SEARCH

SEARCH BY CITATION

Keywords:

  • calibration;
  • composition;
  • configuration;
  • fragmentation;
  • landscape complementation;
  • potential movement zone;
  • resistance

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

1. Capturing the relative influence of landscape composition and configuration in real landscapes remains a challenge. Cost-distance modelling provides an interesting approach to the assessment of landscape complexity in a functional way. However, resistances allotted to landscape elements in cost-distance modelling frequently remain defined on the basis of expert advice. To overcome this weakness, we computed resistance coefficients without a priori knowledge through a calibration/validation method enabling us to test the impact of the matrix heterogeneity on the occurrence of the common toad Bufo bufo, the cycles of which imply migrations between complementary habitats.

2. We used cost-distance modelling to elaborate an integrative parameter of landscape composition and configuration in the form of a potential movement zone. We first applied a calibration procedure that systematically tested different resistance values for each landscape element with a large data set. The robustness of the calibrated resistances was then evaluated using two supplementary validation data sets from contrasted landscapes. Finally, in order to isolate the relative influence of landscape configuration, we compared the predictive power of the calibrated potential movement zone with that of landscape composition only.

3. The landscape matrix strongly influences common toad occurrence: selected resistances were low for forests and meadows and intermediate to high for crops. Within the two validation data sets, the potential movement zone was positively and significantly related to toad occurrence and had a better predictive power than landscape composition.

4.Synthesis and applications. This study provides a tool to manage landscape structure in accordance with the ecological requirements of amphibian populations, especially habitat complementation. This method has minimal biological information requirements and therefore could be widely used to investigate the crucial influence of landscape composition and configuration on a broad range of species, and to design functional ecological networks.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

As human activities increase, more natural areas are converted into agricultural and urbanized landscapes. Such landscape conversions lead to habitat fragmentation which is recognized to be one of the major threats to the long-term persistence of biodiversity (Fahrig 2003). In a fragmented landscape, population viability is expected to depend on functional connectivity (i.e. the permeability of the landscape matrix to movement) which influences the success of migration and dispersal (Baguette 2004). Connectivity is determined first by the resistance which the diverse land uses composing the matrix present to animal movement, and second by the configuration of those land uses (Wiens 2001). Despite their crucial importance for wildlife management, only a few studies have segregated the effects of landscape composition from those of configuration (Fahrig 2003). Recent advances in Geographical Information Systems allow the integration of landscape features into spatially explicit models (Verbeylen et al. 2003), making it possible to link spatial landscape organization to animal movements (Driezen et al. 2007; Epps et al. 2007). More particularly, cost-distance modelling (e.g. Ray et al. 2002) presents the advantages of requiring a restricted set of biological hypotheses and of being easily transferable to landscape managers.

The principle is to build a friction map by assigning a resistance to each cell of landscape according to local land use and its assumed effects on animal movements. Two types of measures are usually derived from this approach: the measure of the least cost path from one point to another (Adriaensen et al. 2003; Verbeylen et al. 2003) or the measure of the accumulated cost surface from a source to its surroundings under a threshold value (Ray et al. 2002; Joly et al. 2003; Compton et al. 2007). Whatever the approach adopted, any false assumption about resistance may result in misleading conclusions. Nonetheless, the accuracy of cost-distance modelling remains largely untested despite its wide use, as resistance is often arbitrarily established on the basis of expert opinion (Driezen et al. 2007; Epps et al. 2007). The reliability of the resistance values thus clearly constitutes the Achilles’ heel of the cost-distance approach (Adriaensen et al. 2003). Resistance could be estimated from behavioural experiments conducted at a fine scale, concerning for example habitat selection (Mazerolle & Desrochers 2005; Stevens et al. 2006) or locomotion performances (Jonsen & Taylor 2000; Stevens et al. 2004). However, such experiments poorly capture at the landscape scale the complex movement processes underlying local population persistence on which cost-distance modelling focuses (Adriaensen et al. 2003). An alternative approach would be to calibrate these resistance values with large and contrasted data sets.

Amphibians are susceptible to the impacts of land-use changes because their ground-dwelling habits and the permeability of their skin result in close exchange with their immediate environment (Rothermel & Semlitsch 2002). Moreover, their life cycles involve seasonal migrations between terrestrial and aquatic habitats (i.e. landscape complementation, Pope et al. 2000), which could compel them to regularly traverse an inhospitable landscape matrix. Areas of intensive agriculture are predicted to expose amphibians to desiccation and chemical agents (Joly et al. 2001; Rothermel & Semlitsch 2002). The presence of amphibians in ponds depends both on the connectivity between populations and on the costs generated by seasonal migration. Amphibians present an opportunity to analyse the effects of landscape structure on population persistence and more particularly that of landscape configuration, which is increased by their need to move between complementary habitats (Fahrig & Nuttle 2005).

In this study, cost-distance modelling is used to test the impact of modifications to landscape connectivity on the occurrence of common toads Bufo bufo Linnaeus. Although our focus species is a generalist, landscape variables determine, as for many amphibians, the distribution and the size of populations (Scribner et al. 2001). We used an integrative parameter based on accumulated cost surface: the area within which common toads can move from the focal pond to their growth and maintenance habitats (i.e. the potential movement zone). The matrix resistance constrains the potential movement zone, which thus constitutes a reliable indicator of local landscape suitability (Ray et al. 2002). Consequently, we expect a positive relationship between the potential movement zone and species occurrence. We adapted a cross-validation method to avoid the arbitrary assignation of resistance values. First, we applied a calibration procedure to estimate resistances, without using a priori knowledge of the influence of land use on toad movements. Then, we tested the validity of the calibrated resistances by predicting toad presence using two different pond data sets, one in the region used for calibration and the second in another region differing in landscape structure. Finally, we tested the relative effect of landscape configuration by separating the predictive power of landscape composition from that of the potential movement zone, which integrates both composition and configuration.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Focus species and sampling method

As demonstrated by Scribner et al. (2001), both presence and population size of the common toad were better predicted by the terrestrial environment than by pond quality. The common toad is a ubiquitous and widespread species in Europe that breeds in a great diversity of wetlands including farmland ponds (Loman & Lardner 2006). The tolerance of tadpoles to the quality of their breeding pond relies on feeding flexibility (Diaz-Paniagua 1989), relative tolerance to fertilizers (Xu & Oldham 1997) and to pesticides (Mandrillon & Saglio 2007), and unpalatability to both native (Glandt 1984) and invasive fishes (Reshetnikov 2003). Therefore, the absence of toads at a pond is more likely to result from the characteristics of the surrounding landscape rather than those of the pond. Moreover, movement and habitat preferences have been well studied in this species. Using genetic markers, Scribner et al. (2001) reported genetic structuring between populations breeding in ponds by at least 2 km distant from one another. This dispersal distance is also consistent with the range of migration distances reported in Sinsch (1988) (i.e. 50–1600 m), while recent reviews have reported non-anecdotic long distances of migration reaching 3 km or more (Smith & Green 2005). We therefore assumed an optimum maximal migration distance of 2 km, but we also examined the sensitivity of our analyses to this hypothesis using both a lower and a higher distance (1 and 3 km respectively). Forests were identified as the usual terrestrial habitat (Denton & Beebee 1994), although we deliberately did not settle an a priori hypothesis about forest resistance to movements in our simulations. Additionally, the placidity and explosive breeding of common toads make them easily detectable when present at ponds. These features reduce the risk of false absence even when the local abundance is low. We monitored the presence/absence of the common toad by visiting ponds at night during the breeding period. Toads were detected and sexed visually using floodlights. As a preliminary survey did not reveal any change in toad occurrence within 14 ponds sampled twice during the breeding period, we decided to sample each pond only once each year.

Study regions and sampled ponds

To explore a wide range of landscape structures, we took advantage of different degrees of agriculture intensity in the lowland landscapes of the Rhone–Alps region (south-eastern France). We selected six areas characterized by the dominance of croplands (two areas), meadows (two areas) and forests (two areas), thus reflecting different landscape compositions (Fig. S1 in Supporting Information) and configurations (Table S1 in Supporting Information).

We sampled 212 ponds distributed among these six areas in 2006 (between 21 and 49 ponds per area). These ponds were selected according to the following criteria: surface exceeding 40 m2, distance from any large urbanized area and mountains, absence of acute pollution and presence of vegetation and of gently sloping banks. Furthermore, ponds were also selected to be separated from one another by at least 2 km to avoid spatial autocorrelation of toad occurrence. These ponds were divided into two sets: 129 ponds constituting the calibration data set and 83 ponds providing the first validation data set. The 129 ponds of the calibration data set were monitored during two consecutive years (2004 and 2005) preceding the present study. As none of the empty ponds in 2006 were previously occupied (during 2004 and/or 2005), the risk of false absence in the 2006 data set is very limited.

A second validation data set located outside the Rhone–Alps region was analysed to avoid pseudoreplication problems (Hulbert 1984). We obtained access to a data set previously used by Ray et al. (2002) that provides male common toad occurrence in ponds of the Canton of Geneva (246 km2, Switzerland) during 1998–1999. We did not use ponds located within the city core but focused on 77 ponds (exceeding 10 m2) located in the countryside. The landscape of this region differs markedly from that of the Rhone–Alps region and falls into one of the two following types: intensively managed agricultural areas including small fragments of meadows and woodlands, contrasting with large forested areas (Fig. S1).

Land-use maps

Five land-use types were considered: standing waters (marshes and water bodies), forested areas, meadows (pastures and open areas), croplands and urban areas (towns and suburbs, industrial areas). Four linear landscape elements were also integrated: roads, rivers, highways and large rivers.

Forests, croplands, meadows and standing waters were identified by automatic classification from satellite images (ASTER/TERRA images with 15-m resolution) using the ENVI software (ITT, Boulder, CO, USA). Linear elements and urban areas were extracted from national maps (BD Carto® from National Geographic Institute, France). Gaps (cells without attribution: <2% in our map) were filled with the spatially closest land use using the NIBBLE function (SPATIAL TOOLS extension in arcview 3.2, ESRI, Redlands, CA, USA). Breaks in linear structures (Rothley 2005) were avoided by the reinforcement of the size of the linear elements, particularly those acting as barriers. For the Geneva data set, we used the map built by Ray et al. (2002). We used arcview 3.2 and its extension SPATIAL ANALYST to store, manage and rasterize all the landscape elements.

Estimation of the potential movement zone

The methodology of cost-distance modelling, originating from graph theory, is briefly described below (for detailed information, see Adriaensen et al. 2003). The resistance of a landscape element expresses the degree by which a potential movement is impeded when compared with the most permeable element (i.e. having a resistance of 1). Given the resistance for each landscape element, a friction map is created where a cost value is assigned to each cell of the grid corresponding to the product of the cell resolution by the resistance of its landscape attribute. A negative growth algorithm was used to calculate the accumulated cost surface and the resulting potential movement zone. The maximal migration distance was set as the starting value of the algorithm. This starting value equals the distance that could be covered in the landscape element offering the lowest resistance. For each cell crossed, the potential for movement was decreased according to the cost value assigned to that cell. Therefore, the potential movement zone takes into account both the composition and the configuration of landscapes. Indeed, two circular landscape areas, each composed of 50% of each of two elements (suitable and unsuitable), but with contrasted configurations (one semicircle of each element vs. a circle of unsuitable element surrounded by a ring of suitable element) would possess different potential movement zones (an extended one for the semicircle configuration vs. a restricted one for the pond embedded in the unsuitable element). The exhaustion of the starting value defines the edges of the potential movement zone that hence represents the terrestrial area which can be reached by toads breeding in/emerging from a focal pond under the hypothesis of a matrix effect (see Supporting Information, Fig. S2).

Statistical analysis

The uneven sex ratio prevents us from properly interpreting the interaction between sex and landscape, although we expect that the sensitivity of common toads to landscape greatly varies between sexes due to sex-biased reproductive investment and nomadic behaviour (Frétey et al. 2004). We thus chose to analyse occurrence data separately for each sex, an approach that is more robust to sex differences in landscape sensitivity, although less powerful than an analysis on the whole data set.

Calibration

The calibration procedure is based on a ‘brute force’ method to examine all possible combinations of resistances for several landscape elements. As their impact is clearly defined, the resistance was kept at the lower bound (i.e. 1) for standing waters, whereas highways and large rivers were set as barriers (i.e. resistance >200). We excluded roads and small rivers from the calibration procedure because their resistance is obviously intermediate but hard to define without risk functions. Indeed, both roads and rivers are linear elements which are not very costly to cross, but the risk associated with crossing them can be very high. Therefore, as they are risky rather than costly to cross, their effects cannot be properly calibrated through cost-distance modelling (see the Discussion section). Moreover, the number of ponds separated by small rivers is restricted (<8% of ponds). We thus attributed an arbitrary resistance value of 12 to small rivers and a value of 7 to roads (Ray et al. 2002). We focused the calibration on the four following landscape elements: forests (constituting both the target terrestrial habitat of common toads and a matrix element through which they move), meadows, crops and urban areas (three matrix elements). We used a sequential procedure to determine the optimal resistance value as well as its 95% confidence interval (95% CI) within the parameter space for each focus landscape element. This process was performed in five steps of increasing resolution.

At the coarsest resolution, five resistance values (1, 25, 50, 75 and 100) were used as cut points (bounds) to parse the parameter space. This resulted in 54 models (i.e. combinations of resistance values for the four landscape elements) for each tested maximal migration distance (i.e. 1, 2 and 3 km). Each model was then evaluated using a logistic regression of the potential movement zone on toad occurrence (male and female). Some models among those tested are likely to be the mirror of others (i.e. inversing resistance values for two negatively correlated landscape elements inevitably leads to different models with the same correlation strength but with opposite slope signs). In this case, only models resulting in a positive slope, which transcribes the effect of an increasing potential movement zone (i.e. of a more permeable landscape) on toad occurrence, are appropriate to elaborate a spatial predictive tool. We selected the optimal resistance of each landscape element among the first five tested values from the model exhibiting the lowest residual deviance. A 95% CI was also constructed around optimal resistance using the approximate chi-squared distribution of the residual deviance (5% threshold 3.84): for each element, we excluded from the 95% CI the resistance values among those tested for which the residual deviance of the resulting model exceeded 3.84 compared with that of the optimal model, thus defining the bounds of the selected parameter space.

Then, the coarse scale was refined using a finer scale of the parameter space to investigate more precise values included in the previously selected space (optimal resistances and their 95% CI). After completion of the multiple refining sequences, a last sequence was performed to obtain the final resistance value and its 95% CI. We used the 95% CI to segregate the effects of landscape elements according to their resistance as the absence of overlap between confidence intervals can be considered as a formal test of divergence between landscape elements. The sensitivity of the calibration method to the maximal migration distance was also examined using evidence ratios (EV) computed on all explored models.

Validation

Validation analyses were performed with potential movement zones constructed with the resistances calibrated for the optimum maximal migration distance (2 km). To assess the validity of the selected resistances, we performed logistic regressions of the ‘calibrated’ potential movement zone on toad occurrence for each data set. We first checked the model adequacy using the Hosmer–Lemeshow goodness-of-fit test (GOF) for continuous explanatory variables (Hosmer & Lemeshow 1989), and we estimated thereafter the predictive power of the potential movement zone using a chi-squared test on deviance (McCullagh & Nelder 1989). Finally, we also checked whether the slope estimated by the logistic regression performed on each validation data set was included in the 95% CI of the slope estimated on the calibration data set. This verification was performed only for the male data set as female occurrence was not available for Geneva.

Moreover, in the Geneva data set, because sampling was exhaustive, the mean distance (670 m) from one pond to another was shorter than in the Rhone–Alps data set. As a consequence, we expected spatial autocorrelation in species distribution. To assess the effect of such an autocorrelation, we tested the predictive power of the potential movement zone on occurrence within sets of spatially independent ponds (resampling method: Holland et al. 2004). As our results remained stable after resampling, we finally decided to work with the entire Geneva data set to avoid reduction in analysis power.

Relative influence of composition and configuration

In order to evaluate the respective influences of landscape configuration, we compared the predictive power of landscape composition alone with that of the potential movement zone, which integrates both composition and configuration. For this purpose, we described landscape composition by the amounts of each main landscape elements (forests, meadows, crops, urban areas, barriers, roads and rivers) in a circular area of 2-km radius centred on the focal pond. The regression model was therefore structured in a conservative way, as follows:

  • image

We examined effect of the landscape composition by evaluating the significance of the explanatory terms using an order-dependent test. Introduced in this way, the effects of landscape composition should be inflated relative to the one actually shared by the potential movement zone. Such a method is therefore conservative as it downplays the predictive power of configuration relative to that of composition. This analysis was performed on each validation data set. The landscape element ‘rivers’ not being available on the map of the Geneva region, we could not incorporate it into the analysis for the Swiss data set. All statistical analyses were performed with R 2.5.0 (R Development Core Team, Vienna, Austria).

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Toad occurrence

In the Rhone–Alps, female toads were detected in 70% of the calibration ponds and in 62% of the validation ponds. Across the same data sets, the occurrence of males was of 90% and 81% respectively. In Switzerland, male toads were present in 52% of ponds.

Calibration

For both males and females, the coarse calibration step segregated resistances for forests, crops and meadows but not for urban areas (Fig. 1). The lowest resistance value was selected for both forests and meadows. Conversely, the lowest resistance was rejected for crops. For urban areas, no resistance stood out against others across the entire range of explored values (Fig. 1). The resistance of this landscape element was therefore kept fixed at the upper bound (i.e. 100) for the subsequent analyses because of its negative impact on common toad populations reported in previous studies (e.g. Hitchings & Beebee 1998). After the final calibration sequence, the optimal resistance was found at the lower bound for both forests and meadows, whereas it was found at a higher value for crops whatever the sex considered (Table 1). The 95% CI for forests and meadows did not overlap that obtained for crops when calibrated on the females’ data set. Using the males’ data set, the 95% CI for forest and crops were well segregated, but the 95% CI for meadows overlapped that obtained for both forests and crops (Table 1). These results were consistent across the three tested maximal migration distances, although one can notice a slight upward shifting of the 95% CI for crops when the maximal migration distance increased.

image

Figure 1.  Resistance values for each landscape element selected by the coarse step calibration (described by the % of representation of each value within the set of selected models). UA, urban areas; Fo, forests; Cr, crops; Me, meadows.

Download figure to PowerPoint

Table 1.   Results of the last calibration sequence: final optimal resistance and its 95% CI for each land-use type
 1 km2 km3 km
FoMeCrFoMeCrFoMeCr
  1. Fo, forests; Cr, crops; Me, meadows.

Females
 Optimal value11711151128
 95% CI[1–2[[1–2[]2–50][1–2[[1–2[]3–90][1–2[[1–2[]4–100]
Males
 Optimal value1110117116
 95% CI[1–2[[1–13[]1–50][1–2[[1–7[]1–90][1–2[[1–7[]1–100]

For females, the lowest maximal migration distance (1 km) was clearly less supported than the other two distances according to their respective evidence ratio (EV2km/1km = 10·9; EV2km/3km = 0·44). For males, the weak relative support of the lowest maximal migration distance is less apparent (EV2km/1km = 3·6; EV2km/3km = 0·59).

Validation

For both validation data sets, the potential movement zone was simulated using optimal resistances obtained by calibration (see Table 1 for details) for the maximal migration distance of 2 km which was supported by literature and EV. For each validation data set, the GOF test of the logistic regression of the potential movement zone on toad occurrence was nonsignificant, whereas the predictive power of the potential movement zone was always significant as indicated by the deviance analysis (for the Rhone–Alps females’ data set, GOF: = 0·64; explained deviance 18%; analysis of deviance: < 0·001; for the Rhone–Alps males’ data set, GOF: = 0·70; explained deviance 11%; analysis of deviance: = 0·003; for the Geneva males’ data set, GOF: = 0·24; explained deviance 11%; analysis of deviance: = 0·001).

For each data set, the occurrence significantly increased with the extent of the potential movement zone (Fig. 2, see Supporting Information Fig. S2). Furthermore, logistic regressions of the potential movement zone on toad occurrence showed that the slope estimated for each validation data set (for Rhone–Alps: 0·42, 95% CI 0·11–0·73; for Geneva: 0.48, 95% CI 0·17–0·79) was encompassed within the 95% CI of the calibrated slope estimate (0·55, 95% CI 0·20–0·90).

image

Figure 2.  Logistic regressions linking the potential movement zone (in km2) computed with the optimal resistance values for maximal migration distance 2 km (for details on resistance values, see Table 1) to: (a) female presence probability in the Rhone–Alps validation data set; (b) male presence probability in the Rhone–Alps validation data set; (c) male presence probability in the Geneva data set.

Download figure to PowerPoint

Relative influence of configuration and composition

The analysis of the relative influences of composition and configuration showed that potential movement zone explained a significant part of the residual deviance after adjustment for the effects of composition variables (Table 2). This analysis therefore showed that potential movement zone had a significantly better predictive power than all the other composition variables. Globally, amounts of forest, crops and linear elements (barriers, roads and rivers when available) only had a gentle predictive power on toad presence, whereas the amounts of meadows and urban areas had a very restrictive one (Table 2), even if these predictive powers fluctuate slightly between the two validation data sets (Table 2).

Table 2.   Analysis of deviance of the sequential regression for the Rhone–Alps and Geneva validation data sets
Explanatory variablesRhone–Alps (females)Geneva (males)
d.f.Residual deviancePd.f.Residual devianceP
  1. The amount of each land-use type was computed within a 2-km radius around each pond. Significance level for each term of the model: *< 0·05, **< 0·01, ***< 0·001.

Null model 109·7  106·4 
Area of forests1107·70·1641101·30·024*
Area of crops1104·10·058196·20·023*
Area of meadows1102·80·241196·10·786
Urban areas1102·70·743196·00·905
Area of barriers198·40·039*194·20·166
Area of rivers193·60·029*1
Area of roads190·10·061191·60·107
Potential movement zone173·20·000039***180·90·001**
Residuals74  69  

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Our results clearly show, first, the influence of functional connectivity on the occurrence of common toads and, second, the alteration of connectivity by agriculture. Confidence in our analysis derived from the concordance among the results obtained (on male and female occurrences, with different maximal migration distances and in different regions).

Biological relevance

Potential movement zone and matrix heterogeneity

The first objective of this study was to evaluate the influence of the landscape matrix heterogeneity on amphibians in the context of habitat complementation. Common toads have to cross the matrix when migrating between breeding pond and forested habitat. Although not set a priori, forest resistance was always found at the lowest bound (i.e. 1) as is expected of a target habitat. Above all, the significantly lower resistance of meadows compared with crops clearly reveals the role of the matrix heterogeneity. High resistance values for crops are corroborated by their avoidance by the common toad (Sinsch 1989) and by the negative association between this species and arable land (Piha et al. 2007). Croplands, which offer detrimental conditions for amphibians, such as exposure to predation, the presence of pesticides, an arid microclimate and ground ruggedness, have a negative impact on connectivity. The low resistance for meadows is consistent with the beneficial conditions (relative ground smoothness and moisture, absence of pesticides) characterizing this landscape element. This result is also supported by correlative studies. For example, Scribner et al. (2001) established a positive relationship between common toad occurrence and the area of meadow in the landscape. However, the calibration surprisingly confused the resistance of forests, characterized by a dual function (habitat and matrix) and of meadows which are a simple matrix element. This lack of discrimination power is likely to originate from our calibration method (see the Improvement of the method section). Overall, these results are corroborated by the resistances attributed to landscape elements on the basis of expert opinions (Ray et al. 2002; Joly et al. 2003). Nevertheless, although urban areas are reported to negatively impact toad populations (e.g. Hitchings & Beebee 1998), setting their resistance to the lower bound rather than the upper bound did not modify the outcome of the projection. Such a flawed calibration is not surprising given the weak representation of this landscape element in our data sets, and the fact that it encompasses a wide variety of land uses (industrial areas, suburbs and villages with gardens).

Stability of the results

Globally, our results were similar for both sexes, although the contrasting effects of crops and meadows are nonsignificant for males. This reveals a lower influence of the matrix heterogeneity on male occurrence than on that of females. This is consistent with the different breeding behaviour of males and females. Indeed, because of their high food requirements and their high reproductive investment, females are expected to be more sensitive to habitat quality (e.g. for Bufo viridis, Sinsch et al. 2007). Furthermore, the operational sex ratio being highly biased in favour of males, they could choose to avoid competition at densely occupied sites. Notably, they have been shown to exhibit higher nomadic behaviour than females (Frétey et al. 2004), which implies the exploration of ponds associated with different landscape qualities. They thus have a higher probability of being observed in unsuitable landscapes resulting in a lower apparent selectivity.

With respect to migration distance, our results remained consistent across the range of maximal migration distances investigated in this study, even at the distance of 1 km that had low supporting EVs. Finally, despite differences in sampling methods between the two data sets (more restricted studied area, higher autocorrelation and smaller ponds included in the Swiss data set), the results remain congruent. Indeed, the impact of the potential movement zone computed with calibrated resistance values was well supported when applied to the different validation data sets, which highlights the wide applicability of this method.

Deciphering the meaning of the potential movement zone

The potential movement zone is an ‘omnibus’ method to evaluate the impact of landscape elements on local populations for several reasons. First, resistance reflects a wide range of ecological processes underlying the connectivity between target habitats (Adriaensen et al. 2003) such as locomotor (Stevens et al. 2004) and physiological constraints (Rothermel & Semlitsch 2002), predation (Russell et al. 2003) or the risk of being run over (Joly et al. 2003). Second, given that forest constitutes the terrestrial habitat, its resistance reflects availability (and perhaps quality) rather than permeability to movement (see the Improvement of the method section).

Finally, cost-distance modelling approaches integrate both landscape composition and landscape configuration. The risk of confounding effects between landscape configuration and water quality in ponds was limited by the avoidance of eutrophic ponds. Besides, most ponds were presumably free of pesticides as they were devoted to fish farming. The potential movement zone can thus be used to track the effect of connectivity. Indeed, a configuration effect was detected despite large landscape compositional effects (in the Geneva data set): the arrangement of the different landscape elements has a largely predominant effect (see Table 2). Landscape configuration is theoretically predicted to have an important influence on populations, yet few large-scale empirical studies confirm this prediction (Harrison & Bruna 1999) as the effects of habitat loss are often not separated from those of configuration (Fahrig 2003). By contrast, our study supports this theory in demonstrating the determining influence of configuration on population distribution.

This conclusion highlights the importance of connectivity between complementary habitats (Pope et al. 2000; Becker et al. 2007) which must be considered at a similar decisive level as connectivity at a regional scale (patchy populations) in landscape management (Fahrig & Nuttle 2005).

Methodological implications

The consistency of our results provides support for the validity of the calibration method as a way of overcoming the caveat of subjective resistance estimation. Even if this method requires both large data sets and intensive computer use, it allows conclusions to be drawn at the landscape scale, unlike empirical estimations of resistance values.

Improvement of the method

First, occurrence data reflect the suitability of habitats only roughly, as local populations can persist for several years in degraded landscapes before going extinct (Piha et al. 2007). Combining less time-lagged parameters of population dynamics, such as abundance, reproduction cues and phenotypic indicators reflecting sublethal effects of fragmentation, such as body condition or stress level, could result in a substantial gain in accuracy. Additionally, pond characteristics might also alter population persistence independently of the potential movement zone. However, such an effect should result in inflated background noise whatever the potential movement zone, while our results indicate that predictive power seems higher for large potential movement zones than for small ones (Fig. 2).

Although the use of an integrative measure is a considerable advantage, the decomposition of the actual contribution of each landscape element would constitute an interesting development. This is particularly true for forests which constitute both a matrix element (used only for movement) and the target habitat of our focus species. Thus, their resistance has a particular meaning as it reflects both permeability to movement and carrying capacity. It would be interesting to determine which part of the variation is due to the sole influence of forest configuration and which is explained by the structure of every other matrix element separating the forest from the breeding habitat.

Moreover, because resistance in cost-distance modelling is expressed in rasterized cells, tested values need to be integers (cells cannot be split). Therefore, the minimal inflation of resistance value is from 1 to 2, which corresponds to a 50% reduction in the achievable distance as it is defined by the ratio between the potential of movement and the resistance value. Nonetheless, it is likely that some landscape elements, particularly meadows, have a resistance close to one but slightly higher than this minimal value. Consequently, to refine calibration, resistance values between 1 and 2 should be investigated.

Finally, because roads are risky rather than costly to cross, cost-distance modelling is not the best way to calibrate their effective impact. Although an a posteriori analysis indicated that results were robust to variation of resistance in a range from 2 to 50 for roads, such a landscape element is likely to have a real impact on amphibian populations (Eigenbrod et al. 2008). Models capable of simulating risks of being run over at the individual scale, such as cellular automaton or more complex individual-based models, could greatly improve the evaluation of the impact of roads.

Conservation implications

Our study demonstrates the efficiency of the potential movement zone to capture both composition and configuration effects of landscape on species distribution. The potential movement zone enables landscape management in accordance with the ecological requirements of local populations, especially habitat complementation. It constitutes an integrative tool to design ecological networks and identify blocking points for connectivity restoration. This study illustrates how a confident parameterization can be implemented in cost-distance modelling to avoid the subjective assignation of resistances. As it requires little biological information, this method could be widely used to track landscape effects, irrespective of the focus species, to examine landscape connectivity and to investigate the crucial influence of matrix composition and configuration on biodiversity.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We thank Philippe Maunoir from the Museum of Natural History of Geneva for access to the Geneva data set. We also thank the editorial team and anonymous reviewers for improving this manuscript. This study was funded by the French Ministry of Environment (MEEDA), the Biodiversity and Global Change programme of the French Institute of Biodiversity (IFB) and the Rhone–Alps region (Environment cluster). We are grateful to all participants to data collection, particularly Raphaël Quesada from Lo Parvi Association. Many thanks are due to Christina Richardson and Brigitte Planade for revising the English language.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Adriaensen, F., Chardon, J.P., Blust, G.D., Swinnen, E., Villalba, S., Gulinck, H. & Mathyssen, E. (2003) The application of least-cost modelling as a functional landscape model. Landscape and Urban Planning, 64, 233247.
  • Baguette, M. (2004) The classical metapopulation theory and the real, natural world: a critical appraisal. Basic and Applied Ecology, 5, 213224.
  • Becker, C.G., Fonseca, C.R., Haddad, C.F.B., Batista, R.F. & Prado, P.I. (2007) Habitat split and the global decline of amphibians. Science, 318, 17751776.
  • Compton, B.W., McGarigal, K., Cushman, S.A. & Gamble, L.R. (2007) A resistant-kernel model of connectivity for amphibians that breed in vernal pools. Conservation Biology, 21, 788799.
  • Denton, J.S. & Beebee, T.J.C. (1994) The basis of niche separation during terrestrial life between two species of toad (Bufo bufo and Bufo calamita): competition or specialisation? Oecologia, 97, 390398.
  • Diaz-Paniagua, C. (1989) Larval diets of two anuran species, Pelodytes punctatus and Bufo bufo, in SW Spain. Amphibia-Reptilia, 10, 7175.
  • Driezen, K., Adriaensen, F., Rondinini, C., Doncaster, C.P. & Matthysen, E. (2007) Evaluating least-cost model predictions with empirical dispersal data: a case-study using radiotracking data of hedgehogs (Erinaceus europaeus). Ecological Modelling, 209, 314322.
  • Eigenbrod, F., Hecnar, S.J. & Fahrig, L. (2008) The relative effects of road traffic and forest cover on anuran populations. Biological Conservation, 141, 3546.
  • Epps, C.W., Wehausen, J.D., Bleich, V.C., Torres, S.G. & Brashares, J.S. (2007) Optimizing dispersal and corridor models using landscape genetics. Journal of Applied Ecology, 44, 714724.
  • Fahrig, L. (2003) Effects of habitat fragmentation on biodiversity. Annual Review of Ecology Evolution and Systematics, 34, 487515.
  • Fahrig, L. & Nuttle, W.K. (2005) Population ecology in spatially heterogeneous environments. Ecosystem Function in Heterogeneous Landscapes (eds G.M.Lovett, C.G.Jones, M.G.Turner & K.C.Weathers), pp. 95118. Springer-Verlag, New York.
  • Frétey, T., Cam, E., Le Garff, B. & Monnat, J.Y. (2004) Adult survival and temporary emigration in the common toad. Canadian Journal of Zoology, 82, 859872.
  • Glandt, D. (1984) Laboratory experiment on the prey–predator relationship between 3-spined sticklebacks, Gasterosteus-Aculeatus L (Teleostei), and common toad larvae, Bufo bufo (L) (Amphibia). Zoologischer Anzeiger, 213, 1216.
  • Harrison, S. & Bruna, E. (1999) Habitat fragmentation and large-scale conservation: what do we know for sure? Ecography, 22, 225232.
  • Hitchings, S.P. & Beebee, T.J.C. (1998) Loss of genetic diversity and fitness in common toad (Bufo bufo) populations isolated by inimical habitat. Journal of Evolutionary Biology, 11, 269283.
  • Holland, J.D., Bert, D.G. & Fahrig, L. (2004) Determining the spatial scale of species’s response to habitat. BioScience, 54, 227233.
  • Hosmer, D.W. & Lemeshow, S. (1989) Applied Logistic Regression. John Wiley & Sons, New York.
  • Hulbert, S.H. (1984) Pseudoreplication and the design of ecological field experiments. Ecological Monographs, 54, 187211.
  • Joly, P., Miaud, C., Lehmann, A. & Grolet, O. (2001) Habitat matrix effects on pond occupancy in newts. Conservation Biology, 15, 239248.
  • Joly, P., Morand, C. & Cohas, A. (2003) Habitat fragmentation and amphibian conservation: building a tool for assessing landscape matrix connectivity. Comptes Rendus Biologies, 326, S132S139.
  • Jonsen, I.D. & Taylor, P.D. (2000) Fine-scale movement behaviors of calopterygid damselflies are influenced by landscape structure: an experimental manipulation. Oïkos, 88, 553562.
  • Loman, J. & Lardner, B. (2006) Does pond quality limit frogs Rana arvalis and Rana temporaria in agricultural landscapes? A field experiment. Journal of Applied Ecology, 43, 690700.
  • Mandrillon, A.L. & Saglio, P. (2007) Herbicide exposure affects the chemical recognition of a non native predator in common toad tadpoles (Bufo bufo). Chemoecology, 17, 3136.
  • Mazerolle, M.J. & Desrochers, A. (2005) Landscape resistance to frog movements. Canadian Journal of Zoology, 83, 455464.
  • McCullagh, P. & Nelder, J.A. (1989) Generalized Linear Models. Chapman & Hall, London.
  • Piha, H., Luoto, M. & Merila, J. (2007) Amphibian occurrence is influenced by current and historic landscape characteristics. Ecological Applications, 17, 22982309.
  • Pope, S.E., Fahrig, L. & Merriam, N.G. (2000) Landscape complementation and metapopulation effects on leopard frog populations. Ecology, 81, 24982508.
  • Ray, N., Lehmann, A. & Joly, P. (2002) Modeling spatial distribution of amphibian populations: a GIS approach based on habitat matrix permeability. Biodiversity and Conservation, 11, 21432165.
  • Reshetnikov, A.N. (2003) The introduced fish rotan (Perccottus glenii) depresses populations of aquatic animals (macroinvertebrates, amphibians, and a fish). Hydrobiologia, 510, 8390.
  • Rothermel, B.B. & Semlitsch, R.D. (2002) An experimental investigation of landscape resistance of forest versus old-field habitats to emigrating juvenile amphibians. Conservation Biology, 16, 13241332.
  • Rothley, K. (2005) Finding and filling the “cracks” in resistance surfaces for least-cost modeling. Ecology and Society, 10, article 4.
  • Russell, R.E., Swihart, R.K. & Feng, Z. (2003) Population consequences of movement decisions in a patchy landscape. Oïkos, 103, 142152.
  • Scribner, K.T., Arntzen, J.W., Cruddace, N., Oldham, R.S. & Burke, T. (2001) Environmental correlates of toad abundance and population genetic diversity. Biological Conservation, 98, 201210.
  • Sinsch, U. (1988) Seasonal changes in the migratory behaviour of the toad Bufo bufo: direction and magnitude of movements. Oecologia, 76, 390398.
  • Sinsch, U.. (1989). Migratory behaviour of the common toad (Bufo bufo) and the natterjack toad (Bufo calamita). Amphibians and Roads (ed. T.E.S.Langton), pp. 113125. ACO Polymer Products, London.
  • Sinsch, U., Leskovar, C., Drobig, A., Konig, A. & Grosse, W.R. (2007) Life-history traits in green toad (Bufo viridis) populations: indicators of habitat quality. Canadian Journal of Zoology, 85, 665673.
  • Smith, M.A. & Green, D.M. (2005) Dispersal and the metapopulation paradigm in amphibian ecology and conservation: are all amphibian populations metapopulations? Ecography, 28, 110128.
  • Stevens, V.M., Polus, E., Wesselingh, R.A., Schtickzelle, N. & Baguette, M. (2004) Quantifying functional connectivity: experimental evidence for patch-specific resistance in the Natterjack toad (Bufo calamita). Landscape Ecology, 19, 829842.
  • Stevens, V.M., Leboulenge, E., Wesselingh, R.A. & Baguette, M. (2006) Quantifying functional connectivity: experimental assessment of boundary permeability for the natterjack toad (Bufo calamita). Oecologia, 150, 161171.
  • Verbeylen, G., Bruyn, L.D., Adriaensen, F. & Mathyssen, E. (2003) Does matrix resistance influence Red squirrel (Sciurus vulgaris L. 1758) distribution in an urban landscape? Landscape Ecology, 18, 791805.
  • Wiens, J.A. (2001) The landscape context of dispersal. Dispersal (eds J.Clobert, E.Danchin, A.A.Dhondt & J.D.Nichols), pp. 96109. Oxford University Press, Oxford.
  • Xu, Q. & Oldham, R.S. (1997) Lethal and sublethal effects of nitrogen fertilizer ammonium nitrate on common toad (Bufo bufo) tadpoles. Archives of Environmental Contamination and Toxicology, 32, 298303.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Fig. S1. PCA on landscape elements composition.

Fig. S2. Illustration of the potential movement zones.

Table S1.Fragmentation indexes of landscape.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
JPE_1665_sm_FigS1.doc105KSupporting info item
JPE_1665_sm_FigS2.doc56KSupporting info item
JPE_1665_sm_TableS1.doc31KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.