The Effects of Habitat Fragmentation on Bird Species Richness
Habitat fragmentation, or the subdivision of continuous habitat into smaller patches, has three components: direct removal of suitable habitat, reduction in patch size, and increasing isolation of the remaining patches (Andren, 1994). The application of Island Biogeography Theory (MacArthur & Wilson, 1963) to terrestrial “habitat islands” (Diamond, 1976; Quinn & Harrison, 1988) predicts that fragmentation should lead to a reduction in the number of species present. In this view, the composition (type and abundance) of habitat drives biological diversity. However, island biogeography may not be falsifiable and often fails to account for species richness- a point of contention long debated in the literature. Terrestrial landscapes (as opposed to homogenous oceanic islands) can be construed as three spatial elements: Patch, corridor, and matrix (Forman, 1995). One of the tenets of landscape ecology is that the shape and juxtaposition of patches affect ecological processes. Therefore, landscape variables that describe patch configuration may be better predictors of species richness than those that describe patch composition.
Landscape Metrics
The degree to which coarseness of the land cover classification system influences landscape metrics was explored by calculating three sets of landscape metrics: land use/land cover using Anderson, et al’s (1976) Level 1 classification, Anderson, et al’s Level 2 classification, and a binary classification (relatively disturbed/undisturbed). Landscape metrics were calculated at several scales: within 1 km of Breeding Bird Survey (BBS) routes, within 5 km of BBS routes, within the spatial extent of each route, within 1:250,000 watershed boundaries, and within each of the five physiographic provinces of Maryland. Land use/land cover data for the State of Maryland (Maryland Office of Planning, 1998) was input into FRAGSTATS (McGarigal & Marks, 1994) for the generation of landscape metrics.
Figure 1: Maryland Land Use/Land Cover from Satellite Imagery and Parcel Data
Figure 2: 1 km Buffer Around BBS Routes
Figure 3: 5 km Buffer Around BBS Routes
Figure 4: Boxes Around the Spatial Extent of BBS Routes
Figure 5: 1:250,000 Watershed Boundaries
Figure 6: Physiographic Provinces in the State of Maryland
Diversity Measures
Data sets of bird species richness for the State of Maryland were obtained from Patuxent National Wildlife Center’s Breeding Bird Survey (BBS). These BBS data (Droege, 1990) consist of 54 routes (Figure 7) that are 24.5 mi long with point counts at every 0.5 miles.
Figure 7: Maryland Breeding Bird Survey (BBS) Routes
The count consists of every bird seen or heard within a 0.25 mi radius within a 3-minute interval. The start and direction of the route is random, but are constrained to follow existing roads.
Results
Coefficients of determination (r2), or squared correlation coefficients, were used determine the degree to which the variation in landscape metrics explain the variation in avian species richness. No significant correlations were found at the scale of 1 km.
| Table 1: Significant Correlations – All Species within 5 km Buffer |
|---|
| Classification | Fragmentation Metric | P value | R2 |
| Anderson Level 2 | None | n/a | n/a |
| Anderson Level 1 | Total core area | 0.014 | 0.298 |
| Anderson Binary |
Total core area |
0.013 |
0.303 |
| Table 2: Significant Correlations – All Species within Box |
|---|
| Classification |
Fragmentation Metric |
P value |
R2 |
| Anderson Level 2 |
Total edge |
0.008 |
0.328 |
| |
Total core area |
0.003 |
0.364 |
| |
Number of core areas |
0.031 |
0.254 |
| |
Patch richness density |
<0.001 | 0.628 |
| Anderson Level 1 |
Total core area |
<0.001 |
0.431 |
| Anderson Binary |
Total edge |
0.032 |
0.253 |
| |
Total core area |
<<0.001 |
0.434 |
| Table 3: Significant Correlations – All Species within 1:250,000 Watersheds |
|---|
| Classification |
Fragmentation Metric |
P value |
R2 |
| Anderson Level 2 |
Mean patch fractal dimension |
0.012 |
0.502 |
| |
Nearest neighbor coefficient of variation |
0.015 |
0.484 |
| |
Patch richness |
<0.001 |
0.696 |
| |
Patch richness density |
<<0.001 |
0.990 |
| |
Relative patch richness |
<0.001 |
0.696 |
| Anderson Level 1 |
Path density |
<0.001 |
0.688 |
| |
Double-log fractal dimension |
0.003 |
0.597 |
| |
Area-weighted mean patch fractal dimenstion |
0.001 |
0.639 |
| |
Core area density |
<0.001 |
0.678 |
| |
Interspersion-juxtaposition index |
0.041 |
0.397 |
| |
Patch richness density |
<<0.001 |
0.994 |
| Anderson Binary |
Patch density |
<<0.001 |
0.828 |
| |
Double log fractal dimension |
0.020 |
0.462 |
| |
Area-weighted mean patch fractal dimension |
0.002 |
0.601 |
| |
Core area density |
<<0.001 |
0.820 |
| |
Mean nearest neighbor distance |
<<0.001 |
0.761 |
| |
Standard deviation nearest neighbor distance |
<<0.001 |
0.784 |
| |
Coefficient of variation nearest neighbor distance |
0.017 |
0.473 |
| |
Patch richness density |
<<0.001 |
0.999 |
| Table 4: Significant Correlations – All Species within Physiographic Provinces |
|---|
| Classification |
Fragmentation Metric |
P value |
R2 |
| Anderson Level 2 |
Mean patch size |
0.035 |
0.753 |
| |
Edge density |
0.047 |
0.717 |
| |
Double-log fractal dimension |
0.012 |
0.846 |
| |
Mean core area 1 |
0.035 |
0.753 |
| |
Mean core area 2 |
0.034 |
0.758 |
| |
Mean nearest neighbor distance |
0.037 |
0.748 |
| |
Standard deviation nearest neighbor distance |
0.007 |
0.880 |
| |
Coefficient of variation nearest neighbor distance |
0.001 |
0.940 |
| |
Patch richness density |
0.004 |
0.910 |
| Anderson Level 1 |
Patch size coefficient of variation |
0.018 |
0.820 |
| |
Area-weight mean shape index |
0.026 |
0.785 |
| |
Double-log fractal dimension |
0.003 |
0.917 |
| |
Area-weighted mean patch fractal dimension |
0.016 |
0.829 |
| |
Core area coefficient of variation 1 |
0.018 |
0.820 |
| |
Core area coefficient of variation 2 |
0.018 |
0.816 |
| |
Interspersion-juxtaposition index |
0.004 |
0.908 |
| |
Patch richness |
0.025 |
0.790 |
| |
Relative patch richness index |
0.025 |
0.790 |
| Anderson Binary |
Patch density |
0.018 |
0.816 |
| |
Mean patch size |
0.004 |
0.910 |
| |
Double-log fractal dimension |
0.008 |
0.876 |
| |
Area-weighted mean patch fractal dimension |
0.047 |
0.717 |
| |
Core area density |
0.024 |
0.795 |
| |
Mean core area 1 |
0.004 |
0.910 |
| |
Mean core area 2 |
0.004 |
0.906 |
Discussion
The results indicate that habitat fragmentation is a scale-dependent process. As predicted by hierarchy theory (O’Neill, 1988), the number of independent variables significantly correlated with species richness increases with “decreasing” spatial scale (fine to broad). In general, the strength of the correlations also increases at broad scales. At finer scales, edge effects explain more variation in species richness than any other fragmentation metric. This suggests that there are two independent processes occurring relevant to overall diversity: edge shape and edge distance (Figure 8).
Figure 8: Edge Shape and Edge Distance
Species richness is greater when core area and edges are high (Figure 8A) than when core area is large is edge is low (Figure 8B). Similarly, species richness is greater when core area is small and edges are high (Figure 8C) than when core area is small is edge is low (Figure 8D). Figure 8A reflects a possible “better” configuration vis-à-vis avian diversity than Figures 8B, C or D: patches suitable for edge tolerant and forest-interior dwelling species. At finer scales, habitat composition does not explain variation in species richness. At fine scales, it is likely that an organism is present if suitable habitat is present and absent if not; the configuration of the habitat is less important than its simple presence/absence.
Measures of connectivity indicate the importance of metapopulation dynamics at broader scales. A landscape at broad scales is likely a mixture of source and sink subpopulations, corresponding to mixtures of habitat patches of varying suitability. Birds are highly motile organisms, and thus measures of connectivity are likely to be important determinants of bird species richness at broad scales. Isolated patches will likely contain fewer species than spatially contiguous patches of suitable habitat. Patch diversity also explains variations in species richness at broad scales. This is consonant with one interpretation of the application of Island Biogeography to terrestrial habitat “islands”: area as proxy for habitat heterogeneity.
Fragmentation metrics did, however change with various classification schemes. The Anderson “Binary” classification (“Disturbed” vs. “Undisturbed”), was expected be more indicative of how species use their environment than the more disaggregate classification systems. To some extent, the results support this hypothesis. While some metrics differed, particularly between Anderson Level 2 and the other two as a group, the overall conditions indicated by the metrics were essentially the same. Patch richness was an exception, presumably because more patch types were possible using Anderson Level 2 and than Level 1 or Binary. This suggests that while individual species may perceive “patchiness” differently, in the aggregate these differences were smoothed when considering all species combined. When examining all species, however, the Anderson Level 2 classification system discriminated between the heterogeneity in species’ perceptions of patchiness. Therefore, habitat composition is a significant determinant of species richness that may remain undetected unless the classification system is fine enough to discriminate this heterogeneity in perception.
| 
