Um

PRa

b

c

d

KHLBP

1h

Ecological Indicators 33 (2013) 82– 95

Contents lists available at SciVerse ScienceDirect

Ecological Indicators

j our nal homep age: www.elsev ier .com/ locate /eco l ind

sing landscape structure to develop quantitative baselines for protected areaonitoring

aola Mairotaa,∗, Barbara Cafarelli b, Luigi Boccaccioa, Vincenzo Leronnia,occo Labadessaa, Vasiliki Kosmidouc, Harini Nagendrad

Department of Agro-Environmental and Territorial Sciences, University of Bari “Aldo Moro” via Orabona 4, I-70126 Bari, ItalyDepartment of Economics, University of Foggia, Largo Papa Giovanni Paolo II, 1, I-71100 Foggia, ItalyInformation & Technologies Institute (ITI), Centre for Research & Technology Hellas (CERTH), 6th km Harilaou – Thermi, 57001 Thessaloniki, P.O. Box: 60361, GreeceAshoka Trust for Research in Ecology and the Environment, Royal Enclave, Srirampura, Jakkur Post, Bangalore 560064, India

a r t i c l e i n f o

eywords:abitat fragmentationandscape heterogeneityiodiversity indicatorsrotected areas monitoring

a b s t r a c t

Changes in habitat extent as well as landscape and habitat structure are often caused by human pres-sure within protected areas and at their boundaries, with consequences for biodiversity and speciesdistributions. Thus quantitative spatial information on landscape mosaic arrangements is essential, formonitoring for nature conservation, as also specified by frameworks such as the Convention on BiologicalDiversity (CBD), and the European Union’s Habitat Directive. While measuring habitat extent is a relativelystraightforward task, approaches for measuring habitat fragmentation are debated. This research aimsto delineate a framework that enables the integration of different approaches to select a set of site- andscale-specific indices and synthetic descriptors and develop a comprehensive quantitative assessment ofvariations in human impact on the landscape, through assessment of habitat spatial patterns, which canbe used as a baseline for monitoring. This framework is based on the use of established methodologiesand free software, and can thus be widely applied across sites. For each landscape and observation scale,the framework permits the identification of the most relevant indices, and appropriate parameters fortheir computation. We illustrate the use of this framework through a case study in a protected area inItaly, to indicate that integrated information from multiple approaches can provide a more completeunderstanding of landscape and habitat spatial pattern, especially related to locations experiencing dis-turbance and pressure. First, identification of a parsimonious set of traditional LPIs for a specific landscapeand spatial scale provides insights on the relation between landscape heterogeneity and habitat fragmen-tation. These can be used for both change assessment and ranking of different sections of the study areaaccording to a fragmentation gradient in relation to matrix quality. Second, morphological spatial patternanalysis (MSPA), provides a pixel based structural characterisation of the landscape. Third, compositionalcharacterisation of the landscape at the pixel level is provided by landscape mosaic analysis. These threeapproaches provide quantitative assessments through indices which can be used singly or in combinationto derive three synthetic descriptors for a comprehensive quantitative baseline representation of land-scape structure that can be used for monitoring: the first descriptor, landscape diversity profiling, basedon the output of landscape mosaic analysis, at the landscape level, complements the evaluation which atthe pixel level can be obtained by more complex modelling; the second descriptor, obtained combining ofthe outputs of MSPA and the landscape mosaic analysis, informs on the local structural pattern gradientacross the landscape space; the third descriptor, derived from the integration of selected LPIs and those

derived from MSPA into a discontinuities detection procedure, allows for the identification of “criticalpoints” of transitions in management where threats to biodiversity and ecosystems integrity may belikely. The framework developed has significant potential to capture information on major landscapestructural features, identify problematic areas of increased fragmentation that can be used to prioritiseresearch, monitoring and intervention, and provide early warning signals for immediate response topressures increasing habitat fragmentation, with the goal of facilitating more effective management.

∗ Corresponding author. Tel.: +39 080 5443021; fax: +39 080 5442504.E-mail address: paola.mairota@uniba.it (P. Mairota).

470-160X/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.ecolind.2012.08.017

© 2012 Elsevier Ltd. All rights reserved.

cal Ind

1

bchttrMDbUf2I2

tvblsorv2epeb

Eocwcaacwaas

(to“CtanoasracL2a2

am

P. Mairota et al. / Ecologi

. Introduction

Monitoring landscape and habitat change in protected areas hasecome a major issue for international nature conservation agen-ies, super-national, national and regional authorities, on the oneand, and local managers on the other. Such monitoring is requiredo assess the impact of conservation policies and evaluate the effec-iveness of financial investments, and assess the effectiveness ofesponse measures within an adaptive management perspective.onitoring is also mandated under the Convention on Biologicaliversity (CBD), wherein 22 biodiversity headline indicators haveeen internationally adopted (Strand et al., 2007). At the Europeannion (EU) and Pan European levels (Council of Europe, 2004), the

ramework suggested by the CBD has been used to develop a set of6 indicators within the Streamlining European 2010 Biodiversityndicators (SEBI 2010) project (European Environmental Agency,009).

In particular, the first three CBD focal areas (i.e., “Status andrends of the components of biological diversity”, “Threats to biodi-ersity”, “Ecosystem integrity, and ecosystem goods and services”),ear a strong relation to landscape structure, which according to

andscape ecology is considered to be a component of biodiver-ity (Jedicke, 2001), and thus regular observation for monitoringf the structure of landscapes containing habitats of concern isequired. Such data are mostly derived by means of Earth Obser-ation (EO) techniques, both satellite and airborne (Strand et al.,007; Buchanan et al., 2008). In addition to indirect measures oncosystem functioning, such as those based on climate, topogra-hy, primary productivity, disturbance (Duro et al., 2007; Nagendrat al., 2012), EO data can be very useful in providing the informationase to derive landscape configuration and composition indicators.

Data on landscape spatial pattern is particularly required in theuropean context where great importance is given to the protectionf habitats of “community interest” for species conservation (Coun-il Directive 92/43/EEC of 21 May 1992 “Habitats Directive”), andhere there is a legal obligation for EU Member States to report on

hanges in conservation status (including changes in habitat extentnd configuration) every six years (Directive 92/43/EEC, articles 11nd 17). Habitat extent and landscape and habitat configuration areritical to monitor, as these provide indicators of human pressureithin and around protected areas (DeFries et al., 2005). Moreover,

s the processes driving landscape transformation in protectedrea-embedded landscapes act at quite fast rates and fine spatialcales, they thus require detailed temporal and spatial monitoring.

Two of the SEBI 2010 proposed indicators, “Ecosystem coverage”relevant to the “Trends in extent of selected biomes, ecosys-ems and habitats” headline indicator of the “Status and trendsf the components of biological diversity” focal area) and theFragmentation of natural and semi-natural areas” (relevant to theonnectivity/fragmentation of ecosystems” headline indicator ofhe “Ecosystem integrity and ecosystem goods and services” focalrea) appear suitable for monitoring through the use of EO tech-iques (Jongman et al., 2011). EO data and associated approachesf spatial analysis are now capable of providing detailed, reliablend frequently updated information at the spatial and temporalcales required for protected area monitoring. These allow for theegular generation of land cover/land use (LC/LU) and habitat mapst different spatial scales based on the adoption of more meaningfullassification taxonomies (Tomaselli et al., 2011), such as the FAO-and Cover Classification System (LCCS, Di Gregorio and Jansen,005), the European Nature Information System (EUNIS, Daviesnd Moss, 2002), or General Habitat Categories (GHCs, Bunce et al.,

011).

Generating baseline measurement of habitat extent is a rel-tively straightforward task. In contrast, producing baselineeasurements of habitat fragmentation that can be used for

icators 33 (2013) 82– 95 83

continued monitoring of changes in landscapes of conservationsignificance constitutes a highly debated issue in the landscapeecological literature, lacking standardised and globally applica-ble approaches. An arsenal of landscape pattern indices (LPI) isavailable for landscape pattern analysis (LPA) on discrete (categor-ical) cartographic data (O’Neill et al., 1988; Forman, 1995; Baskentand Jordan, 1995; Haines-Young and Chopping, 1996), and reli-able computational tools have been developed for this purpose(Baker and Cai, 1992; McGarigal et al., 2002). These have beenwidely applied to a range of landscapes and scales, yet, apart fromfew notable exceptions (e.g., Riitters et al., 1995; McGarigal andMcComb, 1995; Tinker et al., 1998) the selection of specific LPIsfor each site appears idiosyncratic, mainly based on researchers’knowledge, experience, and even personal preferences, often with-out extensive investigation of the range of possible indices. This isproblematic, as although several metrics can capture some aspectsof landscape configuration, and hence return some information onfragmentation, these are not intrinsically capable of disentanglingthe contribution of habitat loss to fragmentation from fragmenta-tion per se (Fahrig, 2003). A further complication is introduced bythe dependence of landscape heterogeneity on the spatial scale ofanalysis (Wu, 2004; Wiens et al., 1993). Therefore, it becomes dif-ficult to propose a single recipe for the selection of a relevant set ofLPIs, as the selection of a set of indices appears to be highly contin-gent on the landscape, as well as the question being asked (Fahrig,2003; Wu, 2004; Taylor et al., 2006).

In addition to LPIs, two approaches have been recently proposedfor the spatial analysis of landscape pattern from categorical maps,that seem to be promising for generating baseline information thatcan be of use for protected area monitoring, i.e., morphological spa-tial pattern analysis, MSPA (Soille and Vogt, 2009), and the analysisof landscape mosaics (Riitters et al., 2009). Both of these represent“local” approaches to landscape analysis, retaining information onthe spatial properties of the surroundings within which pixels in alandscape are situated. Hence, they offer the possibility of explic-itly testing the response of these indicators to changes in the spatialscale of observation, which can be very valuable when using EO datafor monitoring, which is itself a highly scale-dependent approach.

These approaches to LPA, being relatively new, have not beencompared against more standardly used approaches of “global”,i.e. patch, class and landscape based spatial pattern analysis, norused in combination with these widely used approaches. Such acomparison appears extremely useful to investigate the potentialof adopting EO data and associated spatial analysis techniques toprovide baseline data for landscape monitoring, in order to betteraddress conservation challenges.

The objective of this paper is to take a single case study as anillustrative example, and perform quantitative landscape structureanalysis using both “global” LPI analyses and “local” pixel basedapproaches, in order to identify a set of indices and associated spa-tial scales that can be used to generate baseline data on landscapeand habitat spatial arrangements that can serve as CBD/SEBI indi-cators (specifically for “Fragmentation of natural and semi-naturalareas”), to be used within the framework of habitat monitoring inprotected areas and their surroundings. Information derived fromdifferent approaches is compared with reference to its usefulness,and the possibilities of using of these indices for monitoring (eitherindividually or in combination) are explored. While the specificset of indicators and spatial scales derived from this study willbe location specific, the proposed approach for selection of site-specific and scale-specific indicators is standardisable, repeatableand robust, and can be extended to other locations with their own

site-specific characteristics, conservation challenges and relevantspatial scales.

The analysis was intentionally performed by means of wellestablished and consuetudinary techniques and the use of free

8 al Ind

sctctGewlivw

MpdmanWaaif

2

2s

tAs(

4 P. Mairota et al. / Ecologic

oftware for the analysis of categorical data, so as to provide alear guide-map that addresses the operational opportunities andhe technical challenges related to each approach. Among suchhallenges there are the systematic examination of LPIs, objec-ive sampling design, scale considerations, software setting andIS tools. This study is therefore also aimed at guiding practition-rs and regional and local authorities to more informed decisionshile choosing approaches for generating baseline information on

andscape pattern and using this information for future monitor-ng, which, although very important from a conservation point ofiew as well as legally required, nowadays cannot be undertakenithout appropriate technical skills.

The illustrative case study is derived from the Biodiversityulti-SOurce Monitoring System: From Space To Species (BIO SOS)

roject (www.biosos.eu), a three-year research project aimed ateveloping a pre-operational system for cost- and time-effectiveonitoring of changes in the land cover and habitats within and

long the borders of protected areas, in order to judge the effective-ess in protecting and conserving the regions from human impacts.ithin such a framework, the quantitative landscape structure

nalysis carried out, besides being aimed at the identification ofppropriate indicators, is also preliminary to an ongoing empir-cal study focused on the investigation of the impact of habitatragmentation on plant and animal communities in the same site.

. Materials and methods

.1. Study area, focal class and sampling for the illustrative casetudy

The study area is located in Southern Italy within one of the

est sites of the BIO SOS project, the Natura 2000 (N2K) “Murgialta” IT9120007 SCI/SPA, spanning over ∼125,880 ha (Fig. 1). Theame site is also partly designated and managed as a National ParkNP). The spatial extent of the study area (100 km2) covers parts of

Fig. 1. Study extent location map for the illustrative example withi

icators 33 (2013) 82– 95

the N2K site and the NP as well as areas outside both, differing interms of protection regime. Thus it contains two different types ofbuffer zone (interfaces), one in which official bindings of the N2Kapply but the actual legal provisions for the NP are not in force, andthe other with neither formal nor legal protection. Here differentkinds of impact are likely to occur, producing spatially explicit (i.e.,detectable by EO) effects on habitat types (as defined by Annex I to92/43/EEC Directive).

The ecosystem type of conservation concern within this N2K siteis “grasslands”. These cover ∼29,800 ha (24% of the total area of thesite) and represent what remains from the ∼80,000 ha existing atthe beginning of the 20th century. By the early 1990s, grasslandsextent had dropped to 40,000 ha, due to transitions to other landuses (predominantly, transformations to agriculture/urban). Sub-stantial losses of this ecosystem type have occurred between 1990and 2000, mainly due to EU incentives promoting durum wheatproduction, contemporaneous to the enforcement of the 92/43/EECDirective.

Grasslands are represented by two Corine Land Cover (CLC) cat-egories (EEA, 2010): Pastures (231) and Natural grasslands (321),however only the latter occurs within the site. This in the “MurgiaAlta” site corresponds to the LCCS: A12 A2 A10 B4 E5 B12 E6(Tomaselli et al., 2011). As with most European grasslands (EEA,2010), these grasslands can be defined as ‘semi-natural’ becausethey have been developed through a mix of anthropogenic andnatural processes including long periods of grazing by domesticstock, cutting and even deliberate burning regimes, and aremodified and maintained by human activities, mainly throughgrazing and/or cutting regimes (Turbé et al., 2010). They areamong the most species-rich plant communities in Europe andthey host a remarkable set of endemic and orchid species (see

also Wilson et al., 2012). Two habitat types listed in Annex I to92/43/EEC Directive are associated with grasslands at this site:62A0 “Eastern sub-Mediterranean dry grasslands (Scorzonerataliavillosae)” and 6220* “Pseudo-steppe with grasses and annuals of

n the BIO SOS test site Natura 2000 “Murgia Alta” IT9120007.

cal Indicators 33 (2013) 82– 95 85

ti

pgtgsfi

1o(

taewrcea1

soNmit

2f

ha“llmob(awCsm2thiaTfl2itia1

t

P. Mairota et al. / Ecologi

he Thero-Brachypodietea”. More than 100 animal species occurringn the site are listed in European Directives or other red lists.

Agricultural intensification and land abandonment togetherrovide two of the main pressures on biodiversity linked to theserassland ecosystems. The increasing extension of arable crops con-ributes directly to the reduction in the extent of semi-naturalrassland. In addition, pastures become susceptible to shrub inva-ion after the decline of traditional management practices. Habitatragmentation and conversion to bio-fuels or forestry are alsoncreasingly threatening the landscape.

The analysis was carried out on a thematic LC/LU map (2006,:5000 nominal scale), reclassified from the CLC classification tax-nomy into the LCCS by means of site specific correspondence rulesTomaselli et al., 2011; Kosmidou et al., 2011).

A stratified random sampling within a regular standard (Direc-ive 2007/2/EC, INSPIRE, 2010) grid 1 km × 1 km was appliedccording to the internal protocols of the BIO SOS project (Mairotat al., 2011b). A set of thirty 1 km × 1 km landscapes (sample set A)as selected within the spatial extent defined above, so as to rep-

esent all the twenty two LC/LU classes present, hence the differentontexts of the focal class within such an extent (Fig. 2a). To thisnd, a three clusters K-means classifier (Everitt et al., 2011) waspplied to the proportions of each LC/LU class within each of the

km × 1 km landscapes and randomly select from these.In addition to the thirty sample landscapes, a 2 km × 10 km tran-

ect (sample set B) was selected such that it spanned over a gradientf management types and ecosystems, encompassing the whole/S dimension of the spatial extent as well as a mix of manage-ent regimes including areas outside the N2K site and the NP, areas

nside the N2K site and the NP, and areas inside the N2K but outsidehe NP (Fig. 2b).

.2. Proposed framework for quantitative evaluation of habitatragmentation

A comprehensive framework for quantitative evaluation ofabitat fragmentation is proposed, based on both the patch, classnd landscape based “global” approaches of LPA and the pixel basedlocal” approaches of morphological spatial pattern analysis andandscape mosaic analysis. While all approaches quantify aspects ofandscape structure, they provide different and potentially comple-

entary information. In this context, landscape structure and onef its particular aspects, i.e. habitat fragmentation, are evaluatedased on the landscape’s spatial (configuration) and non-spatialcomposition) properties, with no direct link to the behaviouralttributes of any species. This is with respect to the overall goal,hich is the identification of appropriate indices related to theBD/SEBI biodiversity indicator “Fragmentation of natural andemi-natural areas”, thus connecting the analysis to landscapesapped using traditional vegetation associations (Rodwell et al.,

002) and existing European habitat classifications, e.g., the EUNIS,he GHCs and CLCs. These classifications in turn are derived fromuman perception and interpretation of both EO and field data that

s also the scale at which management decisions are implementednd communicated to stakeholders and to the rest of the society.his does not imply the assumption that, given the premise thatunctional processes are linked to pattern, LPA can be used to inferandscape functioning (Tischendorf and Fahrig, 2000; Taylor et al.,006). On the contrary, it is asserted here that LPA is useful as an

ndicator of ecosystem function if performed and interpreted inhe context of the species and of the process under investigationn which appropriate variables related to species functional traits

re considered at the appropriate scale domain (Kotliar and Wiens,990).

A first step of this proposed framework consists in the applica-ion of quantitative techniques referring to traditional LPA, for the

Fig. 2. Selected 1 km × 1 km sample landscapes, sample set A (a), and 2 km × 10 kmsample transect, sample set B (b), within the study extent.

computation of LPIs, (O’Neill et al., 1988; Li et al., 1993; Forman,1995; McGarigal and McComb, 1995; Riitters et al., 1995; Baskent

and Jordan, 1995; Haines-Young and Chopping, 1996; Tinkeret al., 1998; McGarigal et al., 2002; Schindler et al., 2008), mor-phological pattern analysis, MSPA (Soille and Vogt, 2009) andlandscape mosaic analysis (Riitters et al., 2000, 2009). A second step

8 al Ind

caim1b(if

2

gr

siiiecfT2msdiacfaompTcbseabccdpipc

cai(iT((cNasb(S4

of a traditional method based on diversity profiling to syntheti-

6 P. Mairota et al. / Ecologic

onsists in further processing of the indices according to both novelnd well known methods to obtain synthetic descriptors allow-ng comparisons across space and/or time to provide a baseline to

onitor changes over time. These include diversity profiling (Hill,973, 1997), computation of the Similarity Index (SI) as proposedy Estreguil and Caudullo (2010), and detection of discontinuitiesJohnston et al., 1992). The operational details of the specific sett-ngs and tools adopted for the implementation of the software usedor computation are reported in Supplementary Table 1.

.2.1. Landscape pattern indicesThe computation of LPIs allows the characterisation of the

eometric and topological properties of categorical map patternsepresented at a single scale.

For the selection of a parsimonious set of LPIs, systematiccrutiny of the available indices is performed. Twenty-four LPIs arenitially selected from different types, i.e., area/density/edge, shape,solation/proximity, contrast, contagion/interspersion, connectiv-ty, composition (Haines-Young and Chopping, 1996; McGarigalt al., 2002), following McGarigal and McComb (1995), andoncentrating on those describing the spatial attributes of theocal LC/LU/habitat class within the landscape (Supplementaryable 2). Among such indices, the “Effective Mesh Size” (Jaeger,000) is of particular interest here. This index has proven to (a)onotonically decrease with increasing fragmentation and be con-

istent throughout the phases of the fragmentation process asefined by Jaeger (2000) based on Forman (1995); (b) be mathemat-

cally “intensive”, meaning that it can be interpreted as quantifyingn intrinsic landscape feature (Jaeger, 2000); (c) be mathemati-ally “area proportionately additive”, meaning that it is suitableor comparing fragmentation of regions of different extent and forssessing the influence of a part of a region to the fragmentationf the whole region (Jaeger, 2000). In order to compute contrastetrics, edge contrast weights must be defined as objectively as

ossible to approximate functional differences between LU classes.hese are landscape and analysis specific. The definition of edgeontrast weights in this case has included the identification (on thease of expert knowledge) of a set of descriptors (e.g., naturalness,oil cover, plant community type and vegetation structure) consid-red appropriate to represent each class and underline structuralnd functional differences between class pairs. Each descriptor isroken down into levels pertaining to an artificial versus naturalondition gradient for that descriptor itself. At each level, quantifi-ation scores are associated, ranging from 0 to 0.5, considering thatifferent descriptors affect contrast in different ways. For each classair, the sum of the absolute differences for all descriptors is used to

ndicate the contrast for a particular edge between the given classairs, ranging between 0 (low edge contrast) and 0.9 (high edgeontrast) – further details are provided in Mairota et al. (2011b).

Core area type indices are not considered due to the diffi-ulty of objectively identifying edge width for each LC/LU class. Tossess the contribution of landscape configuration to fragmentationndependently from the contribution of habitat relative amountfragmentation per se, Fahrig, 2003), and to reduce the initial set ofndices, the computed LPIs for the sample set A (Supplementaryable 3) are statistically treated as in McGarigal and McComb1995), which to date appears as the most appropriate methodSmith et al., 2009). This has implied the following integrated pro-edure: (1) the NP and IJI indices are removed from the initial set asP coincides with PD at the working grain (1 m) and IJI is undefinednd reported as “N/A” in the FRAGSTATS output for 3 sample land-capes containing less than 3 LC/LU; (2) a further selection is doneased on clear redundancy of some indices, which was expected

Tinker et al., 1998; McGarigal and McComb, 1995; Li et al., 1993;chindler et al., 2008; Riitters et al., 1995) (Supplementary Table); (3) the influence of the percentage habitat area in the landscape

icators 33 (2013) 82– 95

(PLAND) on the other LPIs indices is removed by a regression anal-ysis. In particular, a regression model is fitted for each LPIs indexagainst PLAND. The appropriate regression model (linear models,log-linear models, generalised linear models and additive models)and the goodness of fit are considered and evaluated. Based on theresults of the regression analysis the original values of an indexare retained when it is independent from PLAND, otherwise theresidual values are considered; (4) a cluster analysis based on thePearson’s product-moment correlations (Supplementary Table 5) isthen applied to identify further redundant indices so as to guide theselection of a more parsimonious set; (5) the set of non-redundantoriginal and residual indices is summarised by principal componentanalysis (PCA) into few latent components representing fragmen-tation gradients for the focal class; (6) the difference between thegroups of landscapes associated to each component of the PCA isfinally tested by ANOVA.

2.2.2. Morphological spatial pattern analysisMorphological spatial pattern analysis (MSPA), based on math-

ematical morphology analysis (Soille and Vogt, 2009), allows forthe automatic classification of each pixel of the focal class intoseven mutually exclusive spatial pattern classes (feature classes), or“structural classes” (Riitters et al., 2007). These can be grouped intofour main structural classes, i.e. core, islet, boundary and connec-tor. The proportion in boundary and connector structural classesquantifies the proportion of all edge pixels in a landscape, indicat-ing fragmentation, whereas the proportion in the islet structuralclass quantifies the isolated small patches in a landscape, i.e. thosemore prone to disappear. A critical parameter in MSPA is the “sizeparameter” S (Ostapowicz et al., 2008), which defines the thicknessof the edge classes and affects the relative proportions of MSPAstructural classes, hence the assessment of fragmentation. In orderto objectively identify the most appropriate value for this param-eter, the MSPA analysis is repeated using different values over thetransect, ranging from 5 to 100 pixels, corresponding to effectiveedges (Ostapowicz et al., 2008) varying between 7.1 and 141.4 m atthe working grain. This approach enables variation in the definitionof part of the observation scale, while keeping the extent and thegrain constant (Ostapowicz et al., 2008).

2.2.3. Landscape mosaic analysisLandscape mosaic analysis is based on the calculation of a pixel-

level landscape composition indicator known as “landscape mosaicindex” (Riitters et al., 2009). By means of this index each pixel isclassified according to the LC/LU composition in a fixed-area neigh-bourhood surrounding the individual pixel. Pixel state dynamicsdepend on changes in characteristics of the neighbourhood con-taining that pixel which in turn are likely to be affected by changesin the individual pixels they contain. The percent of landscape inthree generalised land mosaic classes, i.e., natural (unpaved landnot used for agriculture), agricultural and urban, are computed bymeans of moving window analysis and the resulting continuoussurfaces are then combined.

2.2.4. Synthetic descriptors and discontinuities detection at thelandscape level

The composition of the landscape mosaic can be used to describea particular landscape to visualise both interfaces and heterogene-ity gradients and, more importantly, to make comparisons acrossspace and/or time, both at the pixel level and landscape level.Riitters et al. (2009) proposed the use of Markov chain models toassess changes at the pixel level. In this paper, we propose the use

cally describe landscape composition and assess its heterogeneitydegree, which can be used for rapid evaluation of changes at thelandscape level. This method is adopted in community ecology and

P. Mairota et al. / Ecological Indicators 33 (2013) 82– 95 87

n for

if

N

wd˛uaiMtP

Ialcptc

tmplaEeCEIm

Fig. 3. Cluster analysis for indices and residual indices. Abbreviatio

s based on Hill’s numbers (Hill, 1973, 1997). This is a parametricamily given by:

(˛) =(∑

p˛i

)1/(1−˛)(1)

here pi is the species relative abundances and the parameter ˛etermines the measure’s sensitivity to species frequency, with

≥ 0. For given values of ˛, this family links the most commonlysed measures of diversity (richness, evenness, Shannon’s entropynd Simpson’s index). For completeness, the Hurlbert’s evennessndex (Hurlbert, 1971) is also computed, according to Beisel and

oreteau (1997), which can provide further detail on the interpre-ation of the structure of the landscape, when compared with theielou’s evenness index (Pielou, 1966).

Another very promising synthetic descriptor is the Similarityndex (SI) proposed by Estreguil and Caudullo (2010). This is defineds the ratio between the share of a given MSPA class under a certainandscape mosaic type and the total amount of MSPA class. The SIombines information on both landscape mosaic composition andixel configuration and thus appears very useful in characterisinghe state and the context of each non-core MSPA main structurallass.

Finally, discontinuities detection, i.e. the identification of eco-ones and boundaries location in a landscape, by means of the

oving split-window (MSW) technique is proposed to identifyotentially critical locations in relation to fragmentation at the

andscape level. This is well known in landscape ecological studiesnd mainly performed using Euclidean metrics, such as the squareuclidean distance, SED (Ludwig and Cornelius, 1987; Johnstont al., 1992; Mairota and Papadimitriou, 1995; Choesin, 2001;

amarero et al., 2006). A precondition for the applicability ofuclidean metrics is that of using an array of uncorrelated variables.n order to define such an array and therefore describe the frag-

entation of the focal class, some of the selected LPIs and non-core

landscape pattern indices can be found in Supplementary Table 2.

MSPA main feature classes are fed in the MSW analysis and used incombination for the detection of landscape spatial variations.

The window used for the MSW technique consists of 4 “plots”,where a plot is represented by one landscape. A zonal statisticalprocedure is performed using a grid on each continuous surfaceobtained from the computation (by means of a moving squarekernel) of the percent of landscape in the selected LPIs and onthe output raster files relative to the non-core MSPA structuraltypes. Thus each plot is divided into 16 cells in order to obtain,for each plot, a mean value for the variable considered. As the win-dow moves along the transect, a series of values that representsuccessive differences between window halves is produced. Dis-continuities, or boundary locations, occur at maximum SED values,indicating a maximum rate of attribute change.

3. Results

3.1. Landscape pattern indices

Computation of the LPIs, at both “class” and “landscape” lev-els, was carried out on the thirty landscapes sample set A bymeans of the FRAGSTATS free software (McGarigal et al., 2002)(Supplementary Tables 1 and 3). All statistical data analysis wasperformed in R-2.14.0 for Windows (R Development Core Team,2009). The selected LPIs were then also computed on the transectsample set B of the study area.

The regression analysis performed on the set of sixteen nonredundant original indices identified by means of correlation anal-ysis (Supplementary Table 4), allows to discriminate betweenindices inherently independent and those inherently dependent

from habitat extent (PLAND). The first group consists of ten indices:PD, LSI, SHAPE AM, FRAC AM, TECI, ECON AM, AI, CONTAG, SHDI,SIDI. For these the original values were retained. The second groupconsists of six indices: LPI, COHESION, AREA AM, CWED, SPLIT and

88 P. Mairota et al. / Ecological Indicators 33 (2013) 82– 95

Table 1Summary of the nine non redundant indices selected for the principal component analysis.

Cluster analysis group Index Reason for selection based on relevance in the study area and/or interpretability

Class level1 COHESION In the present case, patch cohesion index it is redundant with AREA AM (Supplementary Table 5). It provides a

measure of the physical connectedness of the focal class, and seems therefore complimentary to MESH. In addition, inthe present case it is also related to the AI (cluster analysis group two), which in its turn conveys similar informationthan the LSI (also in group two).

1 MESH In the present case, effective mesh size is redundant with LPI (Supplementary Table 5). It informs on cumulative patcharea distribution and can be interpreted as the size of the patches when the landscape is subdivided into S patches,where S is the value of SPLIT. Moreover, for its mathematical properties this index, monotonically decrease withincreasing fragmentation, quantifies an intrinsic landscape feature and is suitable for comparing fragmentation ofregions of different extent and for assessing the influence of a part of a region to the fragmentation of the wholeregion (Jaeger, 2000).

2 PD Patch density is the only index in this group informing on the number of patches on a per unit area basis which isimportant for comparisons among landscapes of the same of varying sizes.

2 FRAC AM With respect to SHAPE AM, which describes the same property, the fractal dimension index has a bounded range ofvariation.

2 CWED Among the contrast indices computed at the class level, the contrast-weighted edge density returns a standardisedmeasure of edge contrast on a per unit area basis.

2 TECI Total edge contrast index in the present case, is redundant with ECON AM (Supplementary Table 5), with respect towhich returns a measure of the amount of edges in each contrast weight, i.e. it is applied to all edges of thecorresponding patch type.

2 SPLIT The splitting index refers to the intermixing of the focal patch type with the other patch types (considered together),therefore provides information on the texture of the landscape with respect to the focal class. In the present case, it isnot redundant with MESH (Supplementary Table 5) to which is mathematically related.

Landscape level3 CONTAG Contagion is the only index referring to the tendency of patch types to be spatially aggregated to be applicable in the

present case. The similar, yet patch based, IJI which indirectly provides information on the texture of the landscape,was found not applicable to the present case due its mathematical formulation (IJI is undefined and reported as “N/A”in the “basename”.land file if the number of patch types is less than 3).

iversiht no

Mpplffii

nsaicss∼efipwa

tns(flifeLtta

3 SHDI With respect to the Simpson’s dcommunity ecology, even thoug

ESH. For these residual values were obtained by fitting appro-riate regression models against PLAND to remove its effect. Inarticular, a linear model was fitted for LPI and COHESION, a log-

inear model for AREA AM and CWED, a generalised linear modelor SPLIT and an additive model for MESH. The cluster analysis per-ormed using the set of these sixteen, independent configurationndices (made up from both original and residual values), resultedn three groups (Fig. 3 and Supplementary Table 5).

These groups aided the further choice of the final set of nineon-redundant configuration indices which was also based on thetudy area and with reference to interpretability (Table 1). Thesere: patch cohesion index, effective mesh size, patch density, fractalndex weighted mean, contrast weighted edge density, total edgeontrast index, splitting index, contagion index, Shannon’s diver-ity index. A principal component analysis (PCA) was applied to thiset. The first three components were considered, which explained71% of the variance and seemed to indicate independent gradi-nts in configuration for the focal class (Table 2). Interestingly, therst and the third components are characterised by indices com-uted at the class level (PD, FRAC AM, CWED, TECI, COHESION),hile the second component is characterised by indices computed

t the landscape level (CONTAG, SHDI).The first component (PC1) represents a gradient in patch con-

rast, shape and patch density. A group of landscapes (landscapes. 5, 8, 14, 16, 17, 18, 27, 29) associated with this componenthowed relatively high values of the contrast weighted edge densityCWED), indicating a relatively high density of edges involving theocal class, within a relatively high contrast condition. In the sameandscapes the focal class shows a relatively high patch densityndex. This seems to indicate a condition of higher fragmentationor the focal class (more edge and patch density) and of higher het-rogeneity for the matrix in which this is embedded, comprised by

C/LU classes more dissimilar than the focal class (higher edge con-rast). Conversely, the other group of landscapes associated withhis component (landscapes n. 1, 2, 4, 13, 15, 22, 26, 30) showed rel-tively lower CWED and PD (Fig. 4a). An ANOVA between the two

ty index, this non-topological measure of landscape diversity borrowed fromt varying within a bounded range, is more sensitive to rare patch types than SIDI.

groups of landscapes associated with this component indicated asignificant difference (p-value < 0.05).

The second component (PC2) represents an independent gra-dient in landscape contagion (CONTAG) and heterogeneity (SHDI).A group of landscapes (landscapes n. 1, 6, 10, 16, 18, 23, 28) asso-ciated with this component showed relatively high values of thecontagion index and relatively low values of the diversity index. Inthese landscapes the matrix is homogenous and, either dominatedby the focal class (landscape n. 23), or by another single LC/LU class(cereal crops) with focal class highly fragmented (landscapes n. 1, 6,10, 16, 18, 28). The other group of landscapes associated with thiscomponent (landscapes n. 3, 5, 7, 8, 9, 12, 21) showed relativelylow values of the contagion index and relatively high values of thediversity index (Fig. 4a). In these landscapes, the focal class is frag-mented in a heterogeneous matrix. An ANOVA between the twogroups of landscapes associated with this component indicated asignificant difference (p-value < 0.05). In the case of PC1 and PC2,the two groups appear as spatially segregated with respect to theboundary of the N2K/NP.

The third component (PC3) represents a gradient in the patchcohesion index (COHESION) and the fractal index weighted mean(FRAC AM). The two groups associated with lower and higher val-ues for these indices were significantly different in terms of theFRAC AM index based on an ANOVA (p-value < 0.05) (Fig. 4b), butin addition all landscapes associated with PC3 (landscapes n. 2, 3,5, 18, 16, 17, 20, 22, 25, 27) seemed to be characterised by a highdegree of fragmentation of the focal class.

3.2. Morphological spatial pattern analysis

The method was applied on the transect sample B bymeans of the freeware “GUIDOS” (Graphical User Inter-

face for the Description of image Objects and their Shapes,http://forest.jrc.ec.europa.eu/download/software/guidos), wherethe “size parameter” S is operated through the “EdgeWidth”parameter.

P. Mairota et al. / Ecological Indicators 33 (2013) 82– 95 89

Table 2Summary of the principal components analysis.

PCA components

1 2 3 4 5 6 7 8 9

Standard deviation 1.797 1.342 1.150 0.980 0.917 0.629 0.498 0.366 0.262Proportion of variance 0.359 0.200 0.147 0.107 0.093 0.044 0.028 0.015 0.008Cumulative proportion 0.359 0.559 0.706 0.813 0.906 0.950 0.978 0.992 1Loadings:

PD 0.418 −0.263 0.286 0.306 −0.243 0.207 −0.119 −0.403 0.550FRAC AM 0.436 −0.456 0.197 0.266 0.636 0.277CWED 0.474 −0.318 −0.121 0.249 −0.197 −0.742TECI 0.413 −0.355 −0.159 0.113 −0.733 −0.244 −0.228 0.103COHESION −0.278 −0.198 −0.581 −0.246 0.316 0.4 −0.114 −0.403 0.226MESH −0.203 −0.389 −0.879 −0.135SPLIT −0.196 −0.266 0.861 0.224 −0.181 0.23 −0.108CONTAG 0.256 0.595 −0.15 0.153 0.625 −0.373

24

S

moe(impnpvao

3

taaetpimscbofrTtaMpl

3l

twhpos

for the same landscape at different time thresholds can indicatechanges and specify their direction. Analogously, they allow for thecomparison of different landscapes within the same region. Thus,

Table 3Diversity profile for the landscape mosaic of the sample set B.

Hill’s numbers Definitions Values

N0 S (class types number) 19N1 eH 7.342N2 1/� 6.281E1 (Pielou’s evenness

index)H/Hmax 0.677

E2 N1/N0 0.386E3 N1-1/N0-1 0.352E4 N2/N1 0.856E5 N2-1/N1-1 0.833Hulberts’ evenness

index(H − Hmin)(Hmax − Hmin) 0.668

� (Simpson’s index) 0.159

SHDI −0.128 −0.643 −0.2

ignificant values for PCA components 1, 2 and 3 are indicated in bold.

As expected (Ostapowicz et al., 2008), the proportion of theain MSPA structural classes (core, islet boundary and connector)

btained for the complete range of “EdgeWidths” (5–100 pixels)ither decreased (core) or increased with increasing “EdgeWidth”Fig. 5a). For the given working scale and proportion of focal classn the landscape (24% approx.), the “EdgeWidth” of 40 appeared

ost informative. At this value, all main structural class types areresent and the increment rate of edge pixels (boundary and con-ector) drops below the average (Fig. 5b). Moreover, among edgeixels, the share of boundary reaches its highest value. At higheralues of “EdgeWidth”, the observed increase in the proportion ofll edge pixels is sustained by the increase in the proportion of justne among the edge classes, the connector structural class.

.3. Landscape mosaic analysis

This method was applied to the sample set B. For each of thehree generalised LC/LU classes (natural, agricultural and urban)

moving window analysis was conducted in FRAGSTATS, with square 50 m × 50 m kernel. Kernel size was selected based onxpert knowledge, supported by the results of the MSPA, withhe aim of reflecting the scale at which edge effects in terms oflant community composition and structure can potentially occur

n the focal LC/LU class. An appropriate tool in QGIS/GRASS environ-ent (Supplementary Table 1) was used to combine the continuous

urfaces according to the three thresholds (10%, 60% and 100%) indi-ated by Riitters et al. (2009) to partition the tripolar space definedy the three generalised LC/LU classes. The landscape mosaic indexbtained ranges from 1 to 19, with each value identifying a dif-erent landscape mosaic type. These represent gradients in theelative dominance of any of the three generalised LC/LU classes.he landscape mosaic map obtained allows the visualisation andhe description of the interfaces connected with the focal class,s well as of heterogeneity gradients across the whole landscape.ore quantitative information can only be obtained with further

rocessing at both the pixel (Riitters et al., 2009) and the landscapeevels (Section 3.4 below).

.4. Synthetic descriptors and discontinuities detection at theandscape level

The synthetic descriptors and the MSW analysis were appliedo the sample set B. Each one of the 4 plots in the MSW analysisas represented by one 1 km × 1 km landscape of the 1 km × 10 km

alf-transect of sample set B. The zonal statistical procedure waserformed using a 250 m × 250 m grid on each continuous surfacebtained from the computation of the percent of landscape in theelected LPIs, by means of a moving square kernel of 50 m.

−0.319 0.628 −0.105

The composition of the landscape mosaic is effectively describedby the importance diversity curve (Fig. 6) and the diversity pro-file built using the values provided by the landscape mosaic mapfor the computation of Hill’s numbers (Hill, 1973, 1997) (Table 3).These show a rather heterogeneous landscape in which macro-heterogeneities are provided by the dominant landscape mosaictypes, where both agricultural and natural generalised LC/LUclasses are represented. Micro-heterogeneities, instead, are locallyprovided by landscape mosaic types in which the urban generalisedclass is represented in different proportions and combined witheither or, or both, agricultural and natural ones.

Using the landscape mosaic map in combination with the MSPAoutput maps relevant to the main structural classes, to each MSPAnon-core main structural classes pixel, a landscape mosaic type isassigned, allowing for the computation of the SI for each MSPA class(Estreguil and Caudullo, 2010). SI quantifies the proportion of edges(boundary and connector) and isolated fragments in the gradientnatural-anthropogenic context. By plotting the SI of each MSPAclass against landscape mosaic type (Fig. 7) a synthetic quantitativedescription of landscape structure is obtained (composition andconfiguration). This provides information on the kind of potentialpressures affecting each structure class associated with fragmenta-tion, which can be used for rapid change detection and monitoringat the landscape level.

Diversity profile and SI represent both quantitative and qual-itative descriptors of landscape structure and their variations

H (Shannon’s index) 1.994Hmax = ln(S) 2.944Hmin according to Beiseland Moreteau (1997)

0.082

90 P. Mairota et al. / Ecological Ind

Fc

tt

atccafittwtP

can also be useful to combine quantitative and qualitative frag-

ig. 4. Biplots for the principal components PC1 and PC2 (a) and for the principalomponents PC1 and PC3 (b).

hey provide a baseline to make comparisons across space and/orime.

Finally, the moving split window analysis was performed using selection of metrics derived from the LPA, based on the results ofhe PCA (CWED and MESH), and the MSPA non-core main structurallasses (islet, boundary connector). This set of metrics, for the spe-ific landscape and for the observation scale of the study, provided

comprehensive summary of the spatially explicit attributes of theocal class associated with its fragmentation. The MSW analysisndicated the occurrence of sharp discontinuities coinciding withhe midpoints of windows 4 and 5, respectively for the two longi-udinal halves of the sample transect (Fig. 8). None of the metrics

hen used separately for the same analysis provides such informa-

ion unequivocally. At these locations the boundary of the Nationalark internal to the N2K site crosses the transect, and a transition

icators 33 (2013) 82– 95

from a grasslands-dominated landscape to a cereal crop-dominatedlandscape can be observed, indicating the potential of this approachto detect differences in management outcomes.

4. Discussion

4.1. Landscape pattern analysis

The application of the described sampling strategy in the “Mur-gia Alta” site yields a non-uniform sample distribution (Fig. 2a).This is due to an inherent and characteristic spatial aggregationof LC/LU classes in the site (e.g., urban land uses, including manyrare classes, are clumped, whereas “natural” land uses are nega-tively segregated from urban and agricultural land uses). Samplingin a wider or different area does not seem to change the spatialaggregation observed.

According to Didham et al. (2012), the LPA based on LPIscarried out here with the declared purpose of controlling forthe variance attributable to habitat quantity, can be labelled as“landscape-biased”, as opposed to “patch-biased”. As such this isnot sufficient to isolate the intercorrelated part of the variancenot directly attributable to habitat quantity or habitat configura-tion (Didham et al., 2012). However, the aim of the present studyis that of providing a mean of quantitative landscape descriptionwith respect to habitat fragmentation and not (at least at thisstage) that of assessing the effect of landscape pattern (i.e. habi-tat amount and configuration) on any community traits. Therefore,the statistical analysis performed, although traditional (e.g., com-pared to the alternative complexity oriented approach proposed byShuangcheng et al., 2009), appears effective in selecting a site-scalespecific and parsimonious set from a wealth of different indiceseach quantifying a certain component of the spatial pattern (Penget al., 2010). By means of such a set, it is possible to categorisethe sample landscapes according to a habitat-landscape configu-ration gradient independently of habitat quantity, which is alsouseful when sampling at the landscape level for further insightsinto patterns and processes along a disturbance gradient.

In particular, indices computed at the class level and asso-ciated with PC1 (mainly CWED and PD) and PC3 (COHESIONand FRAC AM), as well as MESH, associated with PC5, poten-tially allow to describe how fragmentation proceeds through thedistinct phases (as identified by Jaeger, 2000, based on Forman,1995) by characterising the state of the landscape with respect tothe fragmentation degree of the focal class, in a multi-temporalperspective. The indices computed at the landscape level and asso-ciated with PC2 (CONTAG and SHDI) assist in exploring the relationbetween fragmentation and landscape heterogeneity. In the exam-ined area and at the present observation scale, fragmentation isencountered both in homogenous landscapes (mainly dominatedby crop monocultures) and in more heterogeneous landscapes,where mixed natural, agriculture and/or urban LC/LU occur. Thethirty landscapes of sample set A can thus be divided into threegroups: group 1, comprising 4 landscapes, where the focal classis not fragmented (NF); group 2a and 2b, comprising 4 and 7 land-scapes respectively, where the focal class is moderately fragmentedand fragmented in a homogeneous matrix (MF hom, F-hom); group3a and 3b, comprising 5 and 10 landscapes respectively, where thefocal class is fragmented moderately fragmented and fragmentedin a heterogeneous matrix (MF het, F het).

This provides a baseline for change detection as any shift ingroup membership can be detected and evaluated. In addition, it

mentation indicators, in order to consider the social dimensionand perception of landscape fragmentation by humans, as advo-cated by the European Landscape Convention (Llausàs and Nogué,

P. Mairota et al. / Ecological Indicators 33 (2013) 82– 95 91

0

10

20

30

40

50

60

70

80

90

100

5 10 15 20 25 30 40 50 100

Fo

reg

rou

nd

pro

po

rtio

n (

%)

Edg eWidth (pixel s)

unfragmented

boundary

connector

isolated fragments

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

0

10

20

30

40

50

60

70

80

90

100

5 10 15 20 25 30 40 50 100

Ed

ge

incr

emen

t ra

te (d

fore

gro

un

d p

rop

ort

ion

in e

dg

e /d

Ed

geW

idth

)

Fo

reg

rou

nd

pro

po

rtio

n (

%)

unfragmen ted

isolated fra gment s

edg e

edge incr ement rate

average edge incre ment rate

a

b

sses (a

2soEslin

Ft

4.2. Morphological spatial pattern analysis

EdgeWidth (p ixel s)

Fig. 5. Morphological spatial pattern analysis main structural cla

012). Finally, it provides the expert judgement capability neces-ary to refine the sampling at the landscape level, e.g., by meansf a ranked set sampling design (Gilbert, 1987; Mode et al., 1999;PA QA/G-5S, 2002; Wolfe, 2010). It also provides a means to guideub-landscape level sampling, which is being conducted in this

andscape in a future empirical site-scale specific study, aimed atnvestigating the effects of fragmentation on species or commu-ity traits and at quantitatively partitioning the direct versus the

0.1

1

10

100

1000

AA

NN Na A

An N an Au

uan Nu ua

Ua U

Un

UU un

Au

n

Nu

a

Uan

Shar

e of

area

under

lan

dsc

ape

mosa

ic t

ype

(log h

a)

Landscape mosaic types

ig. 6. Importance/diversity curve of the landscape mosaic types for the sampleransect.

) and edge classes increment rate (b) as a function of EdgeWidth.

indirect (i.e. configuration mediated) effects of habitat quantity onsuch traits, as suggested by Didham et al. (2012).

The morphological spatial pattern analysis (MSPA) has provideda pixel based characterisation of the landscape configuration and

Fig. 7. Similarity Index, obtained combining landscape mosaic types and the non-core main structural types derived from morphological spatial pattern analysis.

92 P. Mairota et al. / Ecological Indicators 33 (2013) 82– 95

F longid

iapdtkehwf

4

eM

ig. 8. Graphical representation of the moving split window analysis on the twoiscontinuities. SED = squared Euclidean distance.

ndicated, for the specific landscape and observation scale, the mostppropriate value for the “EdgeWidth”. This is the most criticalarameter which needs to be set when performing an MSPA. Theetermined “EdgeWidth” value of 40 pixels corresponds to an effec-ive edge of 56.6 m (Ostapowicz et al., 2008). This is close to theernel size (50 m) chosen independently (based on expert knowl-dge) for the landscape mosaic analysis. Once this scale parameteras been identified, the technique becomes an operational toolhich can be easily applied by local authorities to assess habitat

ragmentation change at both the pixel and landscape level.

.3. Landscape mosaic analysis

The landscape mosaic index allows for the application of a gradi-nt perspective to categorical data (McGarigal and Cushman, 2005;airota et al., 2011a). As this index is based on the combination

tudinal halves (a and b) of the sample transect. Peaks indicate the occurrence of

of continuous surfaces of three single-classes (natural, agriculturaland urban), density gradient in a specific neighbourhood providesimportant information about how context patterns of each pixelbelonging to a discrete patch (of the categorical map) vary acrossa multi-class landscape mosaic. Therefore the use of this indexovercomes one of the limitations of the patch mosaic model of land-scapes, which considers all habitats within a patch to be identical,and different patches of the same habitat located in different partsof the same landscape to also be identical, without internal het-erogeneity (Hoechstetter et al., 2008). In particular, this approachenables assessment of the component of spatial variation whichis exogenous to the patch, whereas the assessment of the endoge-

nous component of spatial variation requires a different conceptualapproach and the specific support of EO techniques (Hoechstetteret al., 2008; Cushman et al., 2010; Mondal, 2011; Frazier and Wang,2011).

cal Ind

4l

ordhw

datit1

ddbmotoi(n

lwfdto

5

aslTdaisisp

tfiasasapu

tofb

P. Mairota et al. / Ecologi

.4. Synthetic descriptors and discontinuities detection at theandscape level

Diversity and evenness profiling are gaining momentum in ecol-gy as they provide deeper insights into community structure, thusevealing the contributions of both rare and common species to bio-iversity trends (Studeny et al., 2011). These have been adaptedere for the description of landscape mosaic class composition,ithin the context of change detection and characterisation.

From a theoretical point of view, the assessment of landscapeiversity, i.e. “ecodiversity” is important to understand biodiversitynd its inherent relations to both ecological heterogeneity and cul-ural diversity (Naveh, 1998; Jedicke, 2001). The former reflects thenteractions between abiotic and biotic systems (Nagel, 1998) andhe latter reflects the interactions between man and nature (Naveh,998).

From an operational point of view, the assessment of landscapeiversity provides a quick landscape level evaluation, useful toetect the occurrence of important shifts, which can be then refinedy means of evaluation at the pixel level, based on Markov chainodelling (Riitters et al., 2009). When MSPA classes are overlaid

nto the landscape mosaic map, a quantitative characterisation ofhe local structural pattern gradient across the landscape space isbtained. Such information can be used as proxies for habitat qual-ty in studies aimed at modelling organism and process responsesincluding connectivity) to landscape structure across heteroge-eous landscapes.

The combined use of LPA, namely the quantification of theocal pattern gradient across the landscape space and MSPA

ithin moving split window analysis has also proven use-ul to define a set of indices capable of identifying landscapeiscontinuities, and to indicate “critical points” where threatso biodiversity and ecosystem integrity are more likely toccur.

. Conclusions

The proposed framework conducted for quantitative LPA wasble to capture meaningful complementary aspects of the land-cape structure (composition and configuration) at differentevels (pixel, class, landscape) related to habitat fragmentation.his confirms that landscape structure cannot be exhaustivelyescribed by a small set of indices derived from a singlepproach. Different indices, derived from different approaches,ndicate different properties of spatial organisation of the land-cape, and information from multiple indices requires appropriatentegration to get a more complete understanding of land-cape spatial pattern, especially as related to disturbance andressure.

A landscape- and observation-scale specific set of LPIs was iden-ified for the illustrative case study, while considering a particularocal class. Given the unavoidable site- and scale-specificity of suchndices, the statistical procedure performed is necessary to select

parsimonious set of LPIs capable of describing fragmentation pere. The results show that this procedure (adapted from McGarigalnd McComb, 1995) can be generalised and applied to any site andcale. Further, once the relevant set of LPIs and scale is identified for

given landscape and scale, the set of LPIs can routinely be com-uted at the same scale over time for a quick assessment of changes,sing EO derived LC/LU/habitat maps.

In addition, the selected LPIs provide insights on the rela-

ions between landscape heterogeneity and fragmentation. Thisffers a means for ranking sample landscapes according to aragmentation gradient in relation to matrix quality, that cane extremely useful for monitoring. It also provides objective

icators 33 (2013) 82– 95 93

criteria that can help to optimise the sampling efforts at the land-scape and sub-landscape level, especially when related to empiricalstudies on the causal relationships between fragmentation andspecies/community traits.

The MSPA provides pixel based structural characterisation ofthe landscape and its iteration across a range of “EdgeWidths”(the most critical parameter in this analysis) enabled identifi-cation of the most appropriate scale for a specific site. This isthe precondition for a customary implementation towards quickchange assessment, in combination with other methods. One ofthese is the landscape mosaic analysis, as shown in this study forthe purpose of fragmentation evaluation across space and time.Other studies (Saura et al., 2011) have shown the potential forcombining MSPA with approaches aiming at assessing landscapenetwork connectivity in order to obtain quantitative informa-tion on the relative importance of the individual pixel for thewhole landscape structural connectivity. Both these approachescan potentially be used in a range of monitoring and managementcontexts.

Landscape mosaic analysis has also been proven useful to thisend, by providing an effective characterisation of the compositionof the landscape through landscape diversity profiling. When usedin combination with MSPA, landscape mosaic analysis characterisesthe context of each non-core MSPA main structural class, which isvery useful to link pixel level changes to landscape level changes,and thus critical for change monitoring at multiple scales. In addi-tion, it could be used to get the kind of information needed at thepixel level to solve ambiguities inherent in the automatic transla-tion from LC/LU maps (e.g. LCCS) to GHCs maps (Tomaselli et al.,2011).

The combined use of LPIs derived from the “global approach” toLPA, of a local pattern analysis (i.e., pixel based), and of a tech-nique for the quantification of the local pattern gradient acrossthe landscape space, has also proved appropriate when defininga set of indices capable of identifying landscape discontinuitiesand indicating “critical points”, where threats to biodiversity andecosystems integrity are more likely to occur.

In summary, based on the results from the illustrative exam-ple, the proposed framework has been successful in identifying acomposite set of indices derived different approaches for landscapestructure analysis. This framework appears capable of capturingmain landscape structural features in a given site, so as to return abaseline reference that also identifies problematic areas in termsof increased fragmentation. Such areas might be more criticalin the future (i.e., more prone to human induced changes andtheir negative impacts) and deserve special and more detailedmonitoring effort. Thus the approach proposed may assist in pro-viding early warning signals for immediate response to pressuresincreasing habitat fragmentation. The described framework can beapplied to any site and any observation scale, and provide a robustapproach that can be used for the analysis of change in biodiver-sity indicator with reference to the following CBD-SEBI focal areas(Strand et al., 2007): status and trends of the components of bio-logical diversity, ecosystem integrity, and ecosystem goods andservices.

Acknowledgements

The authors wish to thank the two anonymous reviewers fortheir valuable comments and suggestions. This work was supportedby the European Community’s Seventh Framework Programme,

within the FP7/SPA.2010.1.1-04: “Stimulating the development ofGMES services in specific area”, under grant agreement 263435 forthe project Biodiversity Multi-SOurce Monitoring System: FromSpace To Species (BIO SOS).

9 al Ind

A

fe

R

B

B

B

B

B

C

C

C

C

D

D

D

D

D

E

E

E

E

E

F

F

F

G

H

H

H

H

H

4 P. Mairota et al. / Ecologic

ppendix A. Supplementary data

Supplementary data associated with this article can beound, in the online version, at http://dx.doi.org/10.1016/j.colind.2012.08.017.

eferences

aker, W.L., Cai, Y., 1992. The r.le programs for multiscale analysis of landscapestructure using the GRASS geographical information system. Landscape Ecol. 7,291–302.

askent, E.Z., Jordan, G.A., 1995. Characterising spatial structure of forest landscapes.Can. J. Forest Res. 25, 1830–1849.

eisel, J.-N., Moreteau, J.-C., 1997. A simple formula for calculating the lower limitof Shannon’s diversity index. Ecol. Model. 99, 289–292.

uchanan, G.M., Nelson, A., Mayaux, P., Hartley, A., Donald, P.F., 2008. Delivering aglobal terrestrial observation system through remote sensing. Conserv. Biol. 23,499–502.

unce, R.G.H., Bogers, M.M.B., Roche, P., Walczak, M., Geijzendorffer, I.R., Jongman,R.H.G., 2011. Manual for Habitat and Vegetation Surveillance and Monitoring:Temperate, Mediterranean and Desert Biomes.

amarero, J.J., Gutiérrez, E., Fortin, M.J., 2006. Spatial patterns of plant richness acrosstreeline ecotones in the Pyrenees reveal different locations for richness and treecover boundaries. Global Ecol. Biogeogr. 15, 182–191.

hoesin, D.N., 2001. Use of landscape ecological analysis to determine ecotone loca-tions: an evaluation based on field data. Jurnal Matematika dan Sains 6, 91–99.

ouncil of Europe, 2004. Kyiv Action Plan on Biodiversity Monitoring andIndicators Endorsed by the Task Force and Bureau of the Pan-EuropeanBiological and Landscape Diversity Strategy (PEBLDS), Strasbourg, France,14 May 2004, http://www.strategyguide.org/docs/bureau/may2004/STRA-CO(2004)3f Rev E.doc.

ushman, S.A., Gutzweiler, K., Evans, J.S., McGarigal, K., 2010. The gradient paradigm:a conceptual and analytical framework for landscape ecology. In: Cushman, S.A.,Huettmann, F. (Eds.), Spatial Complexity, Informatics, and Wildlife Conservation.Springer, Tokyo, Japan, pp. 83–108.

avies, C.E., Moss, D., 2002. EUNIS Habitat Classification. Final Report to the Euro-pean Topic Centre of Nature Protection and Biodiversity, European EnvironmentAgency.

eFries, R., Hansen, A., Newton, A.C., Hansen, M.C., 2005. Increasing isolation ofprotected areas in tropical forests over the past twenty years. Ecol. Appl. 15,19–26.

i Gregorio, A., Jansen, L.J.M., 2005. Land Cover Classification System (LCCS): Clas-sification Concepts and User Manual for Software Version 2. FAO Environmentand Natural Resources Service Series, No. 8, Rome.

idham, R.K., Kapos, V., Ewers, R.M., 2012. Rethinking the conceptual foundationsof habitat fragmentation research. Oikos 121, 161–170.

uro, D.C., Coops, N.C., Wulder, M.A., Han, T., 2007. Development of a large areabiodiversity monitoring system driven by remote sensing. Prog. Phys. Geogr.31, 235–260.

EA, 2009. Halting the Loss of Biodiversity in Europe – The 2009 Assessment Basedon the First Set of Biodiversity Indicators Version. 19 December 2008.

EA, 2010. EU 2010 Biodiversity Baseline. Luxembourg. ISBN 978-92-9213-164-7.ISSN 1725-2237. http://dx.doi.org/10.2800/6160.

PA QA/G-5S, 2002. Guidance on Choosing a Sampling Design for EnvironmentalData Collection for Use in Developing a Quality Assurance Project Plan. UnitedStates Environmental Protection Agency. Office of Environmental Information,Washington, DC.

streguil, C., Caudullo, G., 2010. ID 5.3 (Intermediate Deliverable, July2010): Pattern Related Measures and Indicators for Selected Sitesat Varying Spatial Scales in Selected Environmental Strata EBONE.http://www.ebone.wur.nl/NR/rdonlyres/298E8A25-FEB4-4E0C-B7E3-1C725D3F7006/143155/EBONED53 EstreguilandCaudullo2010 final.pdf.

veritt, B.S., Landau, S., Leese, M.D., Stahl, M., 2011. Cluster Analysis, 5th edition.Wiley, Chichester, UK, ISBN 978-0-470-74991-3.

ahrig, L., 2003. Effects of habitat fragmentation on biodiversity. Annu. Rev. Ecol.Evol. Syst. 34, 487–515.

orman, R.T.T., 1995. Land Mosaics: The Ecology of Landscape and Regions. Cam-bridge University Press, Cambridge.

razier, A.E., Wang, L., 2011. Characterizing spatial patterns of invasive species usingsub-pixel classifications. Rem. Sens. Environ. 115, 1997–2007.

ilbert, R.O., 1987. Statistical Methods for Environmental Pollution Monitoring. VanNostrand Reinhold, New York.

aines-Young, R., Chopping, M., 1996. Quantifying landscape structure: a review oflandscape indices and their application to forested landscapes. Phys. Geogr. 4,418–445.

ill, M.O., 1973. Diversity and evenness: a unifying notation and its consequences.Ecology 54, 427–432.

ill, M.O., 1997. An evenness statistic based on the abundance-weighted varianceof species proportions. Oikos 79, 413–416.

oechstetter, S., Walz, U., Dang, L.H., Thinh, N.X., 2008. Effects of topography andsurface roughness in analyses of landscape structure – a proposal to modify theexisting set of landscape metrics. Landscape Online 3, 1–14.

urlbert, S.H., 1971. The nonconcept of species diversity: a critique and alternativeparameters. Ecology 52, 577–586.

icators 33 (2013) 82– 95

INSPIRE Thematic Working Group Coordinate Reference Systems andGeographical Rid Systems, 2010. D2.8.I.1 INSPIRE Specification onCoordinate Reference Systems – Guidelines., http://inspire.jrc.ec.europa.eu/documents/Data Specifications/INSPIRE Specification CRS v3.1.pdf(accessed on March 21, 2012).

Jedicke, E., 2001. Biodiversität, Geodiversität, Ökodiversität. Kriterien zur Analyseder Landschaftsstruktur – ein konzeptioneller Diskussionsbeitrag. Naturschutzund Landschaftsplanung 33, 59–68.

Jaeger, J.A.G., 2000. Landscape division, splitting index, and effective mesh size: newmeasures of landscape fragmentation. Landscape Ecol. 15, 115–130.

Johnston, C.A., Pastor, J., Pinay, G., 1992. Quantitative methods for studying land-scape boundaries. In: Hansen, A.J., di Castri, F. (Eds.), Landscape Boundaries:Consequences for Biotic Diversity and Ecological Flows. Springer-Verlag Inc.,New York, pp. 107–125.

Jongman, R.H.G., Bogers, M.M.B., Lucas, R.M., Mairota, P., Tomaselli, V., Torri, D., 2011.List of Indicators. BIO SOS Biodiversity Multisource Monitoring System: FromSpace TO Species (BIO SOS) Deliverable D2.1.

Kosmidou, V., Petrou, Z., Lucas, R.M., Tomaselli, V., Petrou, M., Bunce, R.G.H., Bogers,M.M.B., Mucher, S., Tarantino, C., Blonda, P., Baraldi, A., 2011. Software for HabitatMaps Production from LC. BIO SOS Biodiversity Multisource Monitoring System:From Space TO Species (BIO SOS) Deliverable D6.10.

Kotliar, N.B., Wiens, J.D., 1990. Multiple scales of patchiness and patch structure: ahierarchical framework for the study of heterogeneity. Oikos 59, 253–260.

Li, H., Franklin, J.F., Swanson, F.J., Spies, T.A., 1993. Developing alternative forestcutting patterns: a simulation approach. Landscape Ecol. 8, 63–75.

Llausàs, A., Nogué, J., 2012. Indicators of landscape fragmentation: the case for com-bining ecological indices and the perceptive approach. Ecol. Indic. 15, 85–91.

Ludwig, J.A., Cornelius, J.M., 1987. Locating discontinuities along ecological gradi-ents. Ecology 68, 448–450.

Mairota, P., Boccaccio, L., Labadessa, R., Leronni, V., Tomaselli, V., Tarantino, C.,Cafarelli, B., 2011b. Pre-evaluation and Rank Sampling. BIO SOS BiodiversityMultisource Monitoring System: From Space TO Species (BIO SOS) DeliverableD6.3.

Mairota, P., Kosmidou, V., Nagendra, H., 2011a. Preliminary Report on LandscapePattern Analysis – State of the Art. BIO SOS Biodiversity Multisource MonitoringSystem: From Space TO Species (BIO SOS) Deliverable D6.2.

Mairota, P., Papadimitriou, F., 1995. Spatial discontinuities in a southern Italiancultural landscape: testing a methodology. In: Griffiths, G.H. (Ed.), LandscapeEcology: Theory and Application. Proceedings of the IALE(UK) Fourth AnnualConference. 19–20 September, 1995, Reading, UK. Colin Gross Printers Ltd.,Garstang, UK, ISBN 0-9524263-1-5, pp. 102–109.

McGarigal, K., Cushman, S.A., 2005. The gradient concept of landscape structure.In: Wiens, J., Moss, M. (Eds.), Issues and Perspectives in Landscape Ecology.Cambridge University Press, Cambridge, UK.

McGarigal, K., Cushman, S.A., Neel, M.C., Ene, E., 2002. FRAGSTATS:Spatial Pattern Analysis Program for Categorical Maps. ComputerSoftware Program Produced by the Authors at the Universityof Massachusetts, Amherst, Available at the following web site:http://www.umass.edu/landeco/research/fragstats/fragstats.html (accessed onMarch 21, 2012).

McGarigal, K., McComb, W.C., 1995. Relationships between landscape structure andbreeding birds in the Oregon Coast Range. Ecol. Monogr. 65, 235–260.

Mode, N.A., Conquest, L.C., Marker, D.A., 1999. Ranked set sampling for ecologi-cal research: accounting for the total costs of sampling. Environmetrics 10,179–194.

Mondal, P., 2011. Quantifying surface gradients with a 2-band Enhanced VegetationIndex (EVI2). Ecol. Indic. 11, 918–924.

Nagel, P., 1998. Landscape diversity – a holistic approach. In: Barthlott, W., Winiger,M. (Eds.), Biodiversity a Challenge for Development Research and Policy.Springer-Verlag, Berlin, Germany, pp. 129–143.

Nagendra, H., Lucas, R.M., Honrado, J.P., Jongman, R.H.G., Mairota, P., Tarantino,C., Adamo, M., 2012. Remote sensing for protected area assessment: monitor-ing habitat area, condition, biodiversity and threats. Ecol. Indic., submitted forpublication.

Naveh, Z., 1998. From biodiversity to ecodiversity – holistic conservation of thebiological and cultural diversity of Mediterranean landscapes. In: Montene-gro, G., Jaksic, F., Rundel, P.W. (Eds.), Landscape Disturbance and Biodiversityin Mediterranean-Type Ecosystems, 136. Ecological Studies Springer, Berlin, pp.23–54.

O’Neill, R.V., Krummel, J.R., Gardner, R.H., Sugihara, G., Jackson, B., De Ange-lis, D.L., Milne, B.T., Turner, M.G., Zygmunt, B., Christensen, S.W., Dale,V.H., Graham, R.L., 1988. Indices of landscape pattern. Landscape Ecol. 1,153–162.

Ostapowicz, K., Vogt, P., Riitters, K.H., Kozak, J., Estreguil, C., 2008. Impact of scaleon morphological spatial pattern of forest. Landscape Ecol. 23, 1107–1117.

Peng, J., Yanglin, W., Yuan, Z., Jiansheng, W., Weifeng, L., You, L., 2010. Evaluating theeffectiveness of landscape metrics in quantifying spatial patterns. Ecol. Indic. 10,217–223.

Pielou, E.C., 1966. Species-diversity and pattern-diversity in the study of ecologicalsuccession. J. Theor. Biol. 10, 370–383.

R Development Core Team, 2009. R: A Language and Environment for Statisti-

cal Computing. R Foundation for Statistical Computing, Wien, Austria, ISBN3-900051-07-0, URL http://www.R-project.org.

Riitters, K.H., Vogt, P., Soille, P., Kozak, J., Estreguil, C., 2007. Neutral model analy-sis of landscape patterns from mathematical morphology. Landscape Ecol. 22,1033–1044.

cal Ind

R

R

R

R

S

S

S

S

S

S

S

P. Mairota et al. / Ecologi

iitters, K.H., Wickham, J.D., O’Neill, R.V., Jones, K.B., Smith, E.R., 2000. Global-scalepatterns of forest fragmentation. Ecol. Soc. (formerly Conservation Ecology) 4,3.

iitters, K.H., Wickham, J.D., Wade, T.G., 2009. An indicator of forest dynamics usinga shifting landscape mosaic. Ecol. Indic. 9, 107–117.

iitters, K.H., O’Neill, R.V., Hunsaker, C.T., Wickham, J.D., Yankee, D.H., Timmins,S.P., Jones, K.B., Jackson, B.L., 1995. A factor analysis of landscape pattern andstructure measures. Landscape Ecol. 10, 23–39.

odwell, J.S., Schaminée, J.H.J., Mucina, L., Pignatti, S., Dring, J., Moss, D., 2002. TheDiversity of European Vegetation: An Overview of Phytosociological Alliancesand their Relationships to EUNIS Habitats. National Centre for Agriculture,Nature Management and Fisheries, Wageningen, 168 pp.

aura, S., Vogt, P., Velasquez, J., Hernando, A., Tejera, R., 2011. Key structural con-nectors can be identified by combining landscape spatial pattern and networkanalysis. Forest Ecol. Manage. 262, 150–160.

chindler, S., Poirazidis, K., Wrbka, T., 2008. Towards a core set of landscape metricsfor biodiversity assessments: a case study from Dadia National Park, Greece.Ecol. Indic. 8, 502–514.

huangcheng, L., Qing, C., Peng, J., Yanglin, W., 2009. Indicating landscape fragmen-tation using L–Z complexity. Ecol. Indic. 9, 780–790.

mith, A.C., Koper, N., Francis, C.M., Fahrig, L., 2009. Confronting collinearity: com-paring methods for disentangling the effects of habitat loss and fragmentation.Landscape Ecol. 24, 1271–1285.

oille, P., Vogt, P., 2009. Morphological segmentation of binary patterns. PatternRecogn. Lett. 23, 687–702.

trand, H., Höft, R., Strittholt, J., Miles, L., Horning, N., Fosnight, E., Turner, W. (Eds.),2007. Sourcebook on Remote Sensing and Biodiversity Indicators. Secretariat of

the Convention on Biological Diversity, Montreal, Technical Series No. 32, 203pp.

tudeny, A.C., Buckland, S.T., Illian, J.B., Johnston Magurran, A.E., 2011. Goodness-of-fit measures of evenness: a new tool for exploring changes in communitystructure. Ecosphere 2, 1–19.

icators 33 (2013) 82– 95 95

Taylor, P., Fahrig, L., With, K.A., 2006. Landscape connectivity: back to the basics. In:Crooks, K., Sanjayan, M.A. (Eds.), Connectivity Conservation. Cambridge Univer-sity Press, Cambridge, UK, pp. 29–43.

Tinker, D.B., Resor, C.A.C., Beauvais, G.P., Kimpfmueller, K., Fernandes, C.I., Baker,W.L., 1998. Watershed analysis of forest fragmentation by clearcuts and roadsin a Wyoming forest. Landscape Ecol. 13, 149–165.

Tischendorf, L., Fahrig, L., 2000. On the usage and measurement of landscape con-nectivity. Oikos 90, 7–19.

Tomaselli, V., Blonda, P., Marangi, C., Lovergine, F., Baraldi, A., Mairota, P., Terzi, M.,Muncher, S., 2011. Report on Relations between Vegetation Types Derived fromLand Cover Maps and Habitats. BIO SOS Biodiversity Multisource MonitoringSystem: From Space TO Species (BIO SOS) Deliverable D6.1.

Turbé, A., De Toni, A., Benito, P., Lavelle, P., Ruiz, N., Van der Putten,W.H., Labouze, E., Mudgal, S., 2010. Soil Biodiversity: Functions, Threatsand Tools for Policymakers. Bio-Intelligence Service, IRD, and NIOO.Report for European Commission (Directorate-General for the Environment)http://ec.europa.eu/environment/soil/pdf/biodiversity report.pdf.

Wiens, J.A., Stenseth, N.C., Van Horne, B., Ims, R.A., 1993. Ecological Mechanisms andLandscape Ecology. Oikos 66, 369–380.

Wilson, J.B., Peet, R.K., Dengler, J., Pärtel, M., 2012. Plant species richness: the worldrecords. J. Veg. Sci. 23, 796–802.

Wolfe, D.A., 2010. Ranked set sampling. Wiley Interdisc. Rev.: Comput. Stat. 2,460–466.

Wu, J., 2004. Effects of changing scale on landscape pattern analysis: Scaling rela-tions. Landsc. Ecol. 19, 125–138.

Further reading

Directive 2007/2/EC, of the European Parliament and of the Council of 14 March2007 establishing an Infrastructure for Spatial Information in the EuropeanCommunity INSPIRE.