13
Multi-criteria, multi-objective and uncertainty analysis for agro-energy spatial modelling Patrizia Tenerelli a, * , Steve Carver b a European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen, T.P. 268, Via E. Fermi 2749, I-21027 Ispra, VA, Italy b School of Geography, University of Leeds, Leeds LS2 9JT, UK Keywords: Energy crops Land capability Land allocation Multi-criteria spatial modelling Uncertainty and sensitivity analysis abstract The worldwide increase in the use of biomass as a Renewable Energy Source raises the issue of intro- ducing crops dedicated to energy production into rural landscapes. The purpose of this paper is to set-up a GIS based multi-criteria approach to assess a range of possibilities for perennial energy crops conversion. The presented method was implemented at the regional level in the Yorkshire and the Humber Region in Northern UK. The rst phase of the study aims to set-up a land capability model for the specic purpose of assessing the potential of different typologies of perennial energy crops, on the basis of specic pedo-climatic and topographic factors. The model output illustrates a range of potentials for energy crop conversion that can be explored in the given landscape. In the second phase a uncertainty analysis of the land capability model was performed through a simulation approach in order to interpret the inuence of assumptions and uncertainty on input data and model parameters. The last phase of the study allows allocating the energy crop conversion area according to specic environmental constraints, nature protection targets, food production priorities and land capability values. The land allocation output gives a rather restrictive energy crop penetration scenario, where more than half of the conversion area is allocated to cropping systems with low land degradation potential. This scenario represents a preliminary regional analysis of the energy crop potential in terms of theoretically available conversion areas. The nal results also show that the areas with highest environmental risks correspond to the areas with both the lowest suitability for energy crop cultivation and the highest model uncertainty. Ó 2011 Elsevier Ltd. All rights reserved. Introduction Given the increasing interest in alternative energy sources, there is a strong need to support energy policy decisions with spatially explicit evidence. Assessing the potential of Renewable Energy Sources (RES) is a research eld with important implications for the energy planning process. Compared to fossil energy sources, RES have a lower energy density per unit of land and a stronger dependence from the physical land factors and constraints, there- fore a more detailed knowledge of the geographical distribution of the resources is needed to develop future scenarios of renewable energy supply. Among RES, traditional biomass supplies the largest share of global energy consumption, while biofuels are growing both in developed and developing countries (REN21, 2010). Biofuels refer to solid, liquid or gaseous materials derived from biomass (UNEP, 2009). In the UK, energy crops for biofuel production are increasingly encouraged under Energy Crops Schemes (ECS) (Natural England, 2009). The UK Biomass strategy sets a target to increase the amount of land allocation to perennial energy crops up to a total of 17% of UK arable land by 2020 (DTI, DFT, DEFRA, 2007). One of the most critical issues of the bioenergy sector is repre- sented by the biomass supply system. The potentially productive land area overheads required for biomass production results in a complex logistic within the whole chain and difcult trade-offs between energy and food production (Berndes, 2006; Giampietro, Ulgiati, & Pimentel, 1997; Ignaciuk, Vöhringer, Ruijs, & van Ierland, 2004; Johansson & Azar, 2007). Moreover, the increasing demand for biomass is expected to affect the actual land use patterns and species diversity within new energy landscapes. Hence, the decision of changing land use for energy purposes should be supported by both a land evaluation under specic environmental conditions, and an assessment of the biological and economic competitiveness of the crops within traditional agricul- tural and natural landscapes. * Corresponding author. Tel.: þ39 0 332 786720. E-mail address: [email protected] (P. Tenerelli). Contents lists available at SciVerse ScienceDirect Applied Geography journal homepage: www.elsevier.com/locate/apgeog 0143-6228/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.apgeog.2011.08.013 Applied Geography 32 (2012) 724e736

Multi-criteria, multi-objective and uncertainty analysis for agro-energy spatial modelling

Embed Size (px)

Citation preview

at SciVerse ScienceDirect

Applied Geography 32 (2012) 724e736

Contents lists available

Applied Geography

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

Multi-criteria, multi-objective and uncertainty analysis for agro-energy spatialmodelling

Patrizia Tenerellia,*, Steve Carverb

a European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen, T.P. 268, Via E. Fermi 2749, I-21027 Ispra, VA, Italyb School of Geography, University of Leeds, Leeds LS2 9JT, UK

Keywords:Energy cropsLand capabilityLand allocationMulti-criteria spatial modellingUncertainty and sensitivity analysis

* Corresponding author. Tel.: þ39 0 332 786720.E-mail address: [email protected]

0143-6228/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.apgeog.2011.08.013

a b s t r a c t

The worldwide increase in the use of biomass as a Renewable Energy Source raises the issue of intro-ducing crops dedicated to energy production into rural landscapes. The purpose of this paper is to set-upa GIS based multi-criteria approach to assess a range of possibilities for perennial energy cropsconversion. The presented method was implemented at the regional level in the Yorkshire and theHumber Region in Northern UK. The first phase of the study aims to set-up a land capability model forthe specific purpose of assessing the potential of different typologies of perennial energy crops, on thebasis of specific pedo-climatic and topographic factors. The model output illustrates a range of potentialsfor energy crop conversion that can be explored in the given landscape. In the second phase a uncertaintyanalysis of the land capability model was performed through a simulation approach in order to interpretthe influence of assumptions and uncertainty on input data and model parameters. The last phase of thestudy allows allocating the energy crop conversion area according to specific environmental constraints,nature protection targets, food production priorities and land capability values. The land allocationoutput gives a rather restrictive energy crop penetration scenario, where more than half of theconversion area is allocated to cropping systems with low land degradation potential. This scenariorepresents a preliminary regional analysis of the energy crop potential in terms of theoretically availableconversion areas. The final results also show that the areas with highest environmental risks correspondto the areas with both the lowest suitability for energy crop cultivation and the highest modeluncertainty.

� 2011 Elsevier Ltd. All rights reserved.

Introduction

Given the increasing interest in alternative energy sources, thereis a strong need to support energy policy decisions with spatiallyexplicit evidence. Assessing the potential of Renewable EnergySources (RES) is a research field with important implications for theenergy planning process. Compared to fossil energy sources, REShave a lower energy density per unit of land and a strongerdependence from the physical land factors and constraints, there-fore a more detailed knowledge of the geographical distribution ofthe resources is needed to develop future scenarios of renewableenergy supply. Among RES, traditional biomass supplies the largestshare of global energy consumption, while biofuels are growingboth in developed and developing countries (REN21, 2010). Biofuelsrefer to solid, liquid or gaseous materials derived from biomass

(P. Tenerelli).

All rights reserved.

(UNEP, 2009). In the UK, energy crops for biofuel production areincreasingly encouraged under Energy Crops Schemes (ECS)(Natural England, 2009). The UK Biomass strategy sets a target toincrease the amount of land allocation to perennial energy crops upto a total of 17% of UK arable land by 2020 (DTI, DFT, DEFRA, 2007).

One of the most critical issues of the bioenergy sector is repre-sented by the biomass supply system. The potentially productiveland area overheads required for biomass production results ina complex logistic within the whole chain and difficult trade-offsbetween energy and food production (Berndes, 2006; Giampietro,Ulgiati, & Pimentel, 1997; Ignaciuk, Vöhringer, Ruijs, & vanIerland, 2004; Johansson & Azar, 2007). Moreover, the increasingdemand for biomass is expected to affect the actual land usepatterns and species diversity within new energy landscapes.Hence, the decision of changing land use for energy purposesshould be supported by both a land evaluation under specificenvironmental conditions, and an assessment of the biological andeconomic competitiveness of the crops within traditional agricul-tural and natural landscapes.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 725

In this study a GIS based multi-criteria approach was applied toassess the area suitable for energy crops introduction in the York-shire and Humber Region of the UK. This region is located inNorthern England and occupies about 15,400 km2, with 76% of theland covered by arable land and grassland. The study area is char-acterized by a heterogeneous landscape with highly productiveland, marginal areas, protected areas (for nature and landscapeconservation) and land sensitivities to erosion and waterlogging.Energy crops are currently planted in the Region under ECS, withhigh potential for expansion (Howes, 2007). Regional targets forenergy crops include not only the reduction of greenhouse gasemission, but also other sustainability and environmental objec-tives, such as biodiversity protection and economic regeneration inrural areas (Howes, 2007).

The aim of this work is to carry out an analysis of the potentialscenarios for biomass supply which takes into account the agri-cultural and environmental issues affecting the successful intro-duction of short rotation forestry and perennial herbaceous energycrops. The implemented methodology is based on a multi-criterialand capability and land allocation mapping process. The param-eter uncertainty and the sensitivity of the model were tested inorder to account for the robustness of the outputs. The land capa-bility, together with the uncertainty analysis and the allocationscenario, provides an integrated evaluation approach which alsoconsiders the stochastic properties of the physical parameters.

Related work

Geographic Information Systems (GIS) are particularly appro-priate for decision support in agro-energy planning and design ofbiomass to energy supply systems. Within the literature, numerousGIS approaches have been applied for the spatial characterization ofbiomass potentials, costs, supply and demand. Specific spatially-explicit models have been designed with regard to the develop-ment of Decision Support Systems (DSS) in the bioenergy sector(Mitchell, Stevens, & Watters, 1999; Noon & Daly, 1996; Overend &Mitchell, 2000; Voivontas, Assimacopoulos, & Koukios, 2001).Several authors applied GIS analyses for locating energy croppotential (Beccali, Columba, D’Alberti, & Franzitta, 2009; Martelliet al., 2002; Tenerelli & Monteleone, 2008; Tenerelli, Pantaleo,Carone, Pellerano, & Recchia, 2007) and for computing the optimalsizing and location of biofuel plants (Edwards, �Súri, Huld, &Dallemand, 2005; Freppaz et al., 2004; Graham, English, & Noon,2000; Ma, Scott, DeGloria, & Lembo, 2005; Towers, Morrice,Aspinall, Birnie, & Dagnall, 1997; Voivontas et al., 2001). Decidinghow to plan biomass collection and harvesting is another challengethat has been addressed by several GIS approaches (Freppaz et al.,2004; Masera, Ghilardi, Drigo, & Trossero, 2006; Noon & Daly, 1996).

In the UK studies on spatial assessments of energy crop poten-tial have been conducted at the regional and national scale.Andersen, Towers, and Smith (2005) implemented a GIS modelbased on land cover, pedo-climatic and topographic parameters forshort rotation coppice potential assessment in Scotland. Richter,Riche, Dailey, Gezan, and Powlson (2008) developed an empiricalyield model for Miscanthus, based on pedo-climatic variables at thenational (UK) and regional (Oxfordshire and North Yorkshire) scale.Lovett et al. (2009) also applied an empirical model for predictingthe yields of Miscanthus, at 1 km resolution; the authors thencompared the spatial variation in yield with the potentially suitableland in England and estimated the energy supply by region. Aylottet al. (2008) developed an empirical model based on pedo-climaticand topographic variables for mapping the yield and supply ofpoplar and willow across England and Wales, at 25 km resolution.Bellarby, Wattenbach, Tuck, Glendining, and Smith (2010) deter-mined the suitability of different bioenergy crops, including oil

crops, cereals, starch crops and solid biofuel crops, under currentand future climatic condition in the UK; the model was based onelevation, temperature and rainfall values, at 5 km resolution.

Most of the studies on energy crop potential in the UK focus onsingle crops, when conducted at sub-national scale (regional orlocal scale), and on multiple crop typologies when performingassessments at national scale. The current work analyses thepotential for a wide range of perennial energy crop typologies ata regional scale. For this purpose a specific land capability conceptwas developed to classify the land quality and flexibility fordifferent energy crop systems. This approach allows having a broadperspective on the regional energy crop potential and an overviewof the possible land allocation conflicts. Thought not founded onempirical models, the present study is based on a comprehensivebackground investigation of the crop ecological requirements andland conservation requisites.

Several studies have focused on energy crop potential, but fewones have performed the analysis of uncertainty and sensitivity ofthe adoptedmodel. Moreover most papers that employ similar landsuitability analysis do not address the uncertainty propagation dueto the input data error and the sensitivity of the model to varyingparameters. The current work undertakes a simulation approach toassess the degree of uncertainty and sensitivity due to thestochastic properties of the variables used for the land allocationmodel which evaluates the energy crop potential.

Method

In this study a spatial model was defined to assess the theo-retically available areas for conversion to perennial energy crops atthe regional scale. The methodology is based on the concept ofevaluating land capability through general diagnostic criteria andsimple rules which can be applied for different sets of crops (croptypologies). The crop typologies were defined according to the life-form (herbaceous or woody), life cycle (cutting cycle) and envi-ronmental impact or possible positive effects in terms of improvedresources management and land conservation.

The general methodology is composed of three different phases(Fig. 1). The first phase aims to identify the energy crops require-ments and apply a land capability model, describing the suitabilityof the land for energy crop growth. The model is based on a multi-criteria evaluation (MCE) which assesses the suitable area and thecapability level for energy crops by considering pedo-climatic andtopographic diagnostic criteria. MCE methods allow multiple andoften conflict criteria to be taken into account and weights to beapplied to input layers depending on the level of importanceascribed to these by the user (Carver, 1991). The second phasesimulates the error propagation and model sensitivity determinedby parameter uncertainty. In this phase Monte Carlo simulationsand jack-knifing techniques (Tukey, 1958) were applied. The lastphase considers different land use and environmental constraintsin order to define the conversion area and integrate the agro-energy use within existing land use patterns, using a specificmulti-objective model.

Land capability for energy crops

The Soil Conservation Service of the US Department of Agri-culture (USDA) (Klingebiel & Montgomery, 1961) providedconceptual definitions of capability classes according to increasingdegree of limitation, in relation to agricultural uses, imposed bypermanent physical properties. Several adaptations of this system,such as the British Land Capability Classification (Bibby & Mackney,1969) and the Canadian Land Capability Scheme (Canada LandInventory, 1970), have been used worldwide. Land capability

Target definition: crop types and ecological requirements

Phase 1Phase 2Phase 3

Selection of diagnostic criteria

Standardization and weigthing

L d bilit i dSelection of constraints

D fi iti f th i

Land capability indexSelection of constraints

Land capability classification

Uncertainty and sensitivity analysis

Binary evaluation

Definition of the environmental constraints

Definition of the conversionarea

Land allocationBinary evaluation

y y

Definition of the conversion goals

Fig. 1. General methodology flowchart.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736726

models differ from land suitability models, which apply to thecapacity of the land to sustain more specific uses (e.g. single cropsuitability).

When introducing dedicated energy crops into an existing landuse mix, a land evaluation, that balances all costs and benefits,including food and non-food production, is needed. Energy cropsdiffer from conventional crops as they are selected andmanaged forbiomass production and energy content, thus having differentcharacteristics and agronomy, leading to different interactions withthe environment (EEA, 2006). A specific land capability for energycrops is thus required in addition to the traditional land capabilityfor agricultural food production.

In this study a land capability approach was applied for perennialenergy crops, including perennial grasses (PG), short rotation coppice(SRC) and short rotation forestry (SRF). SRC have a cutting cycle of3e5 years and regenerate after from stumps (DEFRA, 2004;Hardcastle et al., 2006; Mitchell et al., 1999). The most typicalspecies for SRC in UK are poplar and willow; according to Mitchellet al. (1999) eucalyptus and alder also showed good performances.SRF differs from SRC for the longer cutting cycle (12e15 years), whichimply that SRF can also be used on poorer soils, with less heavymechanization systems. Broadleaved coppice plantations suitable forSRF can also include black alder, hazel, ash, small-leaved lime, sweetchestnut and sycamore. Compared to SRC and SRF, PG dedicated toenergy production are generally less demanding in terms of soilrequirements and are adaptable to a wider range of land types(Lewandowski, Scurlock, Lindvall, & Christou, 2003). Themost typicalPG for energy production in the UK are switchgrass, Miscanthus andreed canary grass (Elbersen, Christian, El Bassam, & Sauerbeck, 2010;Hall, 2003; Theriault, Javorska, Casova, Tucker, & Hansen, 2003).

In order to assess the land capability for energy crops, diagnosticcriteria were chosen according to the following factors: significanceas main ecological requirements for the energy crops, presence ofcritical factors in the study area and availability of knowledge orreferences to evaluate the factor values (Rossiter, 1994). The diag-nostic criteria focus on the main pedo-climatic and topographicconditions which, according to the scientific references, may affectthe plant growth and the mechanized agro-forestry operations. Theselected criteria were then combined in the land capability modelas variables or constraints. The physical properties which define thediagnostic criteria, and the related literature, are described below:

i. Growing Degree Days (GDD): is an indicator of the develop-ment potential for plants based on the heat values. Thedevelopment of a species will occur only if the atmosphericheat temperature exceeds a minimum threshold tempera-ture. The GDD for the UK was calculated by Perry and Hollis(2005). According to several authors, forest species,including SRF, requires between 1375 and 875 day-degreesabove 5.6 �C (Bibby, Heslop, & Hartnup, 1998; Cannell,Sheppard, & Milne, 1988). C4 herbaceous rhizomatousperennials have a more efficient conversion of solar radiationcompared to other plants (Brown, Neilson, Lewandowski, &Jones, 2000) and are not significantly affected by GDD(Heaton, Voigt, & Long, 2004; Lewandowski et al.,2003).However it must be underlined that those species canbe less resistant to low temperatures and can be affected bytemperature in different, site-specific ways. Increasingtemperature and global solar radiation may have a positiveimpact on the growth intensity and the harvestable yields ofMiscanthus, at the same time it can cause an increase ofevapotranspiration and water stress (Richter et al., 2008).

ii. Minimum and maximum annual rainfall: according to Tuck,Glendining, Smith, Housec, and Wattenbachb (2006) mostof the PG dedicated to energy production and SRC requirea minimum of 600 mm of mean annual precipitation, whilethemaximum rainfall varies from 1500 to 2000mm; some PGsuch as Miscanthus and switchgrass (C4 grasses) are reportedto use water more efficiently (Hall, 2003), however their yieldcan be limited by water availability (Richter et al., 2008).

iii. Slope: the slope may affect the mechanized operations anderosion risk. This factor constraint is more important for SRFthan for PG, which are normally harvested with a forageharvester. SomePGasMiscanthus, however, are not planted onslopesmore than 15% (Lovett et al., 2009). In the case of SRC themain difficulties arise for the harvesting operations whichnecessitate specific chippers or cut/bundle (or accumulateloose) machines. Those machines normally are not used onslopes steeper than 20% and their use is questionable on slopesdown to 7% (DEFRA, 2004; Forestry Commission, 1998).Andersen et al. (2005) set a suitability threshold of 12% for SRC.

iv. Soil type (texture, depth, presence of peat layers andflooding):SRC and PG can grow on a wide range of soils (Aylott et al.,

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 727

2008). The optimum soil for SRC should be medium textured,aerated andwith a good supplyofmoisture,while verygravellysoils and low moisture content are generally unsuitable(Forestry Commission, 2002; Paulson et al., 2003). Peaty soils,shallow soils, saltings and marshes are a constraint in forestplantations, both for the rooting and mechanical operations(Bibby et al., 1998). PG can grow on awider range of soils, evenwith low fertility (Lewandowski et al., 2003). Light sandy soils,as well as sloping sites, are not suitable for SRC with a shortcutting cycle (below 5 years), as the cultivation practices canlead to acceleratedwind andwater erosion (DEFRA, 2004). Thesoil conditions which may affect the lodging and windthrowhazard, mainly referable to the presence of shallow soils, gleysoils, peats, soilmoisturedeficits andwaterlogging (Bibbyet al.,1998), were also taken into account.

v. Soil wetness: very dry or very wet soils, as well as water-logged and regularly flooded soils, should be avoided for SRC(DEFRA, 2004; Forestry Commission, 2002; Perry & Hollis,2005). Perennial grasses can grow on dry to poorly drainedsoils, marginal lands and acidic wet soils (DEFRA, 2007;NNFC, 2007; Theriault et al., 2003).

vi. pH: the pH range for SRC and PG varies from 5.5 to 7.5(DEFRA, 2004; DEFRA, 2007; Jobling, 1990), however widerrange had also been found (Paulson et al., 2003).

vii. Soil stoniness: some PG grow on shallow rocky soils(Lewandowski et al., 2003), nevertheless stony soils and areaswith extensive bare rocks should be avoided, either for forestand herbaceous species, because of difficulties for themechanized operations.

ImplementationTable 1 describes the whole GIS dataset used for the spatial

modelling.The implemented multi-criteria GIS model involves a linear

combination of multi-valued variables (factors), overlaid withbinary constraints (Boolean variables). A capability index was thusobtained combining the following standardized factors: annualrainfall, GDD, soil texture, soil wetness, soil pH, soil depth andslope. The following Boolean qualitative criteria were then multi-plied by the capability index, in order to mask the areas where theconstraints are present: soil stoniness, presence of peat layers,flooded areas and water bodies.

We expected the input factors to be correlated. The factorsinterdependency was therefore tested by pairwise correlation,using the Pearson’s correlation coefficient (varying from �1 to 1).Table 2 shows the resulting correlation matrix. The correlation

Table 1Input parameters for the land capability and land allocation model.

Input dataset Source

Interpolated climatic data for 1961e2000 5 km2

resolution (UKCIP02 National Climate Scenarios)Prepared for UKCIP by theand Met Office Hadley Cen

Digital Terrain Model 1:50,000 scale Ordnance Survey Landformdataset via EDINA Digimap

Soil Associations in England 1:250,000 scale National Soil Resources Ins

Land Cover Map (LCM2000) NERC e Centre for EcologyCEH Wallingford

Agricultural Land Capability Natural EnglandFlood Map Environment Agency

Protected Areas Magic Dataset and Local Au

a Soil texture and pH are extrapolated from the horizons’ details on the basis of the e

coefficients are all below 0.7. Relatively high correlations werefound between pH and soil wetness, rain, soil texture, and betweenrain and both wetness and slope. The interdependency betweenthose variables is plausibly due to process-based relationshipsbetween climate, soil and topography; soil and climate, as group offactors, are those showing the highest correlation. Some morecorrelation were expected between interplaying variables such asslope and soil depth, however some correlations might break downat the given observation scale. All the considered factors werefinally used in the land capability model because of the relevantlinkages with the plant species ecological requirement.

The first step of the land capability model was the factorsstandardization. The multi-valued variables were standardizedaccording to their compatibility with the ecological requirement ofthe crops. The numeric variables were standardized following thebenefit or cost functions represented in Fig. 3. The standardizedscores for the nominal variables were assigned by using a rankingapproach (Table 3). Both the quantitative and categorical variablesstandardization was based on expert knowledge.

The final capability index was calculated according to thefollowing:

Si ¼X

fji �Y

cki (1)

where:Si ¼ the capability index for the i-th cell;fji ¼ the score assigned to the j-th factor and the i-th cell;cki ¼ the score assigned to the k-th constraint and the i-th cell.The capability classes were defined by applying equal value cut-

off points (thresholds) to the continuous values of the capabilityindex. The proposed land capability classification assumes thatforest species require a longer growing cycle to reach a maturestage in correspondence with increasing degree of land limitations.Perennial grasses were assumed to grow in most of the capabilityclasses. A total of six classes were defined, wherein higher numbersimply lower land capability for energy crops. The first class hasexcellent land quality and flexibility for different crop systems,while the last capability class excludes any kind of land exploita-tion. Fig. 2 describes the land capability classification customizedfor this study and for the specific purpose of assessing the theo-retical potential of perennial energy crops.

Uncertainty and sensitivity analysis

In GIS modelling several geoprocessing operations can beapplied to derive new data which can be affected by both input andmodel errors. Input errors are due to the error in the source dataset,

Factor or constrains Attribute type

Tyndall Centretre

Growing Degree Days (GDD) QuantitativeAnnual rainfall Quantitative

PanoramaService

Slope Quantitative

titute, 2008 Soil wetness CategoricalSoil texturea CategoricalSoil depth CategoricalpHa QuantitativeSoil stoniness CategoricalPresence of peat layers or waters Categorical

& Hydrology, Land cover classes Categorical

Agricultural land capability class CategoricalPresence of waters or areas floodedby high tides

Categorical

thorities Presence of protected areas Categorical

xpected percentages of series within the soil associations.

Table 2Correlation matrix for the land capability input factors.

Climate Soil Topography

Rainfall GDD Texture Wetness pH Depth Slope

Climate Rainfall 1 �0.36 �0.36 �0.54 �0.55 �0.27 0.47GDD 1 0.25 0.29 0.41 0.12 �0.26

Soil Texture 1 0.20 0.44 0.28 �0.10Wetness 1 0.65 0.01 �0.24pH 1 0.10 �0.29Depth 1 �0.19

Topography Slope 1

Bold numbers represent the highest correlation coefficient values (more than 0.45).

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736728

caused by measurement, variation and data entry; while modelerrors are due to the approximation in the computational model(Heuvelink, 1998). Uncertainty arises when the accuracy of thespatial data, which determines the error, is unknown (Hunter,Robey, & Goodchild, 1994; Zerger, Smith, Hunter, Jones, 2002).Uncertainty propagates trough the model operations from theinput data and parameters to the final output. Handling errors anduncertainty in GIS plays a considerable role in decision-makingwhere it is important to base decisions on probabilistic rangesrather than deterministic results (Abbaspour, Delavar, & Batouli,2003; Aerts, Goodchild, Heuvelink, 2003; O’Brien, 2008).

Simulation is one of the most appropriate approaches to analyseuncertainty propagation through a GIS model, without knowingthe functional form of the errors (Eastman, 2003). Uncertainty andsensitivity analysis are two different simulation approaches whichinvestigate the error propagation and strength of a model. Both theapproaches provide different maps which represent alternatives tothe model output and quantify the uncertainty of the model(Hunter et al., 1994). Uncertainty analysis attempts to identify andquantify confidence intervals for a model output, assessing theresponse to uncertainties in the model inputs (Crosetto, Tarantola,& Saltelli, 2000). Sensitivity analysis is used to partition the resultsunder different conditions of the model components and parame-ters and consequently identifying which are the key determiningvariables (Smith, 2002). Uncertainty and sensitivity analysistogether contribute to understand the influence of the assumptionsand input parameters on a given model (Crosetto et al., 2000).

Several authors have applied uncertainty and sensitivity anal-ysis in spatial modelling with GIS (Abbaspour et al., 2003; Aerts atal., 2003; Crosetto et al., 2000; Heuvelink, 1998; Openshaw,Charlton, & Carver, 1991). Most of the applications deal withmodelling based on Digital Elevation Models, such as hydrologicalmodeling and disaster risk assessment (Hunter et al., 1994; Zergeret al., 2002). Very little research has been conducted on inputdata error propagation in land suitability modelling. Few authorsapplied analysis of sensitivity to weights and ratings variation in

Class SRCt*<3

SRC SRC SRF PG None 3<t<5 t>5 t>10 IIIIII IV V VI

excellent fair

poor

good

no energy plant production

* t = growing cycle (years)

Fig. 2. Land capability classes for energy crops.

models of land suitability for crops (Benke & Pelizaro, 2010; Chen,Yu, & Khan, 2010). Uncertainty analysis has also been applied inthe field of bioenergy planning. Graham et al. (2000) recognize thatwhen estimating biomass potential supply, mapped variables canbe missing or derived frommultiple models and that this introduceseveral uncertainties that are not easily controlled by GIS models.

Difficulties in handling uncertainty in GIS modelling are due tothe fact that iterative or stochastic types of process requiresspecialized programming and significant computing power withlarge CPU time and disk space amount (Zerger et al., 2002).

In this study a simulation approach was used to take intoaccount the stochastic properties of the spatial variables used toassess the land capability for energy crops. Uncertainty and sensi-tivity analyses were performed, on a grid cell base, using iterationsand random perturbing variables. Uncertainty analysis was used toanalyse the propagation of errors from the input data to the finalresults; a Monte Carlo simulationwas used to study the response ofthe final product to perturbed inputs. Two sensitivity analyses wereused to investigate the response of the model to each single inputdata and parameter. The first sensitivity simulation assesses theeffect of each factor on the final product by leaving out the inputsone by one (jack-knifing). The second sensitivity testing appliesa Monte Carlo approach to consider the influence of varying factorweights (weights sensitivity) on the land capability index.

Uncertainty analysisA Monte Carlo simulation was performed to model the error

propagation from the input data to the final product. For each factorthe error was added to the deterministic part of the parameter witha random component and propagated trough the data standardi-zation and linear combination to the land capability output. Theerror was sampled from the normal distribution, starting froma given or assumed RMS error value.

The simulation model is described by the following steps:

i. Generation of the error map (noise) for each data layer usinga random function: each cell was assigned with an error(random variable) sampled from the normal distribution;

ii. Addition of the error map to the expected value of each factor(deterministic variable): for every input data a newmap witha simulated error was created;

iii. Computation of the standardization and linear combinationof the factors: starting from the new input data a new landcapability index which contains a propagated error wascreated;

iv. Repetition of the simulation N times: practically the numberof simulations (N) vary from 50 to 2000 (Heuvelink, 1998;Openshaw et al., 1991; Zerger et al., 2002) according to thecomputational load, the complexity of the model, and thedesired accuracy;

v. Storing the new capability maps from each simulation bysumming them up;

Fig. 3. Standardization functions: a) Growing degree days (GDD); b) pH; c) Rainfall (mm); d) Slope (%).

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 729

vi. Analyse the results: compute the mean value (mu) of the newcapability maps;

vii. Produce statistics: spatial distribution of the error (errormap), cross-correlation analysis between the input factorsand the error map, comparison of the new capability mapwith the expected capability.

Sensitivity analysisSensitivity analysis investigates how much the model responds

to different sources of variation in the input dataset, parametersand assumptions. In this work two sensitivity analyses were per-formed to assess the response of the model output to each factorand to varying factor weights in the linear combination deter-mining the land capability.

The first sensitivity analysis was based on a “one at time”approach (jack-knife). The simulation is described by the followingsteps:

Table 3Nominal variables standardization scores.

Depth Score Texture Score Wetness

Very shallow 0 Loam 1 I 1Shallow 0.2 Clay loam 0.9Deep 1 Sandy clay loam 0.8 II 0.9Deep in places 0.4 Silty clay loam 0.8Shallow in places 0.8 Silt loam 0.6 III 0.7Locally deep 0.4 Sandy loam 0.5Deep, some shallow in places 0.8 Fine sandy loam 0.4 IV 0.4Occasional deep 0.3 Loamy sand 0.3Steep slopes 0 Loamy fine sand 0.2 V 0.1Some shallow 0.6 Clay 0.2Thick 0 Silty clay 0.2 VI 0No soil 0 Fine sand 0.2

i. Computation of the land capability index n times: each timeone of the factor was left out from the linear combination; n isequal to the number of factors;

ii. Analysis of the capability class variation per each left outfactor.

The land capability model assumes that each factor in the linearcombination has an equal weight. This assumption is crucial for theimplemented model as the influence of the different factors variesaccording to specific requirements of single crop species andcannot be generalized at the level of energy crop range on whichthis study is based. A weight sensitivity simulation was performedin order to assess the influence that varying weights may have onthe final result. The simulation is based on a Monte Carlo approachand is described by the following steps:

i. Computation of the land capability index: each time thefactors’ weight vary randomly from 0 to þ50%;

ii. Repetition of the simulation N times (as described in theprevious paragraph, point iv);

iii. Storing the new capability maps from each simulation bysumming them up;

iv. Computation of the mean value (mu) of the new capabilitymaps;

v. Produce comparison statistics.

ImplementationThe main difficulty of applying Monte Carlo simulations for

uncertainty analysis commonly arises from the lack of informationon data quality (Johnston & Timlin, 2000). In this case study theaccuracy assessment was given for the climatic parameters (Bibbyet al., 1998) and the digital terrain model (DTM) (Ordnance

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736730

Survey, 2004, pp. 1e10). According to the accuracy assessment anerror function of�192mmwas applied to the annual rain and�100number of days to the GDD, while a �3 m value was used for theDTM. For the soil data no indications of the accuracy or stochasticproperties were available, a standardized error of 20% was there-fore applied for all the soil parameters assuming a conservativeapproach (Heuvelink, 1998). The error values were then used tosample the error map for each data layer.

Another challenge of implementing Monte Carlo simulations inGIS modelling derives from the data load necessary to runa significant number of simulations. In this study both the MonteCarlo simulation for the uncertainty and sensitivity analysis wererun for 100 iterations. The number of iterations was constrained bythe high computing power required to run the simulation for thewhole study area extent.

The original capability map (expected capability) and theoutputs of the alternative models (perturbed maps) were quanti-tatively related through a cell-by-cell comparison which providesa general assessment of consistency between two maps. Confusionmatrixes were used to cross-tabulate the maps’ classes. For eachsimulation model the thematic correspondence was tested throughthe following per class and overall comparison statistics (Jensen,1996, pp. 257e278; Rossiter, 2004):

i. Relative variation per class: indicates the total variation of thesurface belonging to each category (capability class) withrespect to the original capability map;

ii. Overall accuracy: represents the total agreement between thetwo maps being tested (Congalton, 2001);

iii. Kappa coefficient of agreement (Khat): compares two maps todetermine if they are statistically significantly different(Congalton, 2001). Landis and Koch (1977) characterized thepossible ranges for Khat.

The uncertainty propagation was also assessed trough thespatial correlation between each input factor and the absolute errordistribution, by means of cross-correlation analysis. The cross-correlation index measures the relationship between two grids interms of spatial distribution of an attribute. Cross-correlationanalysis can be applied to assess the role of different variables ascontrol factors for a model output (Luo & Stepinski, 2008). Theindex was calculated as follow (Goodchild, 1986):

C ¼PN

i¼1 cijffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPnk

�zi � zi

�2s�

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPnk

�zj � zj

�2s (2)

where:i ¼ any cell on the first input grid;j ¼ any cell on the second input grid that is offset from i’s

location by the specified x-, y-offset;zi ¼ the value of the attribute cell i;zj ¼ the value of the attribute cell j;zi ¼ the mean value of the attribute of the first grid;zj ¼ the mean value of the attribute of the second grid;cij ¼ the similarity of i’s and j’s attributes:

cij ¼�zi � zi

�� �zj � zj

The index value can range from�1 to 1; a value of 1 implies thatthe grids are highly cross correlated, while a zero value imply thatthey are independent. Negative values indicate that close togethercells have very different attribute values than objects that arefarther apart.

Land allocation

The increasing demand of agricultural land for energy cropsintroduces the need for an effective land allocation, based ondifferent degree of land marginality and objectives priority. Themulti-objective approach is well suited to the issue of food-bioenergy competition in the land planning phase. In this paper,the alternative feasible locations for energy crops were assessedthrough two main steps.

The first step allocates the agro-energy conversion according tothe land use and agricultural land capability classes that are avail-able for the conversion. Themodel assumes that energy crops neverdisplace built-up areas, natural habitats or other land with highecological value, and that only agricultural land that is primarilyarable or grazing land, can be converted into energy crop fields(Fig. 4). Moreover the model excludes from the conversion area allthe highly productive land, corresponding to first and second grade(excellent and very good) in the Agricultural Land Capability Map.This last assumption is based on the concept that energy cropsshould not conflict with food production and crops with a highreturn. Energy crops are therefore be restricted to marginal or lessproductive land (grade three, four and five in the Agricultural LandCapability Map). This concept does also reflect the farmers’ attitudeto land allocation for food and PG under Energy Crops Scheme (ECS)in the UK (Lovett et al., 2009). The following land use classes, asmapped in the LCM2000, were finally included in the conversionscenario:

i. Arable cereals;ii. Horticulture;iii. Improved grassland;iv. Set-aside grass.

In the second step, after excluding the unsuitable land use andagricultural land capability classes, the remaining land was allo-cated to different energy crop typologies (SRC, SRF, and PG)according to alternative land conversion goals, characterized byranked priorities, and to the values of the land capability for energycrops. In the proposed scenario the conversion goals were energycrop production, erosion prevention, biodiversity conservation,flooding and peak flows reduction. The land conversion goals werederived from the following ranked environmental constraints:

1. Erosion risk;2. Presence of protected areas: Wildlife Nature Reserves, Areas of

Outstanding Natural Beauty, Environmentally Sensitive Areas,Heritage Coast, Local Nature Reserves, National Parks, NationalNature Reserve, Special Protection Areas, Sites of Importancefor Nature Conservation, Special Areas of Conservation, RamsarSites;

3. Flooding risk.

The diagram in Fig. 4 shows the model which allocates the landto the energy crop typologies, according to the different environ-mental constraints, conversion goals and land capability classes.

The allocation criteria are based on the assumption that energycrop systems can have different, positive or negative, effects on theenvironment in termsof land conservation, biodiversity preservationand balanced management of the natural resources (EEA, 2007;Hardcastle et al., 2006; McLaughlin & Walsh, 1998; Paulson et al.,2003; Samson, 1991; Theriault et al., 2003). SRC with a longergrowing cycle andSRFhave shownto increasebiodiversity comparedwith arable land (Hardcastle et al., 2006; Paulson et al., 2003; Sage &Tucker, 1998). Those plantations also have a stabilizing effect on thesoil, due to the lower frequency of needed mechanized operations,

Fig. 4. Land allocation model.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 731

which reduces soil compaction and erosion risk (Hardcastle et al.,2006; Makeschin, 1994). Perennial energy grasses have higherinterception rate and water use efficiency than SRC and SRF; more-over some species can grow on poorly drained soils, thus contrib-uting to reduce flood generation (Hall, 2003).

The implementedmodel allocates thedifferent crop typologies bymatching the environmental sensitivity of the land with the landconservation or land degradation potential of the crop systems. Theless intensive systems and less requiring crop typologies are thusadopted in areas with higher environmental constraints. Areas withpresence of erosion risk and protected areas are given the highestconstraint priority in the land allocation matrix. SRC with a highgrowing cycle and SRFwere thus allocated to those areas (priority 1).Perennial grasses are allocated to landwith flooding risk (priority 2).All the areas without environmental constraints are classifiedaccording to the land capability scheme presented in Fig. 2.

The implemented allocation scheme is based on generalecological considerations, without taking into account site-specificsensitivities and land management policies. In particular for pro-tected areas the land management strategies should follow specificconservation and protection goals (e.g. maintain traditional

agriculture, protection of landscape, biodiversity conservation).Local feasibility and Environmental Impact Assessment (EIA)studies are therefore necessary when planning energy cropsintroduction in protected areas. In some specific cases the plantingcannot be allowed, or the planting scale has to be constraint tosmall size fields. In some other less restrictive cases the planting ofeven herbaceous species could be allowed. Planting on grasslandolder than 5 years could be restricted, therefore EIA should also beconducted for planting on grassland and semi-natural areas (Lovettet al., 2009).

Results and discussion

Land capability map

The map in Fig. 5 represents the geographic distribution of theland capability classes for energy crops. The spatial distribution ofthe capability index illustrates that the upland areas of the York-shire Dales and North York Moors National Parks are less suitablefor energy crop cultivation. This is due to the presence of steepslopes, peat or shallow soils and extremes of temperature. After

Fig. 5. Land capability map for energy crops.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736732

ranking the index in capability classes, most of the surface resultedclassified as third (683,258 ha) and second (354,683 ha) class(Table 4). This land capability representation is theoretical and doesnot take into account the land cover and environmental constraintsthat are instead applied in the allocation model.

Uncertainty analysis

The result of parameter uncertainty analysis is presented in theconfusion matrix in Table 4, along with the overall accuracy andKappa index. The percentage of surface area variation due to theperturbed model input is higher for the first and last capabilityclasses, both showing a negative variation. The general effect ofthe parameter uncertainty propagation is a decrease of the capa-bility index, especially for the expected highest classes (first and

Table 4Uncertainty analysis: confusion matrix.

CCa (ha) CCa expected (ha)

1 2 3 4

1 5593 4852 0 02 158 266,131 88,393 03 0 11,220 650,543 21,4944 0 0 30,968 231,1175 0 0 0 25616 0 0 0 0Total 5752 282,204 769,905 255,172Overall accuracy 88.3%Khat 0.823

a CC ¼ Capability Class.

second). A capability index increase occurs only for the expectedfourth and sixth classes. The capability class shift is normally ofone level.

The spatial distribution of the error (Fig. 6) illustrates that thehighest uncertainties take place in the marginal uplands, fringingthe Pennines and North Yorks Moors, and for the highest landcapability on the coast. The result seems sensible with the Vales ofYork and Pickering and the Holderness peninsular being moreconfidently predicted, along with the uplands in the Pennines.

According to the cross-correlation analysis between the inputdata and the error map, the highest correlation occurs with the soilfactors: soil texture showed the highest correlation (0.20), followedby soil depth (0.13) and soil wetness (0.04). Among the climaticfactors, the most correlated to the error distribution is the GDD,with r value of 0.03. The correlation factor for the DTM is also of

Total Variation %

5 6

0 0 10,445 �450 0 354,683 �200 0 683,258 13

2656 0 264,741 �4100,234 20 102,815 5

5392 10,493 15,886 �34108,282 10,514 1,431,827

Fig. 6. Uncertainty analysis: spatial distribution of the error.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 733

0.0.3. All the other factors showed a very low correlation (below0.009). Soil texture and depth can be considered as the mostimportant control factors for the error of the land capability model.Therefore the soil type represents a dominant source of uncertaintyin the computation of the energy crop capability in the study area.The accuracy of the soil map should therefore be considered asa priority when collecting the data for performing land suitabilityassessment for energy crops.

Sensitivity analysis

The analysis of weight sensitivity shows a low sensitivity of themodel to a random variation of the factors’ weight (Table 5), withthe highest impact on the first capability class (highest percentageof variation). The general impact is a decrease of the surface areafrom the expected highest capability classes (first and second class),

Table 5Weight sensitivity analysis: confusion matrix.

CCa (ha) CCa expected (ha)

1 2 3 4

1 8522 1923 0 02 3 341,919 12,761 03 0 2940 676,760 35584 0 0 11,065 250,8625 0 0 0 246 0 0 0 0Tot 8525 346,783 700,699 254,875Overall accuracy 97.5%.Khat 0.962

a CC ¼ Capability Class.

and an increase of the surface area belonging to the lower classes(third, fifth and sixth class). The capability class variation due to thefactor’s weight is always in the order of one class.

Seven different maps of land capability were generated from thejack-knifing analysis, each one being created by leaving out one ofthe factors. The final result shows the relative impact of theexclusion of each factor on the land capability index and the newclass distribution (Table 6). Leaving out the climatic factors resultsmainly in an increase of the surface area belonging to the first twocapability classes, and a decrease of the third class. Leaving out thepH results in an increase of the first three classes and a decrease ofthe last three classes. The main effect of leaving out the texture andthe soil depth is, in both cases, an increase of the third and fourthcapability classes and a decrease of the second class. Leaving outthe wetness affects mainly the forth, and the third classes, withrespectively an increase and decrease of the surface area. The

Total Variation%

5 6

0 0 10,445 �180 0 354,683 �20 0 683,258 3

2814 0 264,741 �4102,791 0 102,815 4

1000 14,886 15,886 6110,508 16,027 1,431,827

Table 6Jack-knifing analysis: capability class variation (ha), per each left out factor.

CCa Expected Left out factor

Rain GDD pH Texture Depth Wetness Slope

1 10,445 71,891 26,067 13,377 �2574 �3868 1789 �29712 354,683 115,107 284,216 2855 �143,107 �160,003 �20,201 �169,6033 683,258 �184,746 �166,406 79,283 71,786 50,839 �119,991 12,4094 264,741 �70,613 �112,686 �63,993 63,611 56,792 125,859 178,4495 102,815 3760 �25,677 �25,660 16,058 57,922 15,203 �84,4566 15,886 64,601 �5514 �5861 �5774 �1683 �2659 66,172

a CC ¼ Capability Class.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736734

topography (slope) strongly affects the second and fourth classes,causing respectively a decrease and an increase of the surface areawhen excluded from the model.

Different ecological factors and weights can be applied whenchoosing single crop species or specific crop typologies. The factorweights sensitivity and the jack-knifing analysis simulate the effectthat specific ecological requirements, corresponding to specificparameters variation, have on the land capability assessment. Thesimulation results showed that, in general, the middle capabilityclasses (third and fourth) are the less sensitive to the input data andparameters deviation in terms of percentage of land surfacevariation.

Land allocation scenarioThe result of the land allocation model is represented by the

map of suitable locations for conversion to specific energy croptypologies (Fig. 7). Table 7 shows the available surface areas pereach plantation type. The conversion scenario shows a high

Fig. 7. Conversio

concentration of areas belonging to the higher energy crop capa-bility classes in the central districts of the study area. In the samedistrict there are also a few less suitable areas (class V), in corre-spondence with the drainage pattern where there is flooding risk.

According to this scenario the total available area for conversionto energy crop is of 4882 km2, this is comparable with the findingsof other energy crop suitability studies in the UK. According toLovett et al. (2009), about 8800 km2 are available for conversion toMiscanthus in Yorks and Humber Region, and about 3400 km2 aresuitable for conversion in the most restrictive scenario, whereenvironmental constraints and food production priorities are alsoconsidered. If we compare the only PG conversion area with thefindings of Lovett et al., a much smaller area, corresponding to468 km2, would be converted to Mischantus plantation. Accordingto this result more than half of the conversion area should beallocated to cropping systems with low land degradation potential(SRC with growing cycle more than five years and SRF). This energycrops penetration scenario can therefore be considered as

n scenario.

Table 7Distribution of suitable areas for each plantation type.

Plantation type Area (Km2) Suitable area%

SRC t < 3 54 13 < SRC < 5 1686 35SRC t > 5 2482 51SRF 191 4PG 468 10Total suitable area 4882

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736 735

restrictive, in accordance with targets of nature protection, landconservation and low competition with highly remunerable crops.

Conclusions

The process of introducing crops dedicated to energy productioninto existing agricultural systems needs to be supported by a properknowledge of the interactions and possible competition with theexisting pattern of rural land use. Therefore, an integration oftraditional land evaluation methods for food-crops with a morespecific evaluation of the land capability for energy crops is neededto support policies and land planning strategies. This paperdescribes and tests the possibility of adopting a unique landcapability approach for different energy crop systems. The evalua-tion process was based on the main physical parameters affectingthe energy crop cultivation, under various uncertainty conditions.The proposed GIS model integrates a multi-criteria approach witha quantitative analysis of error propagation and sensitivity to inputfactors and parameters variation. The model results allow identi-fying the energy crop typologies, among SRF, SRC and perennialgrasses, which best suit the most relevant environmental condi-tions, and the relative range of uncertainty.

The land capability model and the parameter uncertaintyanalysis described in this paper showed that the land which aremore sensitive in terms of environmental risk correspond to theland with both the lowest capability for bioenergy production andthe highest model error. In these areas, the introduction of inten-sive energy crop system would not be sustainable. The weightsensitivity and the jack-knifing analysis showed the impact thatvarying factor numbers and their relative weights may have on theland capability and energy crop potential. Factor weights variationshould be taken into account when selecting single crops specieswith specific ecological requirements.

The land allocation scenario developed in this study is based onthe assumption that relevant positive effects, in terms of landconservation and balanced management of natural resources, canderive from a right choice of energy crop typologies and location ofconversion area. Considering the levels of biodiversity associatedwith the UK agricultural land (Hope & Johnson, 2003), extensivecultivation systems which include energy crops in an integratedplanning system, may provide several benefits in terms of natureconservation.

There is scope for improvement in the multi-objective modelallocating agro-energy land use in existing land patterns. Themodel should ideally integrate spatial interactions and biomasslocation-allocation simulations based on the supply-demand andlogistic-costs factors. The model would ideally allow the analysis ofdifferent scenarios based on policy-economic perspectives (foodversus energy security and nature conservation), and stakeholders’preferences. The different scenario could be finally integrated ina Decision Support System which could sustain the environmentalplanning when implementing different bioenergy routes. In theprocess of choosing the conversion area and the best performingcrops, a detailed benefit-cost study, based on experimental data,

would be needed. Many energy crops are still under experimentalconditions, and the system uncertainty, as well as the farmer’sattitude to risk, should also be considered for each specific case. Thesocial aspects and the energy and economic balance of the wholebioenergy chains go beyond the scope of this study; neverthelessthose issues are crucial in order to define the most promisingbioenergy routes in relation to market, environmental and socialcriteria.

Acknowledgements

This study was undertaken under the auspices of the MarieCurie Early Stage Training Site BIOCONS (European Centre forBiodiversity & Conservation Research) linked to the Rural Economyand Land Use Programme (RELU). The support through the Euro-pean Commission’s Marie Curie Training Fellowships is gratefullyacknowledged.

References

Abbaspour, R. A., Delavar, M. R., & Batouli, R. (2003). The issue of uncertaintypropagation in spatial decision making. In Paper presented at the ScanGIS’2003 e9th Scandinavian Research Conference on Geographical Information Science, June2003, Espoo, Finland.

Aerts, J. C. J. H., Goodchild, M. F., & Heuvelink, G. B. M. (2003). Accounting for spatialuncertainty in optimization with spatial decision support systems. Transactionsin GIS, 7, 211e230.

Andersen, S., Towers, W., & Smith, P. (2005). Assessing the potential for biomassenergy to contribute to Scotland’s renewable energy needs. Biomass and Bio-energy, 2(29), 73e82.

Aylott, M., Casella, E., Tubby, I., Street, N. R., Smith, P., & Taylor, G. (2008). Yield andspatial supply of bioenergy poplar and willow short rotation coppice in the UK.New Phytologist, 178(2), 358e370.

Beccali, M., Columba, P., D’Alberti, V., & Franzitta, V. (2009). Assessment of bio-energy potential in Sicily: a GIS-based support methodology. Biomass andBioenergy, 1(33), 79e87.

Bellarby, J., Wattenbach, M., Tuck, G., Glendining, M. J., & Smith, P. (2010). Thepotential distribution of bioenergy crops in the UK under present and futureclimate. Biomass and Bioenergy, 12(34), 1935e1945.

Benke, K. K., & Pelizaro, C. (2010). A spatial-statistical approach to the visualisationof uncertainty in land suitability analysis. Journal of Spatial Science, 55(2),257e272.

Berndes, G. (2006). The contribution of renewables to society. In J. Dewulf, & H. VanLangenhove (Eds.), Renewables-based technology. London, UK: John Wiley &Sons Ltd.

Bibby, J., & Mackney, D. (1969). Land use capability classification. Technical Mono-graph No. 1: Soil surveys of England and Wales and Scotland. Harpenden: Roth-amsted Experimental Station.

Bibby, J. S., Heslop, R. E. F., & Hartnup, R. (1998). Land capability classification forforestry in Britain. Soil survey monograph. Aberdeen, UK: Macaulay Land UseResearch Institute.

Brown, C. C., Neilson, B., Lewandowski, I., & Jones, M. B. (2000). The modelledproductivity of Miscanthus�giganteus (GREEF et DEU). Ireland Industrial Cropsand Products, 12(2), 97e109.

Canada Land Inventory. (1970). The Canada land inventory: Objectives, scope, andorganization. Report No 1. Ottawa: Department of Regional EconomicExpansion.

Cannell, M. G. R., Sheppard, J., & Milne, R. (1988). Light use efficiency and woodybiomass production of poplar and willow. Forestry, 61(2).

Carver, S. (1991). Integratling multicriteria evaluation with GIS. International Journalof Geographical Information Systems, 5(3), 321e339.

Chen, Y., Yu, J., & Khan, S. (2010). Spatial sensitivity analysis of multi-criteria weightsin GIS-based land suitability evaluation. Environmental Modelling & Software,25(12), 1582e1591.

Congalton, R. G. (2001). Accuracy assessment and validation of remotely sensed andother spatial information. International Journal of Wildland Fire, 10, 321e328.

Crosetto, M., Tarantola, S., & Saltelli, A. (2000). Sensitivity and uncertainty analysisin spatial modelling based on GIS. Agriculture, Ecosystems & Environment, 81,71e79.

DEFRA. (2004). Growing short rotation coppice. DEFRA Publications. Available athttp://www.defra.gov.uk/erdp.

DEFRA. (2007). Planting and growing Miscanthus. Best practice guidelines for appli-cants of DEFRA’s Energy Crop Scheme. DEFRA Publications. Available at http://www.defra.gov.uk.

DTI, DFT, DEFRA. (2007). UK biomass strategy. Available at http://www.mansea.org/pdf/ukbiomassstrategy-0507.pdf.

Eastman, J. R. (2003). IDRISI Kilimanjaro. Guide to GIS and image processing.Worcester, MA: Clark University Press.

P. Tenerelli, S. Carver / Applied Geography 32 (2012) 724e736736

Edwards, R. A. H., �Súri, M., Huld, M. A., & Dallemand, J. F. (2005). GIS-basedassessment of cereal straw energy resource in the European Union. In Paperpresented at the 14th European Biomass Conference & Exhibition Biomass forEnergy, Industry and Climate Protection, October 2005, Paris.

EEA. (2006). How much bioenergy can Europe produce without harming the envi-ronment? EEA. Technical Report, 7. Available at http://www.europabio.org/Biofuels%20reports/eea_report_bioenergy.pdf.

EEA. (2007). Estimating the environmentally compatible bioenergy potential fromagriculture. Technical report, 12. EEA. Available at http://www.eea.europa.eu/publications/technical_report_2007_12.

Elbersen, H. W., Christian, D. G., El Bassam, N., & Sauerbeck, G. (2010). Switchgrass inNW Europe. Final Report FAIR 5-CT97-3701(2010). Available at www.switchgrass.nl.

Forestry Commission. (1998). Harvesting and comminution of short rotation coppice.Technical Note e Technical Development Branch, Forest 8(98).

Forestry Commission. (2002). Establishment and management of broadleaved coppiceplantations for energy. Available at www.biomassenergycentre.org.uk.

Freppaz, D., Minciardi, R., Robba, M., Rovatti, M., Sacile, R., & Taramasso, A. (2004).Optimizing forest biomass exploitation for energy supply at a regional level.Biomass and Bioenergy, 1(26), 15e25.

Giampietro, M., Ulgiati, S., & Pimentel, D. (1997). Feasibility of large-scale biofuelproduction. BioScience, 47(9), 587e600.

Goodchild, M. F. (1986). Catmog. Spatial Autocorrelation, Vol. 47. Norwich: Geo Books.Graham, R. L., English, B. C., & Noon, C. E. (2000). A geographic information system-

based modelling system for evaluating the cost of delivered energy cropfeedstock. Biomass and Bioenergy, 4(18), 309e329.

Hall, R. L. (2003). Grasses for energy reduction e Hydrological guidelines. DTI New andRenewable Energy Programme. Available at http://www.berr.gov.uk/files/file14946.pdf.

Hardcastle, P. D., Calder, I., Dingwall, C., Garret, W., McChesney, I., Matthew, J., et al.(2006). A review of the potential impacts of short rotation forestry. ForestryCommission. Available at www.forestry.gov.uk.

Heaton, E., Voigt, T., & Long, S. P. (2004). A quantitative review comparing the yieldsof two candidate C4 perennial biomass crops in relation to nitrogen, temper-ature and water. Biomass and Bioenergy, 27(1), 21e30.

Heuvelink, G. B. M. (1998). Error propagation in environmental modelling with GIS.London: Taylor & Francis.

Hope, A., & Johnson, B. (2003). English nature e Discussion paper on biofuelsTerrestrial Wildlife Team. Available at www.english-nature.org.uk.

Howes, P. (2007). Yorkshire and Humber vision for biomass. Available at http://www.yhassembly.gov.uk/News/2008/Yorkshire%20and%20Humber%20Vision%20for%20Biomass/.

Hunter, G. J., Robey, M., & Goodchild, M. F. (1994). A toolbox for assessing uncer-tainty in spatial databases. In. Paper presented at the 22nd Annual Conference,November 1994. Sydney: Australasian Urban and Regional Information SystemsAssociation.

Ignaciuk, A., Vöhringer, F., Ruijs, A., & van Ierland, E. C. (2004). Competition betweenbiomass and food production in the presence of energy policies: a partialequilibrium analysis. Energy Policy, 34(10), 1127e1138.

Jensen, J. R. (1996). Introductory digital image processing: A remote sensing perspec-tive (Second ed).. New Jersey, U.S.A: Prentice Hall.

Jobling, J. (1990). Poplars for wood production and amenityIn Forestry CommissionBulletin, Vol. 92. London: HMSO Press.

Johansson, D. J. A., & Azar, C. (2007). A scenario based analysis of land competitionbetween food and bioenergy production in the US. Climatic Change, 82(3-4),267e291.

Johnston, D. M., & Timlin, D. (2000). Spatial data accuracy and quality assessmentfor environmental management. In Paper presented at the 4th internationalsymposium on spatial accuracy assessment in natural resources and environmentalsciences. Delft: Delft University Press.

Klingebiel, A. A., & Montgomery, P. H. (1961). Land capability classification. Agric.handbook 210. Soil conserve. Washington, DC: Ser. V. S. Gov. Press.

Landis, J. R., & Koch, G. G. (1977). The measurement of observer agreement forcategorical data. Biometrics, 33, 159e174.

Lewandowski, I., Scurlock, J. M. O., Lindvall, E., & Christou, M. (2003). The devel-opment and current status of perennial rhizomatous grasses as energy crops inthe US and Europe. Biomass and Bioenergy, 4(25), 335e361.

Lovett, A. A., Sünnenberg, G. M., Richter, G. M., Dailey, A. G., Riche, A. B., & Karp, A.(2009). Land use implications of increased biomass production identified byGIS-based suitability and yield mapping for Miscanthus in England. BioenergyResearch, 2, 17e28.

Luo, W., & Stepinski, T. (2008). Identification of geologic contrasts from landscapedissection pattern: an application to the Cascade Range, Oregon, USA.Geomorphology, 1-4(99), 90e98.

Ma, J., Scott, N. R., DeGloria, S. D., & Lembo, A. J. (2005). Siting analysis of farm-basedcentralized anaerobic digester systems for distributed generation using GIS.Biomass and Bioenergy, 6(28), 591e600.

Makeschin, F. (1994). Effects of energy forestry on soils. Biomass and Bioenergy,1-2(6), 63e79.

Martelli, F., Maltagliati, S., Riccio, G., Bernetti, I., Fagarazzi, C., Fratini, R., et al. (2002).GIS-based planning tool for Greenhouse gases emission reduction through

biomass exploitation in Tuscany. In Paper presented at the 12th EuropeanConference on Biomass for Energy, Industry and Climate Protection, June 2002,Amsterdam RAI, The Netherlands.

Masera, O., Ghilardi, A., Drigo, R., & Trossero, M. A. (2006). Wisdom: a GIS-basedsupply demand mapping tool for woodfuel management. Biomass and Bio-energy, 7(30), 618e637.

McLaughlin, S. B., & Walsh, M. E. (1998). Evaluating environmental consequences ofproducing herbaceous crops for bioenergy. Biomass and Bioenergy, 14(4),317e324.

Mitchell, C. P., Stevens, E. A., & Watters, M. P. (1999). Short-rotation-forestry oper-ations, productivity and costs based on experience gained in the UK. ForestEcology and Management, 121(1), 123e136.

Natural England. (2009). Energy crops scheme establishment grants handbook (3rdEd).. Available at www.naturalengland.org.uk/publications.

NNFC. (2007). Non-food crops database. The National Non-Food Crops Centre.Available at http://www.nnfcc.co.uk/metadot.

Noon, C. E., & Daly, M. J. (1996). GIS-based biomass resource assessment with bravo.Biomass and Bioenergy, 10(2e3), 101e109.

O’Brien, R. (2008). Visualising uncertainty in spatial decision support. In Paperpresented at the 8th International Symposium on Spatial Accuracy Assessment inNatural Resources and Environmental Sciences, June 2008, Shanghai, P. R. China.

Openshaw, S., Charlton, M., & Carver, S. (1991). Error propagation: a Monte Carlosimulation. In I. Masser, & M. Blakemore (Eds.), Handling geographic information:Methodology and potential applications (pp. 78e101). London: Longman Press.

Ordnance Survey. (2004). Land-form PANORAMA. User guide, Vol. 4, Available athttp://gidimap.edina.ac.uk.

Overend, R. P., & Mitchell, C. P. (2000). Modelling biomass and bioenergy. Biomassand Bioenergy, 4(18), 263e264.

Paulson, M., Bardos, P., Harmsen, J., Wilczek, J., Barton, M., & Edwards, D. (2003). Thepractical use of short rotation coppice in land restoration. Land Contaminationand Reclamation, 3, 323e338.

Perry, M., & Hollis, D. (2005). The generation of monthly gridded datasets fora range of climatic variables over the UK. International Journal of Climatology, 25,1041e1054.

REN21. (2010). Renewables 2010 global status report. Paris: REN21 Secretariat.Available at http://www.ren21.net/Portals/97/documents/GSR/REN21_GSR_2010_full_revised%20Sept2010.pdf.

Richter, G. M., Riche, A. B., Dailey, A. G., Gezan, S. A., & Powlson, D. S. (2008). Is UKbiofuel supply from Miscanthus water-limited? Soil Use and Management, 24,235e245.

Rossiter, D. G. (1994). Lecture notes: Land evaluation. Cornell University College ofAgriculture & Life Science Department of Soil, Crop & Atmospheric SciencePress.

Rossiter, D. G. (2004). Technical Note: Statistical methods for accuracy assessment ofclassified thematic maps. Enschede, NL: Technical Report ITC Press.

Sage, R., & Tucker, K. (1998). Integrated crop management of SRC plantations tomaximise crop value, wildlife benefits and other added value opportunities.Available at http://hdl.handle.net/10068/373688.

Samson, R. (1991). Switchgrass: a living solar battery for the prairies. SustainableFarming, 5(3).

Smith, E. (2002). Uncertainty analysis. In A. H. El-Shaarawi, & W. W. Piegorsch(Eds.), Enclcyopedia of environmentrics (pp. 2283e2297). Chichester: John Wiley& Sons Press, Ltd.

Tenerelli, P., & Monteleone, M. (2008). A combined land-crop multicriteria evalu-ation for agro-energy planning. In Paper presented at the 16th Biomass Confer-ence, June 2008, Valencia.

Tenerelli, P., Pantaleo, A., Carone, M. T., Pellerano, A., & Recchia, L. (2007). Spatial,environmental and economic modeling of energy crop routes: liquid vs solidbiomass to electricity chains in Puglia Region. In Paper presented at the 15thEuropean Biomass Conference, May 2007, Berlin.

Theriault, F., Javorska, H., Casova, K., Tucker, M., & Hansen, T. G. (2003). The potentialfor perennial grasses as energy crops in organic agriculture. In EcologicalAGRICULTURE I, SOCRATES European Common Curriculum, Final report. The RoyalVeterinary and Agricultural University Denmark Press.

Towers, W., Morrice, J., Aspinall, R. J., Birnie, R. V., & Dagnall, S. (1997). Assessing thepotential for short rotation coppice in Scotland. In Paper presented at theEnvironmental Impact of Biomass for Energy Conference. The Netherlands: Centrefor Agriculture and Environment Utrecht.

Tuck, G., Glendining, M. J., Smith, P., Housec, J. I., & Wattenbachb, M. (2006). Thepotential distribution of bioenergy crops in Europe under present and futureclimate. Biomass and Bioenergy, 30(3), 183e197.

Tukey, J. W. (1958). Bias and confidence in not quite large samples. Annals ofMathematical Statistics, 29, 614.

UNEP. (2009). Towards sustainable production and use of resources e Assessing biofuels.United Nations Environment Programme. Available at http://www.unep.org.

Voivontas, D., Assimacopoulos, D., & Koukios, E. G. (2001). Assessment of biomasspotential for power production: a GIS based method. Biomass and Bioenergy,2(20), 101e112.

Zerger, A., Smith, D. I., Hunter, G. J., & Jones, S. D. (2002). Riding the storm:a comparison of uncertainty modelling techniques for storm surge riskmanagement. Applied Geography, 22(3), 307e330.