Applied Geography 30 (2010) 394–415

Contents lists available at ScienceDirect

Applied Geography

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

A multi-component GIS framework for desertification risk assessmentby an integrated index

Monia Santini a,*, Gabriele Caccamo b,c, Alberto Laurenti b, Sergio Noce b, Riccardo Valentini a

a Department of Forest Environment and Resources (DISAFRI), University of Tuscia, via S. Camillo de Lellis snc, 01100 Viterbo, Italyb Agriconsulting S.p.A, via Vitorchiano 123, 00189 Roma, Italyc GeoQuest Research Centre, School of Earth and Environmental Science, University of Wollongong, NSW 2522, Australia

Keywords:GIS modellingDesertification riskDegradation modelsSardinia

* Corresponding author.E-mail address: monia.santini@unitus.it (M. San

0143-6228/$ – see front matter � 2009 Elsevier Ltddoi:10.1016/j.apgeog.2009.11.003

a b s t r a c t

This study describes a GIS-based software tool for the qualitative assessment of deserti-fication risk. The tool integrates results from several land degradation models into ArcGISusing the object-oriented programming language Visual Basic. The integration of GIS andmodels is based on different coupling approaches.The proposed methodology was developed and tested on Sardinia Island (Italy). Six drivingfactors of desertification (overgrazing, vegetation productivity, soil fertility, water erosion,wind erosion and seawater intrusion) were model-simulated over two time periods toinvestigate the spatio-temporal evolution pattern of desertification-prone areas. Modelresults were normalized, weighted and combined into an Integrated Desertification Index(IDI) ranging from 0 to 1 (representing the best and the worst conditions, respectively),and classified into five desertification risk levels.The implementation, performance of the methodology and benefits provided by themodelling approach to land management authorities for monitoring processes of landdegradation are fully described in the paper.

� 2009 Elsevier Ltd. All rights reserved.

Introduction

Desertification is an irreversible process (Mainguet, 1994) driven by several interconnected factors and triggered by thecombination of the natural predisposition of the environment with human pressure and climate change (Adamo & Crews-Meyer, 2006; UNCED, 1992). Desertification mainly occurs in drylands (e.g. Dregne, Kassas, & Rozanov, 1991; Kassas, 1995;Sivakumar, 2007), but it affects a wide variety of environments, climates and societies (e.g. Arnalds, 2006).

Given the multi-faceted and multi-disciplinary aspects of desertification, the interest of political institutions and researchcommunities has supported the development of different methods of analysis based on empirical approaches (e.g. FAO/UNEP1984; Geist & Lambin, 2004; Grunblatt, Ottichilo, & Sinange, 1992; Kosmas, Kirkby, & Geeson, 1999; Liu, Gao, & Yang, 2003;Mouat et al., 1997; Sepehr, Hassanli, Ekhtesasi, & Jamali, 2007), remote sensing applications (e.g. Fang et al., 2008; Feoli,Giacomich, Mignozzi, Ozturk, & Scimone, 2003; Gad & Lofty, 2008; Hill, Stellmes, Udelhoven, Roder, & Sommer, 2008; Lira,2004; Roder, Udelhoven, Hill, Del Barrio, & Tsiourlis, 2008; Shalaby and Tateishi, 2007; Wen-yi, When-hua, Su-fang, & Dan,2005; Zhang et al., 2008), and modelling procedures (e.g. Ding, Chen, & Wang, 2009; Hellden, 2008; Ibanez, Martinez, &Puigdefabregas, 2008; Ibanez, Martinez, & Schnabel, 2007; Salvati, Zitti, & Ceccarelli, 2008; Sasikala & Petrou, 2001).

tini).

. All rights reserved.

M. Santini et al. / Applied Geography 30 (2010) 394–415 395

Methodologies initially proposed did not allow for spatially explicit analysis of the problem. Advances in spatial sciencesenabled the implementation of early ‘lumped’ approaches into the Geographic Information Systems (GIS) logic. TraditionalGIS-based methods (e.g. Basso et al., 2000; Costantini et al., 2004; Feoli et al., 2003; Grunblatt et al., 1992; Kosmas et al., 1999;Liu et al., 2003; Motroni, Canu, Bianco, & Loj, 2009; Mouat et al., 1997) used static indicators, which are inadequate to analysedynamic phenomena. In contrast, modelling has been recognized as an important technique to analyse and describe thespatio-temporal evolution of desertification and the effect of socio-economic dynamics and ecological components on landdegradation (e.g. Danfeng, Dawson, & Baoguo, 2006).

Earlier approaches focused on assessing vulnerability, defined as the degree to which a system is likely to suffer fromdamage and loss due to desertification. There is a lack of works focusing on the difference between vulnerability and risk(e.g. Rahman, 2008), i.e., the quantifiable damage or loss in relation to the value (e.g. human well-being, land productivity) ofa region threatened by desertification (MEA, 2005). Givetrun n the large number of people living within or near areas subjectto desertification, this phenomenon must be assessed in terms of risk.

Common desertification vulnerability assessments were often limited to climate index analysis (e.g. Hellden & Tottrup,2008; Millan et al., 2005; Zhai & Feng, 2009), in most cases focusing on meteorological, agricultural or hydrological droughts(e.g. Durao, Pereira, Costa, Corte-Real, & Soares, 2009; Zhai & Feng, 2009). Climate change does not represent the only factordriving land degradation processes, although it contributes to altering critical thresholds beyond which systems can no longermaintain their equilibrium (e.g. Kosmas et al., 1999). The concept of desertification assumed here refers to the final andirreversible state of land degradation processes (e.g. Hill et al., 2008), triggered by the interaction of climate with a number ofeither predisposing or accelerating environmental and anthropogenic components (e.g. Dragan, Sahsuvaroglu, Gitas, & Feoli,2005; Salvati and Zitti, 2008; Wessels et al., 2007). For this reason, rather than using climate average– or climate variability-based indices, this work focused on including them in a more complex evaluation of their interactions with local charac-teristics towards land degradation.

The temporal component of land degradation processes should be incorporated into the risk analysis (e.g. Fang et al., 2008;Haijiang, Chenghu, Weiming, En, & Rui, 2008; Hanafi & Jauffret, 2008; Lu, Mausel, Brondizio, & Moran, 2004; Ringrose,Vanderpost, & Matheson, 1996; Zhang et al., 2008): an area mildly degraded subject to an intense and rapid desertificationprocess is more at risk when compared to an area at higher degradation level but stable over time.

A new methodology is presented here which integrates GIS with environmental models to simulate land degradationprocesses and to provide a comprehensive index of desertification.

Strategies for integrating models with GIS have been discussed and analysed by many researches (e.g. Al-Sabhan, Mulligan,& Blackburn, 2003; Batty & Xie, 1994; Brown, Riolo, Robinson, North, & Rand, 2005; Gemitzi & Tolikas, 2007; Karimi &Houston, 1996), emphasizing two different types of coupling: loose and fully tight. Loose coupling consists of data filetransferring between GIS and an external modelling program. This method enables to exploit the cartographic potential of GISand provides the model developers with the required flexibility in choosing the most appropriate models to integrate and theuser interface to build. The main drawback is that the loose coupling requires the development of further tools to format datafiles properly. With fully tight coupling, the models are entirely developed within a GIS platform using its programminglanguages and analytical functionalities. Neither GIS knowledge nor intermediate file conversion or processing are requiredby the model users. Despite being more refined, such methods do not allow transferring sophisticated model codes into GISlanguages which are often more restrictive. Further intermediate categories of coupling were individuated by other authors(Fedra, 1993; Nyerges, 1993; Sui & Maggio, 1999). Brandmeyer and Karimi (2000) well defined the categories into a five layerpyramid describing the increasing level of integration from one-way data transfer (manual transfer of data to model), to loosecoupling (the data transfer is automatic), to shared coupling (there are one or more Graphical User Interfaces, GUIs, withseparate or common data storage, respectively), to joined coupling (having a single GUI and common data storage), to toolcoupling (consisting in embedded models, single GUI and common data storage).

Besides the two extremes loose and fully tight coupling, an intermediate type used here consists in designing a stand-alonemodel, independent from GIS-based software, but able to process data in GIS formats (e.g. Huang & Jiang, 2002). Thisapproach provides several benefits enabling both to input GIS dataset into executable routines written in faster computinglanguages, avoiding time-consuming loop procedures of less efficient GIS languages, and to readily analyse the output datathrough GIS tools. This method is here referred to as simply tight coupling.

In this paper, the GIS implementation of the proposed desertification risk evaluation tool is described and its application toIsland of Sardinia (Italy) is presented and discussed.

Study area description

The region of Sardinia has an area of 24 089 km2, and is located in the centre of the western part of the Mediterranean Sea(Fig. 1).

The mean annual temperature ranges from 11 �C to 17 �C, while the rainfall varies from 500 mm to more than 1000 mmover the high mountainous areas. Very dry years are often followed by short periods of heavy rainfall, triggering disastrouserosion and flooding events at the local level (Aru, Tomasi, & Vacca, 2006).

The Island is characterized by a high variability of landscapes due to its heterogeneous geologic, morphological, soil andvegetation characteristics. Sardinia is divided into several morphological regions: mountains cover 18.5% of the total area,67.9% of the territory is hilly and plains characterize the remaining 13.6%. Nearly 40% of the Island is covered by meadows and

Fig. 1. Location and topographic/vegetation characteristics of the study area from a LANDSAT 7-derived mosaic collected in May 2005.

M. Santini et al. / Applied Geography 30 (2010) 394–415396

pastures, Mediterranean macchia occupies over 20% of the surface, whereas the hardwood forests cover up to 10% of the totalarea (Vacca, Loddo, Serra, & Aru, 2001).

The main human-related activity causing land degradation in Sardinia is represented by grazing (D’Angelo et al., 2000;Enne, Pulina et al., 2000). Grazing accelerates soil erosion (key element of desertification) by removing vegetation cover,compacting the soil and reducing water infiltration (Aru et al., 2006; Zucca, Canu, & Della Peruta, 2006). Additional causes ofdesertification in Sardinia are local droughts (Fiori, Motroni, Duce, & Spano, 2004), mismanagement and salinization ofgroundwater resources, intense wildfire activity (Aru, 1986; Bajocco & Ricotta, 2008), and decrease in vegetation cover asresult of deforestation and agricultural land abandonment following their overexploitation (Giordano & Marini, 2008;Pungetti, 1995).

Insularity represents another predisposing factor of desertification: according to MEA (2005), insularity strengthens thelinks between ecosystems and local people and improves access to natural resources, resulting in an increased humanpressure and island vulnerability.

A good review on land degradation processes in Sardinia can be found in Enne, Zanolla, & Peter (2000) while recent studies(Costantini et al., 2004; Motroni et al., 2009) on desertification vulnerability assessment in the island represented a goodstarting point to develop the new proposed approach.

Materials and methods

General framework

The desertification risk tool here proposed is based on the analysis of several indicators combined together to providea comprehensive assessment of the phenomenon. Suitable desertification indicators must be able to capture the degradation

M. Santini et al. / Applied Geography 30 (2010) 394–415 397

processes affecting the study area and to incorporate the role of climate and land use components, two key driving factors ofland degradation. After an extensive analysis of both the conditions and the evolution of natural features in the study area(Aru, 1986; Costantini et al., 2004; Motroni et al., 2009), through field investigations and spatial data investigation, a suite ofappropriate indicators of land degradation have been identified on the basis of their applicability and sensitivity to envi-ronmental processes. The following six indicators were considered: overgrazing pressure, vegetation productivity, soilfertility, water erosion, wind erosion and seawater intrusion. These indicators were quantified by selecting the most suitablemodels according to the study area characteristics and data availability. Chosen models were different in type (process-based,empirical and data-oriented) and degree of complexity.

In order to account for spatio-temporal variations in the intensity and distribution of land degradation processes(e.g. Giordano & Marini, 2008), the proposed procedure was applied to two reference periods: as representative of pastconditions the early 1990s were considered; for the most recent conditions the 2005 was the base-year. Focusing on thesetwo different points in time, the datasets used encompassed a wider time scope with the first analysing 1990–1994, and thesecond 2003–2007. These two time frames were chosen based on their similarity in accessibility, quality and resolution ofinput data. Each model was thus run against both reference periods, using the proper temporal datasets.

The model outputs were first standardized (when necessary) to obtain indices of identical value range (Liu et al., 2003).The threshold values that represented the critical conditions for normalizing the outputs were chosen on a land use basis inorder to establish a relative scale of severity, weighting the results by the use of lands in the study area (Al-Adamat, Foster, &Baban, 2003). Finally, the indices were integrated into a single index taken as representative of the desertification risk.

Besides taking into account the changes in terms of desertification severity over time, the final index was formulated torank the relative importance of the different degradation processes in characterizing the land degradation over the studyregion. The analysis combined local expert judgements and statistical methods to determine the weighting of each modelaccordingly to its importance. The desertification risk index was classified into five levels of severity, and then processed toextract hot-spots of desertification, namely the areas seriously affected by risk.

Fig. 2 represents an outline of the approach, which will be detailed further in the next sections.In this work, up to one hundred GIS data layers (in vector and grid format) have been acquired and processed in order to

supply the six models with the necessary inputs. Table 1 lists the core database and its characteristics: from these basicdatasets numerous other layers were derived.

The implementation of the spatial methodology made use of a grid-based system at 50 m cell resolution, as a reasonablecompromise between the finer (e.g. land use, topography) and coarser (e.g. climate) dataset resolutions.

In the following sections, the theoretical principles on which each model is based are discussed, with particular attentionto the parameters used, tuning settings applied, and suitability of the model to work within a GIS environment. Formulationof the integrated desertification index is then introduced and explained.

Model to evaluate overgrazing pressure: CAIAOvergrazing is one of the main factors of land degradation observed in the Mediterranean region (e.g. Yassoglou, 1999). In

Sardinia, an excessive cattle pressure affects the lands by compacting the soil and reducing the vegetation cover, both ofwhich lead to decreased soil structure and stability (Aru et al., 2006).

The model CAIA (Pulina, 2000; Pulina & Zucca, 1999) formulates the land degradation as a function of grazing intensity andland carrying capacity. By combining the estimation of the cattle load sustained by a specific territory with analysis of itsenvironmental characteristics, the model estimates the sustainable grazing load for that same territory. In Sardinia, this

Fig. 2. Flowchart showing the links among temporal databases, modelled degradation processes and final desertification index. Boxes with horizontal shadingrefer to 1990-1994 time period of analysis. Boxes with vertical shading refer to 2003-2007 time period of analysis. Diagonally shaded boxes symbolizecomponents in common between the two periods. Elliptical text frames represent models, and white filled box the output of the procedure.

Table 1Main characteristics of available datasets, listing format, unit, resolution/scale, time period of reference, source and model of application.

Description Type Unit ofmeasure

ResolutionScale

Referenceperiod

Source Models

Corine Land cover 1990 Polygon – 1:1000000 1990 1 AllCorine Land cover 2000 Polygon – 1:1000000 2000 1 AllCorine Land cover 2003 Polygon – 1:250000 2003 2 All

Pedologic mapSoil depth Polygon Categorical (cm) 1:500000 2005 3 CAIARockiness Polygon Categorical (%) 1:500000 2005 3 CAIAStoniness Polygon Categorical (%) 1:500000 2005 3 CAIATexture Polygon Categorical (%) 1:500000 2005 3 CAIA, RUSLE,

CENTURY, WEAMDrainage Polygon Categorical 1:500000 2005 3 CAIAPermeability Polygon Categorical

(cm/h)1:500000 2005 3 CAIA, SHARP

Available water content Polygon Categorical(mm/m)

1:500000 2005 3 CENTURY

Soil organic matter Polygon Categorical (%) 1:500000 2005 3 CAIA, RUSLE, CENTURYDensity Polygon g/cm3 1:500000 2005 3 CENTURY, WEAM

AGEA – crop data Alphanumeric % of Crop foreach unit

1:20000 1995 and 2004 4 CENTURY

AGEA – zootechnicalconsistency

Alphanumeric Number ofcattle head

1:20000 1995–2003–2004 4 CAIA

Forest map Polygon – 1:1000000 2002 2 CENTURYGeolithologic/

permeability mapPolygon – 1:1000000 N/A 3 SHARP

Geologic map Polygon – 1:2000000 N/A 2 SHARPGeologic map Raster – 1:1000000 Various 5 SHARPDigital elevation model Grid m asl 25 m NA 6 CAIA, RUSLE,

SHARPRailway map Line – 1:100000 2003 2 RUSLEGround water resources

(aquifer type)Polygon – 1:5000000 1982 (updated 2006) 5 SHARP

Hydrographic map Line – 1:100000 2003 3 RUSLE

WellsDepth of the staticpiezometric level

Points m N/A Various 5 SHARP

Elevation of the well asl Points m asl N/A Various 5 SHARPPiezometric elevation ofthe water table

Points m asl N/A Various 5 SHARP

Number of aquifer met Points unitless N/A Various 5 SHARPMaximum/meandischarge

Points l/s N/A Various 5 SHARP

Well depth Points m N/A Various 5 SHARPTop and bottom of theaquifer layer

Points m asl N/A Various 5 SHARP

Stratigraphy Points m asl N/A Various 5 SHARP

ClimateMean temperature Points or grid �K 50 km

(grid)1994–2007 7,8,9 MOD17, WEAM,

SHARPDew point temperature Points �K N/A 1994–2007 8 MOD17, WEAMWind speed Points m/s N/A 1994–2007 8,9 WEAMMaximum temperature Points �K N/A 1994–2007 8,9 CENTURYMinimum temperature Points �K N/A 1994–2007 8,9 CENTURY, MOD17Precipitation Points mm/month N/A 1982–2007 2 CAIA, RUSLE,

CENTURY, SHARPTotal precipitationamount

Points mm/day N/A 1994–2007 8,9 RUSLE, CENTURY

Solar radiation Points or grid MJ/Day 50 km(grid)

1975–2007 7 MOD17

M. Santini et al. / Applied Geography 30 (2010) 394–415398

Table 1 (continued)

Description Type Unit ofmeasure

ResolutionScale

Referenceperiod

Source Models

Road maps Line – 1:100000 2003 2 RUSLEPhenologic calendars Alphanumeric – N/A 2006 3 CENTURYFPAR MODIS Grid Unitless 10000 m 2003 – 8days

synthesis10 MOD17

NDVI MODIS Grid Unitless 250 and1’000 m

2003 – 8dayssynthesis

10 MOD17, RUSLE, WEAM

LAI MODIS Grid Unitless 10000 m 2003 – 8dayssynthesis

10 WEAM

NDVI AVHRR Grid Unitless 10000 m 1995 – 7dayssynthesis

11 MOD17, RUSLE, WEAM

1 – European Environment Agency, http://www.eea.europa.eu/publications/COR0-landcover.2 – Regional Administration of Sardinia (RAS).3 – Agriconsulting S.p.A. (Italy), http://www.agriconsulting.it/inglese/htm/intro.htm.4 – Italian Payment Board for Agriculture (AGEA) (Italy), http://www.agea.gov.it/portal/page/portal/AGEAPageGroup/HomeAGEA.5 – High Institute for Environmental Protection and Research (ISPRA) (Italy), http://www.mais.sinanet.apat.it/cartanetms/.6 – Italian Military Geographic Institute (IGMI) (Italy), http://www.igmi.org/.7 – MARS-STAT DATABASE – Joint Research Center (JRC) (Italy), http://mars.jrc.it/mars/About-us/AGRI4CAST/Data-distribution.8 – National Oceanic and Atmopheric Administration (NOAA), http://www.ncdc.noaa.gov/oa/ncdc.html.9 – Central Office for Crop Ecology (UCEA) (Italy), http://www.politicheagricole.it/ucea/forniture/index3.htm.10 – Moderate resolution Image Spectroradiometer (MODIS), https://wist.echo.nasa.gov/wist-bin/api/ims.cgi?mode¼MAINSRCH&JS¼1.11 – Advanced Very High Resolution Radiometer (AVHRR), http://www.class.ncdc.noaa.gov/release/data_available/avhrr/index.htm.

M. Santini et al. / Applied Geography 30 (2010) 394–415 399

methodology has been successfully applied in the past, specifically to extensive and semi-extensive grazing lands (Madrau,Loi, & Baldaccini, 1998).

Any effect relating to the grazing pressure can be evaluated from 20 sub-coefficients, measuring: animal (bovine, ovineand sheep) vegetation consumption and trampling on soil, animal productivity (low, medium or high productive level) andgrazing management (extensive or intensive). The sum of these sub-coefficients identifies the b coefficient which repre-sents a weighing factor for the live weight X of the Fully-Grown Unit (FGU) of a given species present in a territory withsurface area S.

Considering these assumptions the grazing intensity in the territory (actual CAIA) can be evaluated for a dataset withdimension k (number of species of animals for each territorial unit), using the product:

ðX$ B$ bÞ=S (1)

where X is a vector of dimension (1, k), whose values indicate the number of FGU for different species; B is a matrix (k, 20) ofthe twenty sub-coefficients of b; then b is the vector (20, 1) of the sub-coefficients and S is the total surface area (in Ha).

CAIA defines territorial units based on categories of surface elevation, slope, aspect, vegetation cover, rock surface cover,drainage, soil depth, soil texture and available water content in soil. Each attribute is rated with a score ranging from 0 to 1,referring to areas entirely suitable for grazing and inadequate for pasture activity, respectively. All these indices are thensummed for each territorial unit identifying a single index, stretched from 0 to 1 and classified into five levels of increasingvulnerability to grazing (sustainable CAIA; e.g. D’Angelo et al., 2000).

Finally the overgrazing index derives from the ratio between the grazing load actually sustained by a territory (actualCAIA) and potential grazing load capacity (sustainable CAIA). This index varies from 0 (no overgrazing pressure) to 1(maximum overgrazing pressure).

The model structure used here is consistent with the fully tight coupling as the model was implemented using onlyGIS-compatible formats and languages (Arcgis9.x, �ESRI).

Model to evaluate vegetation degradation: MOD17The rate at which light energy is converted into plant biomass is called primary productivity and total amount of energy

converted is called Gross Primary Productivity (GPP). Net Primary Productivity (NPP) is the difference between GPP andenergy lost during plant respiration (Campbell, 1990). GPP can be estimated from satellite remote sensing using visible andnear infra-red spectral wavelengths. In this analysis, the MOD17 algorithm (Running, Nemani, Glassy, & Thornton, 1999) wasused to compute vegetation productivity. The Gross and Net Primary Productivity algorithm is based on Monteith’s (1972,1977) equations, and designed for the MODIS (MODerate resolution Imaging Spectroradiometer) sensor, mounted on TERRAsatellite platform and providing a daily coverage of the globe at 1000 m of resolution. The GPP formula consists of two mainfactors: one represents the Absorbed Photosynthetically Active Radiation (APAR, W m�2), and the other represents theRadiation Use Efficiency (e, kgC M J�1), i.e., the amount of carbon produced per unit energy by specific vegetation types.

Developing these factors the formula becomes:

GPP ¼ APAR$3 ¼ ðPAR$ fPARÞ3 ¼ ðSWrad$ 0:45$ fPARÞ ð3max$mTmin$mVPDÞ (2)

where PAR is the incident Photosynthetically Active Radiation (W m�2) and it is assumed as the 0.45 of the Short Wave

Table 2Reclassification (first column) and description (second column) of the CLC classes (third column) used to set the parameters of MOD17 model. Columns from4 to 8 report MOD17 algorithm parameters as described in Section 3.1.2. Footnotes report the description of CLC categories.

MOD17classes

Description CLC classes 3 (kgC M J�1) Tmin_min (�C) Tmin_max (�C) VPD_min (Pa) VPD_max (Pa)

1 Evergreen NeedleleafForest

3.1.2 0.001008 �8.00 8.31 650 2500

2 Evergreen BroadleafForest

3.1.1.1 0.001159 �8.00 9.09 1100 3900

4 Deciduous BroadleafForest

3.1.1 (residual) 0.001044 �8.00 7.94 650 2500

5 Mixed Forest 3.1.3 0.001116 �8.00 8.50 650 25008 Grassy woodlands 2.2.1–2.2.2–2.2.3 0.000800 �8.00 11.39 930 31009 Wooded grasslands 2.4.4 0.000768 �8.00 11.39 650 31006 Closed shrublands 3.2.4 0.000888 �8.00 8.61 650 31007 Open shrublands 3.2.3 0.000774 �8.00 8.80 650 360010 Grasslands 3.2.1–3.3.3 0.000680 �8.00 12.02 650 350012 Croplands 2.1.1–2.1.2–2.1.3–2.4.1–2.4.2–2.4.3 0.000680 �8.00 12.02 650 4100

3.1.2: Coniferous forest.3.1.1.1: Forests with dominant holm and cork oaks.3.1.1: Broad-leaved forest.3.1.3: Mixed forest.2.2.1–2.2.2–2.2.3: Vineyards – Fruit trees and berry plantations – Olive trees.2.4.4: Agro-forestry areas.3.2.4: Transitional woodland-shrubs.3.2.3: Sclerophyllous vegetation.3.2.1–3.3.3: Natural grassland – Sparsely vegetated areas.2.1.1–2.1.2–2.1.3–2.4.1–2.4.2–2.4.3: Non-irrigated arable land – Permanently irrigated land – Rice fields; Annual crops associated with permanent crops –Complex cultivation patterns; Land principally occupied by agriculture, with significant areas of natural vegetation.

M. Santini et al. / Applied Geography 30 (2010) 394–415400

incoming Radiation (SWRad, W m�2), fPAR (unitless) is the fraction of the PAR absorbed by vegetation, 3max (kgC M J�1) is thebiome-specific maximum conversion efficiency and mTmin and mVPD are two scalar reduction factors, computed by crossingthe minimum temperature (Tmin, in �C) and the Vapour Pressure Deficit (VPD, in Pa) with their lower and upper limits(Tmin_min, Tmin_max and VPD_min, VPD_max, respectively), necessary to each biome type in order to calculate GPP.

In the present application, the MOD17 algorithm was rewritten using the fPAR-MODIS product (Running et al., 1999),supplied as an 8-day composite, after grouping the land cover categories of the Corine Land Cover dataset (CLC hereafter;http://www.eea.europa.eu/publications/COR0-landcover) according to the typical vegetation communities of Mediterraneanecosystems. For the biome types, proper parameters were associated on the basis of their correspondence with the MOD17vegetation classes (Table 2; Running et al., 1999).

The daily data of meteorological stations from different datasets, interpolated to grids, were used to account for Tmin,whereas for SWrad the MARS-STAT database was used (see Table 1). The VPD was calculated including the mean dailytemperature (t_mean) and the mean daily dew point temperature (t_dewpt), according to the following relationship:

VPD ¼ 611�

exp�

19:84�

1� 273:15t mean

��� exp

�19:84

�1� 273:15

t dewpt

���(3)

Both VPD and Tmin where averaged on an 8-day basis, matching the temporal resolution of the fPAR-MODIS product.Furthermore, the algorithm was improved by embedding: (i) a gap-filling procedure in order to replace the fPAR pixels not

satisfying a minimum qualitative level; (ii) a spatial downscaling of the remotely sensed data from the original 1000 m cellsize to 50 m working resolution.

In the gap-filling, any invalid (low quality or missing) pixel in the fPAR image at a given time t was replaced with the meanvalue, over that same pixel, calculated using t-1 and tþ 1 images. If still invalid, the temporal interval was extended to includemaximum t-4 and tþ 4 dates to include only valid fPAR pixels.

For the downscaling, first the ratios on a pixel by pixel basis between fPAR and MODIS Normalized Difference VegetationIndex (NDVI) images (MOD13A2,1000 m resolution) were derived. Second, each ‘ratio’ pixel, resampled to 250 m of resolu-tion, was back-multiplied by the sixteen NDVI pixels at 250 m of resolution (MODIS MOD13Q1 product) contained within thecoarser MOD13A2 pixel. Third, a mixed-pixel approach was used to spatially distribute the fPAR inside the 250 m pixelaccording to the land use type fractional cover dictated by the more detailed CLC dataset, obtaining fPAR maps at 50 mresolution.

As MODIS started to acquire data from 2000, the time period 1990–1994 sourced NDVI data from Advanced Very HighResolution Radiometer (AVHRR, 1000 m resolution) images to derive fPAR using 7-day synthesis according to the relation-ships identified in the literature review (i.e. Chang & Wetzel, 1997). Thus fPAR was downscaled from such resolution to 50 mdirectly using the third step outlined above.

M. Santini et al. / Applied Geography 30 (2010) 394–415 401

For the actual period model run, yearly GPP was computed by summing the 46 MODIS-based GPP maps multiplied byrespective days of coverage (8 for the first 45 maps of the year and 5 for the last one); for the past, the 52 AVHRR-derived GPPmaps of the year were multiplied by 7 and then summed together.

A GIS-tool processing datasets in raster format, exclusively running within ArcGIS9.x software and producing the yearlyGPP map was developed. This is a fully tight coupling in the sense that the model was written using GIS languages, formats andfunctions.

Model to evaluate soil degradation: CENTURYThe loss of organic matter and nutrients is a primary degradation process affecting the soil. This phenomenon interacts

with other soil deterioration processes: e.g., the decrease of vegetation cover, the reduction of water retention capacity andthe increase of erosion risk. The CENTURY model (Parton, Schimel, Cole, & Ojima, 1987) simulates energy balance and thecycle of different nutrients (water, carbon, nitrogen, phosphorus and sulphur) between soil and vegetation and across threemain types of land use: agriculture, grassland and forests. The list of input data and parameters required by the model isreported in Table 3.

In this application, CENTURY was applied to ten vegetation classes: broadleaf forest, needleleaf forest, mixed forest,Mediterranean macchia, grassland/pasture, vineyard, orchard, olive-grove, rice and seeded cultivations. These classes wereinitially extracted through aggregation of the CLC categories. While information regarding the arboriculture areas wasdirectly derived from the CLC, the spatial detail and information contained in the CLC for the forested areas were improvedby cross-analysing it with the Forest Map of Sardinia. The CLC information for the seeded croplands was integrated witha dataset provided by the National Payments Board for Agriculture (AGEA). Specific land management practices were thenassociated to each land use class according to the directives for the Good Agricultural Practices (following the EUCommission Regulation 1750/1999) and to the phenological calendars. Besides geographic site location (rows 1 and 2 inTable 3), the monthly statistics of climate variables (rows 3–6 in Table 3) were interpolated into grids using a meteoro-logical dataset collected from 417 weather stations in the study area between 1982 and 2007. This dataset was subsequentlydivided into two temporal subsets covering, respectively, the 1982–1995 time interval (used to extract climate averagesrepresentative of the earlier reference period) and the 1996–2007 time interval (representative of the later referenceperiod). The soil characteristics (rows 7–13 in Table 3) were extrapolated using the data layers contained in the PedologicMap of Sardinia. The remaining parameters required by the model (row 14–33 in Table 3) were derived from the existing

Table 3List of the main inputs to CENTURY model and their units.

Input Unit of measure

1. Latitude of simulation site cm/month2. Longitude of simulation site �

3. Mean of monthly precipitation amount �

4. Standard deviation of monthly precipitation amount cm/month5. Minimum air temperature at 2 meters �C6. Maximum air temperature at 2 meters �C7. Sand fraction in the soil Unitless8. Silt fraction in the soil Unitless9. Clay fraction in the soil Unitless10. Soil density g/cm�3

11. Wilting point cm H2O per cm of soil12. Field capacity cm H2O per cm of soil13. pH Unitless14. Initial surface C with slow turnover gC/m2

15. Initial soil C with slow turnover gC/m2

16. Initial soil C with intermediate turnover gC/m2

17. Initial soil C with fast turnover gC/m2

18. Initial surface ratio C/N with slow turnover Unitless19. Initial soil C/N ratio with slow turnover Unitless20. Initial soil C/N ratio with intermediate turnover Unitless21. Initial soil C/N ratio with fast turnover Unitless22. Carbon content in the superficial litter gC/m2

23. Carbon content in the soil litter gC/m2

24. Initial C/N ratio of superficial litter with slow turnover Unitless25. Initial C/N ratio of soil litter with slow turnover Unitless26. Initial leaf C in forest ecosystems gC/m2

27. Initial leaf C/N ratio in forest ecosystems Unitless28. Initial wood C in forest ecosystems gC/m2

29. Initial wood C/N ratio in forest ecosystems Unitless30. Initial root C in forest ecosystems gC/m2

31. Initial root C/N ratio in forest ecosystems Unitless32. Initial carbon on death wood gC/m2

33. Initial carbon on death root gC/m2

M. Santini et al. / Applied Geography 30 (2010) 394–415402

literature (e.g. Augusto, Ranger, Binkley, & Rothe, 2002; Giardina, Ryan, Hubbard, & Binkley, 2001; Jackson, Mooney, &Schulze, 1997; Schenk & Jackson, 2002; White, Thornton, Running, & Nemani, 2002). In order to produce a single geo-dataset storing all the required information, climate, soil and land cover layers were combined in Land HomogeneousUnits (LHUs).

Among the numerous outputs calculated by the model, the most important to assess the soil productivity in the first 20 cmof soil thickness is its organic carbon content (gC/m2), chosen here as a proxy of the soil fertility.

CENTURY does not represent a spatially explicit model as it was originally developed to work at single site/point level. Inorder to apply spatial logic to the model, a tool was implemented in Visual Basic which sequentially processes the LHUattribute table, runs the CENTURY model against all the records (polygons) and writes the output to a text table. Finally thetext table was joined back to the LHU layer, exploiting the shapefile polygon geometry to provide a spatial dimension to themodel output. In this case, a loose coupling between the model and GIS was applied.

Model to evaluate soil erosion by water: RUSLEThe water erosion process consists of the removal of soil caused by rain impact, runoff and gravity and is the most

significant erosion phenomenon in the northern part of the Mediterranean basin (Zucca et al., 2006). Different kinds ofmodels were developed to predict erosion, either related to a single event or averaged over the long term, empirical orphysically based (e.g. Verheijen, Jones, Rickson, & Smith, 2009).

The RUSLE (Revised Universal Soil Loss Equation; Renard, Foster, Weesies, McCool, & Yoder, 1997) was chosen because it iswidely used and scientifically sound, as applied and verified for different landscape scenarios, even in very complextopographies. In this model, each variable influencing erosion is associated to an index representing the effect of the variableon erosion according to the magnitude of the specific index. RUSLE is expressed as follows:

A ¼ R$K$LS$C$P (4)

where A (t ha�1 year�1) is the computed soil loss per unit area, R (MJ mm ha�1 h�1 year�1) is the rainfall-runoff factor, K(t h MJ�1 mm�1) is the soil erodibility factor, LS (unitless) is the slope length/steepness factor, C (unitless) is the soil cover andmanagement factor and P (unitless) is the support practice factor, accounting for the management of protection of soil againsterosion.

For this work, the model was refined in order to ensure consistency in the production of the direct inputs for the RUSLEequation. In computing R and K factors, Arnoldus (1980) and Renard et al. (1997) equations were both implemented, whereasestimation of other factors was customized by implementing key processing features better suited to the case study and inaccordance with the available datasets.

The LS factor was computed from the Digital Elevation Model (DEM) integrating Tarboton (1997) and Nearing (1997)algorithms to evaluate hillslope length and steepness contributions in estimating the upslope surface draining water toeach pixel of the map. While the role of P factor is often neglected because of the lack of case studies to assess it, inthis work it was formulated to account for those areas that do not generate any flow (non erodible areas) and/or areasthat stop the flow (breaklines). Across the landscape these features are represented, for example, by the boundariesamong adjacent patches with different land uses (where there are often furrows to drain overland flow of water) or bythe roads and the railway lines (whose border often have bumps). Following the approach of Van Oost, Govers, andDesmet (2000), these two types of behaviour have been directly implemented into the LS factor computing scheme,introducing a zeroing of the drained area where non erodible spots or breaklines occur. This integration of P and LSfactors could also account for the effects of prevention or mitigation actions undertaken against erosion insidea specific land use type (e.g. ditches for water drainage) under the Regional guidelines for the Good AgriculturalPractices.

Concerning the C factor, the following relationship was adopted to derive it:

C ¼ 1� exp�

2:5$NDVI1� NDVI

�(5)

(Van der Knijff, Jones, & Montanarella, 2000), using as NDVI values the monthly averages from MODIS or AVHRR productsdepending on the reference period, and applying them the same gap-filling and downscaling procedures specificallydeveloped in the vegetation productivity model.

The type of GIS-model coupling used for RUSLE reflects the simply tight one, as the model (written in FORTRAN 90/95) isstand-alone, but it reads input and creates output data in GIS formats.

Model to evaluate soil erosion by wind: WEAMWind erosion consists of the removal and transport of particles: the largest particles can be moved through creep and

saltation, whereas the smallest ones move through suspension: this process is most significant for air pollution, damage tocrops and climatic impacts (e.g. Fratini, Santini, Ciccioli, & Valentini, 2009). Wind erosion often occurs when the soil is fine,dry and loose, the terrain is flat, there is no vegetation, the space is wide-open and the wind speed is sufficiently high toremove the soil. Among various existing schemes to evaluate wind erosion by suspension, the evolution of theories by

M. Santini et al. / Applied Geography 30 (2010) 394–415 403

Shao, Raupach, and Leys (1996) into the WEAM (Wind Erosion Assessment Model) was considered. Following Lu and Shao(2001), size-resolved vertical dust fluxes F(d) (gm�2 s�1) are predicted as:

FðdÞ ¼ 0:12$Ca$g$f $rb

p$QðdÞ (6)

where d is the particles diameter (m), Q(d) (kg m�1 s�1) is the size-resolved horizontal flux of saltating particles, Ca isa coefficient of order 1, g (m s�2) is the acceleration of gravity, f is the total volumetric fraction of dust in the sediment, rb is thesoil bulk density (kg m�3) and p (N m�2) is the soil penetration resistance. The last two parameters were assessed hereaccording to the textural information contained in the Pedologic Map of Sardinia. In the present study, the computing ofvertical fluxes F was limited to particles with diameters up to 40 mm considered as falling in the dust range.

The horizontal flux Q(d) can be before modelled as (Lu & Shao, 2001):

QðdÞ ¼(�

c$r$u3� =gh

1� ðu�tðdÞ=u� Þ2i

u� � u�tðdÞ0 u� < u�tðdÞ

)(7)

where the coefficient c is of order 1, r (kg m�3) is the air density, u* (m s�1) is the actual friction velocity and u*t (m s�1) is thethreshold friction velocity.

Concerning the friction velocity, it was thus estimated from:

u� ¼ u,k

ln

zz0

� (8)

where u is the wind speed measured at height z (10 m considering the available weather station dataset), k is the vonKarman’s constant (ca. 0.41), while z0 is the roughness length such that u¼ 0 when z¼ z0. Roughness length was associatedwith land cover classes considering the work of Pineda, Jorba, Jorge, and Baldasano (2004), distinguishing between spring-summer and fall-winter seasons.

Threshold friction velocity u*t, was assumed by Lu and Shao (2001) to be a function of the soil erodibility and was obtainedby correcting the threshold friction velocity u*t0 for bare and loose soil with two independent terms:

u�tðdÞ ¼ u�t0ðdÞ,SR,SH (9)

where

u�t0ðdÞ ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0:0123

ap$g$dþ 1:65$10�4

r,d

!vuut (10)

being ap the ratio between the density of particles and that of the air.In Eq. (9), SR accounts for the differences in the surface roughness with respect to the reference soil while SH accounts for

the differences in the moisture content. A further correcting term, accounting for the degree of soil aggregation and crustingcould be also present in Eq. (9); however, it has been assumed here to be equal to 1, because there is not a reliable way toestimate it at this stage. The procedures adopted to estimate SR and SH were consistent with Fratini et al. (2009).

In this study, great attention was given to the computing of time variable SH from daily soil moisture information firstlyusing the following formula to calculate the soil matric potential J:

j ¼�

R$TM

�ln RH (11)

where R is the gas constant (8.31 J mol�1 K�1), M is the molecular weight of water (ca.18 g mol�1), T (�K) is the air temperatureand RH (unitless) the air relative humidity, both derived by the grid-interpolation of the daily meteorological weather stationdata. The working hypothesis was that the moisture in the first millimetres of soil affected by wind erosion is in equilibriumwith the air moisture near the surface (Ravi & D’Odorico, 2005; Ravi, D’Odorico, Over, & Zobeck, 2004). Starting from the resultof Eq. (11), the volumetric content of water in surface soil (w, in m3 m�3) was determined using different algorithms, rep-resenting its relationship with J, as performed in Fratini et al. (2009), and valid for a broad range of texture information(Saxton, Rawls, Romberger, & Papendick, 1986), which in this study was derived from the Pedologic Map of Sardinia.

Computed emissions F were finally rescaled by a correction term Ev, corresponding to the vegetation cover fractionresponsible of soil protection. The following relationship was used:

Ev ¼ 1� expð�LAI=2Þ (12)

When applying Eq. (12) a random distribution of foliage above the soil and uniform leaf-angle distribution are assumed(e.g. Lu & Shao, 2001). The Leaf Area Index (LAI) data were derived from MODIS LAI product (MOD15A2) at 1000 m ofresolution for the actual period of analysis and applying MODIS LAI and AVHRR-NDVI relationships (Zhangsi & Williams, 1997)for the past period. In both cases, they are near-weekly data (8-day composite for MODIS and 7-day composite for AVHRR),

M. Santini et al. / Applied Geography 30 (2010) 394–415404

representative of each day in the considered interval. The same gap-filling and downscaling procedures already used for fPARand NDVI processing were also applied to LAI data.

The model was not originally a spatial model, running at plot/field scale (Shao et al., 1996). For the purpose of this work,a version enabled to spatially reproduce the phenomenon was implemented using FORTRAN 90/95 programming language, inthe form of a simply tight coupling.

Model to estimate salt water intrusion: SHARPIn coastal aquifers, if fresh and salt water are in equilibrium, the interface between the two domains remains stable. In the

case of over-pumping from coastal wells, it is possible that salt water moves inside the fresh water domain causing salini-zation. This leads to negative consequences if groundwater is used to satisfy human and agricultural needs. Several recentworks refer to this problem in many coastal areas of Sardinia (Aru et al., 2006; Balia, Ardau, Barrocu, Gavaudo, & Ranieri, 2009;Lecca & Cau, 2009).

To simulate the effects of water overexploitation, a modified version of the model SHARP (Essaid, 1990) was used in thisstudy. Among the inputs, the DEM and the geolithologic, geologic and hydrogeological maps were used to assess aquifercharacteristics. The database of the Italian Agency for the Protection of the Environment and the Technical Services (APAT),data-entered in the context of the National Law n. 464/1984, was used to gather necessary information related to thecharacteristics of wells (e.g. number, aquifer isolation, spatial distribution, depth, thickness of the open interval) and for thesuccessive interpolation of the water table level.

Besides the simulation of pumping effects, the calculation for the natural recharge of aquifers was introduced inthe model adopting a simplified water balance accounting for precipitation, evapotranspiration (Turc, 1955) andpotential infiltration based on a coefficient (CIP; Celico, 1988). Turc’s law for the real evapotranspiration (ET, mm)considers that:

ET ¼ Pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0:9þ P2

L2

�r (13)

where P is the annual mean precipitation (mm) and L¼ 300þ 25Tþ 0.05T3, with T (�C) being the annual mean temperature.At this point effective precipitation (Peff), generating runoff and infiltration, is calculated as:

Peff ¼ P � ET (14)

Concerning precipitation and temperature datasets, the annual averages calculated from the interpolated weather stationdata were used, considering the entire temporal interval of 10 years ending at the reference periods of simulation, in order toaccount for two mean climatic years (one for the early nineties and one for the actual period).

Then CIP was codified (Celico, 1988) according to the permeability attribute contained in the geolithologic map (Table 4)where, under the field ‘code’, classes from 1 to 4 indicate an increasing degree of permeability, whereas ‘f’ indicatespermeability by cracking, ‘c’ is the permeability by karstification and ‘p’ represents permeability by porosity.

As model output, maps showing the extension of the intrusion were produced, using indices to specify theunderground conditions: 0.5 indicates that below a given point there is fresh water, 0.75 was used for mixed water(i.e. the interface), 1 for salt water and 0 for a non-coastal aquifer (i.e. not interested by the simulated phenomenon).

In this application, the model, written in FORTRAN 90/95, was adapted to GIS formats. All the input data, such as geometryand hydrodynamic attributes of aquifers (grid layers) or well location, depth and pumping rate (point layers), were manageddirectly in the GIS environment in order to create only grids necessary for the model. This is a simply tight coupling, as themodel is stand-alone, but it reads input and writes output data in GIS formats.

Table 4Potential Infiltration Coefficient (CIP) used to calculate natural recharge in SHARP, codified (Celico, 1988) according to the permeability map of Sardinia.

Code Lithologic types CIP

4f Argillite, clayey/schist, schist, marl, serpentine 0.154p Clay 0.104pf Schist and marl with arenaceous and calcareous levels 0.203p Moraine, lacustrine deposit, clay and sandy clay 0.303f Gneiss, granite, gabbro, quartzite, marly limestone 0.403fc Limestone and dolomitic limestone with marly intercalation 0.503pf Flysch, sandstone, granite, altered granite, syenite 0.502p Fine sand, silt, conglomerate, incoherent molasse 0.602f Volcanic rock, compact limestone 0.702pf Conglomerate and alluvium more or less cemented, travertine, little cemented sandstone 0.802fc Dolomite and dolomitic limestone 0.801p Alluvial deposits with prevalent sand and silt, debris flow, conoids, beaches and dunes 0.90

M. Santini et al. / Applied Geography 30 (2010) 394–415 405

Integrated Desertification Index (IDI)The final phase of the proposed method consisted of deriving an Integrated Desertification Index (IDI) (Santini, 2008)

assuming that:

(i) both assessment of actual state of degradation and changes to the degradation over time are important: here the use ofjust two time periods was constrained by their input data availability and suitability;

(ii) the simulated degradation processes have different significance in favouring the comprehensive land deterioration inthe island;

(iii) the quality (e.g. the level of subjectivity, the systematic or random errors in acquisition or digitalization, the employmentof assessed algorithm for interpolation or more or less sophisticated pre-processing procedures) and the resolution(spatial detail) of the thematic input dataset are not homogeneous over the time scale of analysis and this has to beaccounted for in evaluating the changes from the past to the present;

(iv) the gradual increase of a single index of degradation should be reflected in an accelerated increase of the IDI values: ineffect there are often unknown or hidden dynamics for which degradation processes, when worsening, reinforce eachother.

The output grids from CENTURY, MOD17, RUSLE and WEAM models containing continuous quantities were normalizedchoosing threshold values for each model: beyond such a threshold one can assume degraded conditions for the specificprocess simulated by the model. Within the different model outputs, alternative thresholds were chosen on a land use basis:e.g. it is evident that the same amount of soil loss for erosion has a different meaning when it occurs in agricultural areas thanin forest areas; and similar assumptions are valid for the other processes. This is a way to account for risk and not simply thevulnerability (e.g. Al-Adamat et al., 2003).

In order to use the same scale of values than for CAIA and SHARP models, the normalized quantities were thus stretchedbetween 0 (representing the finest conditions) and 1 (representing fully degraded conditions). Any index from 0 to 1 wasassumed as the ‘degradation index’ for the corresponding simulated process.

The proposed IDI relationship calculated the arithmetic mean of weighted exponential functions of above degradationindices and of their variation from the past, applying weights according to both the importance of the single process and thequality of input data used in its simulation. The formula to derive IDI was the following:

IDI ¼ 1N

XN

i¼1

"ePDIDi

,DIDi

e,

ePIDi,IDi

e

#(15)

where N is the number of considered processes, i represents the single process, IDi represents the degradation index of the ithprocess for the present and DIDi indicates the variation from the past of the ith degradation index. Finally PIDi is the weightassigned to the process i, whereas PDIDi is the weight assigned to the variation of the index i between the two consideredperiods. The weights for the different land degradation processes must range from 0 to 1. In this study, they were chosencombining: (a) local research institutes analysis and technical offices plus information from the relevant literature (Aru, 1986;Costantini et al., 2004; Motroni et al., 2009) and (b) the difference in input data detail and quality for the different models andfor the different time periods (see Table 1).

To address the first issue, the Analytic Hierarchy Process (AHP; Saaty, 1977) was selected as the most appropriate methodfor deriving the weight assigned to each factor. For the second point, a statistical analysis (i.e. Principal Component Analysis,PCA) was conducted taking into account two sets of data: (i) the output maps of the six models separately for each year; (ii)the two (past and present period) output maps, separately for each model. By cross-analysing, the results from AHP and PCAthe weights reported in Table 5 were selected as the most suitable.

The IDI was then classified into five classes of increasing by severity (Table 6), discriminated applying a natural breakalgorithm. This identifies break points by choosing the class divisions which best group similar values and maximize the

Table 5Weights and variations of the indices used in IDI calculation.

Model output and variations Weights

CAIA 1.00Variation CAIA 0.25RUSLE 1.00Variation RUSLE 0.25MOD17 0.80Variation MOD17 0.20CENTURY 0.60Variation CENTURY 0.15WEAM 0.40Variation WEAM 0.10SHARP 0.20Variation SHARP 0.05

Table 6Range of value and description of IDI risk classes.

Risk Class Range of IDI values

Very Low (not subjected areas) 0.00–0.15Low (potential areas) 0.15–0.30Medium (fragile areas) 0.30–0.50High (critical areas) 0.50–0.70Very high (degraded areas) 0.70–1.00

M. Santini et al. / Applied Geography 30 (2010) 394–415406

differences between classes. Recalling the class description for desertification sensitivity from Kosmas et al. (1999), the verylow risk class (1, i.e. Not Subjected Areas) represents areas where the critical factors are very low or absent, with a goodequilibrium between environmental and socio-economic components. The low risk class (2, i.e. Potential Areas) indicatesareas menaced by desertification if poor land management practices are applied or when the influence of adjacent degradedareas causes severe problems: abandoned lands not properly managed could be included. The medium risk class (3, i.e. FragileAreas) represents areas where any change in the delicate equilibrium between the natural environment and human activitiescan lead to desertification. The high risk class (4, i.e. Critical Areas) corresponds to degraded lands characterized bya precarious equilibrium between the natural environment and human activities. Finally very high risk class (5, i.e. DegradedAreas) represents areas already seriously degraded because of past poor management, representing a menace for surroundingareas or with evident desertification processes.

Software design

The framework based on the IDI methodology was designed featuring two main characteristics:

- flexibility, as it allows possibility either to substitute a model simulating a given environmental degradation process withanother one according to the study area, or to work at different spatial scales and resolutions;

- user-friendliness, it contains a straightforward dialog box driving the non-GIS expert users step by step in input dataadding.

The computing made through the IDI tool can be ideally grouped into five sub-modules representing the frameworkfollowed by implementing the integrated index (Fig. 3).

� INPUT DATA module: it controls the input data setting, including the definition of land use layers to be used for normal-ization procedure;

� NORMALIZE module (only for RUSLE, WEAM, CENTURY and MOD17 models): it controls all the statistical operations ona land use basis in order to derive a threshold critical value to normalize the model results on a scale from 0 to 1. Thispermits to obtain all the required unitless degradation indices for the IDI formula (in addition to SHARP and CAIA resultswhich are already dimensionless).

� DIFFERENCE module: it calculates the differences between the indices of the two periods of simulation for each degradationprocess;

� IDI CALCULATION module: the whole Eq. (15) is applied in order to derive the integrated desertification index stretchedbetween 0 and 1, representing the lowest and the highest conditions of risk, respectively. This module requires the weightsto be given to each degradation process and to its index variation.

� CLASSIFY module: the IDI is classified into severity levels as shown in Table 6.

The program was designed to be integrated into ArcGIS.9.x using an add-on toolbar developed by combining ArcObjectand Visual Basic programming languages. This tool supports step by step the end-user through the IDI application providinga user-friendly interface. The tool consists of two main steps (windows): the first one allows selecting or updating the inputparameters, defining the geographic area of interest and deciding whether to calculate the classified IDI or simply thenormalized outputs for all the models. The second window allows for setting relative weights for single indices and theirvariation (PIDi and PDIDi in Eq. (15)), and to choose the output grid name.

Results and discussion

The entire simulation experiment took less than 24 h to complete: the characteristics of the computer used in thisapplication are listed in Appendix A.

Considering the different possibilities of integrating modelling and GIS, the level of coupling, influenced by thecomplexity of the model, impacted on the time required by the single model simulation Level of coupling andcomplexity of the model impact on the time required by the single model simulations, and need to be considered whenintegrating modelling and GIS. WEAM took the longest simulation time and had the highest impact on the internal

INPUT DATA

LUA

CAA

SA

RA

WA

CEA

MA

LUP

CAP

SP

RP

WP

CEP

MP

NORMALIZE

RA

WA

CEA

MA

RP

WP

CEP

MP

DIFFERENCE

ΔCA ΔS ΔR ΔW ΔCE ΔM

IDI

calculation

CLASSIFY

CLASS

INTERVALS

(Table 6)

IDI map

Desertification RISK

map

WEIGHTS

INPUT DATA

LUA

CAA

SA

RA

WA

CEA

MA

LUP

CAP

SP

RP

WP

CEP

MP

NORMALIZE

RA

WA

CEA

MA

RP

WP

CEP

MP

DIFFERENCE

ΔCA ΔS ΔR ΔW ΔCE ΔM

IDI

calculation

CLASSIFY

CLASS

INTERVALS

(Table 6)

IDI map

Desertification RISK

map

WEIGHTS

LUA

CAA

SA

RA

WA

CEA

MA

LUP

CAP

SP

RP

WP

CEP

MP

NORMALIZE

RA

WA

CEA

MA

RP

WP

CEP

MP

DIFFERENCE

ΔCA ΔS ΔR ΔW ΔCE ΔM

IDI

calculation

CLASSIFY

CLASS

INTERVALS

(Table 6)

IDI map

Desertification RISK

map

WEIGHTS

Fig. 3. Flowchart representing the scheme followed to implement IDI through sub-modules. CA, S, R, W, CE, M stand for CAIA, SHARP, RUSLE, WEAM, CENTURYand MOD17 outputs, respectively; LU indicates the land use layer; subscripts ‘‘A’’ and ‘‘P’’ indicate actual and past periods of simulation, respectively, whileunderlines indicate normalized outputs. Finally, ‘D’ prefix represents the variation of indexes between two reference periods.

M. Santini et al. / Applied Geography 30 (2010) 394–415 407

resource of the processor as it processes daily climatic data and calculates numerically two integrations, one for thediameter interval of saltating particles and one for the diameter interval of suspended ones. The CENTURY model, due tothe complexity of processes that it simulates, required a long processing time as well. RUSLE has proven to be a fastmodel, showing that the longest computing time is required by the flow routing algorithm that connects cells from thewatershed boundaries to the outlet of each basin to calculate the LS factor. SHARP ran separately for the ten smallcoastal aquifers considered to be affected by seawater intrusion, therefore, despite the complex physical processes that itsimulates the run was somewhat fast. Finally CAIA and MOD17, consisting of an overlay (summing and/or multiplying) ofvector or grid layers exploiting the GIS utilities, showed to be the fastest models. From the evaluations above, it ispossible to confirm that a loose coupling was the best solution for the more advanced applications (i.e. CENTURY). In thecase of relatively simple models, a simply tight integration with GIS (i.e. RUSLE, WEAM, SHARP) or a fully tight integrationwithin GIS (i.e., MOD17, CAIA) is recommended, as these applications are easier to develop and allow a quick datacombination and interpretation.

Fig. 4. Spatial distribution of areas at high and very high desertification risk, considered as hot-spots of desertification.

M. Santini et al. / Applied Geography 30 (2010) 394–415408

Regarding the IDI, despite the significant dimensionality of the case study, which consisted in a grid of the order of 107

pixels at 50 m of resolution, the running time was contained (taking up to 50 min to complete). This provided the possibilityfor multiple tests of IDI results against different weighting parameters.

From the classified IDI output map, referring to the hot-spot classification by LADA (Land Degradation Assessment inDrylands, http://www.fao.org/nr/lada/), two types of hot-spot were considered:

- areas with high productivity seriously menaced by degradation (class 4, high risk hot-spots);- degraded areas menacing surrounding areas (class 5, very high risk hot-spots).

Fig. 4 shows three clusters of very high risk hot-spots, surrounded by more scattered high risk hot-spots, and located in theextreme north-western, north-eastern and centre-southern part of the island, respectively.

Table 7 shows that about 16.4% of territory is occupied by such hot-spots. In addition, classifying the land use map into fourbroad categories: forest lands, agricultural lands, pasturelands and other lands, where ‘pasture land’ includes all areas usedfor grazing (this was assessed by aerial image interpretation), hot-spot areas emerged as occupying 7.7% of forested lands, 8.1%of agricultural lands, 18.2% of pasture lands and only 5% of other lands (mainly comprising coastal sandy or bare rock areas,which can be defined as ‘naturally’ desertified). This is further proof of the interaction between grazing and low vegetationcover predisposing the territory to further degradation processes (e.g. erosion).

At first evaluation, results of classified IDI were verified by conducting field survey campaigns during two differentseasons, in December 2006 and in May 2007. The validation points were selected according to their accessibility, to thesuggestions by local experts and to the primary results of the desertification risk assessment.

The survey, primarily performed along the hot-spot areas of Fig. 4, showed a good agreement with the classified IDIoutcomes (Table 8). The general landscape conditions during the survey trip (black line in Fig. 5) were observed, while inselected sites (white circles in Fig. 5) a more in-depth examination was carried out, by analysing throughout the plotmainly macroscopic soil, vegetation state, and land management practices. This was then documented by photographsincluded in the validation database. Specifically 14 of 24 surveyed sites confirmed as critical areas, showing clear signs ofdegradation due to either the lack of vegetation cover due to consumption by grazing, or the reduction of soil thicknessbecause of compaction by grazing followed by soil removal for erosion. Then 8 of 24 site points were confirmed as havingno signs of degradation, while 2 areas showed an example of amelioration procedures applied on formerly degraded areas.

Table 7Area (Ha and %) of each hot-spot type derived from IDI for the whole study area and land use category.

Domain Hot-spots 4 (Ha) Hot-spots 5 (Ha) Hot-spot 4 (%) Hot-spot 5 (%) Total (%)

Whole area 3580994 350850 14.90 1.49 16.39Agricultural land 110512 925 7.50 0.60 8.10Forest land 310897 20763 7.10 0.60 7.70Pasture land 2860099 260436 16.70 1.50 18.20Other 100536 10872 4.20 0.80 5.00

M. Santini et al. / Applied Geography 30 (2010) 394–415 409

Both were well detected by classified IDI and confirmed by local land managers. In this case, it was concluded that in 90%of cases the ground-truth matched with the IDI assessment. This was just a first evaluation of model performances andconstrained by the accessibility of sites. Moreover, it was not easy to distinguish during field observation between the twotypes of hot-spots without specific and more in-depth soil, vegetation and land productivity analyses able to verify thesingle IDI components.

Field inspections reinforced the hypothesis that overgrazing and soil erosion are the most important factors affecting thedesertification process in the study area, supporting the choices made in weighting the different degradation processes. Fig. 6shows an example of such severe situation for the Irgoli’s territory (north-eastern sector), where above described features areevidently leading to degraded conditions and manifest as incise channels with considerable soil removal until bedrockoutcropping. In general, degraded zones consist of wide sectors comprising early signs of worsening also in the portion of theterritory not directly interested by the source of degradation (i.e. soil removing or vegetation cover reduction). This is due to‘neighbouring’ interactions and off-site effects, e.g. grazing migration towards the borders of previously grazed lands orburying of fertile soil by eroded sediment deposition.

A detailed study of the complex land cover dynamics in these areas suggests that the misuse and the abuse of existingpastureland and the creation of new pastures on unsuitable marginal terrain favoured severe land degradation processes. The

Table 8List of sites surveyed for validation purpose. Coordinates (decimal degrees) and class of risk derived from IDI application are reported together witha description of the conditions as observed during the field survey.

SITE LONa LATa HAs Description

1 9.032 39.446 4 Agricultural area with several signs of rill incision2 9.046 39.475 no Agricultural area without sign of degradation3 9.053 39.511 5 Area lacking of vegetation cover, devoted to grazing, sloped and

with signs of grazing amelioration procedures4 9.039 39.582 4 Agricultural areas surrounded by highly grazed area5 8.920 39.567 no Signs of ploughing on gently sloped areas, deep soil without stoniness.

No degradation sign6 8.959 39.598 no7 9.025 39.635 4 Sloped ploughed areas with some signs of rill incisions.8 8.931 39.656 no Signs of ploughing on gently sloped areas, deep soil without stoniness.

No degradation sign.9 9.244 40.313 5 Steep area with alternation of sclerophyllous vegetation and bare/grass cover,

showing signs of irregular land-work often in the direction of slope forpasture seeding, also signs of rill/sheet erosion.

10 9.456 40.419 4 Sparsely vegetated (shrubs) and very sloped areas with diffuse sign ofploughing in the direction of slope, some sign of rill/sheet erosion.

11 9.467 40.461 412 9.654 40.454 4 Land interested by grazing, with sparse or low vegetation (shrubs),

degraded soil with high stoniness, with deep incision due to rillerosion, up to outcropping of mother rock.

13 9.693 40.476 514 9.704 40.480 415 8.670 40.533 4 Areas devoted to grazing and intensive and extensive agriculture.

Many signs of soil instability and degradation16 8.628 40.600 417 8.563 40.607 518 8.485 40.660 419 8.471 40.743 no Agricultural land, with gently slope, without signs of degradation20 8.639 40.741 no21 8.289 40.746 no22 8.194 40.735 4 Agricultural areas also devoted to grazing, with medium slope.

With signs of degradation.23 8.204 40.781 4 Agricultural areas also devoted to intense grazing, with high slope.

With signs of degradation.24 8.243 40.831 no Agricultural areas also devoted to grazing, with medium slope.

With signs of degradation.

a coordinates are in decimal degrees.

Fig. 5. Map showing field survey points (white circles), road track (black line) and indication location (white outlined box) of the sector described in Fig. 6.

M. Santini et al. / Applied Geography 30 (2010) 394–415410

agro-pastoral practices, including tillage and fires, are frequently carried out along the maximum slope gradient regardless ofmorphological and soil conditions and often just before rainy periods (Enne, Pulina et al., 2000). Given the loss of vegetationcover and increased of soil erodibility due to these practices, erosion processes become widespread and intense (D’Angeloet al., 2000; Zucca et al., 2006).

Although there is good correspondence between IDI results and field verification, at this time the mesoscale of theapplication at fine spatial resolution limits the possibility of extensive validation. However, as in other vulnerability/risk hot-spots analyses, such a scale represents the first step of a more comprehensive ‘top-down/bottom-up’ procedure, serving toindividuate the most critical situations where more detailed analyses are necessary to upscale the results. A preliminarycoarse scale evaluation allows evaluating differences in exposure and sensitivity; for the most severe situations a finerresolution could be useful to resolve local features. Finally, a ‘regionalization’ procedure can improve previous coarser scaleevaluations (Venevsky, 2006).

It is noteworthy that the two largest hot-spot areas (north-west and centre-south) appear to confirm two large criticalsectors detected by previous work performed at the regional level (Motroni et al., 2009) and carried out via the EnvironmentalSensitive Area (ESA) methodology of Kosmas et al. (1999), based on static indicators and calibrated over the Sardinian casestudy. Such work was completed for about 68% of the island territory at scale 1:100 000 and, if compared with either classifiedIDI results or field survey, shows an evident overestimation of the risk.

It is difficult to compare the ESA-based work with IDI results because of divergence between initial assumptions and finalinterpretation, unless trying to harmonize inputs and outputs.

A more coherent evaluation between the ESA and IDI methods for such an area was conducted by fine-tuning the IDI inorder to compare it to the formulation of the ESA approach. To this aim: (i) only 4 degradation processes were considered(overgrazing, vegetation productivity, soil productivity , water erosion) to meet indicators considered by ESA; (ii) only theexponential factors of actual conditions were kept, fixing DIDi¼ 0 in Eq. 15; (iii) all the considered degradation factors were

Fig. 6. (Top) Aerial photograph over a surveyed location (Irgoli’s area, see white outlined box in Fig. 5) with hot-spot area boundary (white line) overlay: fieldsurvey confirmed the area to be grazed with evident erosion signs (bottom left) leading to total loss of soil and outcropping of mother rock (bottom right).

M. Santini et al. / Applied Geography 30 (2010) 394–415 411

assumed having the same weight (i.e., PIDi¼ 1 in Eq. (15)). Such applications are referred to as simplified-IDI, to which theclassification in Table 6 was thus applied.

The ESA index, ranging from 1 to 1.85, was stretched between 0 and 1 and still classified according to Table 6(reclassified-ESA), in order to match the scale and the ranking of the classified simplified-IDI. Table 9 shows, for the part ofterritory covered by the ESA study, the percentage of area in which the two compared approaches differ for four, three,two, one or no class of risk. Also this suggests where the reclassified-ESA overestimates (positive differences) or under-estimates (negative differences) if compared with the classified simplified-IDI. About 41% and 88% of the sample presentsa perfect and a good agreement, respectively, where ‘good’ means there is maximum one class of difference between thetwo indices.

Table 9Percentages showing the differences between reclassified-ESA and classified simplified-IDI desertification risk class values, after harmonization of accountedprocesses and evaluation scales, over the area extent investigated by Motroni et al. (2009).

Class difference % of Area

�4 0.00�3 0.21�2 3.79�1 20.700 41.221 25.742 7.253 1.044 0.05

M. Santini et al. / Applied Geography 30 (2010) 394–415412

The possibility for IDI to include additional processes, appropriate weights for the different processes, and the rate ofdegradation worsening from the past is obviously an added value if compared with the previous approach.

Conclusions

Desertification affects different types of environments and is a complex issue involving interrelated factors from bothenvironmental and anthropogenic systems (e.g. Adamo & Crews-Meyer, 2006). Thus any decision focusing on desertificationimpact mitigation and prevention requires deep research about the characteristics of the area under analysis. The IDI tool,whose formulation and software structure are here presented, is an innovative tool satisfying the need to not only detectdegraded areas, but also to understand for which of these areas interventions are more urgent. IDI integrates results of varioustypes of models, simulating environmental processes, with GIS, using different degrees of coupling strength, balancing modelsophistication and complexity. In particular, this tool allows the production of desertification risk maps in a manner that isboth easily read and repeated for non expert GIS-user (e.g. Cinderby & Forrester, 2005).

The first application of the methodology to the Sardinia Island case study appeared promising. The field assessmentconfirmed good correspondence between the ground-truth conditions and the desertification risk map, with the simulatedhot-spot areas showing clear signs of degradation when identified in the field. These results point out that IDI is an efficientregional scale tool for detecting areas at risk and taking political actions to mitigate and prevent land degradation processesfrom triggering desertification. Regional scale mapping of areas at risk represents the initial step towards detecting siteswhere more refined investigations and urgent interventions are needed (e.g. Li, Peterson, Liu, & Qian, 2001), and allocatingthe required resources.

Unlike the application described, the use of alternative models is possible when the processes driving desertification inanother territory are different, or when the available input datasets for the selected modelling scheme are limited. It ispossible to apply the methodology at different spatial levels according to the extent of the study area, deciding on the mostrepresentative processes to model according to the scale. Variations in specific process weights allows several classified IDImaps to be quickly produced, compared and verified, also predicting different desertification impacts. This helps decisionmakers to develop suitable land management policies from a range of alternatives in order to mitigate adverse effects.

The tool can be further improved by supplying the models with a more complete dataset following increased productionand availability of thematic maps. This is also useful for validation stage, supporting the advances in desertification assess-ment procedures.

Acknowledgements

This study was developed inside the project: ‘‘Progettazione e realizzazione di un Sistema Informativo Geografico (GIS) peril monitoraggio delle aree della Sardegna a rischio di desertificazione, con la specifica indicazione delle componenti di talerischio, compresa la parametrazione delle stesse’’, funded by the Regional Administration of Sardinia (RAS) and implementedby Agriconsulting SpA (Rome, Italy) and ESRI-Italia (Rome, Italy) (Contract n. 2237 - 23 Dec. 2004). F. Leuzzi, N. Sannio and F.Cotza from RAS and P. Molinari from Agriconsulting are acknowledged for their in itinere careful review of the work. Authorsalso thank I. Iocola, D. Pittalis and D. Soru from InTReGa s.r.l. (Sassari, Italy) for their efforts in the CAIA model implementationand their assistance during the field campaigns; and C. Belli, B. Gioli, and P.L. Porta from Terrasystem (Viterbo, Italy) for theircontribution in implementing the MOD17 and CENTURY models. A special thank to three anonymous reviewers and Editor fortheir suggestions helpful to improve the manuscript. Last but not least, authors are indebted to Sophie Williams fromUniversity of Wollongong for her thorough language editing and proofreading.

Appendix A.

Technical characteristics of the computer used for the simulation.

- Dell Precision PWS 380.- Hard Disk: 150 GB.- RAM: 2 GB.- MultiProcessor: (2) Intel Pentium IV 3.00 GHz.- OS: Windows XP Professional SP2.- Ethernet Card: Broadcom NetXtreme 57xx Gigabit Controller.- LAN: 1 Gbps.- Controller SCSI: Intel 82801 GR/GH SATA RAID Controller.- VGA Card: NVIDIA Quadro NVS 285.- Paging: 4 GB.

References

Adamo, S. B., & Crews-Meyer, K. A. (2006). Aridity and desertification: exploring environmental hazards in Jachal, Argentina. Applied Geography, 26, 61–85.

M. Santini et al. / Applied Geography 30 (2010) 394–415 413

Al-Adamat, R. A. N., Foster, I. D. L., & Baban, S. M. J. (2003). Groundwater vulnerability and risk mapping for the Basaltic aquifer of the Azraq basin of Jordanusing GIS, Remote sensing and DRASTIC. Applied Geography, 23, 303–324.

Al-Sabhan, W., Mulligan, M., & Blackburn, G. A. (2003). A real-time hydrological model for flood prediction using GIS and the WWW. Computers, Envi-ronment and Urban Systems, 27, 9–32.

Arnalds, A. (2006). Desertification in a humid environment: an example from iceland. In R. Lal (Ed.) (2nd ed.).Encyclopedia of soil science, Vol. 1. doi:10.1081/E-ESS-120016594.

Arnoldus, H. M. J. (1980). An approximation of the rainfall factor in the universal soil loss equation. In M. De Boodt, & D. Gabriels (Eds.), Assessment of erosion(pp. 127–132). England: Wiley & Sons.

Aru, A. (1986). Aspects of desertification in Sardinia-Italy. In R. Fantechi, & N. S. Margaris (Eds.), Desertification in Europe (pp. 194–198). Dordrecht: ReidelPublishing Co.

Aru, A., Tomasi, D., Vacca, A. (2006). Aspects of environmental degradation in Sardinia. In Watershed management: Water Resources for the Future.Conference Proceedings, October 22-24th, 2003, Porto Cervo, Sassari, Italy.

Augusto, L., Ranger, J., Binkley, D., & Rothe, A. (2002). Impact of several common tree species of European temperate forests on soil fertility. Annals of ForestScience, 59, 233–253.

Bajocco, S., & Ricotta, C. (2008). Evidence of selective burning in Sardinia (Italy): which land-cover classes do wildfires prefer? Landscape Ecology, 23,241–248.

Balia, R., Ardau, F., Barrocu, G., Gavaudo, E., & Ranieri, G. (2009). Assessment of the Capoterra coastal plain (southern Sardinia, Italy) by means of hydro-geological and geophysical studies. Hydrogeology Journal, 17, 981–997.

Basso, F., Bove, E., Dumontet, S., Ferrara, A., Pisante, M., Quaranta, G., et al. (2000). Evaluating environmental sensitivity at the basin scale through the use ofgeographic information systems and remotely sensed data: an example covering the Agri basin (Southern Italy). Catena, 40, 19–35.

Batty, M., & Xie, Y. (1994). Modeling inside GIS: part 1. Model structures, exploratory spatial data analysis and aggregation. International Journal ofGeographic Information Systems, 8, 291–307.

Brandmeyer, J. E., & Karimi, H. A. (2000). Coupling methodologies for environmental models. Environmental Modelling and Software, 15, 479–488.Brown, D. G., Riolo, R., Robinson, D. T., North, M., & Rand, W. (2005). Spatial process and data models: toward integration of agent-based models and GIS.

Journal of Geographical Systems, 7, 25–47.Campbell, N. A. (1990). Biology. Redwood City, CA: The Benjamin/Cummings Publishing Company, Inc.Celico, P. (1988). Prospezioni Idrogeologiche, Vol. 1 and 2. Liguori Editore. (in Italian).Chang, J. T., & Wetzel, P. J. (1997). Effects of spatial variation of soil moisture and vegetation on the evolution of a prestorm environment: a numerical case

study. Monthly Weather Review, 119, 1368–1390.Cinderby, S., & Forrester, J. (2005). Facilitating the local governance of air pollution using GIS for participation. Applied Geography, 25, 143–158.Costantini, E. A. C., Bocci, M., L’Abate, G., Fais, A., Leone, G., Loj, G., et al. (2004). Mapping the state and risk of desertification in Italy by means of remote

sensing, soil GIS and the EPIC model. Methodology validation in the Sardinia island, Italy. International Symposium: Evaluation and monitoring ofdesertification. Synthetic activities for the contribution to UNCCD, 2nd February 2004. Tsukuba, Ibaraki, Japan: NIES Publication.

D’Angelo, M., Enne, G., Madrau, S., Percich, L., Previtali, F., Pulina, G., et al. (2000). Mitigating land degradation in Mediterranean agro-silvo-pastoral systems:a GIS-based approach. Catena, 40, 37–49.

Danfeng, S., Dawson, R., & Baoguo, L. (2006). Agricultural causes of desertification risk in Minqin, China. Journal of Environmental Management, 79, 348–356.Ding, H., Chen, J., & Wang, G. (2009). A model for desertification evolution employing GIS with cellular automata. International Conference on Computer

Modeling and Simulation, 2009, 324–328.Dragan, M., Sahsuvaroglu, T., Gitas, I., & Feoli, E. (2005). Application and validation of a desertification risk index using data for Lebanon. Management of

Environmental Quality: An International Journal, 16, 309–326.Dregne, H., Kassas, M., & Rozanov, B. (1991). A new assessment of the world status of desertification. Desertification Control Bulletin, 20, 6–18.Durao, R., Pereira, M. J., Costa, A. C., Corte-Real, J. M., & Soares, A. (2009). Indices of precipitation extremes in Southern Portugal – a geostatistical approach.

Natural Hazards and Earth System Sciences, 9, 241–250.Enne, G., Pulina, G., d’Angelo, M., Previtali, F., Madrau, S., Caredda, S., et al. (2000). Agro-pastoral activities and land degradation in Mediterranean areas: case

study of Sardinia. In N. A. Geeson, C. J. Brandt, & J. B. Thornes (Eds.), Mediterranean desertification: A mosaic of processes and responses. Wiley & Sons, Ltd.Enne, G., Zanolla, C., & Peter, D. (2000). Desertification in Europe: Mitigation strategies, land-use planning. In Proceedings of the Advanced Study Course,

Alghero, 30/5-9/6 1999 (pp. 509). European Commission, EUR 19390.Essaid, H. I. (1990). A multilayered sharp interface model of coupled freshwater and saltwater flow in coastal systems: model development and applications.

Water Resources Research, 26, 1431–1454.Fang, L., Bai, Z., Wei, S., Yanfen, H., Zongming, W., Kaishan, S., et al. (2008). Sandy desertification change and its driving forces in western Jilin Province,

North China. Environmental Monitoring and Assessment, 136, 379–390.FAO/UNEP. (1984). Provisional methodology for assessment and mapping of desertification. Rome: Food and Agriculture Organization of the United Nations,

United Nations Environmental Programme. 84 pp.Fedra, K. (1993). GIS and environmental modeling. In M. F. Goodchild, B. O. Parks, & L. T. Steyaert (Eds.), Environmental modeling with GIS (pp. 35–50). New

York: Oxford University Press.Feoli, E., Giacomich, P., Mignozzi, K., Ozturk, M., & Scimone, M. (2003). Monitoring desertification risk with an index integrating climatic and remotely-

sensed data: an example from the coastal area of Turkey. Management of Environmental Quality, 14, 10–21.Fiori, M., Motroni, A., Duce, P., & Spano, D. (2004). A daily water balance estimate for climate risk evaluation at a local scale. In: Acta Horticulturae, Vol. 664.

ISHS. IV International Symposium on Irrigation of Horticultural Crops.Fratini, G., Santini, M., Ciccioli, P., & Valentini, R. (2009). Evaluation of a wind erosion model in a desert area of Northern Asia by Eddy covariance. Earth

Surface Processes and Landforms, 34, 1743–1757. doi:10.1002/esp.1857.Gad, A., & Lofty, I. (2008). Use of remote sensing and GIS in mapping the environmental sensitivity areas for desertification of Egyptian territory. eEarth

Discussion, 3, 41–85.Geist, H. J., & Lambin, E. F. (2004). Dynamic causal patterns of desertification. Bioscience, 54, 817–829.Gemitzi, A., & Tolikas, D. (2007). HYDRA model: simulation of salt intrusion in coastal aquifers using Visual Basic and GIS. Environmental Modelling &

Software, 22, 924–936.Giardina, C. P., Ryan, M. G., Hubbard, R. M., & Binkley, D. (2001). Tree species and soil textural controls on carbon and nitrogen mineralization rates. Soil

Science Society of America Journal, 65, 1272–1279.Giordano, F., & Marini, A. (2008). A landscape approach for detecting and assessing changes in an area prone to desertification in Sardinia (Italy). Inter-

national Journal of Navigation and Observation, 2008. Article ID 549630, 5 pages.Grunblatt, J., Ottichilo, W. K., & Sinange, R. K. (1992). A GIS approach to desertification assessment and mapping. Journal of Arid Environments, 23, 81–102.Haijiang, L., Chenghu, Z., Weiming, C., En, L., & Rui, L. (2008). Monitoring sandy desertification of Otindag Sandy Land based on multi-date remote sensing

images. Acta Ecologica Sinica, 28, 627–635.Hanafi, A., & Jauffret, S. (2008). Are long-term vegetation dynamics useful in monitoring and assessing desertification processes in the arid steppe, southern

Tunisia. Journal of Arid Environments, 72, 557–572.Hellden, U. (2008). A coupled human-environment model for desertification simulation and impact studies. Global and Planetary Change, 64, 158–168.Hellden, U., & Tottrup, C. (2008). Regional desertification: a global synthesis. Global and Planetary Change, 64, 169–176.Hill, J., Stellmes, M., Udelhoven, T., Roder, A., & Sommer, S. (2008). Mediterranean desertification and land degradation. Mapping related land use change

syndromes based on satellite observation. Global and Planetary Changes, 64, 146–157.

M. Santini et al. / Applied Geography 30 (2010) 394–415414

Huang, B., & Jiang, B. (2002). AVTOP: a full integration of TOPMODEL into GIS. Environmental Modelling & Software, 17, 261–268.Ibanez, J., Martinez, J., & Puigdefabregas, J. (2008). Assessing desertification risk using system stability condition analysis. Ecological Modelling, 213, 180–190.Ibanez, J., Martinez, J., & Schnabel, S. (2007). Desertification due to overgrazing in a dynamic commercial livestock-grass-soil system. Ecological Modelling,

205, 277–288.Jackson, R. B., Mooney, H. A., & Schulze, E. D. (1997). A global budget for fine root biomass, surface area, and nutrient contents. Proceedings of the National

Academy of Sciences, USA, 94, 7362–7366.Karimi, H. A., & Houston, B. H. (1996). Evaluating strategies for integrating environmental models with GIS: current trends and future needs. Computers,

Environment and Urban Systems, 20, 413–425.Kassas, M. (1995). Desertification: a general review. Journal of Arid Environments, 30, 115–128.Kosmas, C., Kirkby, M., & Geeson, N. (1999). The MEDALUS project: Mediterranean desertification and land use. Brussels: European Commission. Available from.

http://www.kcl.ac.uk/projects/desertlinks/downloads/publicdownloads/ESA%20Manual.pdf.Lecca, G., & Cau, G. (2009). Using a Monte Carlo approach to evaluate seawater intrusion in the Oristano coastal aquifer: a case study from the AQUAGRID

collaborative computing platform. Physics and Chemistry of the Earth, 34, 654–661.Li, X., Peterson, J., Liu, G., & Qian, L. (2001). Assessing regional sustainability: the case of land use and land cover change in the middle Yiluo catchment of the

Yellow River basin, China. Applied Geography, 21, 87–106.Lira, J. (2004). A model of desertification process in a semi-arid environment employing multi-spectral images. In A. Sanfeliu et al. (Eds.), Progress in pattern

recognition, image analysis and applications. CIARP 2004, Lecture Notes in Computer Science, 3287, 249–258.Liu, Y., Gao, J., & Yang, Y. (2003). A holistic approach towards assessment of severity of land degradation along the greatwall in northern Shaanxi province,

China. Environmental Monitoring and Assessment, 82, 187–202.Lu, D., Mausel, P., Brondizio, E., & Moran, E. (2004). Change detection techniques. International Journal of Remote Sensing, 25, 2365–2407.Lu, H., & Shao, Y. (2001). Toward quantitative prediction of dust storms: an integrated wind erosion modelling system and its applications. Environmental

Modelling and Software, 16, 233–249.Madrau, S., Loi, G., & Baldaccini, P. (1998). Modello per la valutazione della attitudine al miglioramento dei pascoli dei suoli della Sardegna. Monografia.

(In Italian).Mainguet, M. (1994). Desertification: Natural background and human mismanagement (2nd ed).. Berlin/Heidelberg, Germany: Springer-Verlag.Millan, M. M., Estrela, M. J., Sanz, M. J., Mantilla, E., Martın, M., Pastor, F., et al. (2005). Climatic feedbacks and desertification: the Mediterranean model.

Journal of Climate, 18, 684–701.Millennium Ecosystem Assessment (MEA). (2005). Ecosystems and human well being: Desertification synthesis. Washington, DC: World Resources Institute. p. 25.Monteith, J. L. (1972). Solar radiation and productivity in tropical ecosystems. Journal of Applied Ecology, 9, 747–766.Monteith, J. L. (1977). Climate and efficiency of crop production in Britain. Philosophical Transactions of the Royal Society of London, Series B, 281, 277–294.Motroni, A., Canu, S., Bianco, G., & Loj, G. (2009). Monitoring sensitive areas to desertification in Sardinia: the contribute of the Regional Agrometeorological

Service. In A. Marini, & M. Talbi (Eds.), Desertification and risk analysis using high and medium resolution satellite data. The Netherlands: Springer.Mouat, D., Lancaster, J., Wade, T., Wickham, J., Fox, C., Kepner, W., et al. (1997). Desertification evaluated using an integrated environmental assessment

model. Environmental Monitoring and Assessment, 48, 139–156.Nearing, M. A. (1997). A single, continuous function for slope steepness influence on soil loss. Soil Science Society of America Journal, 61(3), 917–919.Nyerges, T. L. (1993). Understanding the scope of GIS: its relationship to environmental modeling. In M. F. Goodchild, B. O. Parks, & L. T. Steyaert (Eds.),

Environmental modeling with GIS. New York: Oxford University Press.Parton, W. J., Schimel, D. S., Cole, C. V., & Ojima, D. S. (1987). Analysis of factors controlling soil organic matter levels in Great Plains grasslands. Soil Science

Society of America Journal, 51, 1173–1179.Pineda, N., Jorba, O., Jorge, J., & Baldasano, J. M. (2004). Using NOAA AVHRR and SPOT VGT data to estimate surface parameters: application to a mesoscale

meteorological model. International Journal of Remote Sensing, 25, 129–143.Pulina, G. (2000). Methodologies to prevent and mitigate land degradation phenomena in mediterranean agrosilvopastoral systems. Proceedings of the

Advanced Study Course, Alghero, May 30th to June 9th, 1999. In G. Enne, C. Zanolla, & D. Peter (Eds.), Desertification in Europe: Mitigation strategies, land-use planning (pp. 509). European Commission, EUR 19390.

Pulina, G., & Zucca, C. (1999). Un nuovo indicatore territoriale per la valutazione dell’impatto del pascolamento in ambiente mediterraneo, Vol. 42. Verona, LV:L’Informatore Agrario. (In Italian)105–109.

Pungetti, G. (1995). Anthropological approach to agricultural landscape history in Sardinia. Landscape and Urban Planning, 31, 47–56.Rahman, A. (2008). A GIS based DRASTIC model for assessing groundwater vulnerability in shallow aquifer in Aligarh, India. Applied Geography, 28, 32–53.Ravi, S., & D’Odorico, P. (2005). A field-scale analysis of the dependence of wind erosion threshold velocity on air humidity. Geophysical Research Letters, 32,

L21404. doi:10.1029/2005GL023675.Ravi, S., D’Odorico, P., Over, T. M., & Zobeck, T. M. (2004). On the effect of air humidity on soil susceptibility to wind erosion: the case of air-dry soils.

Geophysical Research Letters, 31. doi:10.1029/2004GL019485. L09501.Renard, K. G., Foster, G. R., Weesies, G. A., McCool, D. K., & Yoder, D. C. (1997). Predicting soil erosion by water: a guide to conservation planning with the Revised

Universal Soil Loss Equation (RUSLE). In: Agriculture Handbook, Vol. 703. U.S. Department of Agriculture. 384.Ringrose, S., Vanderpost, C., & Matheson, W. (1996). The use of remotely sensed and GIS data to determine causes of vegetation cover changes in southern

Botswana. Applied Geography, 16, 225–242.Roder, A., Udelhoven, T., Hill, J., Del Barrio, G., & Tsiourlis, G. (2008). Trend analysis of Landsat-TM and -ETMþ imagery to monitor grazing impact in

a rangeland ecosystem in Northern Greece. Remote Sensing of Environment, 12, 2863–2875.Running, S. W., Nemani, R., Glassy, J. M., & Thornton, P. E. (1999). MODIS Daily photosynthesis (PSN) and annual net primary production (NPP) product

(Mod17): algorithm theoretical basis document. V. 4.0. Available from. http://modis.gsfc.nasa.gov/data/atbd/atbd_mod16.pdf.Saaty, T. L. (1977). A scaling method for priorities in hierarchical structures. Journal of Mathematical and Statistical Psychology, 15, 234–281.Salvati, L., & Zitti, M. (2008). Regional convergence of environmental variables: empirical evidences from land degradation. Ecological Economics, 68,

162–168.Salvati, L., Zitti, M., & Ceccarelli, T. (2008). Integrating economic and environmental indicators in the assessment of desertification risk: a case study. Applied

Ecology and Environmental Research, 6, 129–138.Santini, M. (2008). A new GIS-based spatial modelling approach for desertification risk assessment in the Mediterranean area. An Italian case study:

Sardinia Island. PhD thesis. University of Tuscia, Italy.Sasikala, K. R., & Petrou, M. (2001). Generalised fuzzy aggregation in estimating the risk of desertification of a burned forest. Fuzzy Sets and Systems, 118,

121–137.Saxton, K. E., Rawls, W. J., Romberger, J. S., & Papendick, R. I. (1986). Estimating generalized soil-water characteristics from texture. Soil Science Society of

America Journal, 50, 1031–1036.Schenk, H. J., & Jackson, R. B. (2002). The global biogeography of roots. Ecological Monographs, 72, 311–328.Sepehr, A., Hassanli, A. M., Ekhtesasi, M. R., & Jamali, J. B. (2007). Quantitative assessment of desertification in south of Iran using Medalus method.

Environmental Monitoring and Assessment, 134, 243–254.Shalaby, A., & Tateishi, R. (2007). Remote sensing and GIS for mapping and monitoring land cover and land-use changes in the Northwestern coastal zone of

Egypt. Applied Geography, 27, 28–41.Shao, Y., Raupach, M. R., & Leys, J. F. (1996). A model for predicting aeolian sand drift and dust entrainment on scales from paddock to region. Australian

Journal of Soil Research, 34, 309–342.Sivakumar, M. V. K. (2007). Interactions between climate and desertification. Agricultural and Forest Meteorology, 142, 143–155.

M. Santini et al. / Applied Geography 30 (2010) 394–415 415

Sui, D. Z., & Maggio, R. C. (1999). Integrating GIS with hydrological modeling: practices, problems, and prospects. Computers, Environment and Urban Systems,23, 33–51.

Tarboton, D. G. (1997). A new method for the determination of flow directions and upslope areas in grid digital elevation models. Water Resources Research,33, 309–319.

Turc, L. (1955). Le bilan d’eau des sols. Relations entre le precipitations, l’evapotranspiration potentielle. Annales Agronomiques, 1. (In French).UNCED. Convention on desertification, 3–14 June, 1992, Rio de Janeiro, Brazil: United Nations Conference on Environment and Development.Vacca, A., Loddo, S., Serra, G., & Aru, A. Soil Degradation in Sardinia (Italy): main factors and processes. Proceedings of 7th International Meeting on Soils

with Mediterranean Type of Climate, 23–28 September 2001, Valenzano, pp. 341–343.Van Der Knijff, J. M., Jones, R. J. A., & Montanarella, L. (2000). Soil erosion risk assessment in Italy. European Soil Bureau, JRC, Space Applications Institute.Van Oost, K., Govers, G., & Desmet, P. J. J. (2000). Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape

Ecology, 15, 577–589.Venevsky, S. (2006). A method for integrated assessment of vulnerability to climate change in Siberian forests: example of Larch area. Mitigation and

Adaptation Strategies for Global Change, 11, 241–268.Verheijen, F. G. A., Jones, R. J. A., Rickson, R. J., & Smith, C. J. (2009). Tolerable versus actual soil erosion rates in Europe. Earth-Science Reviews, 94, 23–38.Wen-yi, F., When-hua, Z., Su-fang, Y., & Dan, L. (2005). Establishing evaluation index system for desertification of Keerqin sandy land with remote sensing

data. Journal of Forestry Research, 16, 209–212.Wessels, K. J., Prince, S. D., Malherbe, J., Small, J., Frost, P. E., & Van Zyl, D. (2007). Can human-induced land degradation be distinguished from the effects of

rainfall variability? A case study in South Africa. Journal of Arid Environments, 68, 271–297.White, M. A., Thornton, P. E., Running, S. W., & Nemani, R. R. (2002). Literature-derived parameters for the BIOME-BGC terrestrial ecosystem model. Data set.

Oak Ridge, Tennessee, USA: Oak Ridge National Laboratory Distributed Active Archive Center. doi:10.3334/ORNLDAAC/652. Available from. http://www.daac.ornl.gov.

Yassoglou, N. J. (1999). Land, desertification vulnerability and management in Mediterranean landscapes. Proceedings of the International Conference heldin Crete, 29 October to 1 November 1996. In P. Balabanis, D. Peter, A. Ghazi, & M. Tzogas (Eds.), Mediterranean desertification: Research results and policyimplications (pp. 87–113). Luxembourg: European Commission – Directorate General Research, EUR 19303.

Zhai, L., & Feng, Q. (2009). Spatial and temporal pattern of precipitation and drought in Gansu Province, Northwest China. Natural Hazards, 49, 1–24.Zhang, Y., Chen, Z., Zhu, B., Luo, X., Guan, Y., Guo, S., et al. (2008). Land desertification monitoring and assessment in Yulin of Northwest China using remote

sensing and geographic information systems (GIS). Environmental Monitoring and Assessment, 147, 327–337.Zhangsi, Y., & Williams, T. H. L. (1997). Obtaining spatial and temporal vegetation data from landsat MSS and AVHRR/NOAA satellite images for a hydrologic

model. Photogrammetric Engineering & Remote Sensing, 63, 69–77.Zucca, C., Canu, A., & Della Peruta, R. (2006). Effects of land use and landscape on spatial distribution and morphological features of gullies in an agro-

pastoral area in Sardinia (Italy). Catena, 68, 87–95.