal Ecosystems in the Western Gulf of California: Tracking Limestone De-posits on the Margin of a Young Sea. University of Arizona Press, pp. 72e82.

Steller, D.L., Hernandez-Ayon, J.M., Riosmena-Rodriguez, R., Cabello-Pasini, A., 2007.Effect of temperature on photosynthesis, growth and calcification rates of thefree-living coralline alga Lithophyllum margaritae. Cienc. Mar. 33.

Tamega, F.T.S., Spotorno-Oliveira, P., Figueiredo, M.A.O., 2013. Catalogue of theBenthic Marine Life from Peregrino Oil Field, Campos Basin, Brazil. InstitutoBiodiversidade Marinha, Rio de Janeriro, p. 140.

Villas-Boas, A.B., Tamega, F.T.D.S., Andrade, M., Coutinho, R., Figueiredo, M.A.D.O.,2014. Experimental Effects of Sediment Burial and Light Attenuation on TwoCoralline Algae of a Deep Water Rhodolith Bed in Rio de Janeiro, Brazil. Cryp-togam. Algol. 35, 67e76. http://dx.doi.org/10.7872/crya.v35.iss1.2014.67.

Wilson, S., Blake, C., Berges, J.A., Maggs, C.A., 2004. Environmental tolerances offree-living coralline algae (maerl): implications for European marine conser-vation. Biol. Conserv. 120, 279e289. http://dx.doi.org/10.1016/j.biocon.2004.03.001.

Paper VIII

Osterloff, J., Nilssen, I., Nattkemper, T. W., 2016a. A computer vision approach for monitoring the spatial and temporal shrimp distribution at the LoVe observatory.

Methods in Oceanography, Vol 15-16, pp 114-128, doi: http://dx.doi.org/10.1016/j.mio.2016.03.002

Methods in Oceanography 15–16 (2016) 114–128

Methods in Oceanography

journal homepage: www.elsevier.com/locate/mio

Full length article

A computer vision approach for monitoring thespatial and temporal shrimp distribution at theLoVe observatory

Jonas Osterloff a,∗, Ingunn Nilssenb,c, TimW. Nattkemper a

a Biodata Mining Group, Faculty of Technology, Bielefeld University, 33615 Bielefeld, Germanyb Statoil ASA, Research and Technology, N-7005 Trondheim, Norwayc Trondhjem Biological Station, Department of Biology, Norwegian University of Science and Technology,

N-7491 Trondheim, Norway

h i g h l i g h t s

• First computational approach to monitor shrimp populations for long time periods with a camera-eqipped

observatory.

• Combination of superpixel representations and a new color change feature enables integration of shape and

color features of shrimp.

• Shrimp abundance heat maps enable for a first time rapid access to changes in shrimp abundances and spatial

distributions in the area.

Article history:

Received 30 June 2015

Received in revised form

22 December 2015

Accepted 23 March 2016

Available online 15 April 2016

Keywords:

Underwater informatics

Image analysis

Shrimp behavior

Environmental monitoring

Marine imaging

a b s t r a c t

This paper demonstrates how computer vision can be applied for

the automatic detection of shrimp in smaller areas of interest with

a high temporal resolution for long time periods. A recorded se-

quence of digital HD camera images from fixed underwater obser-

vatories provides unique opportunities to study shrimp behavior

in their natural environment, such as number of shrimp and their

abundance at different locations (micro habitats) over time. Tem-

poral color contrast featureswere applied to enable the detection of

the semi-transparent shrimp. To study the spatial–temporal char-

acteristics of the shrimp, pseudo-color visualizations referred to

as shrimp abundance maps (SAM) are introduced. SAMs for dif-

∗ Corresponding author.

E-mail address: josterlo@cebitec.uni-bielefeld.de (J. Osterloff).

http://dx.doi.org/10.1016/j.mio.2016.03.002

J. Osterloff et al. / Methods in Oceanography 15–16 (2016) 114–128 115

Pattern recognition

Spatial abundance

Spatial occurrence

ferent time periods are presented, to show the potential of the

methodology.

1. Introduction

A range of fixed long term underwater observatories (FUOs) equipped with a variety of sen-sors, e.g. digital cameras, have for some years become increasingly deployed (e.g. NEPTUNEBarnes et al., 2010, FixO3 Observatories, 2015, DELOS Vardaro et al., 2013, and LoVe Godøet al., 2014).FUOs constitute a rich collection of quantitative and qualitative data for high temporal resolutionmonitoring of smaller areas of interest over longer time periods. Basic knowledge of marine organ-isms living in their natural environment is scarce and FUOs provide a unique opportunity to gaingeneral knowledge about the present species natural variations related to behavior, describing theirspatial and temporal distribution in the various microhabitats covered by the camera frame (approx-imately 1–20m2) over time. Information of such natural variation is essential for detecting long-termchanges, seasonal fluctuations, sudden events, or possible impact from anthropogenic activities suchas oil drilling activities (Nilssen et al., 2015). According to Mortensen and Fosså(2006) the quantita-tive species composition of a reef-associated fauna is hard to study using the traditional monitoring,where point samples of organisms are brought into the laboratory and individuals are counted. Mostmobile species can escape during sampling and transport through the water column. In contrast tothis, imaging provides unique non destructive opportunities for establishment of general knowledgeon species specific behavioral variations.

Behavior is commonly used as an impact parameter for organisms (Buhl-Mortensen et al., 2015;Anderson, 2000). Recently Buhl-Mortensen et al. (2015) performed a manual analysis of a limitednumber (203) of images recorded with a stationary digital camera in a cold-water coral reef to mea-sure possible effects of water flow and drill cuttings particulates on polyp behavior of Lophelia pertusa.

The increasing number of images recorded by FUOs, requires automated or semi-automated an-alytical systems to handle the vast amount of data (Nilssen et al., 2015). While some work has beenpublished about computational segmentation of cold water corals (Purser et al., 2009; Tusa et al.,2014), computational segmentation of the associated fauna is scarce. One group of organisms thatare present in high numbers in Norwegian reefs is the shrimp (Purser et al., 2013). Several shrimpspecies have been identified, dominated by Pandalus spp. (Jonsson et al., 2004; Mortensen and Fosså,2006; Purser et al., 2013). Despite its documented presence shrimp abundance across micro-habitatsin coldwater coral reefs has not been studied in detail (Purser et al., 2013). In contrast to sessile species(e.g. corals), moving species are more difficult to monitor, either by manual annotation or by compu-tational approaches. Existing studies of image analysis applied on shrimp or morphological relatedspecies (Harbitz, 2007; Gorsky et al., 2010; Benfield et al., 2007) have removed the studied speciesfrom their natural habitat before imaging to gain optimized imaging conditions. The only paper weare aware of that considers computational in-situ shrimp detection uses shrimp eye reflection fromthe ROV (remotely operated underwater vehicle) lights in video images as a feature (Purser et al.,2013) and does consider the temporal context. In our study, the shrimp eye reflection approach couldnot be applied, as in the images of the FUO the shrimp eye contrast is not sufficient. Use of color forshrimp detection is challenging due to the shrimp semi-transparency. However, our temporal colorcontrast feature driven approach enables to handle the shrimp color variations. Themethod identifiesand maps the distribution of shrimp in the microhabitats of a cold water coral reef automatically. Thecomputed shrimp abundance is visualized as a shrimp abundance map (SAM).

2. Material

The fixed underwater observatory (FUO) LoVe1 (Lofoten—Vesterålen) is located in the north ofNorway 22 km off the coast at a depth of approximately 260 m (N 68° 54.474′, E 15° 23.145′). The

1 http://love.statoil.com.

116 J. Osterloff et al. / Methods in Oceanography 15–16 (2016) 114–128

FUO is connected via cable to the mainland allowing a two way communication (Godøet al., 2014).The connection also provides power supply to the different sensors mounted on the FUO, includinga digital camera (Canon EOS 550D in a waterproof housing: METAS DSF5210) with a flash (METASDSF4365) attached to an adjustable beamallowing to pan and tilt the camera. The camera is orientatedto face a Lophelia pertusa coral reef with a camera angle of 45°. In European waters Lophelia pertusa iscommonly the key reef building species (Freiwald et al., 1999). The multi spherical structure createsniche for other species (Purser et al., 2013). The image frame was selected based on earlier studies(Mortensen and Fosså, 2006; Purser et al., 2013; Jonsson et al., 2004), indicating that the compositionof the reef microhabitats (live corals, areas with bigger dead fragments and coral rubble where smallcoral fragments are mixed with sand). The distance between camera and coral reef is about 2–3 m

(Godøet al., 2014). The imaged area is about 10 m2, resulting in images with about 2 × 106 pixel

m2 . TheFUO is recording images since September 2013 (Godøet al., 2014), except for amaintenance break frommid-June till the beginning of October 2014. The camera is recording one image every hour (temporalresolution: 1 image h−1) with a resolution of 5184×3456 pixel and 8 bit per color channel.2 The datarecorded by the FUO is publicly available after registration at http://love.statoil.com. For our study wechoose 1140 images3 which were collected in within 6 weeks (1st of May 2014 0:17:19 till 18th ofJune 2014 08:17:19). An example image is shown in Fig. 1. A small subset of N = 80 of these imageswas uploaded to the Bio-Image Indexing and Graphical Labeling Environment (BIIGLE) (Ontrup et al.,2009; Schoening et al., 2009) platform. Using BIIGLE on these N images the positions of shrimp weremanually marked with point labels using a computer mouse. These labels were used as a training setfor our computational shrimp detection algorithm (see Chapter 3). Annotations are publicly accessiblevia https://ani.cebitec.uni-bielefeld.de/biigle/ with the username shrimp and the password shrimpunder the area ‘‘love’’ and transect ‘‘May June Selection’’. The annotators were allowed to place morethan one point annotation on each shrimp, so that many shrimp are represented by a small cloudof marked points (Fig. 4). In total 3459 shrimp point annotations were made on the 80 images. In apreviously performed experiment the intra observer agreement of one annotator was estimated. Theannotator had to re-label 20 images after a time gap of four days, resulting in an agreement of 0.66estimated on in total 3503 shrimp point annotations. This low agreement againmotivates the need foran automation of the detection. Nevertheless, it indicates that shrimp can be identified in the images,as otherwise this agreement would have been much lower.

3. Methods

The different steps of the shrimp detection and location analysis presented in Fig. 2 are describedin detail in the following subsections. After aligning the images spatially in the pre-processing(Section 3.1.1) temporal color contrast features are computed from the superpixel representation ofthe original image (Section 3.1.2). Using these features (Section 3.2), a supervised learning algorithmis trained using the 80 example images (Section 3.3). A pseudo color heat-map, referred to as shrimpabundance map (SAM) (Section 3.5), is derived from the positions found in all 1090 images4 to allowa first impression of the dynamics of shrimp positions.

3.1. Image pre-processing

In general, pre-processing of underwater images is essential, as these type of images often show anoticeable variation in illumination caused by changes in the camera–object distance. A variation inthe object–camera distance will also affect the color of the objects in underwater images as the lightattenuation is wavelength dependent. In this study no illumination or color correction was applied as

2 The operation mode of the camera was changed for the second deployment and is now recording raw image data with

14 bit per color channel.3 21 images of this time period had to be excluded or were not recorded.4 The first 50 images of the 1140 images were only used to compute the first shifting median image, see Section 3.1.2.

Fig. 1. Example image collected at the 1st of May 2014 0:17:19. The camera is facing towards a cold water coral reef with a

horizontal angle of 45° and is pointing 60° north. In the upper left part of the image mostly live corals can be found, in contrast

to the lower right part were mostly dead corals appear. The three blurry patches in the right upper quadrant and in the lower

middle were caused by biofouling, which stayed constant during the actual period.

Fig. 2. The computational shrimp detection and shrimp location analysis. Starting in the upper row from left with the

image collection at the LoVe FUO, image registration (Section 3.1.1) and shrimp annotation within BIIGLE. Continuing

with superpixel segmentation (Section 3.1.3), feature extraction (Section 3.2), machine learning (Section 3.3) (middle row),

automated shrimp detection (Section 3.4) and visualization of the spatial shrimp distribution over time in a shrimp abundance

map (Section 3.5) (lower row).

the image collecting setup is stable in this regard. The object–camera distance is steady and relativelyshort, i.e. 2–3 m (Godøet al., 2014) (Fig. 1). The artificial illumination is evenly distributed throughoutthewhole image and for the total set of images. In order to use the high likeliness of shrimpmovementbetween two images (i.e. one hour time lag) as a feature for discrimination between shrimp and not

shrimp, all images need to be registered (i.e. spatially aligned) as the camera is slightly shifted by thepresent current speed and direction.

3.1.1. Image registration

All 1140 images (heightH = 5184 pixels, widthW = 3456 pixels) are registered to the first imageI(0), the so called reference image, collected on 1st of May 2014 0:17:19 to compensate the small cam-era shiftings. I(0) is initially cropped by 50 pixels on all four image borders, allowing a shift of maximal50 pixels in all four directions for the following images, the so called moving images. The best align-ment shift is computed, using the Canny edge (Canny, 1986)map E(n) of image I(n)with a low threshold

of 30 and a high threshold of 90. Each image I(n) is transformed to a new registered version I(n)

I(n) = ψ(I(n), (xbest, ybest)(n)), (1)

using a transform function ψ , i.e.

ψ(I, x, y) = A ∈ R(H−100)×(W−100) with Aj,k = I50+j−x,50+k−y

that (i) reduces the width and height of the image (see cropping instructions above) and (ii) shifts allpixels by the vector (x, y). The optimal shifting parameters (xbest, ybest)(n) are computed as

(xbest, ybest)(n) = argmax(x,y)∈{−50,...,50}2

{∑j,k

(ψ(E(n), x, y))j,k · (ψ(E(n), 0, 0))j,k}.

3.1.2. Temporal color contrast features

Shrimp are semi-transparent, which causes a color variation of the individual shrimp dependentof the background color. To enhance the contrast between shrimp and background we propose toinclude additional information from previously recorded images. In order to include this in a feature

representation at a pixel (x, y), the temporal color contrast feature C(n)

(x,y) is computed for each pixel inthe CIELab color space (Schanda, 2007) by

(x,y) = I(n)

(x,y) − I(n)

(x,y), (2)

with I(n) as the floating median image.

The floating median image I(n) is also computed in the CIELab color space, resulting in an imageincluding only sessile objects and thereby an image without shrimp (Fig. 3, lower left).

I(n)j,k = I

(mj,k)

j,k (3)

where the position of the median computed on the luminance part of the CIELab color space gives the

image pixel Imj,k used as the pixel (j, k) for the median image:

mj,k = arg medianLj,k

n=n−50,...,n

{I(n)j,k

The result can be visualized as a contrast map C (n) by computing

C (n) = I(n) − I(n)

2+ 127(Fig. 3, lower right). (4)

3.1.3. Super-pixel segmentation

All images I are pre-segmented using the SLIC (Simple Linear Iterative Clustering) superpixelalgorithm (Achanta et al., 2012) with a feature weight of 40 and a step size of 20 pixels, as shownin Fig. 4. The superpixel segmentation enables a reduction of the number of feature extractionpoints. Instead of extracting features on e.g. overlapping quadratic regions, image information of localpixel similarities is used to find regions for feature extractions. The superpixel size was optimizedto cover the size variation of the shrimp, resulting in a size of one superpixel for the smallestshrimp.

Fig. 3. Visualization of the pre-processing steps from the original image to the shrimp contrast map. In the upper left the

original image I can be found. In the upper right the registered image I (see Eq. (1)) is displayed, which is slightly shifted. In the

lower left the median image I (see Eq. (3)) is shown, where no moving objects like the shrimp can be found. In the lower right

the shrimp contrast map C (see Eq. (4)) is displayed.

Fig. 4. SLIC (Simple Linear Iterative Clustering) superpixel segmentation is applied and shown for one example image. White

contours outline the superpixel areas. In the zoomed area examples for annotations of a shrimp (blue color) can be found,

visualizing multiple annotations per shrimp. (For interpretation of the references to color in this figure legend, the reader is

referred to the web version of this article.)

Table 1Description of the features.

LI , aI , bI : Lab mean values computed from I

h0,L,I , . . . , h15,L,I , . . . , h15,b,I : Lab 16 bin histogram values computed from I

LC , aC , bC : Lab mean values computed from C

h0,L,C , . . . , h15,L,C , . . . , h15,b,C : Lab 16 bin histogram values computed from C

Fig. 5. Illustration for the post-processing of the output of the Random Forest (RF) shrimp classification. (a) has a superpixel

size of one and will therefore be excluded, (b) has a superpixel size of five and therefore fulfills the size criteria while (c) will

be excluded as it is to big (11 superpixels).

3.2. Feature extraction

Features are extracted for each superpixel si on I(n) and corresponding C (n) represented by a featurevector xi. These 38 features are mean values and histogram values (16 bins) in the CIELab color space(see Table 1). A label li ∈ (1 = shrimp,−1 = no shrimp) is assigned to each feature vector xireferring to the corresponding superpixel. All 38-dimensional xi are thus either labeled shrimp if oneore more shrimp point labels were placed in the area of the corresponding superpixel or not shrimpotherwise.

3.3. Machine learning

For the automated detection of shrimp the supervised classification algorithm Random Forest (RF)(Breiman, 2001) was selected. The algorithm allows to use a smaller training set, compared to e.g. theDeep Learning algorithms like Convolutional Neural Networks (LeCun et al., 1998). Furthermore it isvery robust against overfitting and can also be used to compute the feature importance (Fig. 14). TheRF is trained and tuned using a stratified four-fold cross validation (Kohavi, 1995). Therefore the set of80 annotated images is divided into four different sets of training and test images. The RF is trained onthe labeled feature vectors (xi, li) of the training set of images. The training parameters of the RF aretuned on feature vectors and their corresponding label belonging to the test set of images and werevaried for the maximum tree-depth (5, 10, 20, 50, 100, 150), the maximum number of trees in theforest (100, 500) and the minimum sample count (5, 10, 20). The class-weights for the shrimp classwere also varied (2, 5, 10, 100). Other parameters were not varied, resolving in a training of in total144 different RFs.

3.4. Post-processing

To optimize the classification performance the classification output is post-processed using a size

criterion. Connected superpixels, referred to as regions r(n)m , that are all classified to be shrimp forming

a shrimp region of size |r| (|r| = # connected superpixel) are filtered to be in range of |r| > 1superpixel, |r| < 11 superpixel, excluding very small (|r| = 1) and very large regions (|r| > 10)of shrimp classifications (Fig. 5).

3.5. Shrimp abundance map

The shrimp abundance map (SAM) visualizes the spatial distribution of the shrimp over a definedtime-period T . The cold water coral reef scene is divided into a grid of equal sized quadrants q(width = 120 pixels, height = 120 pixels). For the quadrants qm the relatively shrimp abundanceα is visualized using a heated object color scale with 9 levels (9-class YlOrRd (Brewer, 2013, 1999)).The color scale is starting at dark red for the lowest abundance and ending at a light yellow for thehighest abundance (e.g. Fig. 10, on the right). The relative shrimp abundance is computed as

αm = log2

⎛⎝ γm − min

m′ γm′

maxm′ γm′ − min

m′ γm′+ 1

⎞⎠ , (5)

where γm is the shrimp abundance for the quadrant qm

γm =∑n∈T

|oni | with sni ∩ qm �= {}.

An average image of the images of the visualized time-period is used as a background image (Fig. 10).

3.6. Accuracy assessment

The assessment of an image analysis system’s accuracy can be a complex subject. In generalmanual annotations of image samples are compared with classification outputs obtained for thesame data and grouped into true positives (TPs), false positives (FPs), false negatives (FNs) and truenegatives (TN). In this two-class classification problem, here shrimp vs. not shrimp, matches of positivemanual annotations and positive classifications are counted as TPs, positive manual annotations witha negative classification output are counted as FNs and positive classification outputs which were notannotated (i.e. not shrimp label) are counted as FPs. True negatives are not considered as no manualnegatives (i.e. not shrimp) were annotated explicitly.

The exact pixel location of the annotation (point labels are used in this study) though is more orless random in a certain extend across the object of interest (OOI), i.e. the shrimp in this case, whichis intended to be annotated, as the OOI has a spatial extent, not to mention that the use of a computermouse also comes with accuracy limitations. This brings up the problem that a classification outputcan be counted as a FP and the annotation as a FN, although both are located on the same OOI, but noton the exact same pixel location. For the purpose of identifying the spatial distribution of shrimp overtime, it is sufficient enough to detect only parts of a shrimp, as long as for each shrimp a part e.g. thetail is detected. A solution can be to allow limited distances between annotations and classificationas it is done by Schoening et al. (2012). However, in case of very small OOI like shrimp this approachbares the potential of an assignment problem for close shrimp.

In our approach labels are assigned to superpixels, representing a pixel area, which extended thepoint label to a superpixel label. Comparing the label li of a superpixel si with the correspondingclassification output oi therefore reduces the previous mentioned location problem. Furthermore ourOOIs consist of several annotations and several superpixels forming a set of annotated superpixels.

We therefore create a binary mask image B(n)L from the labels {l(n)i } and a separate one B

(n)O from the

positive RF outputs {o(n)i } and compare them to determine TP, FP and FN counts.

B(n)L is a binary image, where pixel values are set to B

(n)L = 1 if the corresponding superpixel label

holds l(n)i = 1 (i.e. a shrimp label) and B

(n)L = 0 otherwise. B

(n)O is a binary image, where pixel values are

set to B(n)O = 1 if the classification output holds o

(n)i = 1 for the corresponding superpixel (i.e. shrimp)

and B(n)O = 0 otherwise. The morphological image-dilation function φ is applied to B

(n)O allowing a

neighboring positive annotation to count as a true positive (TP) (Fig. 6(b)). The B(n)L,O is evaluated for all

connected regions r(n)m . Each connected region r

(n)m is counted as a TP, if parts of it are marked in B

Fig. 6. Patches from an evaluation image B(n)L,O computed for the accuracy assessment of the shrimp detection. (a), (b) displaying

both true positives (TPs). Please note that (b) is counted as a TP due to the dilation rule (see text for details). (c) is displaying a

false positive (FP) and (d) a false negative (FN). Pixel values are coded in red for pixels only marked in the binary classification

output image B(n)O , in blue for pixels only marked in the binary label image B

(n)L and in pink for pixels marked in both. White

contours illustrate the superpixel areas. The effect of the dilation function is also visible as a small boarder around the superpixel

classified as shrimp. (For interpretation of the references to color in this figure legend, the reader is referred to the web version

of this article.)

and φ(B(n)O ), as a false positive (FP) if the region is only marked in φ(B

(n)O ) and as a false negative (FN)

if r(n)m is only marked in B

(n)L (Fig. 6).

TP, FP, and FN are finally used to calculate the precision and the recall for the shrimp classificationsystem. The precision ismeasuring the number of correct identified shrimpover all shrimpdetections:

precision = TP

TP + FP. (6)

The recall is measuring how many of the annotated shrimp are correctly identified to be shrimp bythe Random Forest (RF):

recall = TP

TP + FN. (7)

The F1 score is computed and used to optimize the classification system, toweight precision and recallequally, i.e. the harmonic mean between both:

F1 score = 2 · precision · recallprecision + recall

Based on the assumption that the ground truth is correct, i.e. no shrimp are overseen and no ‘‘noneshrimp area’’ is labeled as shrimp, the best accuracywould be a precision of 1 and a recall of 1 resultingin an F1 score of 1.

4. Results

The results of the applied previous described methods are presented in the following sections.

4.1. Registration

The registration was evaluated by visual inspection of a rendered time-laps video. This showedthat the registration process reduced the shift between the images, even though the applied methoddoes not correct for a rotation.

4.2. Superpixel segmentation

The superpixel segmentation was evaluated manually using visual inspection. Most of the shrimpwere segmented into several superpixels (Fig. 4) and only a few very small shrimp were covered byonly one superpixel. In a re-inspection of these one superpixel sized shrimp we observed a low levelof reproducibility in the manual annotation of the shrimp-like objects. Shrimp positions were markedin the training set to consider this phenomenon in the following accuracy assessment.

Table 2The parameters of the Random Forest classifier are optimized using a stratified four-fold

cross validation. The results averaged of the four-fold cross validation for training and

test image sets are given here. The performance of the Random Forests is measured by

recall, precision and F1 score using the assessment method presented in Section 3.6.

Precision Recall F1 score

Training 0.48 1 0.64

Test 0.69 0.52 0.59

Table 3Results of the manual re-evaluation of the test images (Re-ev. test), including the results

from the previous automated evaluation of the test images (Test). The manual re-

evaluation was performed with the conservative ground truth (see text for details).

Precision Recall F1 score

Test 0.69 0.52 0.59

Re-ev. test 0.77 0.91 0.84

Fig. 7. Examples of annotations (outlined in blue) that were excluded in the final manual re-evaluation process as they do

not fulfill the size criteria |rLm| > 1 superpixel. (For interpretation of the references to color in this figure legend, the reader is

4.3. Accuracy assessment of the shrimp detection

The accuracy assessment described in Section 3.6 was applied to optimize the training parametersof the RF to give the highest F1 score. In Table 2 averages for precision, recall and F1 score of thefour-fold cross validation for the best RF parameters are given. The best F1 score was achieved witha tree-depth of 50, a minimal sample count of 20, 10 active variables, a shrimp class weight of 10, amaximal number of trees of 100 and an out of bag error of 0.001 (the last two both used as terminationcriteria during the training).

Manual annotation can be very error-prone (Schoening et al., 2012; MacLeod et al., 2010). Someshrimp will be likely to be overseen during the manual annotation process, as it is expected thatthe performance of the annotator will vary, as we have reported in our intra-observer study (seeSection 2). The classification outputs were therefore manually re-evaluated region wise (Fig. 8).Since we observed a rather low reproducibility for a number of small shrimp (see Section 4.2) were-evaluated the computational results with a second, more conservative ground truth: connected

regions inB(n)L were excluded if they only consist of one superpixel (Fig. 7). This resulted in an improved

F1 score of 0.84 (see Table 3).

4.4. Shrimp abundance maps (SAM)

The number of detected shrimp in each image over the analyzed time period is displayed in thechart of Fig. 9. The locations of the automated detected shrimp in each image were used to generateshrimp abundance maps (SAMs) for different time scales and periods. Fig. 10 displays the relativeshrimp abundance of all 1090 analyzed images (3rd of May 2014 02:17:19 till 18th of June 201408:17:19). In Fig. 11 a SAM obtained by analyzing images recorded at day time between 8 am and

Fig. 8. A selection of re-evaluated classification outputs, displaying true positives (TPs, row 1), false positives (FPs) which are

actually TPs (row 2), FPs (row 3) and in false negatives (FNs, row 4).

Fig. 9. The shrimp occurrence chart displays the number of detected shrimp (# shrimp) per image over the time period from

early May (3rd of May 2014 02:17:19) till middle of June 2014 (18th of June 2014 08:17:19). The number of shrimp is counted

as the number of connected shrimp regions in B(n)O (see Section 3.6).

8 pm is compared with a SAM obtained by analyzing images recorded at night time between 8 pmand 8 am. In Fig. 12 a SAM for the images from 11th of May is compared with the SAM for the imagesfrom 18th of May. Fig. 13 displays a comparison of the first half of the images (first 3 weeks, 3rd ofMay 2014 02:17:19 till 25th of May 2014 21:17:19) with the second half of the images (following 3weeks, 25 of May 2014 22:17:19 till 18th of June 2014 08:17:19).

5. Discussion

A computer vision approach for the automated detection of shrimp in Lophelia pertusa reef hasbeen established. The relative spatial abundance of shrimp over time can be visualized in SAMs andallows the identification of behavioral changes in an intuitive way.

The application of superpixel algorithm for pre-segmentation accelerated the automated detectionand provided a reasonable criterion (|r|) to filter manual annotations and automated classifications.

Fig. 10. The shrimp abundance map (SAM) visualizing the spatial shrimp distribution from early May (3rd of May 2014

02:17:19) till middle of June 2014 (18th of June 2014 08:17:19). The color scale is displayed on the right starting with dark

red for the lowest relative shrimp abundance αm (Eq. (5)), continuing to red and orange indicating a higher relative shrimp

abundance and ending at a light yellow for the highest relative abundance over time. (For interpretation of the references to

color in this figure legend, the reader is referred to the web version of this article.)

Fig. 11. Two shrimp abundance maps (SAMs) comparing the relative shrimp abundance at day time between 8 am and 8 pm

(left) with the relative shrimp abundance at night time between 8 pm and 8 am (right) for the time period early May till middle

of June 2014 (3rd of May 2014 02:17:19 till 18th of June 2014 08:17:19).

Fig. 12. Two shrimp abundance maps (SAMs) comparing the relative shrimp abundance at 11th of May (left) with the relative

shrimp abundance at 18th of May (right).

Using the temporal color contrast feature enabled to discriminate between shrimp and not shrimp,as it combines information from previous recorded images with the current image. This improvementof the ability was validated using a method introduced by Osterloff et al. (2014), comparing different

Fig. 13. Two shrimp abundance maps (SAMs) comparing the relative shrimp abundance of the first 3 weeks (3rd of May 2014

02:17:19 till 25th of May 2014 21:17:19) (left) with the relative shrimp abundance of the next three weeks (25 of May 2014

22:17:19 till 18th of June 2014 08:17:19) (right).

Fig. 14. The feature importance for the best Random Forest. The importance for each variable is displayed, differentiating

between features computed on I colored in blue and features computed on C colored in green. A slightly higher importance

for features computed on C can be observed. (For interpretation of the references to color in this figure legend, the reader is

image processing methods, and by evaluating the feature importance of the best RF (total importance

of 0.59 for C and total importance of 0.41 for I) (Fig. 14). However, features computed on I were notexcluded, since it would have lowered the accuracy, as their importance was still high.

The F1 score of 0.84, estimated in themanual re-evaluation, confirms that we established a reliableclassification system. During themanual re-evaluation itwas also observed thatmost of the remainingFPs were parts of relatively big objects like legs and body parts of crabs, parts of sea urchins, sea liliesand basket stars. Detection sensitivity is dependent on local image quality. Non illuminated areas ofthe images like holes or shadows, caused by the multi-spherical structures of the coral reef, lead to alower detection sensitivity. However, as long as the camera position is fixed, we can presume that thiseffect is constant over time. Obviously the dark spot areas caused by themulti-spherical structure andlight settings, must be handled with care, if someone wants to compare results from different FUOs.

The presented system computes shrimp abundance charts for evaluation of the number of visibleshrimp per image and SAMs for visualization of the relative spatial abundance in the same images overtime. Qualitative differences and similarities can be recognized in an intuitive way e.g. in Figs. 12 and11, respectively. The visual differences between the relative shrimp abundance of the first threeweeksand the following three weeks (Fig. 13), noticed during a manual inspection, were caused by some FPclassification of sea lilies, which moved in the field of view in the second time period. Therefore, inthe next development stage of the system, a pre-detection of big objects (in relation to the shrimp)is advised. Also, an exclusion of images with very low visibility, e.g. caused by high turbidity, in the

pre-processing would increase the reliability of the system. Furthermore, varying inherent opticalproperties (dissolved organic material, plankton and in-organic particulates) of the water will changethe attenuation of light, impacting color and contrast of the images and might affect the detectionsensitivity, as well. Although no such changes were recognized during the analyzed time period, itshould be considered in the pre-processing of an improved system. Variations in the inherent opticalproperties are expected to occur over time. The results from the present study show predominantlyhigher shrimp abundance in the live coral parts compared to the coral rubble or sediment parts of theimage. This distribution corresponds to the findings of the study by Purser et al. (2013). The shrimpdistribution will be investigated in detail in future studies, following statistical analysis of the manualannotation sensitivity and automated detection performances on both areas. As in the study by Purseret al. (2013), no attempt was made in the present study to identify shrimp down to species level.Reliable classification of shrimp requires additional physical sampling and no such sampling has sofar been performed on the LoVe observatory. Even though a more detailed manual classification ofthe shrimp from the images might be possible with additional sampling, automatic separation at thepresent state is not achievable due to shrimp similarity and image resolution.

Knowledge of natural variations is essential to perform a sound environmentalmonitoring (Nilssenet al., 2015). Our next step will be to apply the method described in the present paper on a largerdataset to investigate possible seasonal and yearly variations in the shrimp abundance. Furthermore,we will try to correlate natural variations with the other measured variables, such as turbidity andcurrent speed and direction, available from the LoVe ocean observatory.

Acknowledgment

Financial support was given by STATOIL ASA, Research and Technology, Norway.

References

Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., Susstrunk, S., 2012. Slic superpixels compared to state-of-the-art superpixelmethods. IEEE Trans. Pattern Anal. Mach. Intell. 34 (11), 2274–2282.

Anderson, P.J., 2000. Pandalid shrimp as indicators of ecosystem regime shift. J. Northwest Atl. Fish. Sci. 27, 1–10.Barnes, C.R., Best, M.M.R., Johnson, F.R., Pirenne, B., 2010. Final installation and initial operation of the world’s first regional

cabled ocean observatory (NEPTUNE Canada). Can. Meteorol. Oceanogr. Soc. Bull. 38 (3), 89–96.Benfield, M., Grosjean, P., Culverhouse, P., Irigolen, X., Sieracki, M., Lopez-Urrutia, A., Dam, H., Hu, Q., Davis, C., Hanson, A.,

Pilskaln, C., Riseman, E., Schulz, H., Utgoff, P., Gorsky, G., 2007. RAPID: Research on automated plankton identification.Oceanography 20 (2), 172–187.

Breiman, L., 2001. Random forests. Mach. Learn. 45 (1), 5–32.Brewer, C.A., 1999. Color use guidelines for data representation. In: Proceedings of the Section of Statistical Graphics, pp. 55–60.Brewer, C.A., 2013. ColorBrewer: Color Advice for Maps. URL: http://colorbrewer2.org.Buhl-Mortensen, P., Tenningen, E., Tysseland, A.B.S., 2015. Effects of water flow and drillingwaste exposure on polyp behaviour

in lophelia pertusa. Mar. Biol. Res. 11 (7), 725–737.Canny, J., 1986. A computational approach to edge detection. IEEE Trans. Pattern Anal. Mach. Intell. 8 (6), 679–698.Freiwald, A., Wilson, J.B., Henrich, R., 1999. Grounding pleistocene icebergs shape recent deep-water coral reefs. Sediment.

Geol. 125 (1–2), 1–8.Godø, O.R., Johnsen, S., Torkelsen, T., 2014. The love ocean observatory is in operation. Mar. Technol. Soc. J. 48 (2), 24–30.Gorsky, G., Ohman, M.D., Picheral, M., Gasparini, S., Stemmann, L., Romagnan, J.-B., Cawood, A., Pesant, S., García-Comas, C.,

Prejger, F., 2010. Digital zooplankton image analysis using the zooscan integrated system. J. Plankton Res. 32 (3), 285–303.Harbitz, A., 2007. Estimation of shrimp (Pandalus borealis) carapace length by image analysis. ICES J. Mar. Sci. 64 (5), 939–944.Jonsson, L., Nilsson, P., Floruta, F., Lundälv, T., 2004. Distributional patterns of macro- and megafauna associated with a reef of

the cold-water coral Lophelia pertusa on the Swedish west coast. Mar. Ecol. Prog. Ser. 284 (15), 163–171.Kohavi, R., 1995. A study of cross-validation and bootstrap for accuracy estimation and model selection. In: International Joint

Conference on Artificial Intelligence, Vol. 14, no. 12, pp. 1137–1143.LeCun, Y., Bottou, L., Bengio, Y., Haffner, P., 1998. Gradient-based learning applied to document recognition. Proc. IEEE 86 (11),

2278–2324.MacLeod, N., Benfield, M., Culverhouse, P., 2010. Time to automate identification. Nature 467 (7312), 154–155. 09.Mortensen, P.B., Fosså, J.H., 2006. Species diversity and spatial distribution of invertebrates on deep-water Lophelia reefs in

Norway. In: Proceedings of 10th International Coral Reef Symposium, pp. 1849–1868.Nilssen, I., Ødegård, Ø., Sørensen, A.J., Johnsen, G., Moline, M.A., Berge, J., 2015. Integrated environmental mapping and

monitoring, a methodological approach to optimise knowledge gathering and sampling strategy. Mar. Pollut. Bull..Observatories, F.-P.O.O., 2015. Observatories. URL: http://www.fixo3.eu/observatory/.Ontrup, J., Ehnert, N., Bergmann, M., Nattkemper, T.W., 2009. BIIGLE-web 2.0 enabled labelling and exploring of images from

the Arctic deep-sea observatory HAUSGARTEN. In: OCEANS 2009-EUROPE. IEEE, pp. 1–7.

Osterloff, J., Schoening, T., Bergmann, M., Durden, J.M., Ruhl, H.A., Nattkemper, T.W., 2014. Ranking color correction algorithmsusing cluster indices. In: 2014 ICPR Workshop On Computer Vision for Analysis of Underwater Imagery, CVAUI, pp. 41–48.

Purser, A., Bergmann,M., Lundälv, T., Ontrup, J., Nattkemper, T.W., 2009. Use ofmachine-learning algorithms for the automateddetection of cold-water coral habitats: a pilot study. Mar. Ecol. Prog. Ser. 397.

Purser, A., Ontrup, J., Schoening, T., Thomsen, L., Tong, R., Unnithan, V., Nattkemper, T.W., 2013. Microhabitat and shrimpabundance within a norwegian cold-water coral ecosystem. Biogeosciences 10 (9).

Schanda, J., 2007. Colorimetry: Understanding the CIE System. John Wiley & Sons.Schoening, T., Bergmann, M., Ontrup, J., Taylor, J., Dannheim, J., Gutt, J., Purser, A., Nattkemper, T.W., 2012. Semi-automated

image analysis for the assessment of megafaunal densities at the arctic deep-sea observatory hausgarten. PLoS One 7 (6),e38179. 06.

Schoening, T., Ehnert, N., Ontrup, J., Nattkemper, T., 2009. BIIGLE tools: Aweb 2.0 approach for visual bioimage databasemining.In: Information Visualisation, 2009 13th International Conference, pp. 51–56.

Tusa, E., Reynolds, A., Lane, D., Robertson, N., Villegas, H., Bosnjak, A., 2014. Implementation of a fast coral detector using asupervised machine learning and gabor wavelet feature descriptors. In: Sensor Systems for a Changing Ocean, SSCO, 2014IEEE, pp. 1–6.

Vardaro, M.F., Bagley, P.M., Bailey, D.M., Bett, B.J., Jones, D.O.B., Milligan, R.J., Priede, I.G., Risien, C.M., Rowe, G.T., Ruhl, H.A.,Sangolay, B.B., Smith, K.L.J., Walls, A., Clarke, J., 2013. A Southeast Atlantic deep-ocean observatory: first experiences andresults. Limnol. Oceanogr. Methods 11, 304–315.

Paper IX

Möller, T., Nilssen, I., Nattkemper, T. W., In press. Change detection in marine observatory image streams using Bi-Domain Feature Clustering. International

Conference on Pattern Recognition, Cancún, México

Change Detection in Marine Observatory ImageStreams using Bi-Domain Feature Clustering

Torben Moller∗, Ingunn Nilssen†‡, Tim W. Nattkemper∗∗Biodata Mining Group, Bielefeld University, Bielefeld 33615, Germany

Email: tmoeller@cebitec.uni-bielefeld.de, tim.nattkemper@uni-bielefeld.de†Statoil ASA, Research and Technology, Trondheim 7005 Norway

‡Department of Biology, Norwegian University of Science and Technology, Trondheim 7491, Norway

Email: innil@statoil.com

Abstract—Vision based environmental monitoring using fixedcameras generates large image collections, creating a bottleneckin data analysis. In areas with limited background knowledge ofthe monitored habitat, this bottleneck can often not be overcomeby traditional pattern recognition methods. A new change detec-tion method to identify interesting events such as presence andbehavior of different species is proposed. The change detectionmethod uses the new Bi-Domain Feature Clustering (BDFC).BDFC integrates the location of a feature vector in the featurespace as well as the location in the image into the clustering.Firstly, BDFC is applied to a time dependent representation ofthe image stream to identify regions of similar change. Secondlyit is applied to a time independent representation to group thesechanges into categories. These categories can rapidly be assessedby a human observer to bypass the time consuming inspection ofthe whole data set. To make the posterior browsing of detectedchanges more efficient, a relevance factor computed for eachcategory is proposed.

The approach is demonstrated with experimental runs, usingimages from the Lofoten Vesteralen ocean observatory, showingthe potential to harvest changes of interest and novelties in largeimage collections.

I. INTRODUCTION

Change detection is a commonly used approach to evaluate

images taken from a fixed position, and has already been

of great interest in a variety of scientific fields [1] such

as video surveillance (e. g. traffic) [2]–[5] and remote

sensing [6]–[16] (e. g. deforestation monitoring or monitoring

of destruction/construction in urban areas). In the context of

underwater imagery applications are still limited [17]–[22].

The two most common change detection approaches are

background subtraction (BS) and change vector analysis

(CVA). BS aims at building and maintaining a background

model of the investigated images and then thresholding the

difference between the background model and a target image.

Lots of efforts have been put into background maintenance

as well as in computing a suitable threshold for the pixel

differences. CVA methods basically compute the difference

image between two images before analyzing the vectors of

the difference image (so-called change vectors).

Whereas most change detection algorithms compute a binaryoutput by assigning each pixel to either true or false,

indicating that changes have occurred or not, some algorithms

subdivide the set of changed-pixels into various subclasses.

In [12] and [13] the authors introduce CVA algorithms

detecting and categorizing changes based on the length

and the direction of each change-vector, respectively. Other

change detection algorithms that do not fall within the two

categories described abov are for instance based on temporal

predictive models [23]–[25], significance tests [26] or slow

feature analysis [27]. However, all these methods require a

human observer to identify changes of interest in a threshold

image or a set of transformed images representing changes.

In recent years, a growing number of fixed long-term un-

derwater observatories (FUO, [28]–[30]) have been deployed

to monitor marine habitats. The FUO are in most cases

deployed in areas of particular interest, such as areas with

high biodiversity. The biological diversity and abundance of

species present at these sites represent considerable challenges

to a change detection approach:

(1) Prior knowledge about which species to find in the

monitored areas and/or how they behave is often limited

and can therefore not be used in the design of a change

detection method.

(2) As some species occur and/or move much more frequent

than other species, change detection in pixel values only

is not sufficient to detect relevant changes in the moni-

tored scene.

The contribution of this paper is to introduce a new approach

to detect various changes in an area with very limited a-

priori knowledge. The approach includes the concept of super-

pixel segmentation in feature space and the new Bi-Domain

Feature Clustering (BDFC). In contrast to existing methods,

this approach finds regions of similar change patterns, groups

all detected regions into categories and ranks the categories

according to a proposed relevance based on frequency and

pixel differences.

II. METHODS

The aim in the analysis of images from an FUO is the

identification of regions of interest as connected components

with similar change patterns. As similar changes and patterns

are expected at different time-points, the similarity of change

features at different time-points must be considered as well.

The proposed framework (Fig. 1) integrates these aspects by

Fig. 1. Time dependent features and time independent features are extracted from an image sequence in Section II-B. Superpixel segmentation and thenew Bi-domain Feature Clustering (BDFC) are applied to each of the resulting feature maps in II-A. In Section II-C the change patterns are classified intocategories with semantic meaning and relevant change is discriminated from irrelevant change.

computing two feature representations (Section II-B). The

new clustering procedure (Section II-A) is used to cluster

the time dependent change features and the time independent

change features, respectively. Regions showing similar change

patterns are identified using the time dependent change

features. The regions are assigned to clusters of the time

independent features, which represent categories of change.

The method assumes that all images are in CIELUV [31]

color space. Moreover, we assume that the images are taken

with the same fixed camera at all time-points implying that all

the images have the same height H and the same width W .

The latter assumption allows us to write all images in the

processed image sequence as mappings with a common do-

main P = {(w, h) | 1 ≤ w ≤ W, 1 ≤ h ≤ H}. Consequently,

the image taken at time-point t is a mapping It : P → R3

assigning to each pixel p a CIELUV color It (p). We denote

the L, u and v channels of It as I1t , I2t and I3t , respectively.

To retain spatial information during feature extraction so

called feature maps are used in this paper. A feature map

is a mapping F : P → RN , assigning a feature vector to

each pixel and hence is a straightforward generalization of an

image. A feature map will always be denoted by FA indicating

that it is computed by a specific algorithm A (A ∈ {tmp,int}Section II-B). Note that in this paper, features are extracted

from image sequences not single images.

A. Bi-domain Feature Clustering

Feature maps contain feature vectors and their positions in

the image domain, i. e. the pixel position where a feature vec-

tor is extracted. Traditional clustering algorithms (vector quan-

tization, hierarchical clustering [32], spectral clustering [33])

do not use the image domain as they are not primary designed

for image clustering. To increase the homogeneity in regions

of semantic meaning (Fig. 2), we include the image domain

into the cluster merging and propose the BDFC algorithm that

(i) performs a traditional clustering of the feature vectors to

compute an initial segmentation IbA of the feature map

(ii) performs a bi-domain merging of clusters that (unlike

hierarchical clustering) uses the image domain.

To reduce the influence of noise in the images, we apply super-

pixel segmentation [34] before applying the Bi-Domain Fea-

ture Clustering (BDFC). Superpixel segmentation computes an

image representation with a lower level of detail consisting of

so called superpixels (i. e. connected groups of pixels sharing

similar colors) [34]. Thereby, local redundancy is resolved

and the complexity of the subsequent BDFC is reduced.

However, we apply the approach in the high dimensional

feature representation using a straightforward generalization

of SLIC [34] to the concept of feature maps. Thus, the pixel

set P is segmented based on the change patterns of the pixels

and not based on pixel color.

Let FA be a feature map, M the number of superpixels and

M = {1, . . . ,M} the set of superpixel-indices. A superpixel

segmentation of FA is a tuple (IsA, sA), where

• the index map IsA : P → M assigns to each pixel p

the index of the superpixel containing p in the superpixel

map of feature representation A• sA : M → R

N assigns to each superpixel-index m the

centroid of the feature vectors FA(p) of the pixels p with

IsA(p) = m.

The superpixel feature map, defined by F sA = sA ◦ Is

represents FA on a lower level of detail. The set of all feature

vectors in this map is given by F sA(P) and its cardinality is

apparently limited to M .

BDFC computes a segmentation of the feature map F sA. The

first step in (i) is training a hierarchically growing hyperbolic

self-organizing map (H2SOM [35]) with samples from the

set of feature vectors F sA(P) (Parameters are addressed in

the Result section). A H2SOM is used in this study due to

the advantage of simultaneous clustering and dimensionality

reduction as well as its good performance even for high

dimensional data. However, other clustering algorithms for

vector quantization can in principle be applied as well. Let

UA ={u1A, . . . , u

}be the set of L prototypes computed

by the H2SOM and for 1 ≤ � ≤ L let NA(�) be the set of

all neighbors of cluster � in the H2SOM-topology. To compute

the initial segmentation to gain the index map IbA, each feature

vector in F sA(P) is mapped to the index of its best matching

Fig. 2. The BDFC is illustrated for the time independent feature representation. Features are extracted from an image sequence and clustered by a H2SOM [35].Pixels belonging to one object (see the magnified starfish) are initially assigned to different clusters, as indicated by nine shades of purple/blue in the visualization([36], [37], middle). White contours have been drawn between pixels of the starfish that belong to different clusters, to bring out the fragmentation of thestarfish. After cluster merging (right) based on proximities in the image and feature domain, most pixels of the starfish belong to one cluster.

prototype in the H2SOM clustering result:

IbA(p) = argmin

1≤�≤L

{∣∣F sA(p)− u�

∣∣} (1)

A visualization of the H2SOM result can be found in Figure 2.

To take the proximity of cluster indices in IbA into account

during the cluster merging procedure in step (ii), let

Ω�A =

{p ∈ P ∣∣Ib

A(p) = �}

for every cluster index �. Let r ∈ R with r ≥ 1 (r = 10in this study). We call two pixels p, q close, if and only if

d∞(p, q) ≤ r. With

Ω�,kA =

{(p, q) ∈ P2

∣∣p ∈ Ω�A, q ∈ Ωk

A, d∞(p, q) ≤ r}

we define the image proximity of two clusters � and k by

φA(�, k) =

∑(p,q)∈Ω�,k

A1− d∞(p,q)

r∣∣Ω�A

∣∣+ ∣∣ΩkA

∣∣ . (3)

Note that the choice of the parameter r is not crucial for

the cluster merging, but the restriction of the sum in the

denominator of φA(�, k) increases the computation speed of

φA(�, k). We use the parameter n ∈ N to control the number

of clusters merged in the merging process. In this paper, we

set the parameter to n = 5. We define three conditions to be

fulfilled so two clusters with indices � and k are merged:

(A) � is neighbor of k in the H2SOM grid-topology,

i. e. � ∈ NA(k)(B) the Euclidean distance d2(u

�A, u

kA) is one of the lowest n

distances d2(u�A, ·).

(C) the image proximity φA(�, k) is one of the highest nimage proximities φA(�, ·)

The clusters are merged iteratively, as described in algorithm 1,

until no clusters satisfy (A), (B) and (C). A visualization of

the final result of the BDFC-algorithm is shown in Figure 2.

B. Feature extraction

Let (I1, . . . , IT ) be an image sequence and let the mean

image I = 1/T · (I1 + · · ·+ IT ) model the background of

the scene. To compute change features, we compute a

new representation Jt from each image It, encoding the

while found clusters to merge == true dofound clusters to merge = false

for 1 ≤ k < l ≤ L doif (A) and (B) and (C) then

found clusters to merge = true

|Ωk|·ukA+|Ωl|·u�

|Ωk|+|Ωl|NA(k) = NA(k) ∪NA(�)IbA(p) = k ∀p ∈ Ωl

endend

endAlgorithm 1: Clusters are merged until no pair of clusters

exists that satisfies the conditions (A), (B) and (C). When

cluster l is merged to cluster k, ukA is set to the weighted

mean of the cluster centers. Moreover cluster k absorbs all

neighbors of l in the H2SOM-topology as well as all pixels

former associated to l.

pixel differences to the mean image: Jt (p) =∣∣I (p)− It (p)

∣∣(∀p ∈ P, 1 ≤ t ≤ T ). To account for external effects having

impact on the whole image (e. g. change in lighting) we

standardize each channel Jct (c ∈ {1, 2, 3}) of an image Jt,

i. e. we define Jct (p) to be the standard score of Jc

t (p).To compute the change features representing the temporal

development of change at a given position, we consider the

temporal sequence at position p in channel c, defined by

Jc(p) =(Jc1(p), . . . , J

cT (p)

To separate trend (i. e. long term signal change) from fluc-

tuation (i. e. short term signal change) we apply wavelet

transformation [38] using a Daubechies filter [38]. With WT

and WF the trend and the fluctuation computed by the

wavelet transformation and � the concatenation of vectors

(i. e. (x1, x2)�(y1, y2) = (x1, x2, y1, y2)), we define the

feature map Ftmp by

Ftmp(p) = WT

(J1(p)

� WT

(J2(p)

� WT

(J3(p)

Fig. 3. Connected components of a cluster using the example of a specificcluster �: Non-black color indicates pixels assigned to cluster �. Cluster �consists of 4 connected components as indicated by the colors white (w),green (g), blue (b) and red (r).

Note that Ftmp is based of time dependent feature vectors.

To compute change features that describe time invariant

change patterns, let σ be the permutation arranging the se-

quence Jc(p) in descending order to Jc· :

Jc· =

(Jcσ(1)(p), . . . , J

cσ(T )(p)

Again we use wavelet transformation and define the feature

map Fint by

Fint(p) = WT

(J1· (p)

� WT

(J2· (p)

� WT

(J3· (p)

C. Detecting relevant changes

To find relevant changes in the image sequence, regions

that share a particular change pattern are computed. These

regions are classified into categories and since a relevanceindex is assigned to each of the categories they can be sorted,

permitting a rapid manual posterior browsing. Finally, the

time-points where changes occur are determined for every

region.

To find regions that show similar change features, we apply

BDFC to Ftmp (Figure 1) to compute merged clusters with

centers Utmp ={u1

tmp, . . . , uLtmp

}and an index map Ib

tmp. For a

cluster � (1 ≤ � ≤ L) the set Ω�tmp =

{p ∈ P ∣∣Ib

tmp(p) = �}

composed of pixels showing similar change patterns, i. e. be-

longing to the same merged cluster. As shown for one example

cluster in Figure 3, a set Ω�tmp consists of a number n� of

connected components, referred to as R� ={R�

1, . . . , R�n�

Note that R =⋃

1≤�≤L R� is a segmentation of Ibtmp into

regions.

Next, the regions computed from the time dependent feature

representation are assigned to groups of changes at different

time points, supposed to be caused by the same kind of event

(like passing of one specific species). The groups are computed

from the time independent feature representation and referred

to as change categories. We apply BDFC to Fint (Figure 1) to

compute merged clusters with centers Uint ={u1

int, . . . , uKint

}and an index map Ib

int. Each cluster of the (time independent)

cluster indices k (1 ≤ k ≤ K) defines one change category

denoted by Ck and each region R ∈ R is assigned to a

category κ(R) by

κ(R) = argmax1≤k≤K

∣∣{p ∈ R∣∣Ib

int(p) = k}∣∣ . (7)

Having regions and categories defined, an index is assigned to

every category that estimates the degree of importance for a

change. Without knowledge about the experts’ preferences, we

define the relevance based on frequency and pixel differences.

With the mass center of the categories defined by

|R|∑

1≤k≤K

|{R ∈ R |κ(R) = k }| · ukint (8)

the relevance r(Ck) of a category Ck is defined by

r(Ck) = d2(m,ukint).

This allows us to rank the categories Ck according to the

computed relevance. To further facilitate the non-automatic

posterior browsing, we propose additional filters like the

region size∣∣R�

∣∣ and the length of the prototype u�tmp. The

set of regions satisfying both filters will be denoted by R ⊆ R.

Finally, all time-points where changes occur are computed

for every region R ∈ R. To do so, we define a measure

encoding how well a border of a region matches a contour in

the image It taken at time-point t: Let B be the set of pixels

defining the border of R. For p ∈ B, we denote with ip, op the

(well-defined) pixels in the 8-connected neighborhood from pthat lie on the line through p perpendicular to B. Note that

one of the pixels ip and op is inside R and one is outside R.

We define the coherence of B to It by

dB,t =1

|B|∑p∈B

d1(It(ip), It(op)) (9)

In order to find all time-points where changes occur in the

region R, we compute a threshold τ from the sequence

dB,1, . . . , dB,T . Let σR be the permutation that brings the

sequence dB,1, . . . , dB,T in ascending order, i. e.

dB,σR(t) ≤ dB,σR(t+1) ∀1 ≤ t ≤ T − 1. (10)

m = argmax1≤t≤T−1

dB,σR(t+1) − dB,σR(t), and (11)

τ =dB,σR(m) + dB,σR(m+1)

we say change occurs at time-point T , if and only if dB,t > τ .

Fig. 4. Changes identified by the algorithm. Left: Grouped changes identified by the algorithm data set 1. The most relevant category and the second mostrelevant category are shown on top and bottom, respectively. Right: Changes identified by the algorithm for data set 2. All changes within the eight mostrelevant categories are copied into the background image. Bounding boxes are drawn in different colors indicating different categories, (yellow = medusahead,red = big crab, magenta = starfish, etc.).

III. RESULTS

We evaluate our method on images of the Lofoten-

Vesteralen (LoVe) ocean observatory [30]. LoVe is a fixed

marine observatory monitoring a coral reef in the Norwegian

Sea (N 68◦ 54.474′, E 15◦ 23.145′). The observatory that was

deployed in October 2013 takes one image every 60 minutes.

All images are available online at http://love.statoil.com/. In

the present paper, we evaluate the results of the algorithm

applied to 6 data sets. Each data set is a sequence of 24

images (one day). We use the same parameter settings for

each data set: The superpixel segmentation was performed

setting the weight factor (Section II-A) to 8 and the number

of superpixels to M = 12800 corresponding to a superpixel-

size of 80 pixels in average. For the H2SOM clustering, the

H2SOM with 3 rings and spread factor 8 was trained with

a random set of 0.25 · M samples. The tolerance and the

radius used during cluster merging (Section II-A) were set

to 5 and 10, respectively. For the wavelet transformation in

the feature extraction (Section II-B), we use Daubechies Filter

of length 6. See Section IV for more information on the

wavelet transformation. As proposed in Section II-C, we apply

additional filters to the categorized regions: A region R�i of

pixels mapped to the cluster center u�tmp is considered in the

evaluation if and only if (i) the size of the region is at least

400 pixels and (ii)∣∣u�

∣∣ ≥ 0.2 ·max1≤l≤L

∣∣ultmp

∣∣.As an example Figure 4 shows changes identified by the

algorithm in 2 of the six data sets.

Figure 5 shows the precision of the algorithm in 6 data

sets, each represented by one graph: Each of the N changes

identified by the algorithm was evaluated manually either as

relevant or irrelevant. For 1 ≤ n ≤ N we denote by t(n)the number of true positives, i. e. the number of changes

within the n most relevant changes that are labeled manually

as relevant. The precision is computed as P (n) = t(n)/n and

Fig. 5. Each graph shows the evaluation of the methods result when applied toone of the 6 datasets. With N the number of changes found by the method inone dataset and with n < N , the precision (i. e. the fraction of true positivesin the n most relevant changes) is plotted against n/N .

plotted against n/N . Note that the recall of the method was

not computed, as manual detection of all relevant changes is

not feasible due to the high abundance of changes in the given

image sequences and the absence of a ground truth. However,

most regions of the visual field are considered in the graphs

and the method assigns a high relevance factor to most true

positives. This indicates that most changes of interest belong to

the most relevant changes identified by the proposed method.

IV. DISCUSSION AND CONCLUSION

In this paper we have presented a change detection method

including the new BDFC algorithm, which shows the po-

tential to be used in various future image processing tasks.

The change detection method categorizes and orders various

changes present within a fixed image frame by their relevance.

Thereby, the method enables the identification of a variety

of different changes such as presence and movements of

species within images from a fixed position over time. The

feature extraction method has some range of adaptability and

provides perspectives for future work. A possible approach

is to adapt the change detection to focus on specific species

by applying the feature extraction to feature maps instead of

images. Another approach is to switch to a multiresolution

based wavelet approach in the feature extraction to account for

changes that happen slowly overtime (e. g. a change of color

of a sessile species). However, this paper has shown that the

methods detects movements from unknown species in areas

with limited prior knowledge. With this ability, the method

has the potential to be used in a variety of image detection

scenarios without considerable individual adjustments.

ACKNOWLEDGMENT

Financial support was given by Statoil ASA, Research and

Technology, Norway

REFERENCES

[1] R. Radke, S. Andra, O. Al-Kofahi, and B. Roysam, “Image changedetection algorithms: a systematic survey,” Image Processing, IEEETransactions on, vol. 14, no. 3, pp. 294–307, March 2005.

[2] C. Su and A. Amer, “A real-time adaptive thresholding for video changedetection,” in Image Processing, 2006 IEEE International Conferenceon, Oct 2006, pp. 157–160.

[3] J. K. Suhr, H. G. Jung, G. Li, and J. Kim, “Mixture of gaussians-basedbackground subtraction for bayer-pattern image sequences,” Circuits andSystems for Video Technology, IEEE Transactions on, vol. 21, no. 3, pp.365–370, March 2011.

[4] B. Yin, J. Zhang, and Z. Wang, “Background segmentation of dynamicscenes based on dual model,” Computer Vision, IET, vol. 8, no. 6, pp.545–555, 2014.

[5] A. Ferone and L. Maddalena, “Neural background subtraction for pan-tilt-zoom cameras,” Systems, Man, and Cybernetics: Systems, IEEETransactions on, vol. 44, no. 5, pp. 571–579, May 2014.

[6] J. Morisette and S. Khorram, “An introduction to using generalized linearmodels to enhance satellite-based change detection,” in Geoscience andRemote Sensing, 1997. IGARSS ’97. Remote Sensing - A Scientific Visionfor Sustainable Development., 1997 IEEE International, vol. 4, Aug1997, pp. 1769–1771 vol.4.

[7] M. J. Carlotto, “Detection and analysis of change in remotely sensedimagery with application to wide area surveillance,” Image Processing,IEEE Transactions on, vol. 6, no. 1, pp. 189–202, Jan 1997.

[8] C. Clifton, “Change detection in overhead imagery using neuralnetworks,” Applied Intelligence, vol. 18, no. 2, pp. 215–234, 2003.[Online]. Available: http://dx.doi.org/10.1023/A%3A1021942526896

[9] T. Yamamoto, H. Hanaizumi, and S. Chino, “A change detection methodfor remotely sensed multispectral and multitemporal images using 3-dsegmentation,” Geoscience and Remote Sensing, IEEE Transactions on,vol. 39, no. 5, pp. 976–985, May 2001.

[10] F. Bovolo, L. Bruzzone, and M. Marconcini, “A novel approach tounsupervised change detection based on a semisupervised svm and asimilarity measure,” Geoscience and Remote Sensing, IEEE Transactionson, vol. 46, no. 7, pp. 2070–2082, July 2008.

[11] T. Celik and K.-K. Ma, “Unsupervised change detection for satelliteimages using dual-tree complex wavelet transform,” Geoscience andRemote Sensing, IEEE Transactions on, vol. 48, no. 3, pp. 1199–1210,March 2010.

[12] F. Bovolo and L. Bruzzone, “A theoretical framework for unsupervisedchange detection based on change vector analysis in the polar domain,”Geoscience and Remote Sensing, IEEE Transactions on, vol. 45, no. 1,pp. 218–236, Jan 2007.

[13] F. Bovolo, S. Marchesi, and L. Bruzzone, “A framework for automaticand unsupervised detection of multiple changes in multitemporal im-ages,” Geoscience and Remote Sensing, IEEE Transactions on, vol. 50,no. 6, pp. 2196–2212, June 2012.

[14] L. Bruzzone and D. Prieto, “Automatic analysis of the difference imagefor unsupervised change detection,” Geoscience and Remote Sensing,IEEE Transactions on, vol. 38, no. 3, pp. 1171–1182, May 2000.

[15] ——, “An adaptive semiparametric and context-based approach tounsupervised change detection in multitemporal remote-sensing images,”Image Processing, IEEE Transactions on, vol. 11, no. 4, pp. 452–466,Apr 2002.

[16] L. Bruzzone and F. Bovolo, “A novel circular approach to changedetection in pair of images extracted from image time series,” inGeoscience and Remote Sensing Symposium (IGARSS), 2014 IEEEInternational, July 2014, pp. 1140–1143.

[17] R. D. Macleod and R. G. Congalton, “A quantitative comparison ofchange-detection algorithms for monitoring eelgrass from remotelysensed data,” Photogrammetric Engineering and Remote Sensing,vol. 64, no. 3, pp. 207–216, 1998.

[18] O. Delaunoy, N. Gracias, and R. Garcia, “Towards detecting changesin underwater image sequences,” in OCEANS 2008 - MTS/IEEE KobeTechno-Ocean, April 2008, pp. 1–8.

[19] K. Lebart, C. Smith, E. Trucco, and D. Lane, “Automatic indexingof underwater survey video: algorithm and benchmarking method,”Oceanic Engineering, IEEE Journal of, vol. 28, no. 4, pp. 673–686,Oct 2003.

[20] K. Lebart, E. Trucco, and D. Lane, “Real-time automatic sea-floorchange detection from video,” in OCEANS 2000 MTS/IEEE Conferenceand Exhibition, vol. 2, 2000, pp. 1337–1343 vol.2.

[21] M. Johnson-Roberson, O. Pizarro, and S. Williams, “Saliency rankingfor benthic survey using underwater images,” in Control AutomationRobotics Vision (ICARCV), 2010 11th International Conference on, Dec2010, pp. 459–466.

[22] E. Vahtme, T. Kutser, J. Kotta, and M. Prnoja, “Detecting patterns andchanges in a complex benthic environment of the baltic sea,” Journalof Applied Remote Sensing, vol. 5, no. 1, pp. 053 559–053 559–18,2011. [Online]. Available: http://dx.doi.org/10.1117/1.3653271

[23] S. Kesler and A. Elfishawy, “Adaptive change detection in imagesequence,” in Acoustics, Speech, and Signal Processing, 1990. ICASSP-90., 1990 International Conference on, Apr 1990, pp. 2189–2192 vol.4.

[24] Z.-S. Jain and Y. Chau, “Optimum multisensor data fusion for imagechange detection,” Systems, Man and Cybernetics, IEEE Transactionson, vol. 25, no. 9, pp. 1340–1347, Sep 1995.

[25] Toyama et al., “Wallflower: principles and practice of backgroundmaintenance,” in Computer Vision, 1999. The Proceedings of the SeventhIEEE International Conference on, vol. 1, 1999, pp. 255–261 vol.1.

[26] M. J. Black, D. J. Fleet, and Y. Yacoob, “Robustly estimating changes inimage appearance,” Computer Vision and Image Understanding, vol. 78,no. 1, pp. 8–31, 2000.

[27] C. Wu, B. Du, and L. Zhang, “Slow feature analysis for changedetection in multispectral imagery,” Geoscience and Remote Sensing,IEEE Transactions on, vol. 52, no. 5, pp. 2858–2874, May 2014.

[28] Vardaro et al., “A Southeast Atlantic deep-ocean observatory: first ex-periences and results,” Limnology and Oceanography: Methods, vol. 11,pp. 304–315, 2013.

[29] O. N. Canada. (2014) NEPTUNE in the NE Pacific. [Online]. Available:http://oceannetworks.ca/installations/observatories/neptune-ne-pacific/

[30] O. R. Godø, S. Johnson, and T. Torkelsen, “The love ocean observatoryis in operation,” Marine Technology Society Journal, vol. 48(2), 2014.

[31] J. Schanda, Colorimetry: understanding the CIE system. John Wiley& Sons, 2007.

[32] F. Murtagh, “A survey of recent advances in hierarchical clusteringalgorithms,” The Computer Journal, vol. 26, no. 4, pp. 354–359, 1983.

[33] Y. Yang, Z. Ma, Y. Yang, F. Nie, and H. T. Shen, “Multitask spectralclustering by exploring intertask correlation,” IEEE Transactions onCybernetics, vol. 45, no. 5, pp. 1083–1094, May 2015.

[34] Achanta et al., “Slic superpixels compared to state-of-the-art superpixelmethods,” Pattern Analysis and Machine Intelligence, IEEE Transactionson, vol. 34, no. 11, pp. 2274–2282, Nov 2012.

[35] J. Ontrup and H. Ritter, “A hierarchically growing hyperbolic self-organizing map for rapid structuring of large data sets,” in Proceedingsof the 5th Workshop on Self-Organizing Maps, Paris (France), 2005.

[36] J. Kolling, D. Langenkamper, A. Sylvie, K. Michale, and T. W. Nattkem-per, “WHIDE - A web tool for visual data mining colocation patternsin multivariate bioimages,” Bioinformatics, vol. 28, no. 8, 2012.

[37] K. Gorzolka, J. Kolling, T. W. Nattkemper, and K. Niehaus, “Spatio-Temporal Metabolite Profiling of the Barley Germination Process byMALDI MS Imaging,” PLOS ONE, vol. 11, no. 3, 2016.

[38] I. Daubechies, Ten lectures on wavelets, ser. CBMS NSF regionalconference series in applied mathematics ; 61. Philadelphia, Pa.: Societyfor Industrial and Applied Mathematics, 1992, vol. 61.

Paper X

Nilssen, I., Hepsø, V., Nattkemper, T. W., Johnsen, G., Images in environmental monitoring: Enhanced knowledge through interdisciplinary collaboration.

- two case studies from marine sciences and engineering. Submitted to Marine Pollution Bulletin

Images in environmental monitoring: Enhanced knowledge through

interdisciplinary collaboration

– two case studies from marine sciences and engineering

Ingunn Nilssen1, 2, Vidar Hepsø 1,3, Tim W. Nattkemper 4 and Geir Johnsen 2,5

1 - Statoil ASA, Research and Technology, N-7005 Trondheim, Norway

2 - NTNU AMOS - Centre for Autonomous Marine Operations and Systems, Department

of Biology, Norwegian University of Science and Technology, N-7491 Trondheim,

Norway

3 - Norwegian University of Science and Technology, Department of Petroleum

Engineering and Applied Geophysics, N-7491 Trondheim, Norway

4 - Biodata Mining Group, Bielefeld University, D-33501 Bielefeld, Germany

5 - University Centre in Svalbard, N-9171, Longyearbyen, Svalbard

Corresponding author: Ingunn Nilssen

E-mail: innil@statoil.com

Abstract

Images are increasingly being used for documentation of the marine environment. Based on

humans shared cognitive capabilities in detecting and classifying objects and movement, we

argue that imaging technologies have the potential of acting as a communicative tool (boundary

object) for a general understanding of the environment, including communication and

collaboration. Solutions on how interdisciplinary collaboration on image processing can

contribute to utilise the images inherent knowledge to overcome the human limitations in

perceiving objects and movements other than human’s in time and space are proposed. Two case

studies from the marine environment are used to illustrate the additional knowledge that can be

gained from images through interdisciplinary collaboration. The cases are related to

environmental monitoring and modelling, discussing abilities and challenges related to temporal

and spatial resolution, respectively. Furthermore, future potential use of images in decision

making processes is discussed.

Keywords: Images; environmental monitoring; interdisciplinary approach; communicative tool

1. Introduction

In 1998, Science published an article on the concept of the “third culture”; use of technology

stating that “… technology generates opportunities: new things to explain; new ways of

expression; new media of communications …” (Kelly, 1998). Furthermore, “The third culture

creates new tools faster than new theories, because tools lead to novel discoveries quicker than

theories do”. This rapid development of enabling technology gives a possibility to acquire a more

holistic understanding of the environment and environmental phenomenon (Nilssen et al.,

2015a). On the other hand, according to the technology philosopher Don Ihde, the applied sensor

technology used determines what we see (Ihde, 1979; Ihde, 1991). From this perspective the

rapid technology developments are challenging the human ability to interpret and understand the

data provided in an efficient and effective manner. The limited experience with continuous new

types of technologies might therefore affect our ability to integrate and transform all these new

measurements into new knowledge, which in its turn is used in applications such as models, risk

management or as basis for decisions in general.

The more information gathered with various new technologies and technology applications

and the more acquired knowledge, it has become clear that complex environmental problems

cannot be solved without an effort from interdisciplinary teams of experts (Brown et al., 2015;

Rylance, 2015; Wuchty et al., 2007). This statement is valid for interpretation of complex

datasets on the individual level to interpretation of this acquired knowledge into decisions on

organisational, national and international decision levels (Leedom et al., 2007). However,

interdisciplinary collaboration between disciplines is not new. For instance have archaeologists,

geologists and life sciences such as system biology a fundamental commitment and a tradition of

working across cultures and disciplines. In interdisciplinary work, the disciplines bring along

various preconditions and cultural traditions to solve a given task. Establishment of a sufficient

unified understanding among the actors involved is required to enable reasonable actions to be

taken. The further divided the disciplines are professionally and the more novel or innovative the

task is, the more challenging the establishment of such a sufficient unified understanding is

(Carlile, 2004).

With various professional background and experiences among the actors involved, a success

criterion for establishment of a sufficiently unified understanding of the given task is to find

mechanisms, or tools, that are recognisable or somehow familiar for all actors involved. In

science, for instance, all kind of imaging and visualisation such as graphs, diagrams, table,

simulations and photos are used to disseminate results and knowledge (Burri and Dumit, 2008).

Due to the human cognitive capability, images should in general have an inherent capacity of

being understood by all humans, independent of profession and background. The communicative

potential of images has been reflected by growing numbers of professions since the late 1990’s.

Computer scientists, graphics designers and cognition scientists have all created new fields such

as visualised information, information-graphics and aesthetics, visual diagnostics of methods to

process and display, or abstract, data in the most optimal way for the human observer’s cognitive

skills and background knowledge (Card et al., 1999; Munzner, 2014; Ware, 2012).

In the marine environment use of digital imaging is a relatively new methodological

application for monitoring. While video for habitat mapping has been performed for more than a

decade, use of images for monitoring purposes (time-series or repeated measurements) is still

fairly new, but increases rapidly. Due to the limited experience with image technology in this

environment, there is still a development need for optimised data collection and automated or

semi-automated analyses to handle the vast amount of data (Nilssen et al., 2015a), or simply to

see and exploit the potential provided by the introduction of images into the marine

environmental monitoring. Even though the human brain is good in detecting human motion in

time and space, research shows that we perform worse on movements of animals and object that

are unlike humans and that we have less experiences with (Shiffrar and Thomas, 2013; Smith

and Kosslyn, 2014). Due to the vast amount of images of high temporal and/or spatial resolution

gathered from the, relatively little known environment, an interdisciplinary effort is required to

exploit the inherent knowledge available in the measurements.

The purpose of this paper is to elucidate why images have the potential of being good

communicative and collaborative tools. Furthermore, approaches for how to process the

information from images gathered for underwater environmental monitoring are proposed. The

author’s hypothesis is that interdisciplinary collaboration on processing and interpreting images

has the potential to enhance the knowledge and understanding of the marine environment

considerably since this collaboration enables us to utilize the inherent knowledge available also

within the temporal and spatial domain. How the temporal and spatial obstacles can be dealt with

is exemplified through two interdisciplinary case studies from the marine environment (see

chapter 3 Case studies on images). The first case demonstrates how collaboration between

biologists and computer scientists on time series of time-lapse photos from a multi-sensor ocean

observatory can be used to monitor biological activities and habitat health state with time. In the

second case biologists, information technology personnel, cyberneticists and social scientists

contributed to develop a 2D and 3D Geographic Information System (GIS) model visualising,

among other thing, topography, resources of interest, current speed and direction, water turbidity

and dispersion of discharge from a point source. Models can aid mitigating the human spatial

limitations by providing an overview of the spatial extent of a set of given measurements in a

defined geographical area simultaneously.

2. Material and Methods

In the following chapter the cognitive advantages with using images in environmental

monitoring is explained by (Ihde, 1979) and his theory on instrumental realism. Furthermore, the

potential of enhancing environmental knowledge through interdisciplinary collaboration and

innovation is elaborated in the context of (Carlile, 2004) and his framework for managing

knowledge across boundaries.

2.1. Images and image technology

According to the technology philosopher Don Ihde, our visual perception of what we see is

determined by the measuring sensor technology as it transforms the object of interest or

phenomenon through a two-sided process where some features are magnified and others are

reduced, or even left out (Ihde, 1979; Ihde, 1991). In that sense technologies are non-neutral

(Ihde, 1979). This phenomenon is exemplified in Figure 1, where a variety of measurements with

different spatial resolution covering different areas of the monitored habitat presents an abstract

sketch of various sensors in a habitat. The habitat can for instance be a cold water coral reef

where image and Measurement 2 are exemplified by photos of a coral reef area (see also Figure

2 a) and a magnified photo focusing on shrimp (see also Figure 2 b), respectively. Measurement

1 can for instance represent turbidity measurements, while Measurement 3 can be an echogram

representing the biomass integrated over a certain area (see also Figure 2 d). However, as Figure

1 illustrates, neither individual sensors, nor a combination of sensors will cover all aspects of a

given habitat completely as different sensors amplify and/or reduce different

phenomenon/objects of interests and in most cases these sensors are point samples (Almklov et

al., 2014). Being able to describe isolated variables, objects of interest or an entire habitat we are,

in most cases therefore dependent of models that can extrapolate the measurements (from a

delimited area) to a broader area of comparable composition and environmental condition.

Figure 1: Schematic illustration of a typical multisensory coverage of an investigated habitat, for instance a cold

water coral reef (the habitat is indicated by the irregular outer borders of the figure). The spatial resolution of

the individual sensors will vary. The photo (image) will, dependent of the distance from the camera to the

object of interest, cover approximately 10 m2 of the reef (see also Figure 2 a). Measurement 1-3 can be

turbidity (measurement 1), Measurement 2 can represent a magnified part of the photo, focusing on the

presence of shrimp or sponge (see also Figure 2 b) and Measurement 3 can be an echogram (see also Figure 2

d) measuring zooplankton and fish biomass integrated over a given water column (detection distance and taxa

identification is dependent of the frequency used) or use of models for extrapolation from the findings in the

image to the reef habitat. According to Ihde (1979), also the degree of transformation will vary significantly

dependent on the measuring technology. Technology transformation is further exemplified in Figure 2 a-d.

Ihde (1979)categorises the technologies (the “instrumental intentionality”) in two

dimensions 1) in accordance to how transformed the measurements are from the human vision

(“horizontal” or “vertical instrumental variation”) and 2) the measurements’ degree of

magnification and reduction (“high” or “low contrast”). Contrast in this context is refereeing to

what is magnified in the photo and not its technical quality or properties. Representations close

to the human vision is categorised as horizontal instrumental variations (Figure 2 a, b), while

measurements where the immediate visual recognition have disappeared (Figure 2 c, d) are

categorised as vertical instrumental variations (Ihde, 1979). Since the measurements categorised

as vertical instrumental variation are highly transformed, they have to be read in a particular

scientific tradition (have to be learned) to be interpreted and understood. Furthermore, high

contrast measurements are magnification of selected objects of interest or phenomenon in a

given image (Figure 2 b, d). For instance in Figure 2 b, the photo is magnified on the present

shrimp using their eye reflections for detection and to estimate their abundance. While for the

echogram the technology is integrating the biomass present in a given volume of water (up till

several hundred meters distance, depending on the echo sounder frequency). Low contrast

measurements can for instance be a transformation of the original image (horizontal instrumental

variations of low contrast) to a heat map where the shrimps, or other organisms of interest, are

presented as coloured pixels reflecting their abundance (Figure 2 c).

Figure 2: Examples of images with different degree of instrumental transformation and different degree of

magnification.

Figure 2 a: Photo of cold water coral reef habitat at the Lofoten and Vesterålen (LoVe) ocean observatory. White

parts are living Lophelia pertusa and grey part is sediments with large amounts of dead L. pertusa

fragments (Statoil, n.d.). The photo is a representative for images categorised as horizontal instrumental

variation of low contrast.

Figure 2 b: Magnified part of photo measuring shrimp abundance by their eye reflections (paired or single yellow

dots indicated by arrows) in a cold water coral reef habitat dominated by L. pertusa. Image acquisition and

computational shrimp detection is explained in Purser et al. (2013). The photo is a representative for

images categorised as horizontal instrumental variation of high contrast.

Figure 2 c: Heat map of shrimp abundance at the LoVe ocean observatory where the different colours represent the

shrimp density from no shrimp (dark red) to high density (pale yellow). The algorithms for computing the

shrimp abundances and heat map computation are explained in Osterloff et al. (In Press-a). The heat map is

a representative for images categorised as vertical instrumental variation of low contrast.

Figure 2 d: Echogram (acoustic backscatter) measuring biomass of fish and zooplankton integrated from surface to

approximately 250 m depth at the LoVe ocean observatory. The echogram shows aggregations of Northeast

Arctic cod (Gadus morhua) at 60 – 100 m (depth scale on left) and, presumably, an aggregation of

zooplankton underneath (blue-green “cloud”). The echogram is a representative for images categorised as

vertical instrumental variation of high contrast. (Image provided by Terje Torkelsen, Metas).

2.2. Interdisciplinary work

To succeed within cross disciplinary collaborations it is essential to find appropriate

communicative tools (boundary objects) that are understood by all actors involved (Carlile,

2004). According to Carlile (2004), new knowledge and knowledge development occurs when

individual actors use their already existing knowledge and experiences on new tasks. The

preconditions to succeed are that 1) the actors manage to identify their initial level of mutual

understanding and 2) they, through the development, manage to agree and thereby establish

knowledge by bring the task to a level of high confirmation (Figure 3).

Depending on the complexity of a task, Carlile (2004) divides the framework for managing

knowledge across boundaries into three levels, syntactic, semantic and pragmatic, mentioned in

the order of increased scientific deviation and experiences. When actors are in the syntactic level

(lower part of Figure 3) they share a lexicon of concepts and world views and they can therefore

easily establish a sufficient unified understanding of the task. In such cases development of a

“common language” through transfer of information, orally or for instance by the help of a

power point presentation, is sufficient for the actors to establish a mutual understanding of the

situation, or as in our case to understand that the given image represents a selected area of a cold

water coral reef (see also Figure 2 a). In the semantic area (middle section of Figure 3), the

degree of novelty of the task has increased, and thereby decreases the level of confirmation of

the task. At this stage the actors are further scientifically divided and more effort is required to

establish a sufficient unified understanding of the task. In our case this can be the stage were the

biologists and computer scientists need to collaborate to handle the temporal and spatial

dimensions of the images, for instance to follow change in shrimp abundance with seasons (see

also Figure 2 c and Figure 4). Negotiations between the actors and willingness to change their

domain knowledge may be required at this level (Carlile, 2004). Carlile (2004) describes this as

the level where knowledge has to be translated between disciplines. In the pragmatic level (upper

part of Figure 3) the ability to share and access knowledge between the actors is challenging and

a transformation of the domain-specific knowledge is required to establish a sufficient unified

understanding between the actors involved. In our case, this stage is represented by interpretation

of measured biological activity (documented by the image analysis developed at the semantic

level) with other multisensory measurements and variables such as temperature (time of the

year), currents, and discharges (in areas influenced by industrial activities). Furthermore, models

can be applied to extrapolate the findings from the delimited area of the image frame to the

present habitat in general (Figure 1) or even similar habitats with comparable environmental

conditions.

According to Carlile (2004), it is only when the tasks dealt with in the semantic and

pragmatic level are pushed down to the syntactic level and has been institutionalised in new

methods, tools and representations that innovation has occurred and that new knowledge is

established and visible (arrow on the left in Figure 3). In our case, the time-lapse images (dealing

with the human`s temporal limitations) and models (dealing with the human`s spatial limitations)

are the communicative tools understood by the actors involved based on their low degree of

transformation (See section on Images and image technology) that contributes to bring the tasks

towards a level of high confirmation. It is emphasised that the transition between the levels

characterising the increasing degree of divergence based on scientific background and

experiences (syntactic - semantic and semantic - pragmatic) are diffuse.

Figure 3: The higher degree of novelty involved in a common task, the further the actors are professionally apart and

the more effort is required to establish a sufficient unified understanding. The dotted circles illustrate the

increased effort needed from the actors involved to come up with a successful solution. The corresponding

triangle on the right side is an example of what kind of environmental data the different levels represent from a

photo of a coral reef (syntactic level), to a heat map representing densities of organisms with colours (semantic

level) to use of multivariate data analysis for interpretation of multisensory data (pragmatic level). Results from

collaborations occurring in the semantic and pragmatic level have to be pushed down to the syntactic level and

institutionalised in new methods, tools and representations, before new and mutual knowledge is established

and visualised. Appropriate communicative tools (boundary objects), such as photos and models, are essential

to succeed within cross disciplinary collaborations and understanding. Two examples of practical applications

are given in Figure 7. Figure modified from Carlile (2004).

3. Case studies on images

Based on Ihde (1979) and his theory on instrumental realism, images are to a little degree

transformed or manipulated, compared to the human vision and they should therefore be

relatively easily understood by humans in general. However, research shows that humans have

limitations when it comes to interpreting movements and objects that are unlike humans and that

we have less experiences with both in the temporal and spatial domain (Shiffrar and Thomas,

2013; Smith and Kosslyn, 2014). The following two case studies from the marine environment

are used to demonstrate how interdisciplinary collaboration, in accordance with (Carlile, 2004)

and his framework for managing knowledge across boundaries, can contribute to utilize the

inherent knowledge available in the increasingly number of images gathered from mapping and

monitoring surveys by overcoming the human brain’s limitations in detection objects and

movements in the temporal (Figure 4) and spatial (Figure 5) domain.

3.1. Case study 1; How to overcome human limitations in detecting objects and

phenomenon within the temporal domain: interpreting time-lapse photos

In September 2013 the cabled Lofoten and Vesterålen ocean observatory (LoVe) was

deployed at a cold water coral area, 20 km off the north western coast of Norway (N 68

°54.474’, E 15 °23.145’) at approximately 260 meters water depth (Godø et al., 2014a). Time-

lapse photos (RGB) of the coral habitat, dominated by Lophelia pertusa, are sampled with a

temporal resolution of one hour (Figure 4). In addition, key environmental parameters such as

temperature, salinity, echograms presenting zooplankton and fish biomass in a given water

volume, current speed and direction, sound, concentration of Chlorophyll a, coloured dissolved

organic material and total suspended matter are measured (Godø et al., 2014a; Statoil, n.d.). The

camera used was a Canon EOS 550D in a waterproof housing (MEATAS DSF5210). The photo

covers an area of approximately 10 m2 (Osterloff et al., In Press-a). The cold water coral habitat

holds a high taxonomic diversity and the selected photo frames covers both an area with live

corals and an area with coral rubble. Except from limited long line fishing activities, there is no

industrial activity in the area and the long term deployment provides a unique opportunity to

establish basic knowledge of natural variations in this undisturbed area documenting species-

specific behaviour in their natural environment.

In this case study, the overall aim was to develop algorithms for automatic or semi-automatic

detection of biological activity and health status of the habitat documented in the photos in time

and space. The overall aim was to interpret these results with other environmental factors

affecting species distribution and biomass, such as current speed and direction, temperature (time

of the year) and turbidity. Being able to overcome our limitations in observing objects and

movements in time and space efficiently with respect to time use and human resources,

development of algorithms is crucial. Such an approach calls for an interdisciplinary

collaboration. Furthermore, incorporating images with other multisensory, in this case

environmental variables, also call for interdisciplinary collaboration and application of

multivariate data analyses.

The need for biologists and computer scientists to work jointly was addressed already in the

planning process for the LoVe ocean observatory. To ensure images of such a quality that

development of semi-automatic and/or automatic image analysis could be performed, the

computer scientists were involved already in the designing process of the camera set-up,

facilitating their preferences with respect to distance to object of interest and light conditions.

When the observatory came in operation, the biologists and computer scientists used several

arenas for cooperation to discuss the photos from face-to-face meetings, via video meetings and

to a web based annotation tool (Ontrup et al., 2009). The biologists brought their biological skills

and ecosystem knowledge into the discussion and approached the photos from a viewpoint of

which organisms that were most interesting, what they expected the photos to show, and what

was regarded as unexpected with regards to presence and or behaviour. Examples of the latter are

for instance how frequent the Sea lilies moves, the discovery of colour change of Lophelia

pertusa or the discovery that Tusk (Brosme brosme) is stationary. The computer scientists on the

other hand approached the images and organisms from a technical and mathematical point of

view: Is it possible to develop algorithms enabling automatic or semi-automatic detection of the

specific organisms or to track species-specific behaviour pointed out by the biologists? Is the

photo quality good enough (e.g. is the spatial resolution sufficient to detect and identify different

species or to what extent is the water quality affecting the photos?) and/or are the features of the

organisms so distinctive that they can be discriminated from the surroundings?

At LoVe, to date two approaches have been performed to detect organisms of interest and

bringing the outcome of the interdisciplinary collaboration down to the syntactic level of

understanding and where new knowledge is institutionalised and visible (Carlile, 2004). One of

the approaches has been to use unsupervised computer learning to detect and group interesting

events, such as presence and behaviour of different species. The methodology is based on species

features and change detection (Möller et al., 2015). This approach can for instance be helpful

when operating in more or less unknown areas where knowledge about which species to find or

to document behaviour for species with infrequent appearance. Examples of the latter at LoVe

are detection of star fish and sea urchins. This method provides the biologist with a set of

possible interesting events identified by the computer that they easily can scroll through.

Depending of the scope, the organisms’ location on the reef and how frequent they are present

can be detected in a time-efficient manner.

The other approach has been to identify behaviour of specific species of interest. Examples

of on-going work on natural behaviour of specific species are shrimp abundance (Osterloff et al.,

In Press-a) and monitoring colour change (Osterloff et al., In Press-b). This approach will for

instance be helpful to monitor frequently present or sessile species regarded as key species for a

habitat or to monitoring sensitive or threatened species where change in behaviour or colour

change (Osterloff et al., 2016) is used as an endpoint in laboratory experiments or to measure

anthropogenic impact in the field.

Figure 4: Use of algorithms for automatic detection of shrimp abundance on a cold water coral reef and on the coral

rubble area beneath over time. Figure based on Osterloff et al. (In Press-a).

3.2. Case study 2; How to overcome human limitations in detecting objects and

phenomenon within the spatial domain: 3D modelling

In November 2010, the Integrated Environmental Monitoring project was initiated with

Statoil, Kongsberg Group (Kongsberg Maritime and Kongsberg Oil&Gas), DNV-GL and IBM

as partners. The purpose of this project was to build an overall infrastructure for environmental

data, from sensor carrying platforms, sensors, , data transfer/handling/storage and access to

analysis and visualisation of data to end-users, in this case the decision makers (Hepsø et al.,

2012; Ulfsnes et al., 2014). Visualisation of important parameters and results, such as

topography, presence of resources and dispersion of discharges was developed as part of the

visualisation portal. Both a 2D and a 3D model were developed.

One of the purposes by building the GIS portal/model was to get a visual overview of the

measured parameters, the resources present in the area of interest and the potential impact from

discharges in space and time, all available on one screen (Figure 5). Various environmental data

has different spatial resolution (Figure 1) and in most cases they also measure at different

temporal resolution from seconds to hours. Visualising the different sensors and their various

spatial resolutions in the model increases the total overview tremendously. The portal also has a

time-slider, enabling the user to adjust the temporal resolution of the model. The model enables

the users to establish an understanding of the geographical area of interest. In combination with

this distant overview, the model also enables a close-up (“digital zoom”) of details of the

geographical area; the presence and condition of the, in this case, coral reef structures and how

the discharges of drill cuttings and water-based drilling fluids would disperse (given as a

prognosis in the planning process or as verification with actual measurements during or after the

operation) in the area over time with, at any time, the given current speed and direction.

A model that visually presents important and optimised information to the users with regards

to what impacts to expect and how an environmental monitoring programme in the particular

area should be designed to monitor the predicted impacts was developed through

interdisciplinary collaboration between biologists, information technology personnel,

cyberneticists and social scientists. This information is important with respect to selection of

sensor platform(s) and sensors to use and the corresponding areal cover needed since the type(s)

of sensor platform(s) used should be dependent of topography, resources and the expected extent

of the possible impact from the discharges. Furthermore, the types of sensors used should be

adjusted to the type of resources present, while the type(s) of discharges should decide what kind

of endpoints that should be monitored. By getting all essential parameters visualised

simultaneously at one screen, the model enable us to approach the available information from a

very different perspective than if the same information were given separately, one after the other.

Figure 5: 3D model of the Morvin oil field, presenting the predicted coral risk (coloured lines around the identified

coral structures). The risk is based on the coral reef condition (Poor-excellent) and the modelled sedimentation,

based on measured hydrographical data. Figure based on Ulfsnes et al. (2014).

4. Discussion

Based on the human cognitive capacities to detect and classify objects and movements

(Biederman, 1987; Shiffrar and Thomas, 2013; Smith and Kosslyn, 2014) and the camera’s low

degree of data transformation (Ihde, 1979; Ihde, 1991) this paper has launched images as a

comprehensive tool with the potential of enhancing environmental knowledge. Furthermore, the

essential contribution of interdisciplinary collaboration to enhance innovation and, in our case, to

extract the inherent knowledge available in images and models has been demonstrated through

Carlile (2004) integrative framework for managing knowledge across boundaries. Enhanced

knowledge can be related to the additional information provided by the images or models as such

or it relates to the potentially improved cross disciplinary understanding due to the images or

model’s communicative properties in itself. The latter is further discussed under “Future

applications for use of images in decision making processes”.

Since different scientific disciplines interpret data in their own tradition (Carlile, 2004), cross

disciplinary collaboration need to find collaborative tools, or boundary object(s), that have an

individual case-by-case adjusted meaning that can be understood by each of the actors involved

(Star and Griesemer, 1989). According to (Ihde, 1979), images of horizontal instrumental

variation (Figure 2 a) are to a little extent transformed compared to if we were using our own

eyes. This knowledge can be used to facilitate interpretation and understanding of additional

measurements that are more transformed and abstract for people in general. For instance, if a

transparent version of the heat map (Figure 2 c) is projected on top of the original photo(s), this

would ease the general understanding and interpretation of the data tremendously since the data

(plot on top of the photo) is perceived more like a photo (Figure 6). Furthermore, taking this

approach further, we argue that photos might be used to communicate and explain more abstract

and complex data, such as multisensory data and multivariate data analysis (images in upper

right of Figure 7 b). Even though not a photo, the model from case 2 does appear as a photo and

the graphic presented in this case (Figure 5) will have the same effect on humans; it eases our

ability to interpret all essential information and in addition it makes the spatial extent

manageable.

The human challenge in the spatial domain occurs when the object of interest is of such a

size that the human vision and perception can’t get the overview without having to move around

or closer to the object of interest to obtain a sufficient level of details. Additional aspect to the

geographical extension, areas that in practise are inaccessible for humans such as access to deep

waters, areas with a high risk of explosion for instance during fire on an oil rig are all examples

where use of modelling is highly valid in situation awareness. Presenting results by use of new

approaches can in itself enables completely new ways of utilizing data and opens up for further

engaging and understanding the source information. For instance, Adams (2013) describes use of

digital media, such as 3D models and models combined with photos in archaeology as an

approach that opens up for new way of engaging and understanding the source material. As for

Adams (2013) sketch of an archaeological site, we will through the 3D model get access to a

much higher spatial resolution than a human eye can observe by combining the close-up details

from the camera and reducing the spatial extent to something that can be handled by the human’s

vision on a screen (observing an area of tens to thousands of meters by only moving the eyes).

Another advantage that might be highly valid for the interpretation of knowledge of the object(s)

or phenomenon of interest is that both models and photos can be used to manipulate the real

world by removing unwanted and disturbing elements or features such as biofilm (e.g. bacterial

mats) and kelp or to retract the waters inherent optical properties (concentrations of

phytoplankton, coloured dissolved organic carbon and total suspended matter), having a great

effect on the visibility in the water (Johnsen et al., 2013).

As described above, images should have the potential to be understood, independent of

professional background and experiences. However, even though images reduce profession

specific demands, they are not self-exploratory. Every image has an interpretive tension (Burri

and Dumit, 2008) and issues such as knowledge of the specific object of interest, perception and

interpretation skills, beliefs, goals and expectations will always alter both the accuracy and speed

in selecting meaningful and desired information (Smith and Kosslyn, 2014). To be able to

interpret and utilise the full potential provided by images they need to be prepared and/or

transformed. Development of algorithms for automatic or semi-automatic image analysis are, in

most cases, dependent of humans performing manual labelling of the specific species of interest

from a certain number of photos, needed as a training set of the automated detection. According

to Smith and Kosslyn (2014) attention to only more than one source, for instance looking at more

than one species at the time, the selected information will become imperfect since the attention

then has to be divided between the various sources (different species) present. Due to this, the

human vision is selective, which bias the level of attention towards selected objects that might be

more processed than others (Carrasco et al., 2004; Chun et al., 2011). One way to deal with

reduced attention and selective vision is to group organisms of similar shape/features and

dedicate the attention to one group at the time, finalise the detection of the first one, before

initiation the labelling of the next (Schoening et al., 2016). The unsupervised approach for

grouping events based on features and change detection can help in this process by guiding our

attention to groups of already clustered organisms. Another approach applied to keep attention is

to use information sharing (several screens) for “on the fly” video annotations. In these cases

dedicated personnel annotate dedicated features simultaneously, for instance in the MAREANO

programme biologists and geologists uses the same real-time annotation tool to annotate fauna

and benthic environment from video, respectively (Buhl-Mortensen et al., 2015; Buhl-

Mortensen, pers. com.). Also the fame rate of the videos or time-lapse photos, will influence the

human perception (Shiffrar and Thomas, 2013; Smith and Kosslyn, 2014). Stimuli, in our case

images, presented too quickly (< 0.5 s) seems to preclude the attention of the second (Raymond,

2003). This mechanism (“attentional blink”) does, however, not seem to appear to different

features of the same object (Raymond, 2003). This furthermore underlines the importance of

focusing on one group of organisms at the time. Additional issues to consider when performing

manual labelling is elucidated by (Schoening et al., 2016).

4.1. Future application for use of images in decision making processes

This paper has focused on the human cognitive capacity to understand photos and graphical

presentations (such as presented in Figure 4) and how interdisciplinary collaboration on

processing and interpreting of these can enhance the environmental knowledge by overcoming

our limitations in the temporal and spatial domain. Due to the images potential of acting as a

boundary object, we will also claim that images have the potential of playing an important role in

improving operational decisions in organisations. As Carlile (2004), Leedom et al. (2007) claims

that data can be communicated only if the actors have established a sufficient unified

understanding. In the sensmaking/constructivist tradition knowledge is based on each

individual’s unique experience, expertise and interest (Leedom et al., 2007). Both Carlile (2004)

and (Brown et al., 2015) underline the importance of having a shared mission and a constructing

dialogue between the actors involved and that these actors must be experts within their own field

of expertise, in addition to be open minded and interested in other disciplines. In addition, Brown

et al. (2015) pinpoint institutional support and bridging research, policy and industry as absolute

preconditions for facilitation of interdisciplinary collaboration in daily work.

Using this approach on the two cases it would provide more accurate information on natural

variations of the species dynamic (Case 1), which in its turn can enable a risk based management

regime that are more capable of dividing natural variations from impact from industrial activities

and point discharges when applying the methodology in such areas. This knowledge make

decision makers better prepared to take proper decisions in an operational setting, for instance

related to drilling in a coral area. A 3D model (case 2) will be highly valid both as part of the

planning process upfront a survey or an operation (Ulfsnes et al., 2014) and to perform an

optimised environmental monitoring by ensuring the right spatial and temporal resolution of the

measurements. From a risk based approach, this emphasises the mutual dependencies between

modelling and environmental monitoring (Beyer et al., 2012; Godø et al., 2014b; Nilssen and

Johnsen, 2008; Nilssen et al., 2015a; Nilssen et al., 2015b). Models are used to identify the most

important risks. The environmental monitoring programmes should reflect the identified risks

and effectiveness of the applied mitigation measures. In addition, to improve extrapolations

(Almklov et al., 2014) and accuracy in the modelling (Rye et al., 2012), the environmental

monitoring should also measure parameters essential to improve the models, such as current

speed and direction, turbidity and sedimentation. Such an approach would contribute

considerable to ensure sound decisions. It will, however, require understanding and recognition

from the actors involved both on the individual and organisational level.

5. Conclusions and future perspectives

Increased use of images and models in environmental mapping and monitoring has a

considerable potential to enhance the knowledge about this environment. Interdisciplinary

collaboration is essential to extract the inherent information in time and space provided by these

images and models. The fact that images in general are understood by humans independent of

background and experiences and that they are so plastic that they mission can be adjusted from

case-to-case, indicates that they should be suitable boundary objects in interdisciplinary

collaboration in environmental monitoring. In a future perspective, these properties should also

make images a valid communicative and collaborative tool to ensure better decisions.

However, introducing images into environmental monitoring has also lead to unexpected

challenges that still have to be solved. One such an area is the ontology for species identification,

including taxonomy and routines and standards for proper categorisation and metadata-tagging

of images. Traditional species identification requires physically access to the organisms as the

taxonomy usually is based on morphological details not found in the images. Furthermore,

sufficient levels of details for the categories need to be established. An ontology that is too

detailed can lead to a too complicated system where we easily can get lost, while a too

aggregated system might provide a system that are too general to obtain knowledge of sufficient

accuracy. Together with the development of automatic or semi-automatic images analysis

elucidated in this paper, these aspects are essential to solve before the inherent knowledge

provided by images can be fully implemented and utilised.

6. Abbreviations and definitions

BIIGLE: Bielefeld Image Graphical Labeler and Explorer is a is a Web 2.0 based platform

containing easily uploaded images that can be accessed by collaborating scientists

Boundary object: objects which are both plastic enough to adapt to local needs and the

constraints of the several parties employing them, yet robust enough to maintain a

common identity across sites. They are weakly structured in common use, and become

strongly structured in individual-site use. They have different meanings in different social

worlds but their structure is common enough to more than one world to make them

recognizable, a means of tradition

GIS: Geographic Information System

Heat map: a graphical representation of data where the individual values contained in a matrix

are represented as colors

Horizontal instrumental variation: Measurements within this category are to a little extent

transformed compared to the human vision

Interdisciplinary: collaboration between different disciplines to solve a particular task or mission

Instrumental measurements of high contrast: measurements with a high degree of magnification

Instrumental measurements of low contrast: measurements with a high degree of reduction

LoVe : Lofoten-Vesterålen

RGB: Red, green and blue (the colours used in ordinary photo cameras)

Vertical instrumental variation: Measurements within this category are highly transformed

compared to the human vision and the immediate recognition has therefore disappeared.

These measurements have to be read in a particular scientific tradition.

7. Acknowledgement

This work has been carried out as part of Ingunn Nilssen’s industrial PhD, enabled by Statoil

ASA and the Research Council of Norway through the Industrial PhD programme, NRC project

number 222718. The work was also partly supported by the Research Council of Norway

through the Centres of Excellence funding scheme - AMOS, NRC project number 223254.

We will also like thank Øyvind Ødegård for fruitful discussions on the subject in general.

8. References

Adams, J., 2013. Experiencing shipwrecks and the primacy of vision. In: Adams, J., Rönnby, J.

(Eds.), Interpreting Shipwrecks: Maritime Archaeological Approaches. Highfield press,

pp. 85-96.

Almklov, P. G., Østerlie, T., Haavik, T. K., 2014. Situated with Infrastructures: Interactivity and

Entanglement in Sensor Data Interpretation. Journal of the Association for Information

Systems Vol 15, pp 263-286.

Beyer, J., Myhre, L. P., Sundt, R. C., Meier, S., Tollefsen, K.-E., Vabø, R., Klungsøyr, J., Sanni,

S., 2012. Environmental risk assessment of alkylphenols from offshore produced water

on fish reproduction. Marine Environmental Research Vol 75, pp 2-9,

doi:http://dx.doi.org/10.1016/j.marenvres.2011.11.011.

Biederman, I., 1987. Recognition-by-components: a theory of human image understanding.

Psychological review Vol 94, pp 115-147.

Brown, R. R., Deletic, A., Wong, T. H. F., 2015. Interdisciplinarity: How to catalyse

collaboration. Nature Vol 525, pp 315–317, doi:http://dx.doi.org/10.1038/525315a.

Buhl-Mortensen, L., Buhl-Mortensen, P., Dolan, M. F. J., Holte, B., 2015. The MAREANO

programme – A full coverage mapping of the Norwegian off-shore benthic environment

and fauna. Marine Biology Research Vol 11, pp 4-17,

doi:http://dx.doi.org/10.1080/17451000.2014.952312.

Buhl-Mortensen, P., pers. com. (1 July 2015)

Burri, R. V., Dumit, J., 2008. Social Studies of Scientific Imaging and Visualization. In: Hackett,

E. J., et al. (Eds.), The Handbook of Science and Technology Studies. The MIT Press,

Massachusets Institute of Technology, pp. 297-317.

Card, S. K., Mackinlay, J. D., Shneiderman, B., 1999. Readings in information visualization:

using vision to think. Morgan Kaufmann.

Carlile, P. R., 2004. Transferring, Translating, and Transforming: An Integrative Framework for

Managing Knowledge Across Boundaries. Organization Science Vol 15, pp 555-568,

doi:http://dx.doi.org/10.1287/orsc.1040.0094.

Carrasco, M., Ling, S., Read, S., 2004. Attention alters appearance. Nature neuroscience Vol 7,

pp 308-313, doi:http://dx.doi.org/10.1038/nn1194.

Chun, M. M., Golomb, J. D., Turk-Browne, N. B., 2011. A taxonomy of external and internal

attention. Annual review of psychology Vol 62, pp 73-101.

Godø, O. R., Johnsen, S., Torkelsen, T., 2014a. The LoVe Ocean Observatory is in Operation.

Marine Technology Society Journal Vol 48, pp 24-30,

doi:http://dx.doi.org/10.4031/MTSJ.48.2.2.

Godø, O. R., Handegard, N. O., Browman, H. I., Macaulay, G. J., Kaartvedt, S., Giske, J., Ona,

E., Huse, G., Johnsen, E., 2014b. Marine ecosystem acoustics (MEA): quantifying

processes in the sea at the spatio-temporal scales on which they occur. ICES Journal of

Marine Science Vol 71, pp 2357-2369, doi:http://dx.doi.org/10.1093/icesjms/fsu116.

Hepsø, V., Låte, M., Gramvik, G., Johnsen, S., Nilssen, I., Wesenberg, H., 2012. Integrated

Environmental Monitoring in Daily Operations. SPE Intelligent Energy International.

Society of Petroleum Engineers, Utrecht, The Netherlands.

Ihde, D., 1979. Technics and praxis. D. Reidel Publishing company.

Ihde, D., 1991. Instrumental Realism: The Interface between Philosophy of Science and

Philosophy of Technology. Indiana University Press.

Johnsen, G., Volent, Z., Dierssen, H., Pettersen, R., Ardelan, M. V., Søreide, F., Fearns, P.,

Ludvigsen, M., Moline, M., 2013. Underwater hyperspectral imagery to create

biogeochemical maps of seafloor properties. In: Watson, J., Zielinski, O. (Eds.), Subsea

Optics and Imaging. Woodhead Publishing, pp. 508-540.

Kelly, K., 1998. The Third Culture. Science Vol 279, pp 992-993,

doi:http://dx.doi.org/10.1126/science.279.5353.992.

Leedom, D. K., Eggleston, R. G., Ntuen, C. A., 2007. Engineering Complex Human-

Technological Work Systems: A Sensemaking Approach. DTIC Document.

Munzner, T., 2014. Visualization Analysis and Design. A K Peters/CRC Press.

Möller, T., Nilssen, I., Sumida, P. Y., Osterloff, J., Nattkemper, T. W., 2015. Change Detection

in underwater time laps videos from stationary observatories. GEOHAB, Salvador,

Brazil.

Nilssen, I., Johnsen, S., 2008. Holistic environmental management of discharges from the oil and

gas industry - Combining quantitative risk assessment and environmental monitoring.

SPE International Conference on Health, Safety, and Environment in Oil and Gas

Exploration and Production. Society of Petroleum Engineers, Nice, France.

Nilssen, I., Ødegård, Ø., Sørensen, A. J., Johnsen, G., Moline, M. A., Berge, J., 2015a. Integrated

environmental mapping and monitoring, a methodological approach to optimise

knowledge gathering and sampling strategy. Marine Pollution Bulletin Vol 96, pp 374-

383, doi:http://dx.doi.org/10.1016/j.marpolbul.2015.04.045.

Nilssen, I., dos Santos, F., Coutinho, R., Gomes, N., Cabral, M. M., Eide, I., Figueiredo, M. A.

O., Johnsen, G., Johnsen, S., 2015b. Assessing the potential impact of water-based drill

cuttings on deep-water calcareous red algae using species specific impact categories and

measured oceanographic and discharge data. Marine Environmental Research Vol 112,

Part A, pp 68-77, doi:http://dx.doi.org/10.1016/j.marenvres.2015.09.008.

Ontrup, J., Ehnert, N., Bergmann, M., Nattkemper, T. W., 2009. Biigle - Web 2.0 enabled

labelling and exploring of images from the Arctic deep-sea observatory HAUSGARTEN.

OCEANS 2009 - EUROPE, pp. 1-7.

Osterloff, J., Nilssen, I., Nattkemper, T. W., In Press-a. A computer vision approach for

monitoring the spatial and temporal shrimp distribution at the LoVe observatory.

Methods in Oceanography Vol, pp, doi:http://dx.doi.org/10.1016/j.mio.2016.03.002.

Osterloff, J., Nilssen, I., Nattkemper, T. W., In Press-b. Computational Coral Feature Monitoring

for the fixed underwater Observatory LoVe. IEEE OCEANS, Monterey (CA), USA.

Osterloff, J., Nilssen, I., Eide, I., de Oliveira Figueiredo, M. A., de Souza Tâmega, F. T.,

Nattkemper, T. W., 2016. Computational Visual Stress Level Analysis of Calcareous

Algae Exposed to Sedimentation. PLoS ONE Vol 11, pp e0157329,

doi:http://dx.doi.org/10.1371/journal.pone.0157329.

Purser, A., Ontrup, J., Schoening, T., Thomsen, L., Tong, R., Unnithan, V., Nattkemper, T. W.,

2013. Microhabitat and shrimp abundance within a Norwegian cold-water coral

ecosystem. Biogeosciences Vol 10, pp 5779-5791, doi:http://dx.doi.org/10.5194/bg-10-

5779-2013.

Raymond, J. E., 2003. New Objects, Not New Features, Trigger the Attentional Blink.

Psychological Science Vol 14, pp 54-59, doi:http://dx.doi.org/10.1111/1467-9280.01418.

Rye, H., Loekken, M., Moe, J., Ditlevsen, M. K., 2012. Simulation Of Concentrations And

Depositions Of Particle Matter Caused By Drilling Discharges. Comparison Between

Field Measurements And Simulation Results At Coral Locations. International

Conference on Health, Safety and Environment in Oil and Gas Exploration and

Production. Society of Petroleum Engineers, Perth, Australia.

Rylance, R., 2015. Grant giving: Global funders to focus on interdisciplinarity. Nature Vol 525,

pp 313–315, doi:http://dx.doi.org/10.1038/525313a.

Schoening, T., Osterloff, J., Nattkemper, T. W., 2016. RECOMIA - Recmmendations for Marine

Image Annotation: Lessons Learned and Future Directions. Frontiers in Marine Science

Vol 3, doi:http://dx.doi.org/10.3389/fmars.2016.00059.

Shiffrar, M., Thomas, J. P., 2013. Beyond the Scientific Objectification of the Human Body:

Differentiated Analyses of Human Motion and Object Motion. In: Rutherford, M. D.,

Kuhlmeier, V. A. (Eds.), Social perception : Detection and interpretation of animacy,

agency, and intention. Cambridge, Mass. : MIT Press., pp. 83-108.

Smith, E. E., Kosslyn, S. M., 2014. Cognitive psychology : Mind and brain. Edinburgh : Pearson.

Star, S. L., Griesemer, J. R., 1989. Institutional Ecology, "Translations" and Boundary Objects:

Amateurs and Professionals in Berkeley's Museum of Vertebrate Zoology, 1907-39.

Social Studies of Science Vol 19, pp 387-420,

doi:http://dx.doi.org/10.1177/030631289019003001.

Statoil,n.d. http://love.statoil.com/ (13 february 2015)

Ulfsnes, A., Moskeland, T., Frost, T. K., Hepsø, V., Gramvik, G., Brönner, U., Rylandsholm, P.,

2014. Towards Integrated Environmental Monitoring. SPE International Conference on

Health, Safety, and Environment. Society of Petroleum Engineers, Long Beach,

California, USA.

Ware, C., 2012. Information visualization: perception for design. Elsevier.

Wuchty, S., Jones, B. F., Uzzi, B., 2007. The increasing dominance of teams in production of

knowledge. Science Vol 316, pp 1036-1039.

Appendix

Figure 6: Example on how to “drive” complex data towards a presentation closer to the human perception by projecting a transparent version of the heat map (Figure 2 c) on to the original photo (Figure 2 a).

Figure 7 a: Schematic description of the selected approach for polyp activity annotations: At the

syntactic level the web-based image annotation and labelling system BIIGLE was used in the

interdisciplinary collaboration between computer scientists and biologists. Based on previous

experiences a proposed set of polyp categories were provided by the computer scientists (in

BIIGLE). The biologists wanted to increase the number of categories to get more accurate

measurements of the actual polyp activity. At the semantic level the diverging understanding

and wishes were identified and jointly agreed definitions of categories were established

(exemplified with polyps classified within the different categories; fully expanded (F E),

fully/partly expanded (F/P E), partly expanded (P E), partly expanded/retracted (P E/R) and

retracted (R). At the pragmatic level a so called Gold standard is established based on human

manual annotations. The Gold standard uses only those polyps that were labelled as the same

category by at least two of the three experts. The Gold standard was used as the training set

for the machine leaning and development of algorithms for polyp activity. Feedback loop:

algorithms are established, a new structure (number of categories of polyp activity) and a

gold standard for polyp activity are established (the algorithms are under development, so

images of the automatic polyp categorisation are at present not available).

Figure 7 b: In the syntactic level detection of biological activities in a cold water coral reef and in the

coral rubble underneath the reef over time are performed from photos. In the semantic level

species of interests are identified and an attempt to automatic and/or semi-automatic

detection is done. Different approaches are needed for each species (from upper left to lower

left: level of polyps activity for the reef building cold water coral L. pertusa, detection of L.

pertusa colour, unsupervised change detection over a certain time interval and shrimp

abundance, respectively). When algorithms are developed, other sensors measurements, such

as temperature, and current speed and direction, are used to try to explain the identified

behaviour patterns, here exemplified in the Pragmatic level by a principal component analysis

(PCA) score plot.

Doctoral theses in BiologyNorwegian University of Science and Technology

Department of Biology

Year Name Degree Title

1974 Tor-Henning Iversen

Dr. philos Botany

The roles of statholiths, auxin transport, and auxin metabolism in root gravitropism

1978 Tore Slagsvold Dr. philos Zoology

Breeding events of birds in relation to spring temperature and environmental phenology

1978 Egil Sakshaug Dr. philos Botany

"The influence of environmental factors on the chemical composition of cultivated and natural populations of marine phytoplankton"

1980 Arnfinn Langeland

Dr. philos Zoology

Interaction between fish and zooplankton populations and their effects on the material utilization in a freshwater lake

1980 Helge Reinertsen Dr. philos Botany

The effect of lake fertilization on the dynamics and stability of a limnetic ecosystem with special reference to the phytoplankton

1982 Gunn Mari Olsen Dr. scient Botany

Gravitropism in roots of Pisum sativum and Arabidopsis thaliana

1982 Dag Dolmen Dr. philos Zoology

Life aspects of two sympartic species of newts (Triturus, Amphibia) in Norway, with special emphasis on their ecological niche segregation

1984 Eivin Røskaft Dr. philos Zoology Sociobiological studies of the rook Corvus frugilegus

1984 Anne Margrethe Cameron

Dr. scient Botany

Effects of alcohol inhalation on levels of circulating testosterone, follicle stimulating hormone and luteinzing hormone in male mature rats

1984 Asbjørn Magne Nilsen

Dr. scient Botany

Alveolar macrophages from expectorates – Biological monitoring of workers exosed to occupational air pollution. An evaluation of the AM-test

1985 Jarle Mork Dr. philos Zoology Biochemical genetic studies in fish

1985 John Solem Dr. philos Zoology

Taxonomy, distribution and ecology of caddisflies (Trichoptera) in the Dovrefjell mountains

1985 Randi E. Reinertsen

Dr. philos Zoology

Energy strategies in the cold: Metabolic and thermoregulatory adaptations in small northern birds

1986 Bernt-Erik Sæther Dr. philos Zoology

Ecological and evolutionary basis for variation in reproductive traits of some vertebrates: A comparative approach

1986 Torleif Holthe Dr. philos Zoology

Evolution, systematics, nomenclature, and zoogeography in the polychaete orders Oweniimorpha and Terebellomorpha,with special reference to the Arctic and Scandinavian fauna

1987 Helene Lampe Dr. scient Zoology

The function of bird song in mate attraction and territorial defence, and the importance of song repertoires

1987 Olav Hogstad Dr. philos Zoology Winter survival strategies of the Willow tit Parus montanus

1987 Jarle Inge Holten Dr. philos Botany

Autecological investigations along a coust-inland transect at Nord-Møre, Central Norway

1987 Rita Kumar Dr. scient Botany

Somaclonal variation in plants regenerated from cell cultures of Nicotiana sanderae and Chrysanthemum morifolium

1987 Bjørn Åge Tømmerås

Dr. scient Zoology

Olfaction in bark beetle communities: Interspecific interactions in regulation of colonization density, predator -prey relationship and host attraction

1988 Hans Christian Pedersen

Dr. philos Zoology

Reproductive behaviour in willow ptarmigan with special emphasis on territoriality and parental care

1988 Tor G. Heggberget

Dr. philos Zoology

Reproduction in Atlantic Salmon (Salmo salar): Aspects of spawning, incubation, early life history and population structure

1988 Marianne V. Nielsen

Dr. scient Zoology

The effects of selected environmental factors on carbon allocation/growth of larval and juvenile mussels (Mytilus edulis)

1988 Ole Kristian Berg Dr. scient Zoology

The formation of landlocked Atlantic salmon (Salmo salarL.)

1989 John W. Jensen Dr. philos Zoology

Crustacean plankton and fish during the first decade of the manmade Nesjø reservoir, with special emphasis on the effects of gill nets and salmonid growth

1989 Helga J. Vivås Dr. scient Zoology

Theoretical models of activity pattern and optimal foraging: Predictions for the Moose Alces alces

1989 Reidar Andersen Dr. scient Zoology

Interactions between a generalist herbivore, the moose Alces alces, and its winter food resources: a study of behavioural variation

1989 Kurt Ingar Draget Dr. scient Botany Alginate gel media for plant tissue culture

1990 Bengt Finstad Dr. scientZoology

Osmotic and ionic regulation in Atlantic salmon, rainbow trout and Arctic charr: Effect of temperature, salinity and season

1990 Hege Johannesen Dr. scient Zoology

Respiration and temperature regulation in birds with special emphasis on the oxygen extraction by the lung

1990 Åse Krøkje Dr. scient Botany

The mutagenic load from air pollution at two work-places with PAH-exposure measured with Ames Salmonella/microsome test

1990 Arne Johan Jensen Dr. philos Zoology

Effects of water temperature on early life history, juvenile growth and prespawning migrations of Atlantic salmion (Salmo salar) and brown trout (Salmo trutta): A summary of studies in Norwegian streams

1990 Tor Jørgen Almaas

Dr. scient Zoology

Pheromone reception in moths: Response characteristics of olfactory receptor neurons to intra- and interspecific chemical cues

1990 Magne Husby Dr. scient Zoology

Breeding strategies in birds: Experiments with the Magpie Pica pica

1991 Tor Kvam Dr. scient Zoology

Population biology of the European lynx (Lynx lynx) in Norway

1991 Jan Henning L'Abêe Lund

Dr. philos Zoology

Reproductive biology in freshwater fish, brown trout Salmo trutta and roach Rutilus rutilus in particular

1991 Asbjørn Moen Dr. philos Botany

The plant cover of the boreal uplands of Central Norway. I. Vegetation ecology of Sølendet nature reserve; haymaking fens and birch woodlands

1991 Else Marie Løbersli

Dr. scient Botany Soil acidification and metal uptake in plants

1991 Trond Nordtug Dr. scient Zoology

Reflectometric studies of photomechanical adaptation in superposition eyes of arthropods

1991 Thyra Solem Dr. scient Botany

Age, origin and development of blanket mires in Central Norway

1991 Odd Terje Sandlund

Dr. philos Zoology

The dynamics of habitat use in the salmonid genera Coregonus and Salvelinus: Ontogenic niche shifts and polymorphism

1991 Nina Jonsson Dr. philos Zoology Aspects of migration and spawning in salmonids

1991 Atle Bones Dr. scient Botany

Compartmentation and molecular properties of thioglucoside glucohydrolase (myrosinase)

1992 Torgrim Breiehagen

Dr. scient Zoology

Mating behaviour and evolutionary aspects of the breeding system of two bird species: the Temminck's stint and the Pied flycatcher

1992 Anne Kjersti Bakken

Dr. scient Botany

The influence of photoperiod on nitrate assimilation and nitrogen status in timothy (Phleum pratense L.)

1992 Tycho Anker-Nilssen

Dr. scient Zoology

Food supply as a determinant of reproduction and population development in Norwegian Puffins Fratercula arctica

1992 Bjørn Munro Jenssen

Dr. philos Zoology

Thermoregulation in aquatic birds in air and water: With special emphasis on the effects of crude oil, chemically treated oil and cleaning on the thermal balance of ducks

1992 Arne Vollan Aarset

Dr. philos Zoology

The ecophysiology of under-ice fauna: Osmotic regulation, low temperature tolerance and metabolism in polar crustaceans.

1993 Geir Slupphaug Dr. scient Botany

Regulation and expression of uracil-DNA glycosylase and O6-methylguanine-DNA methyltransferase in mammalian cells

1993 Tor Fredrik Næsje Dr. scient Zoology Habitat shifts in coregonids.

1993 Yngvar Asbjørn Olsen

Dr. scient Zoology

Cortisol dynamics in Atlantic salmon, Salmo salar L.: Basal and stressor-induced variations in plasma levels ans some secondary effects.

1993 Bård Pedersen Dr. scient Botany

Theoretical studies of life history evolution in modular and clonal organisms

1993 Ole Petter Thangstad

Dr. scient Botany Molecular studies of myrosinase in Brassicaceae

1993 Thrine L. M. Heggberget

Dr. scient Zoology

Reproductive strategy and feeding ecology of the Eurasian otter Lutra lutra.

1993 Kjetil Bevanger Dr. scientZoology

Avian interactions with utility structures, a biological approach.

1993 Kåre Haugan Dr. scient Botany

Mutations in the replication control gene trfA of the broad host-range plasmid RK2

1994 Peder Fiske Dr. scient Zoology

Sexual selection in the lekking great snipe (Gallinago media): Male mating success and female behaviour at the lek

1994 Kjell Inge Reitan Dr. scient Botany

Nutritional effects of algae in first-feeding of marine fish larvae

1994 Nils Røv Dr. scient Zoology

Breeding distribution, population status and regulation of breeding numbers in the northeast-Atlantic Great Cormorant Phalacrocorax carbo carbo

1994 Annette-Susanne Hoepfner

Dr. scient Botany

Tissue culture techniques in propagation and breeding of Red Raspberry (Rubus idaeus L.)

1994 Inga Elise Bruteig Dr. scient Botany

Distribution, ecology and biomonitoring studies of epiphytic lichens on conifers

1994 Geir Johnsen Dr. scient Botany

Light harvesting and utilization in marine phytoplankton: Species-specific and photoadaptive responses

1994 Morten Bakken Dr. scient Zoology

Infanticidal behaviour and reproductive performance in relation to competition capacity among farmed silver fox vixens, Vulpes vulpes

1994 Arne Moksnes Dr. philos Zoology Host adaptations towards brood parasitism by the Cockoo

1994 Solveig Bakken Dr. scient Botany

Growth and nitrogen status in the moss Dicranum majus Sm. as influenced by nitrogen supply

1994 Torbjørn Forseth Dr. scient Zoology Bioenergetics in ecological and life history studies of fishes.

1995 Olav Vadstein Dr. philos Botany

The role of heterotrophic planktonic bacteria in the cycling of phosphorus in lakes: Phosphorus requirement, competitive ability and food web interactions

1995 Hanne Christensen

Dr. scient Zoology

Determinants of Otter Lutra lutra distribution in Norway: Effects of harvest, polychlorinated biphenyls (PCBs), human population density and competition with mink Mustela vision

1995 Svein Håkon Lorentsen

Dr. scient Zoology

Reproductive effort in the Antarctic Petrel Thalassoica antarctica; the effect of parental body size and condition

1995 Chris Jørgen Jensen

Dr. scient Zoology

The surface electromyographic (EMG) amplitude as an estimate of upper trapezius muscle activity

1995 Martha Kold Bakkevig

Dr. scient Zoology

The impact of clothing textiles and construction in a clothing system on thermoregulatory responses, sweat accumulation and heat transport

1995 Vidar Moen Dr. scient Zoology

Distribution patterns and adaptations to light in newly introduced populations of Mysis relicta and constraints on Cladoceran and Char populations

1995Hans Haavardsholm Blom

Dr. philos Botany

A revision of the Schistidium apocarpum complex in Norway and Sweden

1996 Jorun Skjærmo Dr. scient Botany

Microbial ecology of early stages of cultivated marine fish; inpact fish-bacterial interactions on growth and survival of larvae

1996 Ola Ugedal Dr. scient Zoology Radiocesium turnover in freshwater fishes

1996 Ingibjørg Einarsdottir

Dr. scient Zoology

Production of Atlantic salmon (Salmo salar) and Arctic charr (Salvelinus alpinus): A study of some physiological and immunological responses to rearing routines

1996 Christina M. S. Pereira

Dr. scient Zoology

Glucose metabolism in salmonids: Dietary effects and hormonal regulation

1996 Jan Fredrik Børseth

Dr. scient Zoology

The sodium energy gradients in muscle cells of Mytilus edulis and the effects of organic xenobiotics

1996 Gunnar Henriksen Dr. scient Zoology

Status of Grey seal Halichoerus grypus and Harbour seal Phoca vitulina in the Barents sea region

1997 Gunvor Øie Dr. scient Botany

Eevalution of rotifer Brachionus plicatilis quality in early first feeding of turbot Scophtalmus maximus L. larvae

1997 Håkon Holien Dr. scient Botany

Studies of lichens in spurce forest of Central Norway. Diversity, old growth species and the relationship to site and stand parameters

1997 Ole Reitan Dr. scient Zoology Responses of birds to habitat disturbance due to damming

1997 Jon Arne Grøttum Dr. scient Zoology

Physiological effects of reduced water quality on fish in aquaculture

1997 Per Gustav Thingstad

Dr. scient Zoology

Birds as indicators for studying natural and human-induced variations in the environment, with special emphasis on the suitability of the Pied Flycatcher

1997 Torgeir Nygård Dr. scient Zoology

Temporal and spatial trends of pollutants in birds in Norway: Birds of prey and Willow Grouse used as

1997 Signe Nybø Dr. scient Zoology

Impacts of long-range transported air pollution on birds with particular reference to the dipper Cinclus cinclus in southern Norway

1997 Atle Wibe Dr. scient Zoology

Identification of conifer volatiles detected by receptor neurons in the pine weevil (Hylobius abietis), analysed by gas chromatography linked to electrophysiology and to mass spectrometry

1997 Rolv Lundheim Dr. scient Zoology Adaptive and incidental biological ice nucleators

1997 Arild Magne Landa

Dr. scient Zoology

Wolverines in Scandinavia: ecology, sheep depredation and conservation

1997 Kåre Magne Nielsen

Dr. scient Botany

An evolution of possible horizontal gene transfer from plants to sail bacteria by studies of natural transformation in Acinetobacter calcoacetius

1997 Jarle Tufto Dr. scient Zoology

Gene flow and genetic drift in geographically structured populations: Ecological, population genetic, and statistical models

1997 Trygve Hesthagen Dr. philos Zoology

Population responces of Arctic charr (Salvelinus alpinus(L.)) and brown trout (Salmo trutta L.) to acidification in Norwegian inland waters

1997 Trygve Sigholt Dr. philos Zoology

Control of Parr-smolt transformation and seawater tolerance in farmed Atlantic Salmon (Salmo salar) Effects of photoperiod, temperature, gradual seawater acclimation, NaCl and betaine in the diet

1997 Jan Østnes Dr. scient Zoology Cold sensation in adult and neonate birds

1998 Seethaledsumy Visvalingam

Dr. scient Botany

Influence of environmental factors on myrosinases and myrosinase-binding proteins

1998 Thor Harald Ringsby

Dr. scient Zoology

Variation in space and time: The biology of a House sparrow metapopulation

1998 Erling Johan Solberg

Dr. scient Zoology

Variation in population dynamics and life history in a Norwegian moose (Alces alces) population: consequences of harvesting in a variable environment

1998 Sigurd Mjøen Saastad

Dr. scient Botany

Species delimitation and phylogenetic relationships between the Sphagnum recurvum complex (Bryophyta): genetic variation and phenotypic plasticity

1998 Bjarte Mortensen Dr. scient Botany

Metabolism of volatile organic chemicals (VOCs) in a head liver S9 vial equilibration system in vitro

1998 Gunnar Austrheim Dr. scient Botany

Plant biodiversity and land use in subalpine grasslands. – Aconservtaion biological approach

1998 Bente Gunnveig Berg

Dr. scient Zoology

Encoding of pheromone information in two related moth species

1999 Kristian Overskaug

Dr. scient Zoology

Behavioural and morphological characteristics in Northern Tawny Owls Strix aluco: An intra- and interspecific comparative approach

1999 Hans Kristen Stenøien

Dr. scient Botany

Genetic studies of evolutionary processes in various populations of nonvascular plants (mosses, liverworts and hornworts)

1999 Trond Arnesen Dr. scient Botany

Vegetation dynamics following trampling and burning in the outlying haylands at Sølendet, Central Norway

1999 Ingvar Stenberg Dr. scient Zoology

Habitat selection, reproduction and survival in the White-backed Woodpecker Dendrocopos leucotos

1999 Stein Olle Johansen

Dr. scient Botany

A study of driftwood dispersal to the Nordic Seas by dendrochronology and wood anatomical analysis

1999 Trina Falck Galloway

Dr. scient Zoology

Muscle development and growth in early life stages of the Atlantic cod (Gadus morhua L.) and Halibut (Hippoglossus hippoglossus L.)

1999 Marianne Giæver Dr. scient Zoology

Population genetic studies in three gadoid species: blue whiting (Micromisistius poutassou), haddock (Melanogrammus aeglefinus) and cod (Gradus morhua) in the North-East Atlantic

1999 Hans Martin Hanslin

Dr. scient Botany

The impact of environmental conditions of density dependent performance in the boreal forest bryophytes Dicranum majus, Hylocomium splendens, Plagiochila asplenigides, Ptilium crista-castrensis and Rhytidiadelphus lokeus

1999 Ingrid Bysveen Mjølnerød

Dr. scient Zoology

Aspects of population genetics, behaviour and performance of wild and farmed Atlantic salmon (Salmo salar) revealed by molecular genetic techniques

1999 Else Berit Skagen Dr. scient Botany

The early regeneration process in protoplasts from Brassica napus hypocotyls cultivated under various g-forces

1999 Stein-Are Sæther Dr. philos Zoology

Mate choice, competition for mates, and conflicts of interest in the Lekking Great Snipe

1999 Katrine Wangen Rustad

Dr. scient Zoology

Modulation of glutamatergic neurotransmission related to cognitive dysfunctions and Alzheimer’s disease

1999 Per Terje Smiseth Dr. scient Zoology Social evolution in monogamous families:

1999 Gunnbjørn Bremset

Dr. scient Zoology

Young Atlantic salmon (Salmo salar L.) and Brown trout (Salmo trutta L.) inhabiting the deep pool habitat, with special reference to their habitat use, habitat preferences and competitive interactions

1999 Frode Ødegaard Dr. scient Zoology

Host spesificity as parameter in estimates of arhrophod species richness

1999 Sonja Andersen Dr. scient Zoology

Expressional and functional analyses of human, secretory phospholipase A2

2000 Ingrid Salvesen Dr. scient Botany

Microbial ecology in early stages of marine fish: Development and evaluation of methods for microbial management in intensive larviculture

2000 Ingar Jostein Øien Dr. scient Zoology

The Cuckoo (Cuculus canorus) and its host: adaptions and counteradaptions in a coevolutionary arms race

2000 Pavlos Makridis Dr. scient Botany

Methods for the microbial econtrol of live food used for the rearing of marine fish larvae

2000 Sigbjørn Stokke Dr. scient Zoology

Sexual segregation in the African elephant (Loxodonta africana)

2000 Odd A. Gulseth Dr. philos Zoology

Seawater tolerance, migratory behaviour and growth of Charr, (Salvelinus alpinus), with emphasis on the high Arctic Dieset charr on Spitsbergen, Svalbard

2000 Pål A. Olsvik Dr. scient Zoology

Biochemical impacts of Cd, Cu and Zn on brown trout (Salmo trutta) in two mining-contaminated rivers in Central Norway

2000 Sigurd Einum Dr. scient Zoology

Maternal effects in fish: Implications for the evolution of breeding time and egg size

2001 Jan Ove Evjemo Dr. scient Zoology

Production and nutritional adaptation of the brine shrimp Artemia sp. as live food organism for larvae of marine cold water fish species

2001 Olga Hilmo Dr. scient Botany

Lichen response to environmental changes in the managed boreal forset systems

2001 Ingebrigt Uglem Dr. scient Zoology

Male dimorphism and reproductive biology in corkwing wrasse (Symphodus melops L.)

2001 Bård Gunnar Stokke

Dr. scient Zoology

Coevolutionary adaptations in avian brood parasites and their hosts

2002 Ronny Aanes Dr. scient Zoology

Spatio-temporal dynamics in Svalbard reindeer (Rangifer tarandus platyrhynchus)

2002 Mariann Sandsund Dr. scient Zoology

Exercise- and cold-induced asthma. Respiratory and thermoregulatory responses

2002 Dag-Inge Øien Dr. scient Botany

Dynamics of plant communities and populations in boreal vegetation influenced by scything at Sølendet, Central Norway

2002 Frank Rosell Dr. scient Zoology The function of scent marking in beaver (Castor fiber)

2002 Janne Østvang Dr. scient Botany

The Role and Regulation of Phospholipase A2 in Monocytes During Atherosclerosis Development

2002 Terje Thun Dr. philos Biology

Dendrochronological constructions of Norwegian conifer chronologies providing dating of historical material

2002 Birgit Hafjeld Borgen

Dr. scient Biology

Functional analysis of plant idioblasts (Myrosin cells) and their role in defense, development and growth

2002 Bård Øyvind Solberg

Dr. scient Biology

Effects of climatic change on the growth of dominating tree species along major environmental gradients

2002 Per Winge Dr. scient Biology

The evolution of small GTP binding proteins in cellular organisms. Studies of RAC GTPases in Arabidopsis thaliana and the Ral GTPase from Drosophila melanogaster

2002 Henrik Jensen Dr. scient Biology

Causes and consequenses of individual variation in fitness-related traits in house sparrows

2003 Jens Rohloff Dr. philos Biology

Cultivation of herbs and medicinal plants in Norway –Essential oil production and quality control

2003 Åsa Maria O. Espmark Wibe

Dr. scient Biology

Behavioural effects of environmental pollution in threespine stickleback Gasterosteus aculeatur L.

2003 Dagmar Hagen Dr. scient Biology

Assisted recovery of disturbed arctic and alpine vegetation –an integrated approach

2003 Bjørn Dahle Dr. scient Biology Reproductive strategies in Scandinavian brown bears

2003 Cyril Lebogang Taolo

Dr. scient Biology

Population ecology, seasonal movement and habitat use of the African buffalo (Syncerus caffer) in Chobe National Park, Botswana

2003 Marit Stranden Dr. scient Biology

Olfactory receptor neurones specified for the same odorants in three related Heliothine species (Helicoverpa armigera, Helicoverpa assulta and Heliothis virescens)

2003 Kristian Hassel Dr. scient Biology

Life history characteristics and genetic variation in an expanding species, Pogonatum dentatum

2003 David Alexander Rae

Dr. scient Biology

Plant- and invertebrate-community responses to species interaction and microclimatic gradients in alpine and Artic environments

2003 Åsa A Borg Dr. scient Biology

Sex roles and reproductive behaviour in gobies and guppies: a female perspective

2003 Eldar Åsgard Bendiksen

Dr. scient Biology

Environmental effects on lipid nutrition of farmed Atlantic salmon (Salmo Salar L.) parr and smolt

2004 Torkild Bakken Dr. scient Biology A revision of Nereidinae (Polychaeta, Nereididae)

2004 Ingar Pareliussen Dr. scient Biology

Natural and Experimental Tree Establishment in a Fragmented Forest, Ambohitantely Forest Reserve, Madagascar

2004 Tore Brembu Dr. scient Biology

Genetic, molecular and functional studies of RAC GTPases and the WAVE-like regulatory protein complex in Arabidopsis thaliana

2004 Liv S. Nilsen Dr. scient Biology

Coastal heath vegetation on central Norway; recent past, present state and future possibilities

2004 Hanne T. Skiri Dr. scient Biology

Olfactory coding and olfactory learning of plant odours in heliothine moths. An anatomical, physiological and behavioural study of three related species (Heliothis virescens, Helicoverpa armigera and Helicoverpa assulta)

2004 Lene Østby Dr. scient Biology

Cytochrome P4501A (CYP1A) induction and DNA adducts as biomarkers for organic pollution in the natural environment

2004 Emmanuel J. Gerreta

Dr. philos Biology

The Importance of Water Quality and Quantity in the Tropical Ecosystems, Tanzania

2004 Linda Dalen Dr. scient Biology

Dynamics of Mountain Birch Treelines in the Scandes Mountain Chain, and Effects of Climate Warming

2004 Lisbeth Mehli Dr. scient Biology

Polygalacturonase-inhibiting protein (PGIP) in cultivated strawberry (Fragaria x ananassa): characterisation and induction of the gene following fruit infection by Botrytis cinerea

2004 Børge Moe Dr. scient Biology

Energy-Allocation in Avian Nestlings Facing Short-Term Food Shortage

2005 Matilde Skogen Chauton

Dr. scient Biology

Metabolic profiling and species discrimination from High-Resolution Magic Angle Spinning NMR analysis of whole-cell samples

2005 Sten Karlsson Dr. scient Biology Dynamics of Genetic Polymorphisms

2005 Terje Bongard Dr. scient Biology

Life History strategies, mate choice, and parental investment among Norwegians over a 300-year period

2005 Tonette Røstelien ph.d Biology Functional characterisation of olfactory receptor neurone types in heliothine moths

2005 Erlend Kristiansen Dr. scient Biology Studies on antifreeze proteins

2005 Eugen G. Sørmo Dr. scient Biology

Organochlorine pollutants in grey seal (Halichoerus grypus)pups and their impact on plasma thyrid hormone and vitamin A concentrations

2005 Christian Westad Dr. scient Biology Motor control of the upper trapezius

2005 Lasse Mork Olsen ph.d Biology Interactions between marine osmo- and phagotrophs in different physicochemical environments

2005 Åslaug Viken ph.d Biology Implications of mate choice for the management of small populations

2005 Ariaya Hymete Sahle Dingle ph.d Biology Investigation of the biological activities and chemical

constituents of selected Echinops spp. growing in Ethiopia

2005 Anders Gravbrøt Finstad ph.d Biology Salmonid fishes in a changing climate: The winter challenge

2005Shimane Washington Makabu

ph.d Biology Interactions between woody plants, elephants and other browsers in the Chobe Riverfront, Botswana

2005 Kjartan Østbye Dr. scient Biology

The European whitefish Coregonus lavaretus (L.) species complex: historical contingency and adaptive radiation

2006 Kari Mette Murvoll ph.d Biology

Levels and effects of persistent organic pollutans (POPs) in -tocopherol – potential biomakers

of POPs in birds?

2006 Ivar Herfindal Dr. scient Biology

Life history consequences of environmental variation along ecological gradients in northern ungulates

2006 Nils Egil Tokle ph.d BiologyAre the ubiquitous marine copepods limited by food or predation? Experimental and field-based studies with main focus on Calanus finmarchicus

2006 Jan Ove Gjershaug

Dr. philos Biology

Taxonomy and conservation status of some booted eagles in south-east Asia

2006 Jon Kristian Skei Dr. scient Biology

Conservation biology and acidification problems in the breeding habitat of amphibians in Norway

2006 Johanna Järnegren ph.d Biology Acesta Oophaga and Acesta Excavata – a study of hidden biodiversity

2006 Bjørn Henrik Hansen ph.d Biology

Metal-mediated oxidative stress responses in brown trout (Salmo trutta) from mining contaminated rivers in Central Norway

2006 Vidar Grøtan ph.d Biology Temporal and spatial effects of climate fluctuations on population dynamics of vertebrates

2006 Jafari R Kideghesho ph.d Biology Wildlife conservation and local land use conflicts in western

Serengeti, Corridor Tanzania

2006 Anna Maria Billing ph.d Biology Reproductive decisions in the sex role reversed pipefish

Syngnathus typhle: when and how to invest in reproduction

2006 Henrik Pärn ph.d Biology Female ornaments and reproductive biology in the bluethroat

2006 Anders J. Fjellheim ph.d Biology Selection and administration of probiotic bacteria to marine

fish larvae

2006 P. Andreas Svensson ph.d Biology Female coloration, egg carotenoids and reproductive

success: gobies as a model system

2007 Sindre A. Pedersen ph.d Biology

Metal binding proteins and antifreeze proteins in the beetle Tenebrio molitor - a study on possible competition for the semi-essential amino acid cysteine

2007 Kasper Hancke ph.d BiologyPhotosynthetic responses as a function of light and temperature: Field and laboratory studies on marine microalgae

2007 Tomas Holmern ph.d Biology Bushmeat hunting in the western Serengeti: Implications for community-based conservation

2007 Kari Jørgensen ph.d Biology Functional tracing of gustatory receptor neurons in the CNS and chemosensory learning in the moth Heliothis virescens

2007 Stig Ulland ph.d Biology

Functional Characterisation of Olfactory Receptor Neurons in the Cabbage Moth, (Mamestra brassicae L.) (Lepidoptera, Noctuidae). Gas Chromatography Linked to Single Cell Recordings and Mass Spectrometry

2007 Snorre Henriksen ph.d Biology Spatial and temporal variation in herbivore resources at northern latitudes

2007 Roelof Frans May ph.d Biology Spatial Ecology of Wolverines in Scandinavia

2007 Vedasto Gabriel Ndibalema ph.d Biology

Demographic variation, distribution and habitat use between wildebeest sub-populations in the Serengeti National Park, Tanzania

2007 Julius William Nyahongo ph.d Biology

Depredation of Livestock by wild Carnivores and Illegal Utilization of Natural Resources by Humans in the Western Serengeti, Tanzania

2007 Shombe Ntaraluka Hassan ph.d Biology Effects of fire on large herbivores and their forage resources

in Serengeti, Tanzania

2007 Per-Arvid Wold ph.d BiologyFunctional development and response to dietary treatment in larval Atlantic cod (Gadus morhua L.) Focus on formulated diets and early weaning

2007 Anne Skjetne Mortensen ph.d Biology

Toxicogenomics of Aryl Hydrocarbon- and Estrogen Receptor Interactions in Fish: Mechanisms and Profiling of Gene Expression Patterns in Chemical Mixture Exposure Scenarios

2008 Brage Bremset Hansen ph.d Biology

The Svalbard reindeer (Rangifer tarandus platyrhynchus)and its food base: plant-herbivore interactions in a high-arctic ecosystem

2008 Jiska van Dijk ph.d Biology Wolverine foraging strategies in a multiple-use landscape

2008 Flora John Magige ph.d Biology The ecology and behaviour of the Masai Ostrich (Struthio

camelus massaicus) in the Serengeti Ecosystem, Tanzania

2008 Bernt Rønning ph.d Biology Sources of inter- and intra-individual variation in basal metabolic rate in the zebra finch, (Taeniopygia guttata)

2008 Sølvi Wehn ph.d BiologyBiodiversity dynamics in semi-natural mountain landscapes - A study of consequences of changed agricultural practices in Eastern Jotunheimen

2008 Trond Moxness Kortner ph.d Biology

"The Role of Androgens on previtellogenic oocyte growth in Atlantic cod (Gadus morhua): Identification and patterns of differentially expressed genes in relation to Stereological Evaluations"

2008 Katarina Mariann Jørgensen

Dr. scient Biology

The role of platelet activating factor in activation of growth arrested keratinocytes and re-epithelialisation

2008 Tommy Jørstad ph.d Biology Statistical Modelling of Gene Expression Data

2008 Anna Kusnierczyk ph.d Biology Arabidopsis thaliana Responses to Aphid Infestation

2008 Jussi Evertsen ph.d Biology Herbivore sacoglossans with photosynthetic chloroplasts

2008 John Eilif Hermansen ph.d Biology

Mediating ecological interests between locals and globals by means of indicators. A study attributed to the asymmetry between stakeholders of tropical forest at Mt. Kilimanjaro, Tanzania

2008 Ragnhild Lyngved ph.d Biology Somatic embryogenesis in Cyclamen persicum. Biological investigations and educational aspects of cloning

2008 Line Elisabeth Sundt-Hansen ph.d Biology Cost of rapid growth in salmonid fishes

2008 Line Johansen ph.d BiologyExploring factors underlying fluctuations in white clover populations – clonal growth, population structure and spatial distribution

2009 Astrid Jullumstrø Feuerherm ph.d Biology Elucidation of molecular mechanisms for pro-inflammatory

phospholipase A2 in chronic disease

2009 Pål Kvello ph.d Biology

Neurons forming the network involved in gustatory coding and learning in the moth Heliothis virescens: Physiological and morphological characterisation, and integration into a standard brain atlas

2009 Trygve Devold Kjellsen ph.d Biology Extreme Frost Tolerance in Boreal Conifers

2009 Johan Reinert Vikan ph.d Biology Coevolutionary interactions between common cuckoos

Cuculus canorus and Fringilla finches

2009 Zsolt Volent ph.d Biology

Remote sensing of marine environment: Applied surveillance with focus on optical properties of phytoplankton, coloured organic matter and suspended matter

2009 Lester Rocha ph.d Biology Functional responses of perennial grasses to simulated grazing and resource availability

2009 Dennis Ikanda ph.d BiologyDimensions of a Human-lion conflict: Ecology of human predation and persecution of African lions (Panthera leo) in Tanzania

2010 Huy Quang Nguyen ph.d Biology

Egg characteristics and development of larval digestive function of cobia (Rachycentron canadum) in response to dietary treatments - Focus on formulated diets

2010 Eli Kvingedal ph.d Biology Intraspecific competition in stream salmonids: the impact of environment and phenotype

2010 Sverre Lundemo ph.d Biology Molecular studies of genetic structuring and demography in Arabidopsis from Northern Europe

2010 Iddi Mihijai Mfunda ph.d Biology

Wildlife Conservation and People’s livelihoods: Lessons Learnt and Considerations for Improvements. Tha Case of Serengeti Ecosystem, Tanzania

2010 Anton Tinchov Antonov ph.d Biology Why do cuckoos lay strong-shelled eggs? Tests of the

puncture resistance hypothesis

2010 Anders Lyngstad ph.d Biology Population Ecology of Eriophorum latifolium, a Clonal Species in Rich Fen Vegetation

2010 Hilde Færevik ph.d Biology Impact of protective clothing on thermal and cognitive responses

2010 Ingerid Brænne Arbo

ph.d Medical technology

Nutritional lifestyle changes – effects of dietary carbohydrate restriction in healthy obese and overweight humans

2010 Yngvild Vindenes ph.d Biology Stochastic modeling of finite populations with individual heterogeneity in vital parameters

2010 Hans-Richard Brattbakk

ph.d Medical technology

The effect of macronutrient composition, insulin stimulation, and genetic variation on leukocyte gene expression and possible health benefits

2011 Geir Hysing Bolstad ph.d Biology Evolution of Signals: Genetic Architecture, Natural

Selection and Adaptive Accuracy

2011 Karen de Jong ph.d Biology Operational sex ratio and reproductive behaviour in the two-spotted goby (Gobiusculus flavescens)

2011 Ann-Iren Kittang ph.d Biology

Arabidopsis thaliana L. adaptation mechanisms to microgravity through the EMCS MULTIGEN-2 experiment on the ISS:– The science of space experiment integration and adaptation to simulated microgravity

2011 Aline Magdalena Lee ph.d Biology Stochastic modeling of mating systems and their effect on

population dynamics and genetics

2011Christopher Gravningen Sørmo

ph.d BiologyRho GTPases in Plants: Structural analysis of ROP GTPases; genetic and functional studies of MIRO GTPases in Arabidopsis thaliana

2011 Grethe Robertsen ph.d Biology Relative performance of salmonid phenotypes across environments and competitive intensities

2011 Line-Kristin Larsen ph.d Biology

Life-history trait dynamics in experimental populations of guppy (Poecilia reticulata): the role of breeding regime and captive environment

2011 Maxim A. K. Teichert ph.d Biology Regulation in Atlantic salmon (Salmo salar): The interaction

between habitat and density

2011 Torunn BeateHancke ph.d Biology

Use of Pulse Amplitude Modulated (PAM) Fluorescence and Bio-optics for Assessing Microalgal Photosynthesis and Physiology

2011 Sajeda Begum ph.d Biology Brood Parasitism in Asian Cuckoos: Different Aspects of Interactions between Cuckoos and their Hosts in Bangladesh

2011 Kari J. K. Attramadal ph.d Biology Water treatment as an approach to increase microbial control

in the culture of cold water marine larvae

2011 Camilla Kalvatn Egset ph.d Biology The Evolvability of Static Allometry: A Case Study

2011 AHM Raihan Sarker ph.d Biology Conflict over the conservation of the Asian elephant

(Elephas maximus) in Bangladesh

2011 Gro Dehli Villanger ph.d Biology

Effects of complex organohalogen contaminant mixtures on thyroid hormone homeostasis in selected arctic marine mammals

2011 Kari Bjørneraas ph.d Biology Spatiotemporal variation in resource utilisation by a large herbivore, the moose

2011 John Odden ph.d Biology The ecology of a conflict: Eurasian lynx depredation on domestic sheep

2011 Simen Pedersen ph.d Biology Effects of native and introduced cervids on small mammals and birds

2011 Mohsen Falahati-Anbaran ph.d Biology Evolutionary consequences of seed banks and seed dispersal

in Arabidopsis

2012 Jakob Hønborg Hansen ph.d Biology Shift work in the offshore vessel fleet: circadian rhythms

and cognitive performance

2012 Elin Noreen ph.d Biology Consequences of diet quality and age on life-history traits in a small passerine bird

2012 Irja Ida Ratikainen ph.d Biology Theoretical and empirical approaches to studying foraging

decisions: the past and future of behavioural ecology

2012 Aleksander Handå ph.d Biology Cultivation of mussels (Mytilus edulis):Feed requirements, storage and integration with salmon (Salmo salar) farming

2012 Morten Kraabøl ph.d Biology Reproductive and migratory challenges inflicted on migrant brown trour (Salmo trutta L) in a heavily modified river

2012 Jisca Huisman ph.d Biology Gene flow and natural selection in Atlantic salmon

Maria Bergvik ph.d Biology Lipid and astaxanthin contents and biochemical post-harvest stability in Calanus finmarchicus

2012 Bjarte Bye Løfaldli ph.d Biology Functional and morphological characterization of central

olfactory neurons in the model insect Heliothis virescens.

2012 Karen Marie Hammer ph.d Biology

Acid-base regulation and metabolite responses in shallow-and deep-living marine invertebrates during environmental hypercapnia

2012 Øystein Nordrum Wiggen ph.d Biology Optimal performance in the cold

2012 Robert Dominikus Fyumagwa

Dr. Philos Biology

Anthropogenic and natural influence on disease prevalence at the human –livestock-wildlife interface in the Serengeti ecosystem, Tanzania

2012 Jenny Bytingsvik ph.d Biology

Organohalogenated contaminants (OHCs) in polar bear mother-cub pairs from Svalbard, Norway. Maternal transfer, exposure assessment and thyroid hormone disruptive effects in polar bear cubs

2012 Christer Moe Rolandsen ph.d Biology The ecological significance of space use and movement

patterns of moose in a variable environment

2012 Erlend Kjeldsberg Hovland ph.d Biology Bio-optics and Ecology in Emiliania huxleyi Blooms: Field

and Remote Sensing Studies in Norwegian Waters

2012 Lise Cats Myhre ph.d Biology Effects of the social and physical environment on mating behaviour in a marine fish

2012 Tonje Aronsen ph.d Biology Demographic, environmental and evolutionary aspects of sexual selection

Bin Liu ph.d Biology Molecular genetic investigation of cell separation and cell death regulation in Arabidopsis thaliana

2013 Jørgen Rosvold ph.d Biology Ungulates in a dynamic and increasingly human dominated landscape – A millennia-scale perspective

2013 Pankaj Barah ph.d Biology Integrated Systems Approaches to Study Plant Stress Responses

2013 Marit Linnerud ph.d Biology Patterns in spatial and temporal variation in population abundances of vertebrates

2013 Xinxin Wang ph.d Biology Integrated multi-trophic aquaculture driven by nutrient wastes released from Atlantic salmon (Salmo salar) farming

2013 Ingrid Ertshus Mathisen ph.d Biology

Structure, dynamics, and regeneration capacity at the sub-arctic forest-tundra ecotone of northern Norway and Kola Peninsula, NW Russia

2013 Anders Foldvik ph.d Biology Spatial distributions and productivity in salmonidpopulations

2013 Anna Marie Holand ph.d Biology Statistical methods for estimating intra- and inter-population

variation in genetic diversity

2013 Anna Solvang Båtnes ph.d Biology Light in the dark – the role of irradiance in the high Arctic

marine ecosystem during polar night

2013 Sebastian Wacker ph.d Biology The dynamics of sexual selection: effects of OSR, densityand resource competition in a fish

2013 Cecilie Miljeteig ph.d Biology Phototaxis in Calanus finmarchicus – light sensitivity andthe influence of energy reserves and oil exposure

2013 Ane Kjersti Vie ph.d Biology Molecular and functional characterisation of the IDA familyof signalling peptides in Arabidopsis thaliana

2013 Marianne Nymark ph.d Biology Light responses in the marine diatom Phaeodactylumtricornutum

2014 Jannik Schultner ph.d Biology Resource Allocation under Stress - Mechanisms andStrategies in a Long-Lived Bird

2014 Craig Ryan Jackson ph.d Biology

Factors influencing African wild dog (Lycaon pictus) habitat selection and ranging behaviour: conservation and management implications

2014 Aravind Venkatesan ph.d Biology Application of Semantic Web Technology to establish

knowledge management and discovery in the Life Sciences

2014 Kristin Collier Valle ph.d Biology Photoacclimation mechanisms and light responses in marine

micro- and macroalgae

2014 Michael Puffer ph.d Biology Effects of rapidly fluctuating water levels on juvenileAtlantic salmon (Salmo salar L.)

2014 Gundula S. Bartzke ph.d Biology Effects of power lines on moose (Alces alces) habitat

selection, movements and feeding activity

2014 Eirin Marie Bjørkvoll ph.d Biology Life-history variation and stochastic population dynamics in

vertebrates

2014 Håkon Holand ph.d Biology The parasite Syngamus trachea in a metapopulation ofhouse sparrows

2014 Randi Magnus Sommerfelt ph.d Biology Molecular mechanisms of inflammation – a central role for

cytosolic phospholiphase A2

2014 Espen Lie Dahl ph.d Biology Population demographics in white-tailed eagle at an on-shore wind farm area in coastal Norway

2014 Anders Øverby ph.d BiologyFunctional analysis of the action of plant isothiocyanates: cellular mechanisms and in vivo role in plants, and anticancer activity

2014 Kamal Prasad Acharya ph.d Biology Invasive species: Genetics, characteristics and trait variation

along a latitudinal gradient.

2014 Ida Beathe Øverjordet ph.d Biology

Element accumulation and oxidative stress variables in Arctic pelagic food chains: Calanus, little auks (alle alle) and black-legged kittiwakes (Rissa tridactyla)

2014 Kristin Møller Gabrielsen ph.d Biology

Target tissue toxicity of the thyroid hormone system in two species of arctic mammals carrying high loads of organohalogen contaminants

2015 Gine Roll Skjervø dr. philosBiology

Testing behavioral ecology models with historicalindividual-based human demographic data from Norway

2015 Nils Erik Gustaf Forsberg ph.d Biology Spatial and Temporal Genetic Structure in Landrace Cereals

2015 Leila Alipanah ph.d BiologyIntegrated analyses of nitrogen and phosphorus deprivation in the diatoms Phaeodactylum tricornutum and Seminavis robusta

2015 Javad Najafi ph.d Biology Molecular investigation of signaling components in sugarsensing and defense in Arabidopsis thaliana

2015 Bjørnar Sporsheim ph.d Biology

Quantitative confocal laser scanning microscopy: optimization of in vivo and in vitro analysis of intracellular transport

2015 Magni Olsen Kyrkjeeide ph.d Biology Genetic variation and structure in peatmosses (Sphagnum)

2015 Keshuai Li ph.d Biology

Phospholipids in Atlantic cod (Gadus morhua L.) larvae rearing: Incorporation of DHA in live feed and larval phospholipids and the metabolic capabilities of larvae for the de novo synthesis

2015 Ingvild Fladvad Størdal ph.d Biology The role of the copepod Calanus finmarchicus in affecting

the fate of marine oil spills

2016 Thomas Kvalnes ph.d Biology Evolution by natural selection in age-structured populationsin fluctuating environments

2016 Øystein Leiknes ph.d Biology The effect of nutrition on important life-history traits in themarine copepod Calanus finmarchicus

2016Johan Henrik Hårdensson Berntsen

ph.d BiologyIndividual variation in survival: The effect of incubation temperature on the rate of physiological ageing in a small passerine bird

2016 Marianne Opsahl Olufsen ph.d Biology

Multiple environmental stressors: Biological interactions between parameters of climate change and perfluorinated alkyl substances in fish

2016 Rebekka Varne Ph.d Biology Tracing the fate of escaped cod (Gadus morhua L.) in aNorwegian fjord system

2016 Anette Antonsen Fenstad Ph.d Biology

Pollutant Levels, Antioxidants and Potential Genotoxic Effects in Incubating Female Common Eiders (Somateria mollissima)

Integrated Environmental Mapping and Monitoring:

Documents

Transcript of Integrated Environmental Mapping and Monitoring:

POWER MAPPING

Environmental Monitoring Report INDIA: Karnataka Integrated ...

New integrated geodetic monitoring system at Stromboli volcano (Italy)

Integrated Algebra

Culture and Emotion in Crisis Communication: Indigenisation of the Integrated Crisis mapping (ICM) Model

Assessment and mapping of debris flow hazard through integrated physically based models and GIS assisted methods

Integrated Electroanatomic Mapping With Three-Dimensional Computed Tomographic Images for Real-Time Guided Ablations

DEVELOPMENT OF A HELICOPTER-BASED INTEGRATED SYSTEM FOR AVALANCHE MAPPING AND HAZARD MANAGEMENT

Mapping and monitoring carbon stocks with satellite observations: a comparison of methods

Integrated Resorts

Integrated planning

Spatial Assessment of Groundwater Quality Monitoring Wells Using Indicator Kriging and Risk Mapping, Amol-Babol Plain, Iran

Integrated road traflc noise mapping in urban Indian context

CASE MONITORING

TEMMS: an integrated package for modelling and mapping urban traffic emissions and air quality

Tools for monitoring the coverage of integrated public health ...

Rockwell Automation Integrated Condition Monitoring

New integrated geodetic monitoring system at Stromboli volcano (Italy

Project Monitoring

DIGISOIL: An Integrated System of Data Collection Technologies for Mapping Soil Properties