Volume No. 10, 2001
Animal Modeling
The goal of this project was to evaluate the accuracy of the second-generation GAP predicted distribution models for Idaho amphibians and reptiles at three spatial scales. We believe that such accuracy assessments are needed to guide appropriate use of the GAP models. Our approach consisted of using intensive herpetological field surveys (conducted for other purposes) to test the amphibian and reptile models at three different spatial scales.
The second-generation predicted distribution models for Idaho amphibians and reptiles (Scott et al. 2002) consisted of the following elements:
1. EMAP hexagons indicating the potential ranges of the species (i.e., where the models were applied);
2. maps of frost-free days indicating suitable thermal conditions;
3. suitable cover-type maps; and
4. buffered aquatic and wetland features for species such as stream- and pond-breeding amphibians (e.g., tailed frogs and long-toed salamanders) and riparian reptiles (e.g., garter snakes).
We conducted amphibian and/or reptile surveys in five areas in Idaho (Figure 1). These surveys were conducted for a variety of organizations, including the Bureau of Land Management, Idaho Army National Guard, Idaho Department of Fish and Game, National Park Service, and USDA Forest Service. The study areas ranged in size from approximately 3,600 to 29,000 ha, in elevation from 250 to 2800 m, and included over 500 sampling sites in a wide range of habitats (lava, grasslands, shrublands, forests, riparian, and wetland areas). Sampling durations varied from one to five field seasons. Amphibian surveys consisted primarily of visual encounter surveys supplemented with listening for calling adults and dip-netting for larvae. Reptile surveys consisted primarily of drift-fence/funnel trap arrays supplemented by visual encounter surveys.
Figure 1. Study area locations.
We used field guides (Nussbaum et al. 1983, Stebbins 1985) and information from the Northern Intermountain Herpetological Database (Idaho Museum of Natural History) to generate a liberal list of the potential species for all of the study areas (Table 1). We plotted the survey results for each sampling site on the GAP predicted maps for each species for each study area. We compared the predictions from the GAP maps (one prediction for each potential species for each study area) with the field survey results at three spatial scales: (1) for entire study areas (~3,600 to 29,000 ha); (2) at sections with sampling sites (259 ha = 1 square mile); and (3) at buffered sampling sites (~2 ha). For each sampling scale/area, we then calculated the number of correct positive predictions, the number of correct negative predictions, the number of incorrect positive predictions, the number of incorrect negative predictions, and overall correct and mistaken classification rates. Classification accuracy equaled the number of correct predictions divided by the total number of predictions.
1. Classification accuracy appeared to increase with the size of the sampling area (Figure 2; Karl et al. 2000). The accuracy of the Idaho amphibian and reptile models was relatively high (~85%) at the scale of entire study areas (~3,600 to 29,000 ha; Figure 2 and Table 2). Accuracy decreased substantially (to ~39%) at the fine (2 ha) and intermediate (259 ha) spatial scales sampling areas (Figure 2). Classification accuracy was higher for amphibian species (90%) than for reptile species (81%; Table 2).
2. Classification error rates decreased with increasing size of the sampling area (Figure 3). Few underpredictions (omission errors) occurred. Most of the errors were due to overpredictions (commission errors).
R2 = 0.26
Figure 3. Classification errors versus sampling unit areas. Solid circles indicate commission error percentages for each study area at three different spatial scales. Open circles indicate omission error percentages. The polynomial regression lines and R2 values are also indicated
3. In other studies (e.g., Burton 2001), multivariate analyses based on data collected in the field had correct classification percentages at the sampling site (2 ha) scale that were less than 75%. This suggests that high classification accuracies (>80%) for GAP models for Idaho amphibians and reptiles will be difficult or impossible to achieve at fine spatial scales, especially for rare species.
1. Using the Idaho amphibian and reptile GAP models at broad spatial scales should provide an accurate list of probable species for large areas such as national forests.
An example of the appropriate use for these models would be the development of a potential species lists for planning an inventory of amphibians and reptiles for a large national park.
Table 2. Classification accuracies by species for the study area spatialscale.
2. Using the Idaho amphibian and reptile GAP models at intermediate and fine spatial scales will considerably overestimate where these species occur. Therefore, these models must be used very cautiously when evaluating how well current reserve areas protect a given species. Depending on the size of the reserves, it may require twice as much area to protect species as indicated by gap analysis.
3. Because our field data-based, multivariate models of occurrence for some species have classification accuracies less than 75% at the site scale, we believe that it is unlikely that the current generation of GAP models can achieve high classification accuracies (>80%) at fine spatial scales for most of these species.
1. Expand analyses to include more study areas and species (e.g., Clearwater National Forest, Hells Canyon National Recreation Area, and Bear Lake National Wildlife Refuge).
2. Analyze the relationship between biophysical (i.e., temperature and moisture) characterization of study sites and accuracy.
3. Examine spatial variation in the accuracy of the predictions (e.g., the effect of the distance of the closest known record on prediction accuracy). Error rates may be higher at ecoregion boundaries.
4. Use error analyses (e.g., Table 2) to revise GAP models.
5. Develop new modeling approaches that increase classification accuracies at intermediate and fine spatial scales (e.g., incorporation of key habitat features such as communal overwintering sites of snakes).
The USGS National Gap Analysis Program provided the funding for analyzing the data. Funding for the field studies was provided by the Aldo Leopold Wilderness Research Institute, Bureau of Land Management, Caribou National Forest, Idaho Department of Fish and Game, Idaho Army National Guard, Idaho State University Graduate Research Committee, National Fish and Wildlife Foundation, National Park Service, The Wilderness Society, and the USGS Biological Resources Division.
We would like to thank Nancy Wright, Jason Karl, Leona Svancara, and Mike Scott for assistance with the GAP models.
Jason Jolley assisted with the GIS analysis. Leona Svancara and Chris Jenkins reviewed the manuscript.
Burton, S.R. 2001. Amphibian declines in southeast Idaho: Using modeling to assess the habitat loss hypothesis. D.A. thesis. Idaho State University, Pocatello, ID.
Karl, J.W., P.J. Heglund, E.O. Garton, J.M. Scott, N.M. Wright, and R.L. Hutto. 2000. Sensitivity of species habitat-relationship model performance to factors of scale. Ecological Applications 10:1690-1705.
Nussbaum, R.A., E.D. Brodie, and R.M. Storm. 1983. Amphibians and reptiles of the Pacific Northwest. University of Idaho Press, Moscow. 332 pp.
Scott, J.M., C.R. Peterson, J.W. Karl, E. Strand, L.K. Svancara, and N.M. Wright. 2002. A Gap Analysis of Idaho: Final Report. Idaho Cooperative Fish and Wildlife Research Unit. Moscow, ID
Stebbins, R.C. 1985. A field guide to western reptiles and amphibians. Houghton.
Return to Table of Contents