Agriculture, Ecosystems and Environment 142 (2011) 18– 28

Contents lists available at ScienceDirect

Agriculture, Ecosystems and Environment

journa l h o me pa ge: www.elsev ier .com/ locate /agee

Comparison of scale and scaling issues in integrated land-use models forpolicy support

H. van Deldena,∗, J. van Vlieta, D.T. Rutledgeb, M.J. Kirkbyc

a Research Institute for Knowledge Systems, P.O. Box 463, 6200 AL Maastricht, The Netherlandsb Manaaki Whenua Landcare Research, Private Bag 3127, Hamilton, New Zealandc School of Geography, University of Leeds, Leeds LS2 9JT, United Kingdom

a r t i c l e i n f o

Article history:Received 23 November 2009Received in revised form 9 March 2011Accepted 14 March 2011Available online 8 April 2011

Keywords:Decision Support System (DSS)ScaleScalingModel integrationPolicy supportLand-use modelling

a b s t r a c t

Recently an increasing number of integrated land-use models have become available that support policymaking. Inevitably, their model components represent processes that act on different scales and thatuse different levels of detail to represent those processes. Therefore, it is a challenge to integrate themproperly. In this paper we analyse and compare scaling issues from four integrated models that areexplicitly spatial and dynamic. All have a strong agricultural component and are developed to supportpolicy making. From these examples we find that scaling issues in model integration typically involvetrade-offs among four factors: (1) the scale at which end users or policy makers require information; (2)the scale at which processes take place and the representation of those processes in a single model; (3)the way to integrate model components representing processes occurring at different scales; and (4) thelimitations posed by practical restrictions such as data limitations and computation speed. Furthermorewe conclude that the complexity of the model components and the spatial and temporal resolutionsapplied in the models are generally related to the size of the study area, while its thematic resolutionis mostly driven by user requirements. Finally we argue that more detail does not necessarily generatebetter results and might even give a false impression of the model’s accuracy.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Today’s world is increasingly more complex and changingrapidly. Numerous processes occurring at different spatial and tem-poral scales act and react upon each other, making it difficult tounderstand and assess the impact of interventions on the human-environment system (Cash et al., 2006). Nonetheless, planners andpolicy makers face the challenge of making decisions in this com-plex system. They are not only confronted with interventions intheir own sector, but must consider the impact of interventions inother sectors as well as a range of external factors not directly influ-enced by policy interventions, such as climate change and globalsocio-economic developments.

In the past decade several integrated models have been devel-oped with the aim to support planning and policy making in thefield of (agricultural) land use (WUR and MNP, 2007; Van Ittersumet al., 2008; Sieber et al., 2008; Van Delden et al., 2010). These mod-els have in common that they are interdisciplinary and are as suchable to assess the impact of policy interventions on a broad range ofsectors. However, as their model components represent processes

∗ Corresponding author. Tel.: +31 43 350 1750; fax: +31 43 350 1751.E-mail addresses: hedwigvandelden@gmail.com, hvdelden@riks.nl

(H. van Delden).

that act on different scales this poses challenges for their integra-tion, because the interaction between the various processes andscales at which they occur is often not straightforward (Van Deldenet al., 2011).

In this paper we analyse four integrated land-use models forpolicy support: the LUMOCAP Policy Support System (Van Deldenet al., 2010), WISE (Rutledge et al., 2008; Huser et al., 2009), theMedAction Policy Support System (Van Delden et al., 2007), and theDeSurvey Integrated Assessment Model (Van Delden et al., 2009)and compare the following scale and scaling issues in developingthese models:

• At what scale(s) does the model allow input and provide resultsto ensure relevance to policy-makers?

• How is a process represented in a model? What is the level ofdetail used to represent a process?

• How are models, representing processes at different scales, con-nected?

• What other limitations related to scale and scaling issues wereencountered in their development?

The paper is organised as follows: section two clarifies the con-cepts and definitions that are being used. Section three describesthe four integrated models and elaborates on the scaling issuesencountered and choices that were made during their develop-

0167-8809/$ – see front matter © 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.agee.2011.03.005

H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28 19

ment. Section four compares and discusses the approaches used.Section five draws conclusions and provides directions for furtherresearch.

2. Concepts and definitions

Scale and scaling issues have been dealt with in many differ-ent disciplines, among others hydrology (Kirkby, 1999), coastalhydraulics (De Vriend, 1991), ecology (Wilson et al., 1999) andsocial sciences (Gibson et al., 2000). In physics, as perhaps theextreme example, scientists have conceptualised time and spacescales ranging over more than 60 orders of magnitude, from the firstmoments of the big bang (10−43 s) to the present time (1017 s), andfrom electrons (10−30 kg) to the universe (1055 kg). It is clear that,although there are attempts to unify understanding across thesescales, the dominant theory in any context is very much constrainedby the inherent scale of the process of interest. Within the morerestricted range of bio-physical and socio-economic processes,dominant system behaviour changes at characteristic thresholds,many of which are associated with typical scales. For example,catchment dynamics are dominated by hillslope processes in smallareas without developed stream channels, and by the behaviour ofthose channels in areas large enough to support them (Smith andBretherton, 1972).

A sharp delimitation of scales is often not possible (Steinhardtand Volk, 2003). Moreover, the scale at which processes operatecannot be seen in isolation, as processes have an impact on or areaffected by processes at other scales as well. For example, Das-gupta (1997 in Gibson et al., 2000) argues that economics triesto explain ‘the various pathways through which millions of deci-sions made by individual human beings can give rise to emergentfeatures of communities and societies’, such as rate of inflation,national income, formation of prices and cultural values. On theother hand he points out that individual decisions at any particu-lar point in time are affected by these emergent features. Anotherexample can be found in land-use change, where impacts of climatechange (scales of decades to centuries) are intertwined with mar-ket price developments and changes in policies (scales of monthsto years).

We define scale as a characteristic dimension in either space ortime or both, of an observation or a process, following Jewitt andGörgens (2000). As already stated by Neef (1963) and describedby Steinhardt and Volk (2003), scale-specific approaches requirescale-specific investigation methods and result in scale-specificinformation and insights. Therefore modellers need to decide onthe representation of processes in a model and as such make deci-sions on the spatial and temporal extent, the spatial level(s) andthe hierarchy by which they are ordered, and the amount of detailincorporated. Levels refer to locations along a scale (Gibson et al.,2000) and detail relates to the spatial, temporal and thematic res-olution(s) and the complexity by which processes are represented.In this paper we define complexity as the number of variables, rela-tions and processes modelled (Van Delden et al., 2011), the spatialresolution as the grid size, the temporal resolution as the length ofthe time steps on which a model operates, and the thematic reso-lution as the number of classes in a categorical map (Castilla et al.,2009).

Linking models that represent processes occurring at vari-ous scales raises questions regarding the hierarchical nature ofthe relationships among them. Scaling is defined as the processof extrapolating or translating information across scales (Blöschland Sivapalan, 1995). Mechanisms through which this can bedone include top-down or a bottom-up approaches. Top-downapproaches are those where the coarser scale influences the finer.Bottom-up approaches are those where coarser-scale processesemerge from processes at the finer scales.

Current approaches to modelling land-use change at a highspatial resolution frequently use bottom-up approaches such asCellular Automata (CA) (RIKS, 2011), Activity-Based Modelling (VanVliet et al., 2011) or Agent Based Modelling (Robinson et al., 2007).Each of these approaches has its own scale issues, which havebeen defined, investigated and described in literature (see forexample Kirman, 1992 and Ménard and Marceau, 2005). Over thepast decade, integrated models linking various aspects of land-use change and allocation, have gained in importance. Examples ofmodels linking land use and economics are presented by Rutledgeet al. (2008), Van Ittersum et al. (2008) and Van Delden et al. (2008);examples of models linking land-use and bio-physical processes bySchulze (2000), Forsman et al. (2003) and Van Delden et al. (2007).From linking different models, new scaling issues arise, while theissues within the individual components remain present.

3. Four integrated models for policy support

For this study we selected four integrated models. All modelsincluded in the analysis are dynamic spatial simulation models thatallow for feedback mechanisms between their components. Timeis represented as discrete steps and interaction between differentmodel components can take place during each time step. They areimplemented within the Geonamica software platform for spatialmodelling and (geo)simulation (Hurkens et al., 2008) and availableas operational software packages. Characteristics of the four modelsare described below and summarized in Table 1.

3.1. LUMOCAP—Dynamic Land Use change MOdelling for CAPimpact assessment on the rural landscape

The LUMOCAP Policy Support System (PSS) is developed toassess the impact of the Common Agricultural Policy (CAP) (CEC,2009) on the land use and landscapes of the 27 countries of theEuropean Union (Van Delden et al., 2010). The system incorporatesmodels for agricultural economics, national and regional interac-tion of population and jobs, land-use allocation, crop choice andsuitability. It uses scenarios for climate change, socio-economicdevelopments and policy alternatives as external drivers. It encom-passes processes operating on four spatial scales: EU 27, national,regional and local, reflected by the four spatial levels of the model(Fig. 1). The LUMOCAP PSS models processes at the local level at twodifferent spatial resolutions: a 1 km resolution for the entire Euro-pean Union and a 200 m resolution for specific case regions. Thetemporal resolution of all models is one year and the time horizonof the system is 2030.

Fig. 1. System diagram of the LUMOCAP PSS. Arrows represent the flows of infor-mation between model components.

20 H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28

Table 1Scale aspects of the various integrated models. Thematic resolution and model complexity are characterized relative to the other models in this paper.

Integrated model Spatial extent Spatial levels Spatial resolutionat local level

Temporal extent Temporalresolution

Thematicresolution

Model complexity

LUMOCAP PSS EU 27(±4,500,000 km2)

Europe, national,regional, local

1000 m 30 years(2000–2030)

Year High foragricultural landuse, low for otherland use classes

Low: focus onagriculture, top-downprocess for agriculturalland use, some detail inprocess representation

WISE ISDSS Region (±25,000 km2) Regional, local 100 m, 200 m,500 m

40–50 years(2006–2050)

Year Medium for allland use classes

Low: large number ofdisciplines, no specificemphasis onagricultural land use,simple processrepresentation

MedAction PSS River basins(300–6000 km2)

Regional, local 100 m 25–30 years(2000–2030 and2004–2030)

Rainfall storm,day, month,year

High foragricultural andnatural land use,low for other landuse classes

High: large number ofdisciplines, detailedprocess representation,bottom-up process foragricultural land use

DeSurvey IAM Countries(50,000–500,000 km2)

National, regional,local

1000 m 30 years(2000–2030)

Month, year Medium foragricultural landuse, low for otherland use classes

Medium: large numberof disciplines,top-down process foragricultural land use,some detail in processrepresentation

The LUMOCAP PSS aims to support policy makers from theEuropean Commission, in particular the Directorate-General forAgriculture and Rural Development. Therefore it needs to provideinformation at the European scale with a focus on the agriculturalsector and impact assessment of EU (agricultural) policies on theeconomy and the environment. These policies affect the decisionsof individual farmers as well as the macro economy. This is reflectedby the inclusion of an agricultural economic model at EU level and acrop choice component with a high number of agricultural classesat local level. The LUMOCAP PSS therefore allows entering policyalternatives at the different levels incorporated in the system, suchas intervention prices at the EU level and the Less Favoured Areapolicy (EC, 1999) at the local level.

The LUMOCAP PSS uses a top-down approach to allocate agri-cultural activities. The total demand for a particular agriculturalactivity is computed at the highest spatial level of the applica-tion (i.e. the EU) by an (agricultural) economic model. Given thelarge geographical extent of the LUMOCAP PSS (EU 27) and thecultural, political and socio-economic differences in the differentEU Member States, three linked agricultural economic models areapplied for three large regions in Europe covering several countrieseach. To downscale the land-use demands per crop type as wellas figures for population and jobs from socio-economic scenarios,national and regional interaction models have been applied thatdistribute the EU and national totals to Member States and regions,respectively, based on their relative attractiveness for differentagricultural activities. Factors that contribute to the attractivenessare a combination of bio-physical, social and financial elements: (1)land suitability that serves as a proxy for yield, (2) the number oflarge farms to account for economies of scale for relevant crops (e.g.cereals), (3) jobs in agriculture, (4) relative advantages related tothe way policies (subsidies) are implemented and (5) tradition orexisting practices. Within the NUTS 2 regions1, the Metronamicaland-use model (RIKS, 2011) allocates area demands for generalland-use categories to cells of 1 × 1 km. A separate crop choicemodel then distributes the total area for each crop type over thelocations with agricultural land-use in each NUTS 2 region, basedon existing practices, the suitability of the location and the locally

1 NUTS stands for Nomenclature of Territorial Units for Statistics in the EuropeanUnion; NUTS 2 regions are administrative regions that typically match provinces.

applied policies. The suitability of locations is calculated at locallevel (impacted by global climate change) and used in both theland-use and crop-choice model at local level and in the calculationof the agricultural production in the models operating at regional,national and EU levels.

In the LUMOCAP PSS, the land-use allocation model has theoption to run at a 200 m resolution within the wider context ofthe 1 km resolution used for the rest of the EU. To do so, thismodel is applied four times, once at the coarser and three timesat the finer resolution, while their results are boundary conditionsto each other. The interaction between the two resolutions takesplace through the neighbourhood effect, which specifies the influ-ence of neighbouring land uses on the allocation of other land uses.To enable a seamless integration, small cells are aggregated to thecoarser resolution using a majority aggregation, while larger cellsare disaggregated to the finer resolution for the computation of theneighbourhood effect. For the finer resolution, a 1000 meter cell isconsidered as 25 cells of 200 meter each with the same land useas the large cell. A more elaborate discussion on this integration ispresented in Van Vliet et al. (2009).

3.2. WISE—Waikato Integrated Scenario Explorer

The Waikato Integrated Scenario Explorer (WISE) is an inte-grated spatial decision support system (ISDSS) that aims to supportlong-term integrated policy development and planning in theWaikato region in New Zealand by taking into account cultural,social, environmental and economic well-being (Rutledge et al.,2008; Huser et al., 2009). The system incorporates models for eco-logical economics, population, water quality, hydrology, land-useallocation and biodiversity, which are represented at four spa-tial levels: regional, district (i.e. sub-regional level), sub-catchmentand local. For the comparison to the other approaches these lev-els have been combined into two spatial levels in Fig. 2. Drivers inthe integrated model are climate scenarios, socio-economic drivers(e.g. fertility, mortality and migration rates, exports and consump-tion patterns) and policy alternatives (zoning regulations, impactrestrictions and construction of infrastructure). The temporal reso-lution of all models is one year and the time horizon of the systemis 2050. The spatial resolution at local level varies between 500 m(hydrology), 200 m (land-use allocation) and 100 m (biodiversity).

H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28 21

Fig. 2. System diagram of WISE. Arrows represent the flows of information betweenmodel components.

As WISE is intended to support the Waikato Regional Council,named Environment Waikato, the scale of interest from a user per-spective is regional. Because of the administrative structure, therewas an explicit request to include the district level, governed by thedistrict councils. The system was intended to be used in the devel-opment of the Long-Term Council Community Plan (EnvironmentWaikato, 2009) and information at a high spatial resolution wasvalued to show the spatial implications of autonomous develop-ments and policy alternatives to local communities. On the otherhand, simulation times in the order of minutes were requested forits use in participatory sessions.

The Waikato region is a rural region and issues related to thedevelopments within this region, such as urban sprawl, an expand-ing agricultural sector and an appreciation to regain indigenousvegetation, cannot be related to a single land-use type. This isreflected in the selection of land-use categories as both urban, agri-cultural and natural land is subdivided into 3–5 sub-classes, eachwith their own behaviour.

The Waikato region in New Zealand has a prosperous dairy-ing sector, which expects a continued increase in demand due toan increase in global consumption, especially as a result of shift-ing dietary habits in many developing countries. Conversion ofland to dairy farming will likely meet some of that demand. WISEacknowledges the important interaction between regional eco-nomic development and scarcity of land. The economic model thatoperates regionally (i.e. the highest spatial level) generates demandfor land used by economic activities, while an age-cohort modeloperating at district level generates demand for the residentialland; conversely the local land-use allocation model, Metronamica,determines the actual land supply. This model simulates a compe-tition for space in which the strongest actors (represented by landuses) will occupy the best locations for their specific use. If actorsare not strong enough in economic or political terms, or if suitablelocations are lacking, land will not be allocated to the particularuse. This generates a negative feedback to the economic model,which adjusts the actual output to the supply side for the affectedeconomic sectors as well as any related economic sectors throughindirect effects via an economic input–output table.Results fromthe land-use model at local level are subsequently used in the ter-restrial biodiversity component, the hydrological model and thewater quality model. The biodiversity component in WISE calcu-lates at a finer resolution (100 m) than the land-use model (200 m).No special conversion between the two resolutions was requiredbecause a 200 meter cell is simply considered as 4 cells of 100meters each with the same land use.

The hydrology model in WISE maps the land-use classes toa canopy storage capacity using a lookup table. As the hydrol-ogy model calculates at a coarser resolution than the land-usemodel, one hydrology cell will cover several land-use cells. For

each hydrology cell, an average of the canopy storage capacity forthe corresponding land-use cells is calculated. In this calculation,the land-use cells are weighted according to the area within eachhydrology cell, to account for the fact that the grid cells at the tworesolutions do not line up exactly.

The water quality model calculates flow rates and nutrient (Nand P) loadings per stream reach, using land-use areas per sub-catchment. As such, WISE incorporates two sub-regional units:districts and sub-catchments. District boundaries are selected insuch a way that cells at local level are either completely within orcompletely outside the region. The water quality model uses poly-gons to provide the sub-catchment boundaries. These polygons cancross local cells and in these circumstances only the area of a cellthat fits within the region is included. This choice was made becauseinformation from the socio-economic models needed to be disag-gregated to the local level and the information from the land-usemodel to the water quality model could be aggregated.

While developing WISE there was the option to use a represen-tation of land environments for reporting on terrestrial biodiversityusing either 100 classes at 100 m resolution or 500 classes at 25 mresolution. Although the latter option was preferable for its detail,the former was chosen nonetheless to increase processing speed.

3.3. MedAction—Mediterranean Action

The MedAction PSS supports regional development to mitigatedesertification, focusing on sustainable farming, water resourcesand land degradation in arid and semi-arid regions (Van Deldenet al., 2007). The system consists of several sub-models that areintegrated in a single system (Fig. 3): a stochastic weather gen-erator, process models for hydrology, plant growth, salinisation,erosion and sedimentation, a rule-based model for transitionsin natural vegetation type groups, a utility-based and farm-levelmodel for crop choice including financial, bio-physical and socialcharacteristics, a land-use allocation model, and a land manage-ment component for irrigation, terracing and planning of cropsowing, ploughing and harvesting. The local level is representedby 100 meter grid cells; the land-use allocation model includesthe coarsest temporal resolution and calculates on a yearly basis,while the hydrology module has a variable time step in the order ofminutes determined by the intensity and duration of the showersand includes thus the most detailed temporal resolution. The time

Fig. 3. System diagram of the MedAction PSS. Arrows represent the flows of infor-mation between model components.

22 H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28

horizon of the system is 2030. External drivers include climate sce-narios and land demand from international and national economicand demographic developments. The system includes a wide rangeof policy options, among others subsidies, taxes, zoning regula-tions, reforestation, decisions on water use and extraction, waterpricing and construction of infrastructure. Impacts are assessed bymeans of a range of policy-relevant indicators that are aggregatedinto three headline indicators: water shortage, environmentally-sensitive areas and long-term agricultural profits.

Climate change and management of agricultural and naturalland were perceived as crucial drivers for desertification during dis-cussions with policy makers. This required incorporation of impactsof global climate change together with local management practicesin one model. Policies found relevant included EU policies such asthe CAP as well as local policies related to restrictions in using nat-ural resources and stimulation measures to change the behaviourof farmers, inhabitants, tourists and other actors.

To meet user requests, the system needed to include a highlevel of spatial and thematic detail. A main challenge identified wasthe simulation of land-use change at a high thematic resolution.Because the processes driving different types of land-use changediffer, we were unable to identify one model that was able to rep-resent all processes. The MedAction PSS therefore uses a two-stepapproach for allocating land uses to cells. In the first step, the gen-eral land-use classes such as residential land use, industry, tourism,agriculture, and natural vegetation, are allocated using Metronam-ica. In the second step, a utility-based crop choice model allocatescrop types within locations with agricultural land use. Similarly,a rule-based model determines natural vegetation types for thenatural vegetation class. The combined result is a highly detailedland-use map, featuring some 25 different land-use classes.

To represent the behaviour of farmers in such a way that sim-ulates their actual decision-making process, farmers’ decisions aremodelled from the bottom up: For each grid cell a farmer decidesif he continues farming at that location, what crop he grows andhow the land is managed. Choices depend on changing bio-physical,social, economic and institutional conditions. A stochastic utilityfunction is used to simulate crop choice. It considers the expectedprofit of the current crop and its possible substitutes, the farm-ing history, the anticipation of future profits, and social conditionsthat constrain farmers’ choices. If farming no longer appears prof-itable, plots are abandoned and become available for other landuses. Similarly, expanding agricultural activity may displace naturalland.

3.4. DeSurvey Integrated Assessment Model

The DeSurvey Integrated Assessment Model (IAM) (Van Deldenet al., 2009) is the successor of the MedAction PSS. It is a mod-elling framework that incorporates a large number of models withdifferent levels of complexity operating at different temporal andspatial resolutions. Within the framework coarse and fine scaleapplications can be developed depending on the area extent to bemodelled, the data availability and the (policy) questions/issues tobe supported. Whenever this paper refers to the DeSurvey IAM, itrefers to the coarse scale country-wide applications and in partic-ular the one to Spain, for which the system diagram is shown inFig. 4. The DeSurvey IAM includes models for macro-economics,regional interaction of population and jobs, allocation of land useand activities, crop choice, suitability, vegetation growth, hydrol-ogy, and erosion, which are represented at three spatial levels:country, regional and local. External drivers include climate change,demographics and economic drivers. Policy options include subsi-dies, zoning regulations, restrictions on water use and constructionof infrastructure. The DeSurvey IAM uses a 1 km resolution at thelocal level. The temporal resolution of the bio-physical components

Fig. 4. System diagram of the DeSurvey IAM. Arrows represent the flows of infor-mation between model components.

is monthly, while the land-use components operate yearly. Theintegrated model simulates changes from 2000 until 2030.

Coarse scale applications of the DeSurvey IAM are intended fornational administrations. User workshops clarified the importanceto link to the UNCCD’s National Action Programmes and to pinpointhot spots for desertification under various future conditions. Fur-thermore rural depopulation and urban sprawl were identified ascrucial problems in preserving the countryside. Both processes arethe result of incremental changes in activity on a location ratherthan sudden changes in the land use. In the development of thesystem, we needed to ensure an appropriate representation of thisprocess.

In the DeSurvey IAM, two versions of the land-use model canbe applied. The first option is the original version of Metronam-ica, which simulates location changes in discrete land-use classes.The second option is the Activity-Based version of Metronamica(Van Vliet et al., 2011), which provides information on activities(people, jobs) allocated to a cell in addition to the predominantland-use class and thus further information for policy making. Forexample knowing the number of people residing in areas prone toflooding is more relevant for risk management than merely stat-ing that there is a flood risk at a particular residential location.Moreover, the Activity-Based model provides information on ruraldepopulation which simulates the required processes related torural depopulation, land abandonment and desertification. Thesepopulation trends would remain undetected when a model onlyincludes land use classes.

Another important scaling issue that needed to be solved in thedevelopment of the DeSurvey IAM as well as the MedAction PSSdescribed in Section 3.3, was the interaction between fast (bio-physical) processes such as erosion and soil moisture changes andslow (socio-economic) processes related to land use change andcrop choice. In both systems this is solved by introducing the con-cept of dynamic suitability. Dynamic suitability maps representthe aptness of the location for the different plant types. They areupdated annually on the basis of results from the bio-physical mod-els. They are input again to the land-use and crop-choice models.Response curves for the different plant types, describing the rela-tion between plant growth and the physical variables, are appliedto transform the model output into suitability values (Fig. 5).

The left-most column shows the output maps of the physicalmodels considered as vital factors for plant growth, comprising of

H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28 23

Fig. 5. Dynamic suitability maps linking bio-physical models with models for land use change. Plant-specific suitability response curves (2) are applied to factor maps (1),which are the output of the physical models, and transformed into suitability maps per factor (3). The latter are combined into a suitability map (4) per crop type using aminimisation function.

soil moisture, soil salinity, fertile soil depth, and temperature. Thesecond column shows the plant-specific response curves for thesefactors, which have been obtained from literature and through cal-ibration. They return a suitability value (y-axis) for a specific planttype as a function of a factor value (x-axis). The third column showsthe result of this interpretation of bio-physical factors, which isthen combined into a single map assuming that the factor with thelowest suitability limits plant growth. Because all physical modelsrun on a sub-annual time step, model output is aggregated on anannual basis. For soil salinity and fertile soil depth, the latest valuesare used because they represent the current state of those factors.For soil moisture and temperature, the average is taken over thegrowth period for that specific plant type.

4. Comparing different approaches for dealing with scaleand scaling issues

As can be seen from the description above, the four integratedsystems differ in the range of spatial and temporal scales for whichprocesses are included, the levels at which models represent theseprocesses, and the amount of detail in their representation. Thissection compares how different scale and scaling issues have beendealt with in the different applications and discusses the reasonsfor these decisions as well as their benefits and limitations.

4.1. Policy relevance

Design choices for all four systems started with the selection ofthe scales relevant for the intended users. The LUMOCAP PSS aimsto support policy making on a European level, the DeSurvey IAM atnational level, and WISE and the MedAction PSS at regional level.From the four examples we can find a relation between the scale ofthe entire application and the resolution selected for the local level.This resolution is partly reflected by the available data, but mostlyby the interest of the users. For policy makers at EU and nationallevel a coarse resolution highlighting the major developments wasfound sufficient, while for regional policy makers more spatial andtemporal detail was required to see the impact of targeted spatialpolicy alternatives.

Compared to the other applications, the DeSurvey IAM offers theuser more detailed information on activities related to land use. Thisresulted in a better representation of the rural depopulation pro-

cess, but also in additional uncertainty, because initial maps withactivities needed to be created as a starting point for the simulation.

The selection of land-use classes was to a large extent basedon user requests. Depending on the policy context detail was pro-vided in the agricultural, natural and/or residential categories. Thefocus on particular land-use types is also reflected in the selectionof models. The LUMOCAP PSS has a strong focus on the agricul-tural sector and hence this sector is modelled with a dedicatedagricultural economic model at EU level and a crop choice modelat local level, while the MedAction PSS includes models for natu-ral vegetation dynamics and crop choice to capture desertificationprocesses. Furthermore models have been selected and adaptedto allow calculating the impact of policy alternatives on policy-relevant indicators. This meant that policy interventions had to beincluded at appropriate spatial levels of the model and similarlythat output was presented in a way relevant to the policy context.For the DeSurvey IAM there was a user request to point out hot spotsfor desertification in a map at local level and in both the DeSurveyIAM and the LUMOCAP PSS local information had to be aggregatedto regional, national and European indicators to match the interestof the national and European policy users.

High computational capacity requirements and long runningspeeds are often seen as factors limiting the use of integrated mod-els. The time required to run the four applications ranges fromminutes (WISE) to hours (LUMOCAP PSS) and mainly depends onthe size of the area and the complexity of the processes calculated atlocal level. Because WISE is intended to be used in participatory ses-sions, model complexity and temporal and spatial resolutions wereselected to allow for simulating a 30–50 year period within a fewminutes. Although a coarser thematic resolution would also havecontributed to an increased running speed, this was not seen as adesired solution by the users. The other applications include an areaextent that is very large (LUMOCAP PSS) or complex process calcu-lations (MedAction PSS and to a lesser degree the DeSurvey IAM)making the systems unsuitable for calculating various scenarioswithin a workshop session.

4.2. Representation of processes in models

As a rule of thumb a process was represented at the scale andresolution that reflects the scope of the model and at which resultscan be used meaningfully. When users require information at a

24 H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28

coarse scale, models can often be less complex, e.g. there is noneed to model demographics per region if only national informa-tion is desired. On the other hand, results at a coarser scale are oftencaused by actions and processes on a finer scale. In such cases pro-cesses need to be modelled with a certain level of detail to simulatethe system’s behaviour correctly. For example land-use patternsare typically generated from local interaction and hence known tobe path dependent (Brown et al., 2005). This requires a modellingapproach that uses fine spatial and temporal resolutions even whenthe result is only required at a coarser level.

4.2.1. Spatial extent and resolutionFor all models the spatial resolution relates to the size of the total

study area. The MedAction PSS and WISE are applied to a relativelysmall region, and represented on a 100 meter and 200 meter gridfor their respective land uses. The DeSurvey IAM and the LUMOCAPPSS, which cover complete countries and EU 27, respectively, usecell sizes of 1 km2. During the development of the latter a discussiontook place on the interpretation of cell states as the predominantland use at a given location. While it is reasonable to assign onecrop type to a 1 ha cell, such as in the MedAction PSS, for a 1 km2

cell the land use is inherently more often a mix of crop types or landuses. In the LUMOCAP PSS agricultural cells are therefore occupiedby a number of crop types, their shares adding up to 1 km2.

We also found a relation between the spatial resolution andthe detail in the processes represented. The DeSurvey IAM and theMedAction PSS include models for similar bio-physical processes,but the scale of the application asks for a different representation.The PATTERN model (Mulligan, 1996) underlying the MedActionPSS, was originally designed as a fine-scale model with a detailedprocess representation and a spatial resolution of 1–10 m andmodified to work at the 100 m resolution (Mulligan, 2004). Thismodification was required for the MedAction PSS to support plan-ning and policy making at river basin level (up to 6000 km2). ThePESERA model (Kirkby et al., 2008) that underlies the DeSurveyIAM was originally designed to model soil erosion across Europe ata resolution of 1 km2, which is sufficient for national level policysupport.

On the regional and national level, space is also partitioned,although not in grid cells. Except for the MedAction PSS all mod-els subdivide their total area into regions. WISE, which covers initself one region of New Zealand, is further subdivided in districts tosimulate population dynamics. The LUMOCAP PSS at the other handuses two subdivisions: national and regional. These two levels wereneeded to simulate population and job dynamics, because theseprocesses take place on two levels: migration between nations, andwithin each nation. Hence the size of the region determines thetype of process (international migration, versus local migration)that can be simulated. The LUMOCAP PSS, the DeSurvey IAM andWISE use administrative boundaries for modelling socio-economicprocesses. WISE includes a subdivision in sub-catchments to betterrepresent the regions relevant for hydrology, in particular to simu-late run-off processes. The entire area modelled in the MedActionPSS is also a sub-catchment, which has been selected for exactlythe same reason. Because the MedAction PSS does not work withadministrative regions all relevant socio-economic data had to beconverted to information at catchment scale.

4.2.2. Temporal extent and resolutionAll applications have a temporal extent of 25–50 years, which

was selected because of their relevance for policy support. They usea temporal resolution of one year for simulating land-use changesbecause this represents the scale at which land-use dynamics takeplace (or so we perceive it). However, some bio-physical pro-cesses require a finer temporal resolution to reflect seasonal orevent-driven changes. A guiding principle has been to select those

resolutions that best reflect the real-world processes. Wheneverdata or computational restrictions would limit this approach, infor-mation was upscaled to resolutions at which these problems wereovercome. For this reason we selected models with different tem-poral resolutions for incorporation in the MedAction PSS and theDeSurvey IAM. In the MedAction PSS bio-physical models run twicea day: for the day-time period and the night-time period. Whenrainfall storms occur, the hydrology, salinisation and erosion mod-els start to operate on a minute basis for the duration of the rainfallevent. This allows for taking into account the impact of intenserainfall with a short duration on soil sealing, run-off and erosionprocesses (Mulligan, 2004; Van Delden et al., 2007). The coarserDeSurvey IAM uses monthly averages for calculating the wateravailability, but also uses the monthly extreme rainfall event toensure incorporation of rainfall storms for a correct calculation oferosion.

4.2.3. Thematic resolutionThe level of detail in the different applications was very much

decided in anticipation of their expected use. This resulted in alarge number of agricultural land-use classes in all applications.Additionally, WISE subdivides the residential classes to simulatepotential impacts of different spatial planning strategies, such ascompact growth or urban sprawl. Similarly the MedAction PSSincludes more detail in natural vegetation classes to answer ques-tions relevant to desertification.

Because each individual model is developed for a specific pur-pose it might not automatically be suited for application beyondthis purpose, which poses the question if a combination of mod-els is required. All four models include Metronamica, a constrainedcellular automata model that simulates land-use allocation (RIKS,2011). However, this model is less suited for modelling transitionswithin agricultural and natural vegetation classes. Therefore theMedAction PSS, the LUMOCAP PSS and the DeSurvey IAM use dedi-cated models to allocate crop types within agricultural areas and inthe case of the MedAction PSS also a dedicated model for simulatingdynamics in natural vegetation.

Adding extra model components to provide extra informationhas added value when data and process information on the specificsof the sector are available. However, without this extra informationa simplified approach might be preferred, to avoid a false senseof accuracy, as detail is sometimes confused for accuracy. Hence,the decision on the amount of detail should be weighted againstuser requirements, data availability and reliability/robustness ofindividual stand-alone models.

4.2.4. Model complexityModel complexity increases with the number of variables and

processes incorporated and by linking various model components.More complex models may generate unnecessary detail, while sim-ple models may omit essential processes. It is not enough to applyOccam’s razor to just one model component, as the requirementsalso apply to the integrating linkages between components (Clarke,2004).

In most approaches, integrating additional model componentsprovides extra information and a better representation of thereal-world interactions between different processes and scales byrepresenting feedback loops. A clear example of this is the Activity-Based version of Metronamica that provides additional informationon jobs and population at the cellular level and is – by includingthis information in the local dynamics – better able to representcertain land use change processes such as rural depopulation. Rais-ing the number of processes included in a model or linking modelstypically increases uncertainty in the results and special cautionshould be paid to the risk of error propagation (Van der Sluijs et al.,2005; Jakeman et al., 2006; Van Delden et al., 2011). However, cou-

H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28 25

pling models can also provide more accurate results because of theimproved representation of the process. This is the case in WISEby linking an economic model and a land-use allocation model.Because of the incorporation of limited resources, particularly land,the economic model provides more realistic results, as it incorpo-rates not only the demand side, but also the supply side. Withoutsimulating the competition for space, the supply of available loca-tions would have to be estimated or guessed and would thereforebe less accurate.

The DeSurvey IAM and the MedAction PSS include models forsimilar bio-physical processes, although their complexity differsbased on the scale of the application, the available data and therequired running speed. The assumptions that underlie the pro-cesses that are modelled are also fundamentally different, whichresults in a different representation of the processes. PESERAassumes that each cell is drained by a stream or ditch, which limitsits finest applicable resolution. PATTERN on the other hand sim-ulates the flow of water and sediments from upstream cells todownstream cells, which allows e.g. for simulating the sedimenta-tion of reservoirs. Another example is the biomass growth, which iscalculated in PESERA as one value per location, while in PATTERN aplant is divided in various fractions (root, wood, live stem, leaf andyield). In the latter case biomass developments are calculated foreach of these fractions with various growing speeds based on nat-ural conditions and management practices. See Kirkby et al. (2009)for a more elaborate comparison of both models.

4.3. Model integration

Typically, integrated models consist of a number of pre-existingmodels originally designed to operate independently, i.e. as a stand-alone model. These models will inevitably vary in the detail andcomplexity with which they represent processes, as some modelsaim for detailed analysis and others for broad investigation, so thatthere is rarely a simple or unique choice of scale. Linking mod-els across scales introduces scaling issues as has been pointed outin section 2. Here we discuss the various scaling issues that haveoccurred in the four applications, the solutions chosen and theirbenefits and limitations.

4.3.1. Linking models operating on different temporal resolutionsThe MedAction PSS and the DeSurvey IAM model socio-

economic processes and land-use allocation annually, while thebio-physical models operate at finer temporal resolutions. In bothapplications information from these detailed temporal resolutionsis integrated into annual maps representing the suitability forgrowing certain plant types. Based on the type of information a dif-ferent integration mechanism is used. For aspects that contributeto or limit the growth of plants during certain periods of the year(e.g. the soil moisture during important growing stages or the tem-perature in the blossoming period) the average is taken over theseimportant periods. Because suitability maps are used in the cropchoice and land-use models for long-term decisions, we use a threeto five year average of these values. For aspects that influence thecurrent condition of the location, such as the fertile soil depth orthe salinity, we take the latest values available.

Additionally, the timescales for which individual models aresuited can also vary to a large degree. In our applications, mosteconomic models have a time horizon of a few years to a decade,while a number of bio-physical models have a time horizon of sev-eral decades. At present many integrated models for policy support,including those described in this paper, use intermediate time hori-zons relevant for planning (e.g. 25–50 years into the future) thatstretch the limits of the economic models but are rather short-sighted for some important bio-physical processes.

4.3.2. Selection of spatial levelsThe LUMOCAP PSS, the DeSurvey IAM and WISE include models

that represent processes operating at different spatial scales, whichare reflected by models operating at various spatial levels. In eachof these models, a decision needed to be made on the number ofspatial levels that needed to be included. Although this decision isoften made based on the understanding of the processes, the userneeds and the data available, in the DeSurvey IAM we tested theimpact of including or omitting a regional model on the calibra-tion results. We found that using an approach that disaggregatednational figures for population and jobs via a regional interactionmodel to local level gave much better results than a direct alloca-tion of national figures to the local level (RIKS, 2010). This can beexplained by the interaction between cities and regions that playsan important role in the decision in what parts of the country peo-ple will live and work. Local dynamics were not able to capturethese migration processes in a sufficient manner.

In WISE users preferred to model population dynamics andmacro economics on a district level. However due to time anddata limitations only population was modelled at the district level,while macro-economics was modelled at the regional level. Inte-gration of population into the macro-economic model was simplydone by using the sum of the people in all districts. However,using economic information in the migration between districts wastherefore not possible, because this information was only availableat regional level. For future developments it was recommended tosimulate processes at the same level to allow for mutual interaction.

4.3.3. Bottom-up versus top-down approachesTop-down approaches are found in all four systems. For exam-

ple, all systems have land-use demands defined or modelled atthe European, national or regional level, while these demands areallocated locally. At present many integrated models disaggregateinformation from higher to lower spatial levels (e.g. Eururalis, WURand MNP, 2007 and CAPRI-Spat, Britz and Mittenzwei, 2008).

However, not many models allow for a feedback from local levelto higher spatial levels. The LUMOCAP PSS and WISE both includethis recursive relation. In the LUMOCAP PSS changing local condi-tions (impacting on yield) have an impact on agricultural economicsat EU level and agricultural economics have an impact on localland use by providing the area demand per crop type. In WISE theeconomic model is an important driver for land-use change in pro-viding land-use demand for a range of economic activities whilethe availability of suitable land can restrict growth for differenteconomic sectors.

Bottom-up approaches towards linking models are most pro-found in the MedAction PSS, where many decisions are taken at thelevel of the individual cell, representing a unit on which a farmermakes a decision. The combined effect of these decisions eventu-ally affects the result at the regional level. For example, each farmermakes decisions on their land management practices, which has animpact on the regional water balance and the regional agricultureeconomy. Furthermore, the farmer uses a utility function to deter-mine if the plot will stay in agricultural practice and if so, whatcrop will be grown on the plot. In case the land is abandoned, itreturns to natural vegetation and the natural vegetation model sim-ulates the transition in natural vegetation type groups based on thebio-physical characteristics of the location and the human-inducedinterventions, such as grazing. The impact of the land abandonmentis than reflected in the regional agricultural development.

The decision for modelling processes top down or bottom upmainly depended on the scale of the entire application and therelated data availability, resulting in a top-down allocation of landuse in the LUMOCAP PSS and a bottom-up approach of the sameprocess in the MedAction PSS. The most prominent problem expe-rienced in the bottom-up approach was the availability of spatially

26 H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28

explicit data on farms and especially data related to the socialaspects of farmers (Volk et al., 2010). Because of this problem weused a top-down approach for the other systems, which posesa conceptual question as crop choice decisions are made locally.From what level (EU, national, regional) a downscaling approachfor agricultural land use is still valid is a question that needs fur-ther attention. For the general land-use classes in the LUMOCAPPSS, calibration over the period 1990–2000 shows that a down-scaling approach from EU and national level generates good results(Van Delden et al., 2010). However, because the detailed crop typemap at local level was constructed based on information from dif-ferent data sources, a similar procedure could not be carried outfor the detailed crop types leaving the above-mentioned questionunanswered.

4.3.4. Linking different model conceptsIn the sections above, scaling issues experienced were the result

of integrating different process representations. However whenunderlying assumptions differ between models, the problem is alsoon the conceptual validity of this link. Many disciplines have theirown specific way to construct a model and linking them is notevident.

In developing coupled human-natural systems we often experi-ence discrepancies between (economic) models that are developedto compute an equilibrium state and (bio-physical) simulationmodels that simulate future development in subsequent time stepsbased on a number of drivers, never reaching equilibrium (VanDelden et al., 2011). Coupling these models is often technically pos-sible but conceptually not sound (Van Delden and McDonald, 2010).The macro-economic model in WISE requires that demand and sup-ply are in equilibrium for a single year. However, for the dynamicinteractions with the land-use model it was necessary to dividethe demand and supply calculations over two time steps. This isconceptually not ideal, but neither is the alternative of iteratingbetween the economic model and the land-use model within onetime step. Whatever option is chosen, there is always a violation ofone of the underlying modelling paradigms. In the LUMOCAP PSS,which also links an (agricultural) economic model with a dynamicland-use allocation model, this problem was not directly expe-rienced, because the agricultural model used is an econometricmodel that uses yearly time steps. However, because some parame-ters in the econometric model are derived from other – equilibriumbased – approaches, this method still requires careful investigationon its conceptual validity. Furthermore caution should be paid inusing trend extrapolations of historic data for scenario studies witha temporal horizon of 25–50 years into the future.

Another example where model concepts can be in conflict isin the integration of discrete models (such as cellular automataland-use change models) and models calculating a continuousdevelopment (such as growth of biomass or yield), which can cause(unwanted) shock effects in the integrated model. Whenever theseshocks represent events that occur in reality (such as the conver-sion of an agricultural area into a residential area or a flood orfire event that takes place), the integrated models are normallyable to deal with these transitions. However, when transitions aremostly artefacts of the modelling approach (e.g. in models simulat-ing transitions in natural vegetation type groups) such behaviouris undesirable (Van Delden et al., 2007, 2011).

4.4. Data limitations

Data is frequently incomplete, inconsistent or inaccurate, whichcan require interpolation or a coarsening of scales to match theavailable data (Kirkby, 1998, 1999, 2005). Therefore often modelsare selected or adapted based on data availability. An example ofthis is the crop choice model in the MedAction PSS that makes a

decision on a cellular level, rather than on the level of individualfarmers, because spatial information on the latter was not available.

Scaling issues regarding data were also found in the aggrega-tion of land-use information. The four integrated models describedabove all include a land-use model that requires spatially explicitland-use data. This data was often available at a finer resolutionthan required; therefore, we needed to aggregate land use infor-mation. The LUMOCAP PSS and the DeSurvey IAM used a majorityaggregation to obtain land-use information at the desired cell sizefor the general land-use classes. This method was chosen becauseland uses on the land-use maps reflect the predominant land use ona location, and this interpretation is violated in other aggregationmethods.

In the development of the DeSurvey IAM, the LUMOCAP PSSand the MedAction PSS we experienced difficulties in obtaining adetailed crop-type map that would match the information availableat regional level. For this reason local land-use data was processedby filling in the total agricultural area from the land-use map withregional crop-type statistics using various disaggregation tech-niques. Although this resulted in more consistency between theregional and local level, this also created additional uncertainty tothe local level land-use map.

All integrated models suffered from an incompatibility betweenstatistical data and map data as the statistical data on crop areasdid not match the classes on the land use maps. For example, it isnot clear what kind of crop types are included in a land use classnamed ‘heterogeneous areas’ and how one should deal with classesthat only appear in the statistical data but not on the land-use mapor vice versa. Furthermore, when classes did match, area totalsper crop type as provided by the statistical data for administra-tive regions were often very different from those computed froman aggregation of the land use maps. The latter aspect can be partlyexplained, because the land use maps show gross land area includ-ing farms and rural roads, while the statistical data uses net areabased on land parcels. During the development of the LUMOCAP PSSit was decided in discussion with users to use as much as possibleoriginal data sources. This meant that data of both sources was usedas initial data in the model. Conversion factors were derived fromthe differences found between the two sources and applied duringeach step of the model to downscale regional land-use demandsto local grid cells and to aggregate information on yield from thelocal to the regional level. The disaggregation of regional infor-mation posed serious problems, because insufficient informationwas available at local level to do this properly. The aggregation toregional level was less of a problem as the average yield per croptype could be computed from the local level. Limitations to this cal-culation were mainly imposed by the difference between gross andnet land as described above. This resulted in lower average yieldscalculated at local level than provided by the data at regional level.

5. Conclusions and recommendations

In this study we discussed scale and scaling issues that wereencountered in the development of four integrated land-use mod-els that have to aim to support policy impact assessment. Fromthis comparison we conclude that decisions made on scale andscaling issues are typically the result of a trade-off between thescale at which users – such as policy makers and planners – requireinformation, the scale at which a process can best be represented,the linkage between models that represent processes at differentscales, and practical constraints that data limitations and computa-tion speed are posing on the system. The applications we compareddiffer in their intended use partly by the size of their study area,as applications that cover EU 27 are inherently more appropriatefor analyzing large-scale developments, whereas small-scale appli-cations are more interesting for local planners or policy analysts

H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28 27

assessing the implications of spatially targeted policies or explor-ing the actual process of land-use change. We found that users oftenrequire information at a scale that is not necessarily the appropri-ate scale to model the underlying process. A typical example is theCommon Agricultural Policy, for which European policy makers areinterested in the impact on a national or continental scale, whilethis effect is essentially the result of the individual farmer’s reactionto this policy. Hence the underlying processes need to be consid-ered at finer scales than what is requested on the output side ofthe system and subsequently model results need to be upscaled tomeet user requests.

We also conclude that the complexity of the model componentsand the spatial and temporal resolutions applied in models is gen-erally related to size of the study area. Besides the computationaladvantage, this also hints that after a threshold is passed the repre-sentation of a process can be upscaled. On a European or nationallevel the behaviour of farmers can be approximated by their aggre-gate behaviour, while on a small-scale application, actors need tobe represented as such. However, not all processes can be modelledcorrectly with a simpler representation and we argue that implica-tions of upscaling processes should be investigated carefully beforeincluding them in (integrated) models. Remarkably, the thematicresolution did not seem to be dependent on the scale of the applica-tion, but more on the requirements from the user, as did the rangeof information provided by the model.

Some scale issues remain a challenge. First of all, the appropri-ate representation of a process is not always obvious. We arguethat testing the same model at different resolutions or trying outmore or less complex models representing the same process wouldcontribute to better decisions on the representation of processes.Furthermore, we conclude that more detail does not necessarilygenerate better results. Using more land-use classes, more com-plex process representations, or smaller grid cells does not implythat the underlying processes are represented more accuratelyand can even give a false impression of accuracy. Finally, as mod-els require data, the availability thereof can limit the scale atwhich a process is modelled. In some cases this requires data pro-cessing such as aggregation or reclassification. In other cases itsimply means that a process cannot be modelled at the appropri-ate scale. The latter is particularly the case for the smallest scales,because data on the level of plots, farms or individuals is seldomavailable.

When a selection of models and scales is made, the next chal-lenge is to connect them and deal with the scaling issues this mightbring about. We found that downscaling socio-economic informa-tion from a high spatial level to the local level often required severalintermediate steps to capture the processes appropriately. Oftenadministrative levels (countries, provinces) are selected in thisdownscaling approach, because of their match with user interestsand data availability. Not all processes can however be modelled ata high spatial level and still provide realistic information at locallevel after a downscaling procedure. It is questionable whetheragricultural land-use models operating at high spatial levels can beused for local land-use allocation and, if they can, how many inter-mediate steps should be taken and if the approach can be entirelybased on a disaggregation of model results or if it should includebottom-up approaches as well. Calibration and validation of sucha procedure using local crop type maps would be an appropriateway to evaluate this, but these maps are not currently available forlarge areas such as the EU.

As for the integration between various model components,we would like to stress that this is more than an upscaling ordownscaling of model results. It is a challenge technically, butmore important conceptually, because we need to deal with pro-cesses from several disciplines, operating at a range of scales andrepresented by various types of models of which the underly-

ing assumptions are sometimes at odds with each other. Mostintegrated models inherit components that have been developedfor another purpose. While this brings about the advantage ofreusability we suggest that in the early stages of the developmentof integrated models, sufficient time is spent on the conceptualchallenges related to the interactions that transcend disciplinaryboundaries and the related decisions on the scale and scalingaspects. Based on this, appropriate models can be sought or devel-oped, which will lead to a higher compatibility of the modelcomponents of the integrated model.

Ultimately, there is no panacea for solving scale and scalingissues. There is always a balancing act between the user, the sci-ence and the data. Finding appropriate ways of dealing with theseoften conflicting interests is a major challenge for future researchand practice of integrated modelling for policy support.

Acknowledgments

The New Zealand Foundation for Research, Science, and Tech-nology funds the Creating Futures project (ENVW0601) in whichWISE is being developed and the European Commission hasfunded the development of the LUMOCAP PSS (FP-6 SSPE-CT-2005-006556), the MedAction PSS (FP-5 EVK2-2000-22032) and theDeSurvey IAM (FP-6 IP 003950).

References

Blöschl, G., Sivapalan, M., 1995. Scale issues in hydrological modelling: a review.Hydrological Processes 9, 251–290.

Brown, D.G., Page, S., Riolo, R., Zellner, M., Rand, W., 2005. Path dependence andthe validation of agent-based spatial models of land use. International Journalof Geographical Information Systems 19 (2), 153–174.

Britz W., Mittenzwei, K., 2008. Spatial down-scaling as a tool to improve multifunc-tionality indicators in economic models. In: Findley, P.N. (Ed.), EnvironmentalModelling: New Research. Nova Publishers, ISBN: 978-1-60692-034-3.

Castilla, G., Larkin, K., Linke, J., Hay, G.J., 2009. The impact of thematic resolution onthe patch-mosaic model of natural landscapes. Landscape Ecology 24 (1), 15–23.

CEC, 2009. Communication from the Commission to the European Parliament andthe Council: A Simplified CAP for Europe–A Success For All. COM (2009) 128final of 18.3.2009. Brussels, Belgium.

Cash, D.W., Adger, W., Berkes, F., Garden, P., Lebel, L., Olsson, P., Pritchard, L., Young,O., 2006. Scale and cross-scale dynamics: governance and information in a mul-tilevel world. Ecology and Society 11 (2), 8.

Clarke, K.C., 2004. The limits of simplicity: towards geocomputational honesty inurban modelling. In: Atkinson, P., Foody, G., Darby, S., Wu, F. (Eds.), GeoDynam-ics. CRC Press, Boca Raton, FL, pp. 215–232.

De Vriend, H.J., 1991. Mathematical modelling and large-scale coastal behaviour.Journal of Hydraulic Research 29, 727–740.

EC, 1999. Council regulation 1259/1999 on establishing common rules for directsupport schemes under the common agricultural policy, 17.05.1999.

Environment Waikato, 2009. One Waikato, many communities: draft long-term council community plan 2009-2019. Acquired on 30.3.2009:http://www.ew.govt.nz.

Forsman, A., Grimvall, A., Scholtes, J., Wittgren, H.B., 2003. Generic structuresof decision support systems for evaluation of policy measures to reducecatchment-scale nitrogen fluxes. Physics and Chemistry of the Earth 28,589–598.

Gibson, C.C., Ostrom, E., Ahn, T.K., 2000. The concept of scale and the human dimen-sions of global change: a survey. Ecological Economics 32, 217–239.

Hurkens, J., Hahn, B.M., Van Delden, H., 2008. Using the Geonamica® softwareenvironment for integrated dynamic spatial modelling. In: Sànchez-Marrè, M.,Béjar, J., Comas, J., Rizzoli, A., Guariso, G. (Eds.), Proceedings of the iEMSsFourth Biennial Meeting: Integrating Sciences and Information Technology forEnvironmental Assessment and Decision Making. International EnvironmentalModelling and Software Society, Barcelona, Spain, pp. 751–758, ISBN: 978-84-7653-074-0.

Huser, B., Rutledge, D.T., Van Delden, H., Wedderburn, M.E., Cameron, M., Elliott, S.,Fenton, T., Hurkens, J., McBride, G., McDonald, G., O’Connor, M., Phyn, D., Poot,J., Price, R., Small, B., Tait, A., Vanhout, R., Woods, R.A., 2009. Creating futures:towards an integrated spatial decision support system for local government inNew Zealand. In: Anderssen, R.S., Braddock, R.D., Newham, L.T.H. (Eds.), Proceed-ings of the 18th World IMACS Congress and MODSIM09 International Congresson Modelling and Simulation. Modelling and Simulation Society of Australia andNew Zealand and International Association for Mathematics and Computers inSimulation, Cairns, Australia, pp. 2370–2376, ISBN: 978-0-9758400-7-8.

Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development andevaluation of environmental models. Environmental Modelling and Software 21,602–614.

28 H. van Delden et al. / Agriculture, Ecosystems and Environment 142 (2011) 18– 28

Jewitt, G.P.W., Görgens, A.H.M., 2000. Scale and Model Interfaces in the Context ofIntegrated Water Resources Management for Rivers of the Kruger National Park.Report 627/1/00. Water Research Commission, Pretoria, South Africa.

Kirkby, M.J., 1998. Landscape modelling at regional to continental scales. In: Her-garten, St., Neugebauer, J.J. (Eds.), Process Modelling and Landform Evolution.Springer Verlag, pp. 189–208.

Kirkby, M.J., 1999. Translating models from hillslope (1 ha) to catchment (1000km2) scales. In: Diekkrüger, B., Kirkby, M.J., Schröder, U. (Eds.), Regionalizationin Hydrology, vol. 254. IAHS Publication, pp. 1–12.

Kirkby, M.J., 2005. Theory, organization and scale. In: Anderson, M.G., McDonnell,J. (Eds.), Encyclopedia of Hydrological Sciences, Part I. John Wiley, Chichester,doi:10.1002/0470848944.hsa003.

Kirkby, M.J., Irvine, B.J., Jones, R.J.A., Govers, G., PESERA Team, 2008. The PESERAcoarse scale erosion model for Europe. I.—Model rationale and implementation.European Journal of Soil Science 59 (6), 1293–1306.

Kirkby, M.J., Irvine, B.J., Mulligan, M., 2009. Report on comparison of fine (basedon PATTERN model) and coarse scale (based on PESERA model) approachesto land condition modelling. A comparison based on application of the mod-els to the Guadalentín catchment, S.E. Spain. DeSurvey project, Deliverable1.4.1.3 (available on http://www.geog.leeds.ac.uk/research/groups/pesera; lastaccessed 08-03-2011).

Kirman, 1992. Whom or what does the representative individual represent? Journalof Economic Perspectives 6 (2), 117–136.

Ménard, A., Marceau, D.J., 2005. Exploration of spatial scale sensitivity in geographiccellular automata. Environment and Planning B 32, 693–714.

Mulligan, M., 1996. Modelling hydrology and vegetation change in a degraded semi-arid area. PhD Thesis. King’s College London, University of London.

Mulligan, M., 2004. MedAction: Development, testing and application of the climate,hydrology and vegetation components of a desertification policy support sys-tem. Final report for work undertaken as part of MedAction: policies to combatdesertification in the Northern Mediterranean region supported by the EC-DGXIIunder contract EVK2-2000-22032. Brussels, Belgium.

Neef, E., 1963. Dimensionen geographischer Betrachtungen. ForschungenFortschritte 37, 361–363.

RIKS, 2011. Metronamica–Model descriptions. RIKS, Maastricht (available onhttp://www.riks.nl/products/Metronamica).

RIKS, 2010. Report on the calibration and validation of the DeSurvey IntegratedAssessment Model (MP2) applications. DeSurvey project, Deliverable 1.2.3.11(available on http://www.riks.nl/projects/DeSurvey; last accessed 08-03-2011).

Robinson, D.T., Brown, D.G., Parker, D.C., Schreinemachers, P., Janssen, M.A., Huigen,M., Wittmer, H., Gotts, N., Promburom, P., Irwin, E., Berger, T., Gatzweiler, F.,Barnaud, C., 2007. Comparison of empirical methods for building agent-basedmodels in land use science. Journal of Land Use Science 2 (1), 31–55.

Rutledge, D.T., Cameron, M., Elliott, S., Fenton, T., Huser, B., McBride, G., McDon-ald, G., O’Connor, M., Phyn, D., Poot, J., Price, R., Scrimgeour, F., Small, B., Tait,A., Van Delden, H., Wedderburn, M.E., Woods, R.A., 2008. Choosing regionalfutures: challenges and choices in building integrated models to support long-term regional planning in New Zealand. Regional Science Policy & Practice 1 (1),85–108.

Schulze, R., 2000. Transcending scales of space and time in impact studies of climateand climate change on agrohydrological responses. Agriculture Ecosystems andEnvironment 82, 185–212.

Sieber, S., Müller, M., Verweij, P., Haraldsson, H., Fricke, K., Pacini, C., Tscherning, K.,Helming, K., Jansson, T., 2008. Transfer into decision support: the sustainabilityimpact assessment tool (SIAT). In: Helming, K., Perez-Soba, M., Tabbush, P. (Eds.),Sustainablility Impact Assessment of Land Use Changes. Springer-Verlag, BerlinHeidelberg, pp. 107–128, ISBN: 978-3-540-78647-4.

Smith, T.R., Bretherton, F.P., 1972. Stability and the conservation of mass in drainagebasin evolution. Water Resources Research 8 (6), 1506–1529.

Steinhardt, U., Volk, M., 2003. Meso-scale landscape analysis based on landscape bal-ance investigations: problems and hierarchical approaches for their resolution.Ecological Modelling 168, 251–265.

Van Delden, H., Luja, P., Engelen, G., 2007. Integration of multi-scale dynamic spatialmodels of socio-economic and physical processes for river basin management.Environmental Modelling and Software 22 (2), 223–238.

Van Delden, H., Gutiérrez, E., Van Vliet, J., Hurkens, J., 2008. Xplorah, A multi-scaleintegrated land use model. In: Sànchez-Marrè, M., Béjar, J., Comas, J., Rizzoli, A.,Guariso, G. (Eds.), Proceedings of the iEMSs Fourth Biennial Meeting: Integrat-ing Sciences and Information Technology for Environmental Assessment andDecision Making. International Environmental Modelling and Software Society,Barcelona, Spain, pp. 827–834, ISBN: 978-84-7653-074-0.

Van Delden, H., Kirkby, M.J., Hahn, B.M., 2009. Towards a modelling framework forintegrated assessment in arid and semi-arid regions. In: Anderssen, R.S., Brad-dock, R.D., Newham, L.T.H. (Eds.), Proceedings of the 18th World IMACS Congressand MODSIM09 International Congress on Modelling and Simulation. Modellingand Simulation Society of Australia and New Zealand and International Asso-ciation for Mathematics and Computers in Simulation, Cairns, Australia, pp.3563–3569, ISBN: 978-0-9758400-7-8.

Van Delden, H., McDonald, G., 2010. Towards the integration of economic and landuse change models. In: Swayne, D.A., Yang, W., Voinov, A.A., Rizzoli, A., Filatova,T. (Eds.), Proceedings of the iEMSs Fifth Biennial Meeting: Modelling for Envi-ronment’s Sake. International Environmental Modelling and Software Society,Ottawa, Canada, pp. 1540–1548, ISBN: 978-88-903574-1-1.

Van Delden, H., Stuczynski, T., Ciaian, P., Paracchini, M.L., Hurkens, J., Lopatka, A.,Shi, Y., Gomez Prieto, O., Calvo, S., Van Vliet, J., Vanhout, R., 2010. Integratedassessment of agricultural policies with dynamic land use change modelling.Ecological Modelling 221 (18), 2153–2166.

Van Delden, H., Seppelt, R., White, R., Jakeman, A.J., 2011. A methodology for thedesign and development of integrated models for policy support. EnvironmentalModelling and Software 26, 266–279.

Van der Sluijs, J.P., Craye, M., Funtowicz, S., Kloprogge, P., Ravetz, J., Risbey,J., 2005. Combining quantitative and qualitative measures of uncertainty inmodel-based environmental assessment: the NUSAP system. Risk Analysis 25,481–492.

Van Ittersum, M.K., Ewert, F., Heckelei, T., Wery, J., Alkan Olsson, J., Andersen, E.,Bezlepkina, I., Brouwer, F., Donatelli, M., Flichman, G., Olsson, L., Rizzoli, A.,Van der Wal, T., Wien, J.E., Wolf, J., 2008. Integrated assessment of agriculturalsystems—a component-based framework for the European Union (SEAMLESS).Agricultural Systems 96, 150–165.

Van Vliet, J., Navarro, C., Gutiérrez, E.R., Hurkens, J., Van Delden, H., 2009. XplorahMunicipio, a multi resolution spatial decision support system. In: Haunert, J.-H.,Kieler, B., Milde, J. (Eds.), Proceedings of the 12th AGILE International Confer-ence on Geographic Information Science. IKG, Leibnitz Universität, Hannover,Germany, ISBN: 2073-8013 (available on http://www.agile-online.org/).

Van Vliet, J., Hurkens, J., White, R., Van Delden, H. An activity based cellularautomaton model to simulate land use dynamics. Environment and PlanningB (doi:10.1068/b36015).

Volk, M., Lautenbach, S., Van Delden, H., Newham, L.T.H., Seppelt, R., 2010.How can we make progress with decision support systems in landscapeand river basin management? Lessons learned from a comparative analysisof four different decision support systems. Environmental Management 46,834–849.

Wilson, J., Low, B., Costanza, R., Ostrom, E., 1999. Scale misperceptions andthe spatial dynamics of a social-ecological system. Ecological Economics 31,243–257.

WUR, MNP, 2007. Eururalis 2.0 CDrom. Published by Alterra Wageningen UR.Wageningen, The Netherlands.