e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

available at www.sciencedirect.com

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

Cost distance dened by a topological function of landscape

J.C. Folt ete a, , K. Berthier b , J.F. Cosson b
a b

Th eMA-UMR 6049 CNRS, University of Franche-Comt e, 32 rue M egevand, F-25030 Besanc on, France Centre de Biologie et de Gestion des Populations, INRA, Campus International de Baillarguet, F-34980 Montferrier-sur-Lez, France

a r t i c l e
Article history:

i n f o

a b s t r a c t
Distance is a basic concept in the domain of animal species motion. Cost distances, rather than Euclidian distances, are more and more used in order to have a more realistic measure, on the basis of resistance values assigned to each landscape class. We propose here a method to compute resistance values by using topological functions of landscape, i.e. by taking account of the proximity of habitat/non-habitat edges, with continuous functions. An example is given when comparing cost distances and the propagation of water vole in the massif of Jura (France). The comparison with usual cost distances gives information about the ecological assumptions. The results show also that the statistical behaviour of the distances depending of the parameters of the functions allows to precise the inuence of edges in terms of spatial range. 2007 Elsevier B.V. All rights reserved.

Received 22 May 2006 Received in revised form 29 June 2007 Accepted 9 July 2007 Published on line 20 August 2007 Keywords: Cost distance Resistance Edge Habitat Spread Water vole

1.

Introduction

Distance is a basic concept inherent to any geographical space. This notion is a key factor in population ecology and especially in animal movement analysis. In the well-known theory of isolation by distance (Wright, 1943), it is considered a central factor playing of role on species invasions or species extinctions. According to this theory, the genetic difference between populations increases with geographical distance when considering a spatial scale higher than the average dispersal distance of the species. The relationship between genetic distance and geographical distance (Rousset, 1997, 2000) is then a proxy used to infer the individual movement ability of a given species (Arter, 1990; Michels et al., 2001; Arnaud, 2003; Coulon et al., 2004; Berthier et al., 2005; Broquet et al., 2006). In landscape ecology, spatial distance is of a primary importance in the concepts of connectivity and fragmentation (Forman

and Godron, 1986; Forman, 1995). Distance is also explicitly included in the formulation of some landscape metrics such as the proximity index proposed by Gustafson and Parker (1994) and in the use of graph theory applied to the spatial relations between landscape patches (Urban and Keitt, 2000; Bunn et al., 2000). The application of diffusion-reaction models in the context of population movements (Turchin, 1998; Okubo and Levin, 2001) requires the use of the distance between places and a central point (as a release point). All these examples of analysis based on the notion of spatial distance show the great importance of its measurement. Euclidian distance, as the crow ies when applied to the at space of maps, is the simplest measure of a distance. However its use assumes a neutral spatial framework, which is not very realistic when representing effective spatial accessibility. In the context of animal population movements, the assumption of a homogeneous space where all places are

Corresponding author. E-mail addresses: jean-christophe.foltete@univ-fcomte.fr (J.C. Foltete), berthier@ensam.inra.fr (K. Berthier), cosson@ensam.inra.fr (J.F. Cosson). 0304-3800/$ see front matter 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolmodel.2007.07.014

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

105

equally accessible is rarely justied (Matthiopoulos, 2003): the habitat of the studied species, the conguration of habitat patches, resources, corridors or displacement constraints lead to a very heterogeneous space and nally challenges the relevance of Euclidian distance. Following these observations, several authors have shown the interest of least-cost distances (or cost distances) that take into account the spatial heterogeneity. These distances are computed from raster data only (for example from a given cell to a target cell) and allow to allot a movement resistance value to each landscape class called a friction. Cost distances are sometimes called effective distances (Ferreras, 2001), in contrast to the Euclidian distance. The use of cost distance computations is more and more popular in ecology (Knaapen et al., 1992; Yu, 1996; Pain et al., 2000; Bunn et al., 2000; Halpin and Bunn, 2000; Ray et al., 2002; Chardon et al., 2003; Adriaensen et al., 2003; Ray and Burgman, 2006), since they are implemented in several GIS softwares (ESRI, 1996, 2001; Eastman, 1998; Ray, 2005). The choice of the resistance values assigned to each landscape class is usually based on a specic knowledge of the mobility behaviour of the studied population. A resistance of 1 is quite often assigned to the habitat class and to the displacement corridors; a higher level is given to the other classes in relation to their degree of hostility (presence of potential predators) or their ability to limit the movements (physical barriers, lack of resources). Verbeylen et al. (2003) have shown a simple method to determine an optimal combination of resistance values when using the cost distance in prediction model of species abundance. However, the assignment of uniform resistance values to all landscape classes implicitly means that a given landcover class involves a uniform displacement behaviour. This is in contradiction with the well-known edge effect in ecology (see for example Paton, 1994; Reese and Ratti, 1988; Fagan, 1999). Following this effect, the proximity to different landcover classes can lead to a modication of this behaviour for several reasons: change of accessibility to resources, increase of predator risk, and so on. For example, for a species subject to predators living outside its habitat, the land-cover class dening this habitat does not necessarily correspond to a uniform context of mobility, because of the varying ability of predators to reach these areas. In an ecological context where the main moving factor of a given species is the accessibility to resources, the inhospitable land cover classes may be partially crossed by the individuals, provided they remain near the boundary, because in this case their resources remain accessible. In consequence, we argue that proximity to the limit between hospitable and inhospitable areas can play a great role in the permeability of landscape to animal movements. However, the classical approach of cost distances does not take this phenomenon of progressive permeability into account: it simply replicates the discrete structure of land cover classes in the spatial distribution of resistance values. The aim of the present paper is to show that a specic method of allocation of resistance can be implemented in order to improve the relevance of cost distance when the limits between certain land cover classes play an effective ecological role. Starting from a simplied landscape divided into two classes habitat (H) non-habitat or inhospitable (NH) a method is proposed to assign variable resistance values to

each class. Such values are dened by a function of the position in relation to the boundary between H and NH and by applying what we call a topological function of landscape. An additional class designated neutral class or N is added to include areas which are not under the inuence of the previous boundary. After the presentation of the data set and the used method (2) we will show the results obtained in the context of the spread of a rodent population in a part of the Jura massif (France) (3).

2.
2.1.

Materials and methods

Research context and data set

In the Jura massif, the grasslands are regularly swarming with a rodent species, the common vole (Arvicola terrestris). The spread of the vole populations makes a travelling wave with a period of about 5 years (Giraudoux et al., 1997). This e et al., 1999) and travelling wave ruins the grasslands (Quer promotes the transmission of diseases (Delattre et al., 1998). Furthermore, the use of pesticide leads to many environmental consequences especially by hitting vole predators (Delattre et al., 2000). In this context, biologists and geographers seek to understand the role of landscape structures on the vole spread. As several epicentre zones have been identied (Duhamel et al., 2000), we focus here on the plateau of Nozeroy, a zone inuenced by a single epicentre in order to analyze the spread phenomenon without interferences. This zone is a plateau of approximately 200 km2 composed of a matrix of grassland surrounded by large forests of conifers, which may be considered as almost impassable barriers for vole populations (Fig. 1). Different spatial congurations of grasslands are present in the plateau: certain parts contain very large and continuous grassland patches (openeld) whereas other parts can be considered as bocage structures, i.e. fragmented by hedges and some linear forest elements. The relief is rather homogeneous, at an altitude ranging between 700 and 900 m and without signicant topographic accidents. Starting on April 2002, estimations of A. terrestris density were done every 6 months in a set of cells of observation in a regular grid of 1 km2 . One cell out of two was analyzed, giving a total of 92 sites of observation. For each site, the density was estimated on the basis of vole tumuli count along two 250 m diagonal segments (Fig. 2a). Using this count, a surface index, expressed as a percentage, was calculated as described in Giraudoux et al. (1995); a resulting value of 0 means the absence of vole and a value of 100 means a maximal vole density. This paper analyzes the initial phase of the invasion of the plateau by the A. terrestris. At the start of the observation (April 2002), the high densities are located in a very small area in the north of the zone. The unique site of observation with the maximal value of 100 is considered here as the local epicentre of the diffusion (Fig. 2b). The challenge is to explain the spatial distribution of densities at the next date of observation (September 2002) through a function of spatial distance from the epicentre. As different types of spatial distance may be used (Euclidian distance, cost-distances), the resulting spa-

106

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

Fig. 1 Experimental zone: the plateau of Nozeroy (Jura, France).

tial spread represents an interesting test of these types of cost distances. Because of the presence of vole predators nesting in hedges and linear forest elements, we assume that the permeability of grassland areas (the habitat of A. terrestris) varies depending on the distance to these landscape elements. Landscape classes were dened from remotely sensed data. An Indian Remote Sensing LISS-III image acquired in September 1997 was merged with the panchromatic band (IRS-1c PAN) to have a multispectral image of 7 m-spatial resolution. Then a maximum likelihood classier was applied to obtain a categorical image with four land cover classes: grasslands, wooded areas, bare grounds and water. Because of the relative stability of land cover structures in this zone, the temporal delay between demographic data and landscape data is considered acceptable for cross-analysis.

2.2.

General framework

The assumptions presented above lead to implement a continuous function of resistance instead of the dichotomy between the uniform absence of constraint of H and the uniform constraint of NH. This function depends on the boundary between H and NH; it is applied to each cell of a grid to dene resistance surfaces that are used to compute the cost distances. The global principle of the method is thus based on a particular denition of the resistance values. The third spatial class N is added to account for areas which are not concerned by the opposition between H and NH. It is

called neutral class because it is not linked to the ecological assumptions about the progressive permeability of certain landscape classes, but it is not necessarily neutral in terms of resistance. For example, class N may be represented by a land cover type which is impassable for the species in question and which is not in keeping with the notions of resources or vulnerability to predators. It may also be a class of potential movement (resistance value of 1) without any relationship to an exogenous inuence. The resistance value of the class N depends on each ecological context and its specicity lies in its independence from the spatial inuence of the other classes H and NH. The method presented here is therefore specic only with relation to the allocation of resistance of the other landscape classes, and that is why the presentation of the computations will not mention class N. The computation of cost distances is accomplished through four successive steps. In the rst step, the minimal distance between each cell and the HNH boundary is computed by gradually increasing a buffer zone from the cells located at this boundary. An image of boundary distances is then generated (Fig. 3b). The transformation of these distance values by a topological function (that will be detailed later) allows us to obtain an image of resistance values (Fig. 3c). For a given source zone which can be a single cell or a set of cells, this image is then used for computing the minimal cost by the use of the classical growth algorithm implemented for example in Arcview (ESRI, 1996) and Idrisi (Eastman, 1989) (Fig. 3d).

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

107

Fig. 2 Demographic data for analysing the spread of vole swarming.

108

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

Fig. 3 Successive steps of computation of cost distances.

At the end of this process, the minimal cost can be computed for any cell of the image. The creation of the image of boundary distances and the cost distances computation are invariant steps but the denition of the topological function require several parameters. The choice of these parameters is explained below.

simple denition of f0 as follows: f0 (p = H) = 1

f0 (p = NH) = rm

(2)

2.3.

Different kind of topological distance

In order to yield relative resistance values in relation to the NNH boundary, several continuous functions can be considered according to biological assumptions. Here these functions are based on two parameters when allocating a specic status (for example impassable) to the neutral class N: rm : the maximal resistance value that corresponds to the greatest constraint level of the inhospitable class; dm : the maximal distance of inuence of the boundary which corresponds to a given spatial range. We use a sigmoid-shaped function which smoothes the transitions between the extreme resistance values. Starting from this generic shape, several functions can be precisely dened by changing the position of the function in relation to the boundary and by modifying the values of rm and dm (Fig. 4). The usual denition of resistances values for the computation of cost distances is called function 0. It will constitute a reference function in order to evaluate the relevance of other functions. For a cell of landscape class p, the resistance is given by r = f0 (p) where f0 is a discrete positive function: f0 (1) = ; f0 (2) = ;. . .; f0 (k) = and f0 (i) 1i. With only two classes H and NH and a resistance of 1 for the class H, it only necessary to assign a resistance value rm to the class NH, giving a very

The following functions depend both on the classes H and NH and on the distance d to the boundary between them. For the function f1 , the class H is assigned a uniform resistance of 1 whereas the class NH takes an increasing resistance as one goes away from the HNH boundary, up to the maximal distance where the resistance becomes uniform: f1 (d, p = H) = 1,

r (r 1) 1 d2 m m dm f1 (d, p = NH) =
rm

if d dm if d > dm

(3)

Using this function assumes that the areas of NH close to H (i.e. not more distant than dm ) can be used for movements, for example thanks to a possible and rapid access to the resources. If the individual goes beyond the distance dm , this accessibility becomes more and more difcult and the environment becomes rapidly more restrictive, yielding an avoidance behaviour. On the other hand, the resistance value of 1 of the H areas means that very few movement constraints are added to the physical distance. In the function f2 the class NH is assigned a uniform value of constraint rm . In the areas close to the HNH boundary the class H presents a strong resistance which decreases as one goes away from this boundary until the value becomes 1 for

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

109

here least cost; (2) the distance of the least-cost trip, used for example in Broquet et al. (2006) and called least distance. The rst measure explicitly includes the notion of cost given in a somewhat abstract unit whereas the second, expressed in a metric unit, is perhaps easier to interpret. Both will be used in this study. The comparison of combinations of parameters is done by using a Pearson correlation coefcient (r) between resulting least cost or least distances and the distribution of vole density in September 2002. As densities are expected to decrease with the distance from the epicentre, the weakest correlations will be obtained from the more relevant type of distance, i.e. the distance whose spatial distribution is the most realistic representation of the spread phenomenon.

3.

Results

The knowledge about vole swarming leads to assume that grasslands will always be preferred to other landscape classes and wooded areas involve a repulsion effect. Hence grasslands constitute the habitat of voles (H), wooded areas are the habitat of most vole predators that gives inhospitable areas (NH), and the other classes are dened as neutral areas (N).

3.1. Relationships between population density and distances to the epicentre

Correlation coefcients are calculated between the density values of September 2002 and the different compared set of distances. The different resistance values rm allocated to the unfavourable landscape class are 5, 10, 20, 40, 60, 80 and 100. The neutral class is successively affected a resistance value of 1 (favourable), rm (also depending on each parameter set) and + (symbolic value representing an impassable barrier). The rst result concerns the Euclidian distance, which shows a strong relationship, with an r value of 0.76 (p < 0.0001). The cost distances are then applied following the usual manner (f0 function). With a neutral class considered as an impassable zone, correlations are globally in the same area as the values obtained from the Euclidian distance (Table 1). For the least cost, such a relationship decreases when the rm value increases. For the least distance, the level of the correlation is very stable and not signicantly different from the correlation obtained from the Euclidian distance. For both kinds of distance (least cost and least distance) the change of the resistance value allocated to the neutral class does not show additional results.

Fig. 4 Functions used to compute the resistance values.

the distance dm .

1 + (r 1) 1 d2 m f2 (d, p = H) = d2 m
1 f2 (d, p = NH) = rm

if d dm , if d > dm , (4)

In comparison with the preceding function, this function corresponds to a relative vulnerability of the habitat near the boundary. It can represent for example the pressure of predators living outside the habitat of the species in question. In this case the distance dm means the range of potential displacement of these predators. The habitat areas located far from the inhospitable areas will be in consequence favorite transit areas. In the example above both functions can be compared to test the assumptions of permeability of inhospitable areas or of vulnerability of habitats. The cost distances are thus computed from the epicentre point by dening values of rm and dm for all others sites of observation. Starting from a given combination of parameters, several distances can be recorded for a trip between two points: (1) the accumulation of costs, which is the most common computed measure and which is called

Table 1 Relationship between usual cost distances and vole density rm value
5 10 20 40 60 80 100

Least cost
0.7646 0.7635 0.7613 0.7580 0.7542 0.75023 0.7462

Least distance
0.7641 0.7657 0.7644 0.7641 0.7642 0.7644 0.76455

110

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

Fig. 5 Relationships between f1 and f2 -based cost distances and vole density.

Next functions f1 and f2 are applied using the same set of rm values and by successively increasing distances dm of 5 pixels (35 m), from 5 (35 m) to 50 pixels (350 m). Results obtained with function f1 are similar to the earlier results for all values of distance dm and resistance rm , whatever the kind of distance used (Fig. 5a). On the other hand, the application of function f2 leads to higher absolute values of Pearson correlations with some combinations of parameters, especially for least cost (Fig. 5b). Globally the curves are very similar and all show the same shape with the minimal correlation for dm = 20 (140 m). When inspected in more detail, it becomes apparent that the level of relationship increases from resistance rm = 5 to 40 and becomes quite stable for the upper values. For the smaller distance, the difference between the correlations obtained with function f2 and using Euclidian distance or the usual cost distances are not as high as seen previously. Several maps are used to compare the different distances from a spatial point a view. First, a spatial interpolation of the vole density is applied by means of an ordinary kriging (Cressie, 1993: 119). We note that the spatial precision of the ground observations and the spatial resolution of the interpolation grid (7 m) are not in agreement. Consequently, even though the accuracy of the density computed locally cannot be validated, the resulting map can be considered as the reference spatial distribution in order to evaluate the ability of cost distances to represent the spreading process (Fig. 6a). Second, the logarithmic transformation of the Euclidian distances from the epicentre shows regular aureoles (Fig. 6b) which do not match the previous distribution. However, the application of usual cost distances (Fig. 6c) leads globally to

a similar distribution, where the shape of aureoles is weakly modied near wooded patches. Only the cost distances based on function f2 provide a global distribution which is more radically different from an isotropic shape (Fig. 6d). In this case, the preferential orientation North-East/South-West is reproduced even if several details in the maps Fig. 5a and d strongly differ.

4.

Discussion and conclusion

The resistance values allocated to the landscape classes for the cost distances computations have been dened here by using topological functions of landscape. The main assumption explaining this proposal is the role of edges as interface in the operation of ecosystems. Several kinds of functions are possible, corresponding to assumptions on the movement forms of the studied species. The case of the spread of vole has provided an example where the method is applied by comparing distances from an epicentre to density measured after the start of a diffusion process. Compared to the Euclidian distance, the costs dened by uniform resistances do not contribute to increase the explanatory power of the resulting distances. In the same way the costs dened by function f1 do not bring any change. This means that the wooded class cannot be considered as partially permeable areas. On the other hand, the use of function f2 leads to a higher level of statistical relationship, when the statistical criterion used is the least cost. This means that the grasslands near the wooded areas (in a spatial range of

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

111

Fig. 6 Cost distances generated by function f2 in the studied zone.

112

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

140 m) are avoided, and this can be interpreted as some kind of spatial vulnerability. Predator constraints and accessibility to the resources can be modelled as resistance function increasing when approaching an edge with a wooded element. This result is consistent with precedent conclusions obtained et al., from other analysis based on different methods (Foltete 2005). For all the types of cost distance, the resistance value allocated to the neutral class has never played a role in the global relationship between distances and densities. This fact is probably the consequence of a very weak surface taken up by this class; in other applications its role might depend on its surface in the studied zone. Concerning the results obtained with the function f2 , it is interesting to note that both least cost and least distance criteria provide different levels of statistical relationship. However, the present analysis does not allow interpreting this difference from a general point of view, because the greater relevance of the least cost criterion may be linked to the specic context of the present example. The diffusion process of vole swarming involves a very complex combination of reproduction, colonization, relation to predators and resources, and cannot be compared to a simple movement of a set of individuals seeking to minimize their distance of moving. The cost unit depends on the series of resistance values, making its ecological meaning difcult to interpret.

The fundamental contribution of the topological functions of landscape used in the computation of cost distances is lies in the possibility they offer to successfully model a kind of barrier effect which is not directly visible in the land cover map. To be more precise, the classical approach of cost distances involves no difference of habitat permeability between a large zone and a corridor surrounded by inhospitable patches, while the method presented here allows to distinguish between these two cases. In other words, even if vole populations live in the landscape matrix (i.e. in the studied zone where nearly all grassland areas are inter-connected), the specic allocation of resistances according to the proposed method tends to gradually decrease the level of permeability near inhospitable patches until absolute barriers are generated when the density of these patches exceeds a given threshold. A theoretical example of this case is illustrated in Fig. 7: it shows how the inuence of some landscape elements on potential trips of individuals may be different according to the method used for allocation of resistances. This means that the use of topological functions of landscape which does not rely only on strict localization but takes into account the spatial context results in a modication of the scale of inuence of landscape on the movement. Globally the method used here allows to substantiate some assumptions about the landscape-species relationships and to produce new knowledge elements. It can complete the

Fig. 7 Cost methods and relative barrier of landscape.

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

113

usual analysis of edge effects based on ground data by making use of GIS tools, especially when generalization of edge effect patterns are difcult (Paton, 1994; Batary and Baldi, 2004). The method also provides useful data with the prospect of use of cost distances in other approaches: for example, the cost distances can be incorporated into certain diffusion model instead of Euclidian distances; the resistance maps can be used in individual-based movement models. The present example is based on a single point source but can easily become widespread with any set of spatial targets. In the same way other types of function can also be used. It remains to bring up an important limit inherent to the method. The main interest of cost distance computations is the possibility of taking into account the spatial heterogeneity of the landscape matrix, especially in the case of species living in habitat patches. The present method required a simplication of landscape with only two classes where resistance can be dened by a topological function, and another class reserved to a neutral behaviour towards the limit between the two previous classes. It is theoretically possible to take into account more neutral classes by allocating several different resistances in relation to land cover, but the number of resistance values to be dened is in reality difcult to manage, and working on a very simplied list of land cover classes is much more practical. One can certainly imagine an extension of the topological functions to more classes in order to combine several ecological effects but it would lead to a very complex method. In addition, the presence of multiple edge effects proves to be difcult to interpret (Fletcher, 2005) and the validation of the resistance values which is time-consuming in the usual computations (Verbeylen et al., 2003) would be even longer in this case. Consequently, the present method can be applied to relatively simple landscape contexts or to landscapes that can be simplied without signicant loss of information.

Acknowledgements
Funding was provided by the Regional Council of the Franche and the French Ministry of Environment. The authors Comte are grateful to F.P. Tourneux for helping with image preprocessing and to R. Barzel for reviewing the language of the manuscript.

references

Adriaensen, F., Chardon, J.P., De Blust, G., Swinnen, E., Villalba, S., Gulinck, H., Matthysen, E., 2003. The application of least-cost modelling as a functional landscape model. Landscape Urban Plann. 64, 233247. Arnaud, J.F., 2003. Metapopulation genetic structure and migration pathways in the land snail Helix aspersa: inuence of the landscape heterogeneity. Landscape Ecol. 18, 333346. Arter, H.E., 1990. Spatial relationship and gene ows paths between populations of the alpine snail Arianta arbustorum. Evolution 10, 966980. Batary, P., Baldi, A., 2004. Evidence of an edge effect on avian nest success. Conserv. Biol. 18, 389400.

Berthier, K., Galan, M., Foltete, J.C., Charbonnel, N., Cosson, J.F., 2005. Genetic structure of the cyclic fossorial water vole (Arvicola terrestris): landscape and demographic inuences. Mol. Ecol. 14, 28612871. Broquet, T., Ray, N., Petit, E., Fryxell, J.F., Burel, F., 2006. Genetic isolation by distance and landscape connectivity in the American marten (Martes americana). Landscape Ecol. 21, 877889. Bunn, A.G., Urban, D.L., Keitt, T.H., 2000. Landscape connectivity: a conservation application of graph theory. J. Environ. Manage. 59, 265278. Chardon, J.P., Adriaensen, F., Matthysen, E., 2003. Incorporing landscape elements into a connectivity measure: a case study of Speckled wood buttery. Landscape Ecol. 18, 561573. Coulon, A., Cosson, J.F., Angibault, J.M., Cargnelutti, B., Galan, M., Morellet, N., Petit, P., Aulagnier, S., Hewison, A.J.M., 2004. Landscape connectivity inuences gene ow in a roe deer population inhabiting a fragmented landscape: an individual based approach. Mol. Ecol. 13 (9), 28412850. Cressie, N.A.C., 1993. Statistics for Spatial Data. John Wiley, New York. Delattre, P., Duplantier, J.M., Fichet-Calvet, E., Giraudoux, P., 1998. publique. Cah. Pullulation de rongeurs, agriculture et sante Agric. 7, 285298. Delattre, P., Chaste, B., Silvy, C., 2000. Lutte biologique et rongeurs. Dossiers de lenvironnement de lINRA 19, 3547. e, J.P., Delattre, P., Giraudoux, P., 2000. Duhamel, R., Quer Landscape effects on the population dynamics of the fossorial form of the water vole (Arvicola terrestris sherman). Landscape Ecol. 15, 8998. Eastman, J.R., 1989. Pushbrooms Algorithms for Calculating Distances in a Raster Grid, vol. 9. Autocarto, Baltimore, pp. 288297. Eastman, J.R., 1998. Idrisi for Windows. Clark University, Graduate School of Geography, Worcester. ESRI, 1996. Arc View Spatial Analyst, Redlands. ESRI, 2001. Using ArcGISTM Spatial Analyst, Redlands. Fagan, W.F., 1999. How habitat edges change species interactions. Am. Nat. 153, 165182. Ferreras, P., 2001. Landscape structure and asymmetrical inter-patch connectivity in a metapopulation of the endangered Iberian lynx. Biol. Conserv. 100, 125136. Fletcher, R.J., 2005. Multiple edge effects and their implications in fragmented landscapes. J. Anim. Ecol. 74, 342352. Foltete, J.C., Berthier, K., Cosson, J.F., 2005. Paysage et pullulation animale: les facteurs paysagers de la vitesse de propagation des vagues de pullulations du campagnol terrestre. Cybergeo 306. http://193.55.107.45/articles/306.pdf. Forman, R.T.T., 1995. Land Mosaics Cambridge. University Press, Cambridge. Forman, R.T.T., Godron, M., 1986. Landscape Ecology. John Wiley, New York. Giraudoux, P., Pradier, B., Delattre, P., Deblay, S., Salvi, D., Defaut, R., 1995. Estimation of water vole abundance by using surface indices. Acta Theriol. 40, 7796. e, J.P., Deblay, S., Giraudoux, P., Delattre, P., Habert, M., Quer Defaut, R., Duhamel, R., Moissenet, M.F., Salvi, D., Truchetet, D., 1997. Population dynamics of fossorial water vole: a land use and landscape perspective. Agric. Ecosyst. Environ. 66, 4760. Gustafson, E.J., Parker, G.R., 1994. Using an index of habitat patch proximity for landscape design. Landscape Urban Plann. 29, 117130. Halpin, P.N., Bunn, A.G., 2000. Using GIS to compute a least-cost distance matrix: a comparison of terrestrial and marine ecological applications. In: Proceedings of the 20th Annual ESRI User Conference, San Diego, CA, pp. 2630, http://gis.esri.com/library/userconf/proc00/professional/ papers/PAP890/p890.htm.

114

e c o l o g i c a l m o d e l l i n g 2 1 0 ( 2 0 0 8 ) 104114

Knaapen, J., Scheffer, M., Harms, B., 1992. Estimating habitat isolation in landscape planning. Landscape Urban Plann. 23, 116. Matthiopoulos, J., 2003. The use of space by animals as a function of accessibility and preference. Ecol. Model. 159, 239268. Michels, E., Cottenie, K., Neys, L., De Gelas, K., Coppir, P., De Meester, L., 2001. Geographical and genetic distances among zooplankton populations in a set of interconnected ponds: a plea for using GIS modelling of the effective geographical distance. Mol. Ecol. 10 (8), 19291938. Okubo, A., Levin, S.A., 2001. Diffusion and ecological problems. Springer, New York. Pain, G., Baudry, J., Burel, F., 2000. LandPop: un outil detude de la structure spatiale des populations animals fragmentees. Revue Internationale de Geomatique 10, 89106. Paton, P.W.C., 1994. The effect of edge on avian nest success: how strong is the evidence? Conserv. Biol. 8, 1726. e, J.P., Garel, J.P., Rous, C., Pradier, B., Delattre, P., 1999. Quer ats causes par les pullulations du Campagnol Estimer les deg terrestre (Arvicola terrestris) en prairie naturelle. Fourrages 158, 133147. Ray, N., 2005. Pathmatrix: a geographical information system tool to compute effective distances among samples. Mol. Ecol. Notes 5, 177180.

Ray, N., Burgman, M.A., 2006. Subjective uncertainties in habitat suitability maps. Ecol. Model. 195, 172186. Ray, N., Lehmann, A., Joly, P., 2002. Modeling spatial distribution of amphibian populations: a GIS approach based on habitat matrix permeability. Biodivers. Conserv. 11, 21432165. Reese, K.P., Ratti, J.T., 1988. Edge effect: a concept under scrutiny. Trans. North Am. Wildl. Nat. Resour. Conf. 53, 127136. Rousset, F., 1997. Genetic differentiation and estimation of gene ow from F-statistics under isolation-by-distance. Genetics 145, 12191228. Rousset, F., 2000. Genetic differentiation between individuals. J. Evol. Biol. 13, 5862. Turchin, P., 1998. Quantitative analysis of movement. Sinauer, Sunderland. Urban, D.L., Keitt, T.H., 2000. Landscape connectivity: a graph theoretic approach. Ecology 82, 12051218. Verbeylen, G., De Bruyn, L., Adriaensen, F., Matthysen, F., 2003. Does matrix resistance inuence Red squirrel (Sciurus vulgaris L. 1758) distribution in a urban landscape? Landscape Ecol. 18, 791805. Wright, K.A., 1943. Isolation by distance. Genetics 28, 114138. Yu, K., 1996. Security patterns and surface model in landscape ecological planning. Landscape Urban Plann. 36, 117.