Quantification and site-specification of the support practicefactor when mapping soil erosion risk associated with oliveplantations in the Mediterranean island of Crete

Christos G. Karydas & Tijana Sekuloska &

Georgios N. Silleos

Received: 2 May 2007 /Accepted: 14 January 2008 / Published online: 5 March 2008# Springer Science + Business Media B.V. 2008

Abstract Due to inappropriate agricultural manage-ment practices, soil erosion is becoming one of themost dangerous forms of soil degradation in manyolive farming areas in the Mediterranean region,leading to significant decrease of soil fertility andyield. In order to prevent further soil degradation,proper measures are necessary to be locally imple-mented. In this perspective, an increase in the spatialaccuracy of remote sensing datasets and advancedimage analysis are significant tools necessary andefficient for mapping soil erosion risk on a fine scale.In this study, the Revised Universal Soil LossEquation (RUSLE) was implemented in the spatialdomain using GIS, while a very high resolutionsatellite image, namely a QuickBird image, was usedfor deriving cover management (C) and supportpractice (P) factors, in order to map the risk of soilerosion in Kolymvari, a typical olive farming area inthe island of Crete, Greece. The results comprised arisk map of soil erosion when P factor was takenuniform (conventional approach) and a risk map when

P factor was quantified site-specifically using object-oriented image analysis. The results showed that theQuickBird image was necessary in order to achievesite-specificity of the P factor and therefore to supportfine scale mapping of soil erosion risk in an olivecultivation area, such as the one of Kolymvari inCrete. Increasing the accuracy of the QB imageclassification will further improve the resulted soilerosion mapping.

Keywords Soil erosion risk . RUSLE . P factor .

QuickBird imagery . Object-oriented image analysis

Introduction

Soil erosion and olive cultivation

The Mediterranean region is particularly prone to soilerosion due to the fact that is subject to long dryperiods followed by heavy bursts of erosive rainfallon steep slopes with shallow and fragile soils.Although soil erosion is a natural phenomenon,human activities such as agriculture can accelerate itfurther. As an anthropogenic impact accelerated soilerosion in the Mediterranean regions is highlyassociated with intensification of olive culture, whichwas introduced by the EU Common AgriculturalPolicy as a support regime for olive oil productionsince the 1980s in Greece, Spain and Portugal andearlier in Italy. This regime resulted in the expansion

Environ Monit Assess (2009) 149:19–28DOI 10.1007/s10661-008-0179-8

DO00179; No of Pages

C. G. Karydas (*) : T. SekuloskaDepartment of Environmental Management,Mediterranean Agronomic Institute of Chania,P.O. Box 85, 73100 Chania, Greecee-mail: xkarydas@maich.gr

G. N. SilleosDepartment of Applied Informatics,University of Macedonia,Thessaloniki, Greece

of intensive plantations, marginalisation of low-inputfarms, and adaptation of inappropriate practices bythe olive farmers, thus causing serious land degrada-tion (Beaufoy 2000; Fig. 1). Farming practices can beappropriate for instance when establishing new ormaintaining existing terraces within or between oliveplantations in sloppy terrains. This measure togetherwith the existence of rural roads and paths perpen-dicular to the slope direction can contribute tosplitting up the slope length, thus slowing down therun-off process and improving water penetration(Beaufoy 2000; Gomez et al. 2003).

RUSLE model

In order to assess risk of soil erosion, various methodsmay be implemented distinguished as expert-based ormodel-based methods. The main difference betweenthem is that expert-methods allow only qualitativeassessment, while model-based methods focus onquantifying the risk of soil erosion. For bothapproaches, estimations are based on the factorialscoring technique (Morgan 1995), by which theprocesses and factors influencing the rate of erosionare ranked, thus providing a series of erosionindicators which then can be weighted accordingly(Gobin et al. 1999).

One of the most widely applied models for assessingsoil erosion is the Universal Soil Loss Equation(USLE), developed by Wischmeier and Smith in1978 (Morgan 1995). As an empirical model, USLE

takes into consideration several determining factorsand more specifically the soil erodibility factor, therainfall factor, the length-slope factor, the covermanagement factor and the support practice factor.USLE was developed mainly for soil erosion estima-tion in croplands or gently sloping topography. Theevolved Revised Universal Soil Loss Equation(RUSLE) follows the same formula as USLE, buthas got several improvements in the determiningfactors and a broader application to different situa-tions, including forests, rangelands and disturbedareas compared to USLE (Trojacek et al. 2004).RUSLE estimates soil loss from a hill-slope caused byraindrop impact and overland flow (commonly termed“inter-rill” erosion), plus rill erosion; it does notestimate gully or stream-channel erosion.

RUSLE is a non data-demanding model, thereforeit can be fed by data usually available in institutionaldatabases, such as low or medium spatial resolutionsatellite images, limited rainfall data (extrapolatedover a study area), or geologic maps; in the samesense, implementation of RUSLE is not very expen-sive as well. On the contrary, other models such as thePESERA model are highly data-demanding, thusrendering many applications non-feasible in practice.Moreover, the role of RUSLE is that of a conservationmanagement tool, where relative comparisons amongareas are more significant than precise assessments ofthe absolute soil loss in a particular location.

Mapping soil erosion risk

With regard to supporting technology in soil erosionrisk assessment, use of Geographic InformationSystems (GIS) can provide the means for calculatingand mapping the different indicators used, concludingthen to the final environmental risk. Lately, EarthObservation (EO) data of very high spatial resolution(VHR) and advanced image processing techniques,such as Object-oriented analysis (OOA) have openedup a new insight for mapping landscape features, suchas terraces and roads, thus creating the basis forquantitative assessment of farming practices as anindicator in soil erosion risk assessment (Karydaset al. 2005a). OOA has been used in several domainsof environmental GIS, such as natural resourcemanagement, urban mapping, forest fire mapping,landscape mapping, land-use mapping, and biodiver-sity mapping. OOA is suggested as an appropriate

Fig. 1 Olive plantations are the most extended cultures inCrete (Greece). Newly established plantations (indicated by theellipses in the picture) contribute essentially to intensification ofthe cultivation and thus to soil erosion

20 Environ Monit Assess (2009) 149:19–28

technique for high-textured data, such as those ofVHR satellite images (Blaschke et al. 2005; Baatzet al. 2002).

Main aim and objectives

The main aim of this research was to propose atechnique for estimation of the support practice factorP when assessing risk of soil erosion which isassociated with olive cultivation in a Mediterraneanagricultural area. This overall aim was achievedthrough the following specific objectives:

& Mapping of natural or artificial features which arepreventive to soil erosion at the landscape level,such as terraces, roads, and rural paths.

& Quantification of P factor site-specifically basedon the landscape features.

& Mapping of the overall risk of soil erosion byusing the (site-specific) quantified P factor.

& Comparison of the mapping results derived when a(site-specific) quantified P factor is used with thosederived when a (conventional) uniform P factor isused.

Study site

The municipality of Kolymvari was selected as atypical example of an olive cultivation area in Crete,Greece (Fig. 2). The overall landscape of the area ischaracterized partially as gently undulating, moun-tainous or semi-mountainous. A very undulating reliefis present in the peninsula of Rodopos (North part ofthe municipality), which enjoys special protection

under NATURA 2000 (Directive 92/43/EEC, GreekHabitat Project Natura 2000; Baourakis et al. 2003).The protected area accounts for 54.7% of the totalterritory of the municipality.

The study site for the soil erosion risk mapping wasin the South part of the municipality, which comprisesmainly tree crops covering an area of 6,800ha. Theolive tree, with deep historical roots, is the mostextensively grown crop there; small divisions ofvineyards, greenhouses, fruit plantations and vegeta-bles can be found on a smaller extent. Regarding theclimate, Kolymvari is a typical Mediterranean area,with the mean annual air temperature around 17°C,January and July being the coldest and the warmestmonths, respectively. The average annual rainfall is722.6 mm, with January (143.8 mm) and July(0.6 mm), being the wettest and the driest months,respectively. The exact area on which the methodologywas implemented was delineated by excluding all non-agricultural areas from the total municipality territo-ries. In this task the GIS layer of CORINE land coverof the year 2000 was taken as the basis for locating thenon-agricultural uses, which then were further refinedwith visual photo-interpretation of a very high spatialresolution satellite image.

Dataset and methodology

Dataset

The dataset comprised mainly raster type data; rasterdata is an abstraction of the real world where spatialdata is expressed as a matrix of cells (or pixels), withspatial position implicit in the ordering of the pixels.

Fig. 2 The island of Creteis in the East side of theMediterranean Sea. Thestudy site is located in theNW side of the island

Environ Monit Assess (2009) 149:19–28 21

Some vector maps (points, line, or polygons) wereused only for zoning or interpretation purposes. Morespecifically the dataset consisted of the following:

& Avery high spatial resolution (VHR) satellite image,namely a QuickBird image, in two modes: Panchro-matic mode (grey-scale; PAN) and Multi-Spectralmode (MS), covering approximately 16.5 by16.5 km, acquired on November 16, 2002 with a14° off-nadir angle. The spatial resolution was0.64 m in the PAN mode and 2.50 m in the MSmode. The two images were orthorectified and fusedin the lab prior to further analysis. Orthorectificationis a procedure of removing the effects of topographyfrom an image, while fusion (or pansharpening orresolution merge) is the merge of a high spatialresolution PAN image with a high spectral resolu-tion MS image, in order to take advantage of bothmodes and expand visual photo-interpretationpotential of the image (Fig. 3). A set of grey-scale

orthophotos with a spatial resolution of 1 m wereutilised as reference in orthorectification of thesatellite images.

& A Digital Elevation Model (DEM, i.e. a rasterformat representation of the surface elevation) of

Fig. 3 The fused QuickBirdimage used in the study and –in the inset – a close view ofit, where individual olivetrees are visible. The studysite was defined by theboundaries of the agriculturalland derived from theCORINE land cover mapand enhanced with visualphoto-interpretation ofthe image

Table 1 The K factor estimated from the geologic formationsfound in the area by expertise

Parent material (geology) Soil typea K factor

Alluvial deposits SL–L 0.15Limestone C, SiC 0.4Peridotid CL–C 0.5Granite S–SL 0.2Schists L 0.7Gneiss S, LS, L 0.3Tertiary deposits SL–L 0.15

aC Clay, CL clay loam, L loam, LS loamy sand, S sand, SCsandy clay, SL sandy loam, Si silt, SiC silty clay, SiCL silty clayloam, SiL silt loam

22 Environ Monit Assess (2009) 149:19–28

the area, more specifically a 4 m spatial resolutiondataset, which was derived with interpolationfrom elevation point-grid of 20 m.

& A geological map scanned from paper map-sheets& Land cover in vector format (polygons).& Meteorological tabular data (temperature and

rainfall) from two local weather stations.

Overall methodology

Risk assessment and mapping of soil erosion wascarried out by implementing RUSLE methodology.RUSLE is a computation method that may be used forsite evaluation and planning purposes and also forassisting the decision process of selecting erosioncontrol measures. It provides an estimate of theseverity of erosion and also numerical results thatcan validate the benefits of planned erosion control

measures in the risky areas (Silleos 1990). Theformula used by RUSLE is:

A ¼ R� K � LS � C � P; ð1Þ

where A: eroded soil in ton ha−1 year−1; R: rainfallerosivity factor; K: soil erodibility factor; LS: combi-nation of slope-length factor (L) and slope-steepnessfactor (S); C: cover and management factor; and P:erosion control (or support) practice factor.

RUSLE was applied in Kolymvari in the spatialdomain using GIS; more specifically, all RUSLEfactors were derived as geographic layers in the rasterformat after processing of original raster data and weremultiplied together for calculating the final risk maps.In this procedure, support practice factor (P) was givenspecial attention; more specifically, P was quantifiedby mapping landscape features which are indicative

Fig. 4 The map of the LSfactor in the study site,derived from the availableDEM

Environ Monit Assess (2009) 149:19–28 23

of existing support measures (terraces) or conditionspreventive to soil erosion (rural roads and paths).

In order to compare results from a site-specificapproach for P (which was the target) with a uniformapproach (which is a conventional method), RUSLEwas implemented twice: in the first case, factor P wasassigned values derived from the classification of theQuickBird image and the slope map, while in thesecond case a uniform value of 0.8, was assigned to Pthroughout the study site; this value was selected byexpertise as representative for the study site as awhole.

Derivation of RUSLE factors

In the following, the derivation of all RUSLE fac-tors is presented, while emphasis is given to the Pfactor.

Rainfall erosivity factor (R) Assuming a high corre-lation between elevation and rainfall intensity, themethodology comprised two consecutive steps:

& Use of the available DEM to estimate theelevations of the two available weather stations;and

& Calculation of precipitation value in every gridcell of the DEM, after a linear regression ofprecipitation versus elevation.

More specifically, the equation used for derivingthe precipitation grid was the following:

R ¼ a� hþ b; ð2Þ

where R: rainfall (mm), h: DEM value (m), and a, b:linear coefficients.

Fig. 5 The map of the Cfactor in the study sitederived from the QuickBirdimage using a fuzzysigmoid transformation ofthe NDVI values

24 Environ Monit Assess (2009) 149:19–28

Soil erodibility factor (K) Soil erodibility representsthe inherent resilience of soil to detachment andtransport by rainfall. Values of factor K are directlyrelated to sensitivity to erosion. Erodibility varieswith soil texture, aggregate stability, infiltrationcapacity and organic and chemical content (Chmelovaet al. 2002). The erodibility textural/parent materialparameter is based on a combination of the dominantsoil surface texture and the type of parent material.Taking into account the strong relationship betweensoil types and geological formation and the missingdetailed soil data in the study site, the K factor valueswere estimated in two consecutive steps: texture typewas derived from the geological formation (parentmaterial) using the available geologic map and thenvalues of the K factor were assigned to every texturetype by expertise (Table 1).

Slope length and steepness factor (LS) The factorsSlope Length and Slope Steepness refer to the

topographic (or relief) influence on erosion intensity.The factors L and S were derived from the DEM in araster format using an extension of ArcView 3.2software package, namely ‘TOPOCROP v. 1.2,’which implements the following formulas (Mooreet al. 1991; Schmidt et al. 2003):

L ¼ 1:4As

22:13

� �0:4

; ð3Þ

S ¼ sin b0:0896

� �1:3

; ð4Þ

where β: slope angle in degrees and As: specificcatchments area or upslope area per unit width ofcontour (m2/m; Burrough 1986). The resulting rastercan be shown in Fig. 4.

Cover management factor (C) Cover managementfactor (C) for each grid cell is dimensionless ranging

Fig. 6 The map of the Pfactor derived from theQuickBird image using anobject-oriented classifica-tion of the existing terracesand rural roads

Environ Monit Assess (2009) 149:19–28 25

from 0 to 1, where by definition C = 1 under standardfallow conditions. While surface vegetation cover isadded to the soil, the C factor value approaches 0. Inthis research, C factor calculation was based on the useof the Normalised Difference Vegetation Index (NDVI)a broadly accepted index of sensitivity to presence andgood health of green vegetation; the NDVI values werecalculated from the QuickBird image. Two distinctsteps were implemented (Silleos 2000):

& Derivation of the NDVI thematic layer from theMS QuickBird image band with ‘ERDAS Imagine8.7’ image processing software package.

& Reclassification of the NDVI values, using thesigmoid fuzzy-logic membership function foundin ‘IDRISI Kilimanjaro’ image processing soft-ware package. Reclassification with a fuzzyfunction served as a means for taking intoaccount uncertainty parameters, such as thecanopy structure, ground growth density, a

littered layer, and the conditions of the vegeta-tion, which is not always related to its soilprotective function (Yang et al. 2003; Fig. 5).

In cases where arable land predominates in anarea, C factor is calculated for different seasons,because vegetation conditions change rapidlythroughout a year. However, this was not necessaryin this research, because olive trees are permanentcrops; therefore, their canopy’s conditions do notchange significantly throughout a cultivation year.Only annual pruning might change their canopy,but because RUSLE is a method for comparativeassessment, this fact could affect the result onlyuniformly.

Support practice factor (P) Support practice factor isdefined as the ratio of soil loss after a specific supportpractice to the corresponding soil loss after up anddown cultivation. It has been very common to

Fig. 7 The map of soilerosion risk using a rastergeographic layer withdifferentiated valuesfor P factor in the RUSLEformula

26 Environ Monit Assess (2009) 149:19–28

evaluate current farming practices by experts, eitherwith field observations or using air photographs(Strand et al. 2002). Lately, analysis of VHR satelliteimagery is also gaining attention. Among severalimage analysis techniques are visual photo-interpre-tation and object-oriented analysis of VHR satelliteimages (Karydas et al. 2005b). Visual photo-interpre-tation of an image is a method whereby separateobjects and pattern elements are identified by thehuman eye, and regions of relatively uniform com-position and appearance are accurately defined(Lueder 1959). Object-oriented image analysis(OOA) is a method of classifying meaningful land-scape objects (instead of pixels) apparent in an image,derived after grouping neighbouring pixels accordingto a heterogeneity criterion (Baatz et al. 2002). Boththe abovementioned techniques were used for assess-ing P factor, though object-oriented approach wasemphasised; while visual analysis was used to achievean overall interpretation of the image, OOA was usedfor semi-automatically mapping terraces and ruralroads (Karydas et al. 2005a).

More specifically, the P factor was derived withobject-oriented analysis (classification) of the Quick-Bird image, resulting in the quantification of supportpractices through a thematic map (instead of assigninga uniform P value for the entire study area or assigningdifferent values by expertise, as usual). In technicalterms, after terraces were mapped with OOA, a bufferzone of 60m was delineated around them consideringthat this distance was the mean positive influence rangeof the terraces as means for preventing erosion; then,different P values were assigned to the buffersaccording to the local slope. Regarding the rural roads,only objects lying across the slope direction weremapped, considering only these as the roads having aprotective character to erosion; then, a buffer zone of30m was delineated around them and a P value of 0.6was assigned to the buffers (Fig. 6). The specificranges for buffer zones and their P values wereselected based on the literature, the knowledge of thestudy site, and teams’ domain expertise.

Results and discussion

In this research soil erosion risk was mapped in thesemi-mountainous, olive cultivation environment ofCrete using a QuickBird image, a DEM, a geologic

map, and rainfall data from two local stations. Themain achievement of the research was the quantifica-tion and site-specification of the support practice factorP in the spatial domain and its incorporation in theRUSLE formula as a grid raster geographic layer. Thefinal results of this research comprised the following:

& A map of soil erosion risk which was derivedwhen differentiated values for P factor were inputin the RUSLE formula (Fig. 7).

& A map of soil erosion risk which was derivedwhen a uniform value 0.8 for P factor was input inthe RUSLE formula.

In both cases the erosion values were reclassifiedinto five severity classes and more specifically thefollowing: Very slight (ERC1): 0–5 ton ha−1 year−1,Slight (ERC2): 5–10 ton ha−1 year−1, Moderate(ERC3): 10–20 ton ha−1 year−1, Severe (ERC4): 20–40 ton ha−1 year−1, Very severe (ERC5): >40 ton ha−1

year−1. Reclassification is consistent with the RUSLEmodel’s role as a conservation management tool,where relative comparisons among areas are moresignificant than assessment of the absolute soil loss ina particular location.

The implementation of the methodology and themapping results showed that:

& Fine scale mapping of the roads and paths withobject-oriented classification of a QuickBird imagewere satisfactory. This was assessed by visualinterpretation and classification evaluation tech-niques in the environment of the object-orientedanalysis (OOA) software (‘eCognition’).

& Fine scale mapping of the terraces with object-oriented analysis were a more difficult task. It wasnoted that terraces were underestimated whenclassified in a QuickBird image with OOA.

& The DEM was necessary not only for derivationof the LS factor, but also for the estimation of theP factor; the slope map (a derivative of the DEM)was used as an input parameter in the calculationof the P value for every feature (terrace, path, orroad).

& Input of site-specific P values in the RUSLE modelresulted in the spatial differentiation of areas thatwere conventionally classified as highly risky (hotspots) into the slight–moderate erosion risk classes;in other words, site-specificity worked as a

Environ Monit Assess (2009) 149:19–28 27

correcting tool for the values derived uniformly.Accurate mapping is of crucial importance whenused at the operational and planning level. In thiscase, 3% of the total study site (i.e. 204ha) that wasmapped as highly risky in the uniform approachwas finally mapped as non-highly risky by the site-specific approach. Obviously, the area which wasmisclassified by the uniform approach is evenbigger, because the current Object-oriented classi-fication model did not achieve to map all therelevant landscape features (especially the terraces)throughout the study site. An improved OOAmodel is necessary and is planned to be developed.

The main conclusion is that a high spatial resolutionsatellite image, such as the QuickBird image, wasnecessary and efficient in order to map P factor andthus contribute to the refinement of soil erosion riskassessment. The results of a fine soil erosion mappingcan be used for appropriate management measures atthe field level instead of the area level as usual.

Acknowledgements This research was carried out in theframework of project ‘ISOTEIA’ (INTERREG IIIB CADSES,code: 3B093). For more, http://www.isoteia.org.

References

Baatz, M., Benz, U., Dehghani, S., Heynen, M., Holtje, A.,Hofmann, P., et al. (2002). eCognition user’s guide.Munchen: Definiens Imaging GmbH: Digital.

Baourakis, G., Drakos, P., Karydas, C., Papadantonakis, N., &Stamataki, E. (2003). Setting up and implementation ofsustainable and multifunctional rural development modelbased on organic and competitive agriculture. Chania:Mediterranean Agronomic Institute of Chania.

Beaufoy, G. (2000). The environmental impact of olive oilproduction in European Union. European Forum onNature Conservation and Pastoralism, pp. 73.

Blaschke, T., Lang, S., & Moller, M. (2005). Object-basedanalysis of remote sensing data for landscape monitoring:Recent developments. In Anais XII Simpósio Brasileiro deSensoriamento Remoto, Goiânia, Brasil, 16–21 Abril 2005(pp. 2879–2885). Brazil: INPE.

Burrough, P. A. (1986). Principles of geographical informationsystems for land resources assessment. New York: OxfordUniversity Press.

Chmelova, R., & Sarapatka, B. (2002). Soil erosion by water:Contemporary research methods and their use.Geographica,37, 23–30.

Gobin, A., Kirkby, M., & Govers, G. (1999). Pan-European soilerosion risk assessment. http://perswww.kuleuven.ac.be/~u0001760/PESERA/PESERA.htm.

Gomez, J. A., Batani, M., Renschler, C. S., & Fereres, E. (2003).Evaluating the impact of soil management on soil loss inolive orchards. Soil Use and Management, 19(2), 127–134.

Karydas, C. G., Sekuloska, T., & Sarakiotis, I. (2005a). Finescale mapping of agricultural landscape features to be usedin environmental risk assessment in an olive cultivationarea. IASME Transactions, 2(4), 582–589.

Karydas, C. G., Sekuloska, T., & Sarakiotis, I. (2005b). Use ofimagery to indicate landscape features important whenassessing environmental risk caused by olive farming andolive oil production. In Proceedings of 2005 IASME/WSEAS International Conference on Energy, Environ-ment, Ecosystems and Sustainable Development. Athens:Polytechnic School of Athens.

Lueder, R. R. (1959). Aerial photographic interpretation.Principles and applications. New York: McGraw-Hill.

Moore, I. D., Grayson, R. B., & Ladson, A. R. (1991). Digitalterrain modelling: A review of hydrological, geomorpho-logical, and biological application. Hydrological Processes,5, 3–30.

Morgan, R. P. C. (1995). Soil erosion and conservation.Harlow: Longman.

Schmidt, F., & Persson, A. (2003). Comparison of DEM datacapture and topographic wetness indices. Precision Agri-culture, 4(2), 179–192.

Silleos, N. G. (1990). Mapping and evaluation of agriculturallands. Thessaloniki: Giahoudi-Giapouli.

Silleos, N. G. (2000). Introduction to remote sensing andgeographical information systems. Thessaloniki: Giahoudi-Giapouli.

Strand, G.-H., Dramstad, W., & Engan, G. (2002). The effect offield experience on the accuracy of identifying land covertypes in aerial photographs. International Journal of AppliedEarth Observation and Geoinformation, 4, 137–146.

Trojacek, P., & Kadlubiec, R. (2004). Detailed mapping ofagricultural plots using satellite imagers and aerial orthphotomaps. In R. Goossens (Ed.) Remote sensing in transition(pp. 253–257). Rotterdam: Millpress.

Yang, C.-C., Prasher, S. O., Landry, J.-A., & Ramaswamy, H. S.(2003). Development of an image processing system and afuzzy algorithm for site-specific herbicide applications.Precision Agriculture, 4(1), 5–18.

28 Environ Monit Assess (2009) 149:19–28