Environmental Monitoring and Assessment (2005) 109: 199–225 DOI: 10.1007/s10661-005-6266-1
c Springer 2005
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION AND ITS CONSEQUENCES ON LONG-TERM PERSISTENCE OF ANIMAL POPULATIONS J. P. AURAMBOUT1,∗ , A. G. ENDRESS1 and B. M. DEAL2 1
Department of Natural Resources and Environmental Sciences, University of Illinois, 1101 West Peabody Drive, Urbana, Illinois, U.S.A.; 2 Department of Urban and Regional Planning, University of Illinois, 611 East Lorado Taft Drive, Champaign, Illinois, U.S.A. (∗ author for correspondence, e-mail:
[email protected])
(Received 16 February 2004; accepted 12 November 2004)
Abstract. The increasing use of the landscape by humans has led to important diminutions of natural surfaces. The remaining patches of wild habitat are small and isolated from each other among a matrix of inhospitable land-uses. This habitat fragmentation, by disabling population movements and stopping their spread to new habitats, is a major threat to the survival of numerous plant and animal species. We developed a general model, adaptable for specific species, capable of identifying suitable habitat patches within fragmented landscapes and investigating the capacity of populations to move between these patches. This approach combines GIS analysis of a landscape, with spatial dynamic modeling. Suitable habitat is identified using a threshold area to perimeter ratio. Potential movement pathways of species between habitat patches are modeled using a cellular automaton. Habitat connectivity is estimated by overlaying habitat patches with movement pathways. The maximum potential population is calculated within and between connected habitat patches and potential risk of inbreeding within meta-populations is considered. The model was tested on a sample map and applied to scenario maps of predicted land-use change in the Peoria Tri-county region (IL). It (1) showed area of natural area alone was insufficient to estimate the consequences on animal populations; (2) underscored the necessity to use approaches investigating the effect of land-use change spatially through the landscape and the importance of considering species-specific life history characteristics; and (3) highlighted the model’s potential utility as an indicator of species likelihood to be affected negatively by land-use scenarios and therefore requiring detailed investigation. Keywords: animal dispersal, connectivity, environmental planning, habitat fragmentation, LEAM, population management, spatial dynamic modeling
1. Introduction Natural habitat is significantly decreasing with the increasing proportion of the landscape used by humans. Conversion of land for agricultural and urban development has turned large continuous unbroken patches of wild habitat into numerous small patches, isolated from each other among a matrix of inhospitable land-uses. Urbanization is considered a major cause of habitat fragmentation (Tigas et al., 2002), recognized as a major threat to biological diversity and considered to be the primary cause of the current species extinction crisis (Wilcox and Murphy,
200
J. P. AURAMBOUT ET AL.
1985; Gibb and Hochul, 2002; Schmiegelow and M¨onkk¨onen, 2002), and therefore threatens the survival of many species worldwide. Numerous research projects are underway to assess landscape changes as a response to human activity, particularly to urbanization. However, most studies focus on “human oriented” interests and very few consider the environmental consequences of the modified landscapes. The purpose of this study is to create a general model, adaptable for specific species, capable of (1) determining the location of habitat patches likely to sustain populations, (2) estimating population size, and (3) assessing the degree of connectivity and potential gene flow between the patches. This method, applied to a changing landscape, indicates changes in species-specific patch connectivity and determines the impact of land-use change on population isolation and therefore on population fragmentations, which could be used as an indicator of habitat fragmentation.
2. Habitat Fragmentation and Its Impact on Species Population The process of habitat fragmentation involves three factors, which have important repercussions on plant and animal species that originally occupied large continuous areas of wild habitat (Schmiegelow and M¨onkk¨onen, 2002; Gehring and Swihart, 2003). First, fragmentation leads to large patches breaking into numerous smaller patches with a net habitat loss. This results in decreased amounts of resources and fewer shelter areas available to wild species, in turn leading to a general reduction in the number of individuals that can be hosted. Second, by opening core areas, fragmentation of continuous habitat patches leads to a dramatic increase in edges (Sih et al., 2000). Edges provide distinct microclimatic conditions from the core; they may become less suitable for some species. Edges also contribute to higher predation rates by favoring generalist predator influx (Schmiegelow and M¨onkk¨onen, 2002), which in turn greatly impacts the population of resident species. Third, habitat fragmentation results in the geographic isolation of “habitat islands” among a matrix of urban or agricultural land-uses. The mobility of some organisms might thus be restricted (Andreassen et al., 1996) thereby isolating some populations. Small isolated populations can be threatened by inbreeding, which represents a serious problem for their survival, and could lead, in the case of severe inbreeding, to population extinctions (Templeton et al., 1990; Schmitt and Seitz, 2002). Moreover, small populations are more sensitive to stochastic events, such as fires or epidemic outbreaks, which could drive a local population to extinction. As the isolation of habitat patches increases, the probability of their recolonization decreases (Parker and Mac Nally, 2002). Therefore, long-term persistence of isolated populations cannot be assumed. Nonetheless, not all species have the same sensitivity to habitat fragmentation. Naturally rare, sedentary species, with specialized habitat requirements show a significant decline whereas abundant
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
201
mobile generalist species are less affected (Mac Nally et al., 2000) or even favored, in the case of edge specialists (Tscharntke, 1992; Gehring and Swihart, 2003). Also the degree of isolation of habitat patches might depend on the migration capacity of each species living within them. As a consequence, habitat fragmentation cannot be generally described and should be specified for every individual species.
3. The LEAM Tool The Land-Use Evolution and Impact Assessment Model (LEAM) developed at the University of Illinois by Deal (2001) represents a spatial decision support system evaluating human development patterns. It describes land-use changes resulting from spatial and dynamic integration among economic, ecological and social systems in a region (Deal, 2001). This tool utilizes USGS land-use data sets (at a 30 × 30 m resolution), along with census and economical data to produces GIS grid maps of landscape transformation resulting from policy related inputs (Deal, 2001). Our model is a complement, considering the impact of development on habitat fragmentation, to the LEAM tool. It is designed to use land-use change GIS raster maps produced by the LEAM tool to estimate habitat fragmentation and model its impact on animal population persistence in the landscape. However, it is not limited to LEAM outputs and can potentially be applied to any land-use map. Our model incorporates species-specific habitat requirements, dispersal capacity and movement preferences. It is based on the spatial analysis of land-use maps to identify locations of suitable habitats for specific species. It also estimates, from landscape features, the maximum population possibly hosted in each patch and the potential of these populations to exchange individuals. Population estimations are generated purely from landscape considerations. However, population variations are very complex and their dynamic can be influenced by factors others than the landscape in which they live. Consequently, our population estimates should be considered more as population indicators than exact estimates. The inherent limitation to our approach, using land-use to predict species presence in the landscape, is that it disregards other ecological factors such as prey availability or presence of predators or mates, which can influence species habitat selection. However, prey abundance was shown, for some species, to be related to habitat type (Chamberlain et al., 2003). And we believe the location of prey species depending on primary producers can be readily identified from land-use maps. Therefore it might be possible to indirectly account for food availability by successively modeling the habitat of a species’ prey. Nevertheless even the consideration of these two factors might be insufficient to exactly predict species habitat and the obtained results could potentially be improved by the consideration of local inputs such as more detailed land-use maps or population census.
202
J. P. AURAMBOUT ET AL.
4. Methods Our approach is composed of three successive steps that aim at (1) identifying suitable habitat patches within a landscape based on specific species habitat preference and requirements; (2) modeling species dispersal through the landscape, based on its specific movement behavior across different land-uses; and (3) determining connectivity between patches and providing an estimation of the population size capable of gene exchange between connected patches. 4.1. TEST
LANDSCAPE AND ORGANISM
4.1.1. Test Landscape In order to describe our method in a visually explicit way, we applied every step of our approach to a test landscape map. This landscape, shown in Figure 1, was extracted from the 30 × 30 m resolution 1995 National Land Cover Data (NLCD) Illinois land-use map. The test site, located north-west of the city of Peoria (IL) near Kewanee at the intersection of Henry, Bureau and Stark Counties, is a rectangle 601-cells long and 506-cells wide and covers a surface of 27369.54 ha. This site was selected because it illustrates the typical pattern of a fragmented landscape where forest habitat patches are isolated within a matrix of agricultural land-uses.
Figure 1. Test landscape.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
203
In order to account for roads and highways, which can have an effect on species movement patterns, but are not accounted within the NLCD classification, this test map was merged with the grid maps of (a) roads (obtained from the Illinois Department of Natural Resources (Illinois Streets and Highways) at 1:100 000 by County 1994). (b) highways (obtained from the Illinois State Geological Survey (Interstates and Toll Roads, 1996) and Illinois Department of Natural Resources (Illinois State Roads, 1994; U. S. Highways 1994)). 4.1.2. Test Organism Habitat fragmentation does not affect all species in the same way. Our model was designed to estimate habitat requirements for species sensitive to the fragmentation of their habitat. In order to test the approach, we applied it to a hypothetical species for which all dispersal behavior and natural habitat requirements were known. The species was assumed to have a territory identifiable from the raster land-use map (larger than one 30 × 30 m2 grid cell). We defined our focal species as an exclusive forest specialist with limited dispersal, equivalent to a weasel, a squirrel or a small woodland bird. 4.2. STEP 1: IDENTIFICATION
OF SUITABLE HABITAT PARCHES
4.2.1. Criteria of Habitat Suitability In order to identify, from a landscape map, patches of habitat suitable to support a population of a considered species, a selection criterion should be defined. Considerations of patch area alone may be insufficient to estimate habitat suitability for a species, since patches of equal area but different shape may present a different proportion of edge and therefore may not be equally able to support a given animal population (Helzer and Jelinski, 1999). An estimation of the amount of core habitat per patch would represent a good estimator of habitat suitability for edge-sensitive species. However, Schumaker (1996); Helzer and Jelinski (1999) reported that the distance through which edge disturbance infiltrates core habitat varied widely (between 8 and 240 m) and also that the edge effect varied between geographical regions. Helzer and Jelinski (1999) emphasized the need for a relative measure of core, such as perimeter to area ratio, also shown by Schumaker (1996) to be correlated with habitat quality, to account for the amount of patch area exposed to edges without requiring a subjective estimation of edge depth. The perimeter to area ratio, reflecting the area and shape of a patch, was later shown to be, for grassland birds, a good predictor of both individual species presence and overall species richness (Helzer and Jelinski, 1999). The area of a patch is usually much larger than its perimeter. To avoid working with very small numbers, we therefore decided to use the area to perimeter ratio (APr) (the inverse of the perimeter to area ratio) as a selection criterion to identify patches of suitable habitat.
204
J. P. AURAMBOUT ET AL.
The threshold ratio value above which a habitat patch was considered suitable was defined as the APr of the smallest perfectly circular shape able to sustain a population of at least one individual. In practice, both minimum area and APr could be estimated, for a particular species, by using the method defined by Helzer and Jelinski (1999) who calculated in a logistic regression model the correlation of area and perimeter to area ratio with species occurrence and used an incidence value of 50% to define minimum requirement values. The use of the APr as a criterion of habitat selection also offers a major computational advantage over the direct identification of core habitat. This approach does not require extensive calculations and allows suitable habitat to be readily identified across large landscapes (several counties). 4.2.2. Habitat Selection The selection, from a land-use grid map, of suitable habitat patches, using the APr criteria was performed through a succession of GIS calculations using ESRI ArcView GIS 3.3. The first step identified land-uses to be included in the species territory, from the land-use classification. In our test map, forested land-uses served as focal habitat structure, while others served as the matrix within which forested patches were dispersed. The corresponding grid cells were then selected and converted into a Shapefile. This transformation allowed the removal of boundaries between land-use cells of the same type and creation of habitat patches: a collection of 30×30 m cells occupied by forest that touched at either their sides or corners. For each obtained patch, the APr was calculated and patches presenting a ratio ≥ than the defined threshold value were converted to a shapefile and identified as suitable habitat. We chose to convert the selected features from the land-use grid into a shapefile (which smoothes the grid structure and provides a more realistic patch shape) rather than a coverage (which retains the grid-like structure) because the shapefile can account for perfectly circular shapes. This method was consistent with our APr criterion being defined relative to the smallest perfectly circular shape able to sustain a population of at least one individual. In order for this approach to be used with a coverage (retaining the raster structure of the grid land-use map), the area to perimeter threshold ratio would need to be redefined (a perfect circular shape cannot exist in raster format) based on the cell resolution of the land-use grid (since the perimeter of a patch is modified by the rasterization process). 4.2.3. Identification of Core and Edge Habitat The capacity of particular species to survive and successfully reproduce in a natural habitat patch varies according to their susceptibility to edge disturbance. For example, edge-sensitive species frequently exhibit elevated death rates within edges while for other species edges represent essential feeding grounds. Knowing the amount of core and edge constituting a habitat patch might therefore be essential in the estimation of the species population size it can host.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
205
Consequently, even though the estimation of edge depth is susceptible to influence by the type of species considered or the location of the geographical region investigated, edge identification could provide, at local scale, essential information on the viability of populations within the suitable habitat patches. Therefore, we adapted our approach to account for the amount of core and edge habitat available in each of the selected habitat patches, based on a user-defined edge depth. However, this calculation of core and edges was performed only on habitat patches that fit the APr criterion and was not used to determine the suitability of habitat patches. To calculate the amount of core and edge present in the suitable habitat patches, several GIS transformations and calculations were performed. First, the shapefile of suitable habitat patches was converted, in ArcView 3.3, into a grid (using a cell resolution of 30×30 m). The cell values of this grid were reclassified to a value of 1 to create the “patch grid.” The “patch grid” was then converted to coverage format in ArcInfo. The amount of core in each patch was identified by buffering the patch coverages by a negative distance equal to the edge depth. Because of an “error” generated by the ArcInfo-buffering process, some “fake” core areas generated in gaps within the suitable habitat were later removed from the coverage file. The obtained coverage, corresponding to the core habitat present in the suitable habitat patches was then converted to a grid (using a 30 × 30 m cell resolution) whose cell values were reclassified to a value of 2. Finally, this grid was overlaid, in ArcView 3.3, with the “patch grid.” The final “core-edge” grid displayed the proportions of core (with a cell value of 2) and edge habitat (with a cell value of 1) present in each suitable habitat patch. 4.2.4. Model Behavior We applied our approach to the test landscape in order to identify potential flows or drawbacks as well as observe its response to different values of habitat selection criteria and edge depths. 4.2.4.1. Suitable Habitat Patch Selection. The suitable habitat identification method was applied to the test landscape for three different APr thresholds corresponding to the ratio of perfectly circular shapes of 1 ha (Figure 2a), 2.5 ha (Figure 2b) and 5 ha (Figure 2c). We observe, by comparing Figure 2a–c, a decrease in the amount of patches identified as suitable habitat as the value of the threshold increases; larger more clumped patches are also progressively selected. We also observe that every patch identified as suitable habitat using the APr of a circular shape of 5 ha was also identified in selections using the ratio of smaller circular shapes. Therefore the model’s behavior conforms to our expectations. We summarized in Table I the total area, number of patches and average patch area for patches identified with an APr: (1) less than the ratio of a 1-ha circle, (2) between the ratio of a 1-ha and 2.5-ha circle, (3) between the ratio of a 2.5 and a 5-ha circle and (4) more than the ratio of 5-ha circular shape. A majority of the habitat patches have an APr smaller than the one of a 1-ha circle. Therefore such APr habitat
206
J. P. AURAMBOUT ET AL.
(a)
(b) Figure 2. Identification of suitable forest habitat patches presenting APr ≥ than the ratio of perfectly circular habitat patches of (a) 1 ha, (b) 2.5 ha and (c) 5 ha. Forest patches identified as suitable for the species appear in black while non-suitable forest patches appear in dark grey. Non-forested land-uses appear light grey and roads appear in white. (Continued on next page)
207
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
(c) Figure 2. (Continued)
requirements strongly influence the capacity of a species to occupy the landscape. The average areas of patches with an APr fitting the APr requirements of 1–2.5 ha circles and 2.5–5 ha circles were much larger than those of the corresponding perfectly circular shapes. This pattern is similar to the one observed by Helzer and Jelinski (1999) and reflects a high degree of irregularity in the shape of our habitat patches. However in Figure 2b, some patches in the north-eastern corner of the map contain clumped parts large enough to include suitable habitat patches, but are not selected as suitable because they are connected to narrow strings of habitat that decrease their overall APr. This reveals an important drawback of this approach: if a clumped patch of habitat is connected to very fragmented strips of habitat, the TABLE I Number of patches, total area and average patch size per class of suitable habitat patch area (≤1, 1–2.5, 2.5–5 and ≥5 ha)
Number of patches Area occupied Average patch size
≤1 ha
1–2.5 ha
2.5–5 ha
≥5 ha
2263 758.25 0.33
63 705.51 11.20
17 432.72 25.45
11 581.94 52.90
208
J. P. AURAMBOUT ET AL.
APr of the entire patch will be greatly decreased and the patch will be rejected from the selection. This problem might be overcome, in a very scattered landscape, by manipulating the land-use map and manually “cutting” the connection of large patches with the corridors and selecting only the clumped part of the habitat. 4.2.4.2. Core and Edge Habitat Estimation. The ability of our method to identify core and edge habitat was examined on the test landscape for patches of suitable habitat having an APr ≥ to the ratio of a 2.5 ha perfectly circular shape. Four levels of edge depth were investigated: 0 m (Figure 3a), 60 m (Figure 3b), 120 m (Figure 3c), and 240 m (Figure 3d) (the maximum edge depth reported by Schumaker (1996)). We observe from Figure 3a through Figure 3d that as the edge depth increases, the amount of core identified quickly decreases to be almost inexistent in Figure 3d, which shows most suitable habitat patches as small and narrow. The narrow, unclumped structure of our test landscape is further emphasized by Figure 4, which shows dramatic successive drops in core habitat (60, 88 and 99% loss) as edge depth increases from 0 to 240 m. This procedure therefore allows users to identify detailed landscape features within suitable habitat patches and could be (1) applied to identify the location of suitable habitat for core or edge specialist species; (2) used to determine potential corridors (low quality habitat) versus higher quality habitat; or (3) applied to land-use change scenario maps to compare the consequences of land-use decisions on the landscape structure. The fragmentation of habitat impedes species movement and can prevent metapopulations from exchanging individuals. However, determining the location, size and structure of suitable habitat patches provide no information concerning the connectedness of the landscape. Therefore in order to estimate the impact on species of habitat fragmentation, connectivity between patches needs to be considered. 4.3. STEP 2: M ODEL
SPECIES DISPERSAL IN THE LANDSCAPE
4.3.1. Creation of the “Habitat” Map In order to model species movement through the landscape, origin points are required, from which and where to moving species will disperse. These locations were identified in the “core-edge” grid. In order to account for the land-use matrix surrounding the suitable habitat patches, we merged (in ArcView 3.3) the “coreedge” grid on top of the initial land-use grid map of the considered landscape, creating a “habitat” grid map. This habitat map, displaying both the fraction of core and edge of suitable habitat patches as well as the different land-uses surrounding them, was used in our modeling of habitat connectivity. 4.3.2. Conceptual Model Schumaker (1996) suggested habitat connectivity and animal dispersal should be considered using factors influenced by animal natural history, in preference to mathematical formulas. In order to integrate these concerns in our approach, we
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
209
(a)
(b) Figure 3. Identification of core and edge habitat within suitable forest patches presenting an APr ≥ than the ratio of a 2.5 ha perfect circular shape, for edge depths of (a) 0 m, (b) 60 m, (c) 120 m and (d) 240 m. Core habitat appears in black, edge habitat in dark grey, other habitats in light grey and roads in white. (Continued on next page)
210
J. P. AURAMBOUT ET AL.
(c)
(d) Figure 3. (Continued)
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
211
Figure 4. Variations in total areas of core and edge habitat, for increasing edge depths (0, 60, 120 and 240 m), within suitable forest patches presenting an area APr ≥ than the ratio of a 2.5 ha perfect circular shape.
developed a cellular automaton to model the complex species dispersal movements over the “habitat” grid, by incorporating species-specific movement patterns in response to land-uses. This modeling approach is similar to the one described by Deal et al. (2000) and Rafferty et al. (in review). The state associated to each grid cell was identified as one of the variables within a dynamic dispersal model created under STELLA 7.0.1. The incorporation of the STELLA model equations within each grid cell, and the calculation of simultaneous local interactions between each of the model variables, is done through the use of the Spatial Modeling Environment (SME) (Maxwell et al., 2002). This method, defined in more detail elsewhere (Deal et al., 2000; Rafferty et al., in review), allows each grid cell to interact with its eight immediate neighbors. 4.3.3. Dispersal Model The dispersal of the species outside of suitable habitat patches was simulated in each land-use grid cell by a dispersal model created in STELLA 7.0.1. The first task performed by the dispersal model is a reclassification of the values of the “habitat” grid map into three categories: (1) suitable habitat, (2) barrier habitat and (3) crossable habitat. (1) Suitable habitat cells represent origin cells out of which the species can disperse. In our model, these cells correspond to the core and edge cells identified in the “habitat” map.
212
J. P. AURAMBOUT ET AL.
(2) Barrier habitat cells represent an insurmountable limit to the species movement and act as a barrier to dispersal. (3) Crossable habitat cells represent habitats that can be readily crossed by the species and constitute a virtual road for the dispersing species. The reclassification of the land-use values is based on a user defined “landscape diffusion coefficient” look-up table, where the user identifies each land-use as suitable, crossable or barrier habitat for the considered species (based on behavioral determinants (obtained from scientific literature on species movement and dispersal or from the observation of experts) of whether the species is likely to cross a certain land-use or avoid it). Our approach models the dispersal process by attributing to each grid cell, and for every time step of the model run, a state of either “reached” (by dispersing individuals) or “empty.” At the initial time step, the state of the cells identified as suitable habitat patches is set to “reached” while the state of every other cell is set to “empty.” Once the state of a cell is set to “reached,” it remains unchanged. During each of the subsequent time steps, every cell of the land-use grid evaluates the state of its eight neighbors to determine whether they are “reached” or “empty.” If a cell is identified as “crossable” and has at least one neighbor with a “reached” state, its state changes to “reached” at the next time step. However, if the cell is identified as “barrier” its state remains empty. This process, simulating the progressive dispersal of species away from their suitable habitat patches into the landscape matrix is repeated until the end of the simulation. For the last iteration of the simulation, the model produced a text file, later imported into a grid file, using ArcView3.3 to obtain a “migration map.” This migration map identified all possible cells of the “habitat map” that can be reached by individuals dispersing out of the suitable habitat patches. The distance reached by the dispersing individuals increases by one grid cell per time step. Therefore the total number of iterations of the simulation should be determined as a function of the dispersal ability of the considered species. In our model, we fixed the final simulation time step as being equal to the average “straight line” dispersal distance of the considered species divided by grid cell resolution. This ensured the dispersal pattern covered any reachable cell located within the dispersal range of the species. In a single simulation, the selective dispersal approach allowed us to investigate all potential paths able to be used by the species while dispersing outside of suitable habitat cells. At the end of the simulation, the impact on the dispersal pattern of species, of each specific land-use type surrounding suitable habitat, could be evaluated. Different species have different dispersal capacities to move through nonfavorable habitat. In the dispersal model, the distance migrated away from suitable habitat is proportional to the number of time steps in the simulation. It can be easily modified to fit the dispersal capacity of the investigated species by increasing or
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
213
decreasing the number of time steps per simulation. By definition, each “crossable” cell could be crossed with a probability of one. In reality, migration across a matrix of heterogeneous land-uses could take place via different paths as species may prefer to traverse certain land-uses more than others. Also land-uses might affect the death rates of individuals with resultant significant impacts on the probability of the focal species reaching particular locations. Finally the species movement pattern and dispersal distance might be influenced by the presence in the landscape of stepping stones. Our approach, using a cellular automaton, could be modified to account for these more complex movement behaviors by increasing the complexity of the STELLA model utilized. However, these considerations are outside of the scope of the current model. The current objective of the model is to investigate the potential movement of populations and therefore genes between patches, rather than the actual probability of migration success of the species. This approach seeks to account for the hypothesis that even limited exchanges of individuals between metapopulations increase their genetic diversity and decrease their chance of inbreeding depression (Simberloff and Cox, 1987; Templeton et al., 1990). Therefore a population of the focal species was assumed to be able to send genes as far as the diffusion model allowed it to reach.
4.4. S TEP 3: ESTIMATION
OF CONNECTIVITY BETWEEN SUITABLE
HABITAT PATCHES
Suitable patches of habitat were considered connected if they could exchange individuals (i.e. if individuals from one suitable habitat patch could disperse into another). In order to assess habitat connectivity between the suitable habitat patches identified from the “habitat map,” we performed, in ArcView3.3, a series of GIS calculations. In the first step, the “migration map” was converted from grid format to shapefile. This process allowed each separate patch of the “migration map” to receive a distinct identification number. Each of these patches represents a dispersal halo encompassing all locations reachable by dispersing individuals. Therefore if two suitable habitat patches are located within the same dispersal halo, they are likely able to exchange populations and form a “genetically connected entity.” In the second step, the “migration map” shapefile was converted to grid format and the value associated with each grid cell was chosen as the shapefile patch identification number. The obtained grid was then multiplied with the “patch grid.” The obtained grid map was converted to a shapefile to create the “connected map.” These steps were designed to associate the identification number of the dispersal halo within each patch of suitable habitat. Since patches that could exchange individuals had the same identification number, the “connected map” was finally used to assess habitat connectivity between patches.
214
J. P. AURAMBOUT ET AL.
TABLE II Landscape diffusion coefficients, obtained through reclassification of the NLCD land-use values, into crossable and barrier cells, based on the movement behavior of the focal species Crossable land-uses
Barrier land-uses
24: Roads 31: Bare rock/sand/clay 33: Transitional 41: Deciduous forest 42: Evergreen forest 43: Mixed forest 51: Shrubland 71: Grassland/herbaceous 81: Pasture/hay 83: Small grain 84: Fallow 91: Woody wetland 92: Emergent herbaceous wetland
11: Open water 12: Perennial ice and snow 21: Low intensity residential 22: High intensity residential 23: Commercial 25: Large roads such as state and national highways 32: Quarries, strip mines 61: Orchards 82: Row crops 85: Urban, recreational grasses
Note. The numbers correspond to the NLCD numeric code for each land-use.
4.5. TESTING
OF THE DISPERSAL MODEL BEHAVIOR
The portions of the model investigating species dispersal and estimating habitat connectivity were applied to the test landscape, for a species that only moved through “crossable” land-uses as identified in Table II. Suitable habitat patches were selected as having an APr ≥ than the ratio of a perfectly circular shape of 2.5 ha. Four dispersal migration distances were considered: 0 m (Figure 5a), 600 m (Figure 5b), 1200 m (Figure 5c) and 1800 m (Figure 5d). As the species were allowed to migrate over longer distances, more suitable habitat patches were observed to connect. The number of isolated patches, representing “genetic islands,” in which no outside immigration might occur, decreased from 28 in Figure 5a to 9 in Figure 5c. However, the increase in dispersal distance from 1200 (Figure 5c) to 1800 m (Figure 5d) did not lead to a decrease in the number of isolated patches. These results indicate how scattered the suitable habitat patches are among the unsuitable land-use matrix of the test landscape. To test the model’s response to different “landscape diffusion” inputs, we reran the model for identical settings, but assumed the land-use corresponding to row crops, previously considered a “barrier” (Table II), became “crossable.” The resulting map, based on a 1800 m dispersal, is displayed in Figure 6. We observe strong differences in migration patterns between Figure 6 and Figure 5d. These can be explained by the land-use composition of the landscape.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
215
(a)
(b) Figure 5. Estimation of patch connectivity for species migrating distances of (a) 0 m, (b) 600 m, (c) 1200 m and (d) 1800 m through non-forested habitat. Suitable forest patches present an APr ≥ to the ratio of 2.5 ha circular shape and species movements follow Table II landscape diffusion coefficient. Suitable forest patches appear in black, non-forest crossable habitat within reach of the species appears in dark grey, non-forested land-uses not reachable appear in light grey and roads appear in white. (Continued on next page)
216
J. P. AURAMBOUT ET AL.
(c)
(d) Figure 5. (Continued)
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
217
Figure 6. Estimation of patch connectivity for species migrating distances of 1800 m through nonforested habitat. Suitable forest patches present an APr ≥ than the ratio of 2.5 ha circular shape and species movements follow Table II landscape diffusion coefficient with row crops as crossable. Suitable forest patches appear in black, non-forest crossable habitat within reach of the species appears in dark grey, non-forested land-uses not reachable appear in light grey and roads appear in white.
We also observe that the modification in the landscape diffusion input led to a decrease in the number of isolated patches. Therefore, migration patterns, and potential gene exchange between patches, appear highly dependent on the landscape pattern and “landscape diffusion” input. This indicates the specificity of the model and underscores the importance of individual species response to landscape features in habitat fragmentation issues. 4.6. ESTIMATION
OF MAXIMUM CARRYING CAPACITY AND MAXIMUM CONNECTED GENE POOL
The selection of suitable patches using the habitat area to perimeter ratio allowed us to choose patches that can support populations of the focal species (at least one individual), but provided no information on the total population that could be hosted in each patch. We used a set of hypotheses to extrapolate the carrying capacity of potential species and therefore populations from the area of suitable habitat patches.
218
J. P. AURAMBOUT ET AL.
First, we assumed the home range of the focal species encompassed a single patch of habitat and only habitat patches larger than the minimal home range area could support populations. Second the considered species was assumed to be very territorial and only one individual was considered to live in each home range (but the model could be adapted for less territorial species). Third, since individuals of a species might not always establish in the optimal location within a suitable patch, but can be pushed out to less suitable locations by more dominant individuals, we assumed that suitable habitat patches could be fully occupied by the considered species. From these hypotheses, we estimated the maximum carrying capacity of suitable habitat patches by dividing their surface area by the average home range of the considered species. Since the proportion of core and edge can influence population size and varies within habitat patches, however, the resulting values for each patch are overestimates of the real population that could be supported. Isolated populations can be at risk of extinction through the process of inbreeding if the number of individuals they contain becomes too low. In order to account for risks of inbreeding, we considered the extent of genetic pools present within the landscape’s genetically connected entities. To estimate, from the “connected map,” the amount of suitable area available within each “genetically connected entity,” we used the dissolve function in ArcView 3.3. Suitable habitat patches presenting the same identification number were fused in one category and given a value corresponding to their summed-area. This allowed us to associate to each suitable habitat patch the value corresponding to the area of genetically connected patches to which they belong. The size of genetic pools was then determined by dividing the summed area of genetically connected patches by the species average home range. 5. Application: Estimation of Fragmentation Impacts as a Result of Land-Use Change Policies We applied the above described approach to two land-use maps, representing the same landscape at two successive stages of development. Through this application, we evaluated the repercussions of urban development on (1) the amount and spatial arrangement of suitable habitat patches; (2) the suitability of the connected patches to host populations of a species. We also investigated (3) the relationship between habitat destruction and the loss of suitable habitat. 5.1. STUDY
AREA
The investigated landscape, displayed in Figure 7, represents a group of three counties, Peoria, Tazewell, and Woodford surrounding the city of Peoria, IL (U.S.A). The land-use composition of this landscape was considered for two different time periods, 1995 and 2025.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
219
Figure 7. Landscape map of the Peoria Tri-county region.
The first land-use map, corresponding to year 1995, was extracted from the 30 × 30 m resolution 1995 NLCD Illinois grid map, using the boundaries of the three considered counties. To account for roads and highways, this map was merged with the grid maps of Illinois roads and highways (as described in 4.1.1). The second land-use map, provided by the LEAM team, was obtained from a 30 yr simulation run of the LEAM tool (Deal, 2001) configured to account for urban development within Facility Planning Areas (FPAs), resulting from a high economic development scenario of the Peoria region. Both land-use grid maps were composed of 5268747 grid cells, covering a surface of 474187 ha. Contrary to previous work on habitat fragmentation using artificially generated landscapes (Schumaker, 1996), our focus site was obtained from a real landscape and contained complex spatial arrangements of various land-uses, not produced by a mathematical algorithm. The use of the LEAM tool provided an advantage over other methods simulating natural habitat loss in that it uses socioeconomic drivers to determine land-use change and urbanization and did not consider a random pattern of habitat destruction (performed in most models). It therefore provides a more accurate estimation of potential habitat loss triggered by landscape changes.
220
5.2. FOCUS
J. P. AURAMBOUT ET AL.
SPECIES
We investigated the consequences of habitat loss resulting from urban development on a fictitious species assumed to be initially present in the study area. The fictitious focal species was assumed to be a territorial, exclusive forest specialist very sensitive to fragmentation. The equivalent of a small bird or mammal, this species was assumed unable to fly long distances from one habitat patch to the other, and is thus required to physically cross land-use cells in order to disperse through the landscape. The species was assumed to require a 5-ha home range and occupy suitable habitat patches presenting an APr ≥ than the ratio of a 5 ha perfectly circular shape. This is equivalent to the habitat requirements calculated by Helzer and Jelinski (1999) for Western Meadowlarks. The species was assumed to move freely inside suitable habitat patches and to disperse through unsuitable land-uses by moving exclusively through “crossable” habitats defined in Table II (but assuming row crops crossable). Assumed behavior of the species included their (1) avoidance of “human occupied” landscapes, (2) inability to swim across rivers, and (3) average dispersal through unsuitable “crossable” habitat was limited to 1200 m. Gene flow was assumed to occur between connected patches of suitable habitat and a population size of 500 individuals was assumed sufficient for maintaining genetic variation and long-term persistence (Tscharntke, 1992). Our approach considered habitat preference and movement characteristics of a fictitious species, but could be parameterized for potentially any terrestrial organism unable to fly over long-distances. 5.3. R ESULTS
AND DISCUSSION
We observe in Figure 8 that urban development as predicted by the LEAM tool leads to a decrease of 19.5% in the surface areas of habitat suitable for our focal species, and by 2025 the landscape has 5894 ha less of suitable habitat than in 1995. The loss of suitable habitat was not distributed homogeneously in the landscape, but occurred mainly in the center and north part of the map, which is consistent with an increase in urbanization around the city of Peoria. Habitat loss resulted in the shrinking of initially large suitable habitat patches as well as in the disappearance of numerous smaller patches. We observe the areas of forest lost to development between 1995 and 2025 (7687 ha) is larger than the actual areas of suitable habitat lost (5894 ha). However, most of the developed forest cells appear to be outside of suitable habitat patches and the areas converted from forest to development within suitable habitat patches is much smaller (1835 ha). Consequently there appear to be a non-equal relation between the areas actually lost to development and the amount of suitable habitat lost. To illustrate this point, we focused our analysis on a smaller site located at the center of the study area. Figure 9 shows the summary map of the changes in habitat suitability triggered by development for this site. The development occurring within initially suitable habitat (habitat lost) is highly scattered within the patches.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
221
Figure 8. Summary map of changes in suitable habitat (presenting an APr ≥ than the ratio of a 5 ha perfectly circular shape) between 1995 and 2025 land-use maps for a species migrating 1200 m based on Table II landscape diffusion coefficient with row crops as crossable. Remaining suitable habitat appears in black, lost suitable habitat appears in dark grey, non-forested land-uses appear in light grey and unsuitable forest patches appear in white.
The consequences of this development pattern are that although only 35% of the initially suitable habitat is destroyed by development (256.59 ha out of 738 ha), 96% of the original habitat becomes unsuitable for the focus species which can persist only in 4% (30.6 ha) of its original range. Consequently the location of developing habitat cells relative to suitable habitat patches appears to have a strong influence on the population they can host. Focusing only on the amount of available habitat, while ignoring its spatial arrangement and distribution, may lead to important overestimations of suitable habitat areas. Investigating the effect of land-use change spatially and quantitatively, across the landscape, together with inclusion of life history characteristics for the species are therefore essential to determine species habitat usage. However, considerations of the amount and spatial arrangement of habitat within suitable patches is not sufficient to provide indications of the likelihood of the considered species to persist in them over a long period of time and potential risks of inbreeding depression should be considered.
222
J. P. AURAMBOUT ET AL.
Figure 9. Focused location in the summary map.
When accounting for habitat connectivity between patches and considering the capacity of a “genetically connected entity” to host large number of individuals of the species (Figure 10), we observed in both cases that fewer than half of the individuals potentially present in the landscape (41% in the 1995 map and 47% in
Figure 10. Total number of individuals of the species that can be hosted per class of genetically connected entities (can host 500 individuals) among forest habitat patches with an APr ≥ than the ratio of a 5 ha circular shape for Peoria tri-county 1995 and 2025 land-use maps. The species migration distance is 1200 m and follows Table II landscape diffusion coefficient with row crops as crossable.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
223
the 2025 map) belonged to a “genetically connected entity” large enough to host more than 500 individuals and therefore prevent inbreeding depression (Tscharntke, 1992). Thus the majority of populations potentially present in the landscape is isolated among groups of patches too small to provide a sufficient genetic pool. In the long term they may face moderate (500–250 individuals per genetically connected entity) to high risks (0–250) of extinction. In this context, urban development occurring between 1995 and 2025 at this site seems to un-evenly affect population groups since meta-populations at moderate risk of inbreeding (hosting between 250 and 500 individuals) suffer more habitat loss than other groups. However, this disproportionate loss of habitat between “genetically connected entities” at moderate and high risk of inbreeding is only apparent in Figure 10 since habitat patches with moderate risk of inbreeding turn into patches with high inbreeding risk when they lose part of their habitat, and therefore increases the number of individuals present in low quality habitat. Nevertheless, the predicted urban development decreases habitat availability within genetically connected entities that could host populations with low or moderate risks of long term inbreeding, while some patches hosting populations at high risk of inbreeding remain untouched (Figure 8). The predicted development pattern for 2025 therefore appears not to be the most suited to optimize long-term persistence of our species in the landscape. In a context of long-term planning for species survival, it might be more appropriate to implement development policies that favor clumped urbanization, within patches of genetically connected entity hosting small populations and which future is already impaired, instead of within patches that could host populations on the long term. The above results and comments are likely to be applicable to other fragmentation-sensitive species. Therefore, the described approach, applied to different species could provide data that would help environmental managers consider spatial components of species habitat requirements as well as habitat connectivity. In particular it could be used to test the potential repercussions of urban development (for example, creation of a highway or shopping center) or logging policies on populations of sensitive species. Through the production of impact scenario maps and tables, it could help decision-makers choose policies having the least impact on the focal species. This approach could also be easily adapted for different species by changing the “landscape diffusion” input, habitat requirements and average dispersal distance and thereby indicate a potential ecosystem response to the changes. Our approach could also assist wildlife planners when they consider land acquisition to create corridors and allow population movements between existing reserves. For example, “connected map” displaying “genetically connected entities” could be used to identify neighbor suitable isolated patches that, if connected together, could increase the shared genetic pool and decrease the extinction risk of the populations they sustain. This approach, therefore, could help select optimal patches to be connected and could be used to decide which land acquisitions would facilitate effective population movements between patches.
224
J. P. AURAMBOUT ET AL.
This approach could be complemented by incorporating a spatial dynamic population model for the focal species, which could provide more accurate estimations of population variations within the landscape as well as more detailed dispersal patterns. Additionally hypotheses could be tested, concerning the migration distance of some species or the type of land-uses they are capable of crossing. The model could be run for several sets of hypotheses and field experiments could be conducted to determine which provide the best match to the observed data. Finally this approach could also be applied to species in phases of geographical expansion, such as invasive species, to help determine potential migration routes.
6. Conclusion We developed, as part of the Land use Evolution and impact Assessment Model (LEAM), an approach combining GIS analysis of a landscape with the spatial dynamic modeling capacity of a cellular automaton to (1) identify habitat patches likely to sustain populations of fragmentation-sensitive species within a landscape; (2) model their dispersal patterns and estimate habitat connectivity between patches; and (3) estimate population sizes and their long-term propensity to suffer inbreeding depression. This approach incorporates species-specific movement behavior components and can be adapted to model different species. When applied to a fictitious forest specialist species on maps depicting a changing landscape, our approach showed that (1) urbanization by destroying habitat and impeding species movement could lead to important modifications of species presence in the landscape and increase their risk of inbreeding depression and (2) the spatial pattern of habitat destruction within and among patches has a tremendous influence on the areas of habitat made unsuitable for the species. We suggest this approach could be used as a tool to estimate the long-term consequences of different land-use change scenarios on habitat sensitive species. We also suggest this model, and the results it generates, could be a useful tool to (1) help land-use planners identify patches of high-biological value where development should be minimized; (2) help wildlife planners choose land patches to acquire by testing their capacity to increase habitat connectivity and maximize species populations; (3) test scientific hypotheses concerning species dispersal and habitat usage and (4) estimate the colonization path of invasive species.
Acknowledgments We thank the LEAM group (http://www.leam.uiuc.edu) for providing funding and computer power for this project. We also thank Z. Sun, J. Terstriep, Y. Kim and W. Choi for their help, suggestions, and review.
A SPATIAL MODEL TO ESTIMATE HABITAT FRAGMENTATION
225
References Andreassen, H. P., Halle, S. and Ims, R. A.: 1996, ‘Optimal width of movement corridors for root voles: Not too narrow and not too wide’, J. Appl. Ecol. 33, 63–70. Chamberlain, M. J., Leopold, B. D. and Conner, L. M.: 2003, ‘Space use, movements and habitat selection of adults bobcats (Lynx rufus) in Central Mississippi’, Am. Midl. Nat. 149, 395–405. Deal, B.: 2001, ‘Ecological urban dynamics: The convergence of spatial modelling and sustainability’, Build. Res. Inform. 29, 381–393. Deal, B., Farello, C. A., Lancaster, M., Kompare, T. and Hannon, B.: 2000, ‘A dynamic model of the spatial spread of an infectious disease: The case of fox rabies in Illinois’, Environ. Model. Assess. 5, 47–62. Gehring, T. M. and Swihart, R. K.: 2003, ‘Body size, niche breadth, and ecologically scaled responses to habitat fragmentation: Mammalian predator in agricultural landscape’, Biol. Conserv. 109, 283– 295. Gibb, H. and Hochul, D. F.: 2002, ‘Habitat fragmentation in an urban environment: Large and small fragments support different arthropod assemblages’, Biol. Conserv. 106, 91–100. Helzer, C. and Jelinski, D. E.: 1999, ‘The relative importance of patch area and perimeter-area ratio to grassland breefing birds’, Ecol. Appl. 9, 1448–1458. Mac Nally, R., Bennett, A. F. and Horrocks, G.: 2000, ‘Forecasting the impacts of habitat fragmentation on birds in the box-ironbark forests of central Victoria, Australia’, Biol. Conserv. 95, 7–29. Maxwell, T., Villa, F. and Costanza, R.: 2002, ‘The Spatial Modeling Environment’, International Institute for Ecological Economics, Center for Environmental Science, University of Maryland System, Solomons, MD. Available at http://www.uvm.edu/giee/SME3. Parker, M. and Mac Nally, R.: 2002, ‘Habitat loss and the fragmentation threshold: An experimental evaluation of impacts on richness and total abundances using grassland invertebrates’, Biol. Conserv. 102, 217–229. Rafferty, J. P., Aurambout, J. P. and Deal, B.: in review, ‘Modeling the population response of the raccoon (Procyon lotor) to land use change in North Aurora, Illinois’, Ecol. Model. Schmiegelow, F. K. and M¨onkk¨onen, M.: 2002, ‘Habitat loss and fragmentation dynamic landscapes: Avian perspectives from the boreal forest’, Ecol. Appl. 12, 375–389. Schmitt, T. and Seitz, A.: 2002, ‘Influence of habitat fragmentation on the genetic structure of Polyommatus coridon (Lepidoptera: Lycaenidae): Implications for conservation’, Biol. Conserv. 77, 291–297. Schumaker, N. H.: 1996, ‘Using landscape indices to predict habitat connectivity’, Ecology 77, 1210– 1225. Sih, A., Jonsson, B. G. and Luikart, G.: 2000, ‘Do edge effects occur over large spatial scale?’ Tree 134–135. Simberloff, D. and Cox, J.: 1987, ‘Consequences and costs of conservation corridors’, Conserv. Biol. 1, 63–71. Templeton, A. R., Shaw, K., Routman, E. and Davies, S. K.: 1990, ‘Genetic consequences of habitat fragmentation’, Ann. Miss. Bot. Gard. 77, 13–27. Tigas, L. A., Van Vuren, D. H. and Sauvajot, R. M.: 2002, ‘Behavioral responses of bobcats and coyotes to habitat fragmentation and corridors in an urban environment’, Biol. Conserv. 108, 299–306. Tscharntke, T.: 1992, ‘Fragmentation of Phragmites habitats, minimum viable population size, habitat suitability and local exctinction of moths, midges, flies, aphids and birds’, Conserv. Biol. 6, 530– 536. Wilcox, B. A. and Murphy, D. D.: 1985, ‘Conservation strategy: The effect of fragmentation on extinction’, Am. Nat. 125, 879–887.
