Thus, we asked how climate change would affect fragmentation, net primary productivity, and ecological characteristics (such as species composition) of communities. Since fragmentation of habitats often underlies any fragmentation of mammalian distributions, we first analyzed habitat distributions as a prerequisite to assessing the fragmentation of animal distributions. Net primary productivity is the foundation upon which many other characteristics of an ecosystem rest, and small changes in this variable may have significant consequences for the viability of consumer communities within an area. To understand the way in which communities respond to perturbations such as climatic shifts, it also is necessary to understand the processes that govern their current structure. Therefore, we also have focused on detecting patterns within communities of different species diversity.

Fragmentation was assessed for the current distribution of forest and grassland habitats, those habitats occurring under the hotter/wetter climatic scenario (R30), and the hotter/drier scenario (CCC). In each case, habitats were analyzed separately, after being grouped into broad community types (e.g., forest and grassland), and after being grouped by Holdridge environmental categories (e.g., tropical, subtropical, and temperate). Several indices were used to provide a robust assessment of spatial patterns. For the fractal index, D, 10 grids with cells of increasing sizes were overlaid on the map of habitat types, the number of cells covering the range of the habitat was counted, and D was estimated (for each habitat) as the slope of a linear regression equation in which the size of the box predicted the number of occupied cells. Harmonic means of the distances among patches of each habitat were calculated directly from GIS maps: the distance to the nearest polygon of the same type was calculated for each polygon, and then these data were used to calculate the mean. Abundance of each habitat was assessed by overlaying a single grid on the habitat maps for each scenario and simply counting the number of boxes in which each habitat occurred. Finally, the perimeter/area ratio of each habitat type was computed.

Community composition was analyzed for the 59 non-aquatic species of rodents occurring in forests and grasslands in Texas. The ranges of all rodent species (determined as above) were overlaid and the resulting polygons counted for species number. The taxonomy and life history characteristics of the species occurring in each species diversity polygon were determined, as well as the areas occupied by each. Our primary metric was the proportion and area of communities of a given species diversity occupied by a particular species. For communities of a particular size we had a single number for each species reflecting its occupancy of those areas. Thus, each species formed a row in an occurrence matrix where each column was a unique community size. Occurrence patterns were then analyzed under each individual scenario, and statistical analyses were used to test for changes in the distributions resulting from climatic change. We also tested for the effects that taxonomic group, size category, food type, and substrate preference had on the current distribution of species.

Net primary productivity (NPP) may be estimated from the percent cover in an area. We will employ a commercially available NDVI (normalized differential vegetation index) data set to estimate the cover of the study area. This raster data set will be converted to a vector ARC/INFO coverage compatible with our other data sets at the research center at Indiana University. The resulting cover will then be used to assign current cover indices to each of the weather polygons used in the other analyses. The meteorological information from each weather polygon will be used to develop a predictive equation where NDVI is the outcome variable and the weather variables are independents. The weather data from each climate change scenario can then be employed in this equation to predict cover under various conditions. The resulting coverages will be overlaid with the current coverage to determine a net change in NPP for each weather polygon. Statistical analysis of these areas will allow us to predict the net effects of each climatic scenario on the cover of the region as a whole, as well as the impact it has on geographical and environmental subsets of the area.

(Fragmentation) There were significant increases in the abundance or coverage of tropical habitats from current conditions regardless of the level of precipitation (CCC {drier}: Kruskal-Wallis 2(appr) = 3.99, p <0.05; R30 {wetter}: 2(appr) = 4.40, p < 0.04), and complete losses of warm temperate environments. Thus, regardless of changes in overall abundance of habitats, the models predict substantial changes in vegetative composition. There was also significantly higher variance among the coverages of tropical environments after climatic perturbations (drier conditions: Levine's median test, t = 4.564, p < 0.02; wetter conditions: t = 3.114, p < 0.05), although the variances between the CCC and R30 models did not differ. Compared to current conditions, therefore, we would expect a greater disparity in the sizes of different tropical habitats. However, there were no differences in the mean distributions of forests, despite the movement of individual habitat types (8 of 17 current habitats disappeared in a dry climate, 13 of 17 disappeared under wet conditions). We also compared the total proportion of boxes occupied by each habitat type under each scenario using log-likelihood ratios. Forest cover decreased significantly under the drier CCC climates (G = 9.72, p < 0.005) and increased significantly under the wetter R30 (G = 217.7, p < 0.001). Grassland decreased significantly under both climate change scenarios (CCC: G = 292.9, p < 0.001; R30: G = 895.6, p < 0.001). Thus, there were substantial changes in the abundances of habitats present after climatic changes, but these new types were distributed in essentially the same manner before and after climatic changes.

The degree to which the biomes or environment categories were fragmented (D) after climatic change did not differ from current conditions, nor between the 2xCO2 scenarios (Kruskal-Wallis tests, p > 0.25 in all cases). The variance in fragmentation differed only for forests under CCC, where forest fragmentation was less variant (Levine's test; t = 2.723, p < 0.05). The average isolation (i.e., the harmonic means among polygons within each environment or habitat) was also not significantly different among any pairs of scenarios, nor did the variance in this statistic differ among scenarios.

There were no significant differences in mean fragmentation or isolation across all individual habitat types among the three scenarios (Kruskal-Wallis tests, p > 0.18 in all cases). Likewise, there was also no difference in the magnitude of D between pairs of habitats occurring under both CCC and R30 (sign test, n = 13, p = 0.08). The variance in fragmentation among individual habitats did not differ significantly (Levine's test; CCC: t = 0.43; R30: t = 0.60). Likewise, the average isolation of the habitats under each scenario did not differ, nor did they have any statistically significant change in variance.

D for a particular habitat was significantly affected by the abundance of the habitat (Pearson's product-moment correlation; current: r2 = 0.48, p < 0.05; CCC: r2 = 0.55, p < 0.006; R30: r2 = 0.64, p < 0.004). However, since there were several cases where the overall abundance of a habitat type changed but its fragmentation (D) did not, we feel confident this statistic measured some unique component of the spatial complexity present. Nevertheless, we also calculated fragmentation using the perimeter/area relationship of each habitat. Again, however, there was no significant effect of scenario on the distribution of this measure of fragmentation (Kruskal-Wallis, 2(appr) = 0.68, p > 0.70) or on its variance (Levine's tests, ,t, < 0.50).

(Community Structure) The distribution of sizes of rodent communities was altered after climatic change occurred. Currently, the lowest diversity communities contain 9 species and the most diverse contain 41 species. Under both climatic scenarios the range of community sizes expanded, to low diversity communities with 7 species and high diversity communities with 46 species. The additional high diversity communities were derived primarily from those communities with high diversities under current conditions: the areas occupied by 36-40 species fell from 73.4x103 km2 to 11.4x103 km2 under CCC and to 4.3 x 103 km2 under R30, while communities with 41-46 species occupied 39.1x103 km2 under CCC and 16.2x103 km2 under R30. Under wetter conditions the geographic location of these high diversity communities remained unchanged, while under the dry scenario there was a northward movement. Under current conditions, the areas of greatest diversity occurred in transitional dry--moist forest. Under both climate change scenarios, lower montane grassland-forest transitional habitats were the loci of greatest diversity. There was also a substantial redistribution under R30 into communities containing a relatively low number (15) of species.

Despite these differences, however, the overall distributions were similar. There was a statistically significant similarity between the area of Texas occupied by communities of a particular species richness under each of the different scenarios (Pearson's product-moment correlation; current with CCC: = 0.379, p < 0.05; current with R30: = 0.563, p < 0.01; CCC with R30: = 0.552, p < 0.01). In addition, none of the distributions of community sizes were normally distributed (Kolmogorov D statistic; Current: D = 0.1239, p < 0.01; CCC: D = 0.1711, p < 0.01; R30: D = 0.2642, p < 0.01). Thus, while each scenario was similar, none was structured in a random (normal) pattern. This observation implies that the communities are structured by non-random processes that do not operate equally over the breadth of observed community sizes.

However, the way individual species were distributed among communities of a particular size was similar across different climatic conditions: the proportion of polygons occupied (PO) by the rodent species were significantly correlated with each other (p < 0.05 in all cases). Thus current communities of a particular species richness would be expected to have the same general composition regardless of whether the climatic shifts resulted in a wetter or drier environment. Within each climatic scenario there were generally differences in the composition of communities that had different diversities. For example, there was no correlation between the PO of CCC communities containing 10 and 21 species. There was also no effect of similarity in community size: those communities that had similar numbers of species were no more likely to be correlated than those with vastly different species diversity. Species present in only a single community class did not generally occur in the highest diversity communities. Widely distributed species (those occurring across a wide range of species richness classes) were generally similar species under all conditions.

Occurrence patterns under current conditions were not well-predicted by categorical variables reflecting taxonomic affiliation, body size, food type, or preferred substrate. The only significant predictor (from analysis of variance) was food type, and this variable did not predict patterns across all species diversity intervals. Rather, only communities with intermediate levels of diversity showed a significant effect of food type. In these communities, the differences were likely caused by differences between granivores and omnivores, the only two categories to have significant differences among their means (Scheffe's multiple comparisons test, p < 0.05).

After climatic change, the way species are distributed within communities of a particular species diversity generally remains the same. However, the habitat types in which diversity is greatest differs from current conditions. In both climate change scenarios diversity is greatest in the same habitat type (grassland-forest transitions), implying that the precipitation differences among the models does not directly affect diversity. The basic distribution patterns, however, suggest that internal processes may qovern the composition of individual communities. Indeed, there are few consistencies to discriminate among the species that occupy species-rich or species-poor communities, and patterns among such categories are not well predicted by differences in life-history attributes. We infer that detailed scrutiny of particular species and their possible interactions may help to clarify the factors determining local communitv diversity.

We hypothesize these counterintuitive results may be due to characteristics of Texas that make it a biologically diverse area: steep environmental gradients, both a north/south temperature continuum and an east/west moisture gradient. Under shifting climates, the intersection of these influences allows for a continuous progression to new habitat types, but retains the spatial diversity of the original condition. Thus, climatic change may affect fragmentation of more environmentally homogeneous grasslands and forests to a greater extent than more environmentally diverse areas. Since biological diversity at a regional scale is generally enhanced by environmental diversity, specialized species occurring in low diversity regions may be particularly sensitive to climatic change.