nena masthead
SENA Home Staff & Editors For Readers For Authors

Monitoring-based Assessment of Gap-analysis Models
Jill A. LaBram, Amanda E. Peck, and Craig R. Allen

Southeastern Naturalist, Volume 6, Number 4 (2007): 633–656

Full-text pdf (Accessible only to subscribers.To subscribe click here.)

 



Access Journal Content

Open access browsing of table of contents and abstract pages. Full text pdfs available for download for subscribers.

Issue-in-Progress: Vol. 23 (1) ... early view

Current Issue: Vol. 22 (3)
SENA 22(3)

Check out SENA's latest Special Issue:

Special Issue 12
SENA 22(special issue 12)

All Regular Issues

Monographs

Special Issues

 

submit

 

subscribe

 

JSTOR logoClarivate logoWeb of science logoBioOne logo EbscoHOST logoProQuest logo


2007 SOUTHEASTERN NATURALIST 6(4):633–656 Monitoring-based Assessment of Gap-analysis Models Jill A. LaBram1, Amanda E. Peck1, and Craig R. Allen2,* Abstract - Gap-analysis models of vertebrate species richness are primarily created based on literature and expert review to predict individual species’ occurrences and overall richness of vertebrates. Such models need validation based on empirical data to assess their accuracy. We describe and apply a new technique for assessing the accuracy of spatially explicit models. We evaluated the accuracy of South Carolina gap-analysis vertebrate models of predicted occurrence for reptile, amphibian, and mammal species on the Savannah River Site, SC, by comparing the agreement between gap-analysis models with models derived from multi-year monitoring data. We determined the species model agreement, commission and omission errors, and spatial correspondence in both single-species and richness models, and spatial correspondence of nodes of high richness. Average species agreement (accuracy) between models was 63%, with similar commission and omission error rates. Where there was spatial correspondence in single-taxon analyses, up to 15% of species identities differed in richness maps. Further refinement of vertebrate models will improve their accuracy, critical for the application of gap analyses to conservation decision-making. Introduction The gap-analysis program (Scott et al. 1993) identifies potential areas of high biodiversity in the United States by creating predictive models of vertebrate species’ potential habitat based upon remotely sensed vegetation data and other auxiliary information (Scott et al. 1987). Gap analysis was originally designed to identify possible “gaps” in the coverage of ecological reserve networks, but efforts have broadened to identify candidate conservation areas (Kiester et al. 1996). Such potential reserves should be designed to represent the full range of biodiversity within the region of interest (Margules and Pressey 2000). Although gap analysis does not recommend methods for reserve design, it develops some of the information needed and assesses the degree to which mapped elements may be represented in existing conservation areas (Jennings 2000). As such, gap analysis may help focus biodiversity conservation efforts and guide more rigorous field-based surveys of biological diversity. Gap-analysis models for vertebrates are primarily based on literature and expert review to predict individual species’ occurrences and overall richness of vertebrates. Because of this, gap analysis serves as a first step 1South Carolina Cooperative Fish and Wildlife Research Unit, Department of Forestry and Natural Resources, Clemson University, Clemson, SC 29634. 2US Geological Survey, Biological Resources Division, Nebraska Cooperative Fish and Wildlife Research Unit, University of Nebraska, Lincoln, NE, 68583. *Corresponding author - allencr@unl.edu. 634 Southeastern Naturalist Vol. 6, No. 4 in conservation decision-making that should be followed with focused biological surveys to validate gap-analysis models. The most common accuracy-assessment method used for gap-analysis models is a comparison of the predicted species within a National Park or National Wildlife Refuge to park or refuge checklists of breeding species (Boone and Krohn 2000). In general, the accuracy assessment of animal spatial models is crude and poorly developed, and requires quantification of both commission and omission errors. Omission errors (occurrence when absence is predicted) are relatively easy to document, but commission errors (absence when occurrence is predicted) are more difficult to estimate. Additionally, these different errors may have weighted costs associated with the ecological “value” of the species in terms of conservation priorities. Failure to correctly predict positive locations may be more “costly” than commission errors (Fielding 2002). For example, high omission error could possibly lead to the exclusion of species from conservation plans. Recent accuracy assessments of gap-analysis vertebrate models have stressed the importance of distinguishing actual commission errors (species is not present on the site) from apparent errors (field inventories are inevitably incomplete and falsely omit some species occurrence) and the importance of a priori species ranking. Species ranking places common, density-dependent species above rare ones in terms of likelihood of the model being correct (Boone and Krohn 1999, Shaefer and Krohn 2002). Boone and Krohn (1999) developed a multivariate method to correct commission errors in gap-analysis models by predicting the likelihood that a species would be sampled in future surveys, called likelihood of occurrence ranks. They demonstrated that variables such as size of survey site, duration of surveys, natural history of the species, and the quality of species-distribution models influence the validity of accuracy assessments. Vertebrate monitoring programs allow for the validation of model predictions. Multi-year sampling decreases errors associated with spatial and temporal variability in animal-habitat use and increases the odds of detecting less-common species. Thus, they provide rigorous data with which to assess the accuracy of gap-analysis models at the locations where sampling occurs. Similarly, spatially extensive sampling increases detection probabilities. Ultimately, assessment techniques must balance the spatial and temporal scales of sampling protocols, the extent and grain of species models, and the reality of logistical and financial constraints. Figure 1 (opposite page, upper figure). The land-cover classification for the 78,000- ha Savannah River Site (SRS), SC, modified from Imm (1997). We sampled herpetofauna and mammals in five replicates of each of the seven land-cover classes for three and five years, respectively. Figure 2 (opposite page, lower figure). Land-cover classification of the 78,000-ha Savannah River Site (SRS) area as classified by the South Carolina Gap Analysis Program. 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 635 636 Southeastern Naturalist Vol. 6, No. 4 We utilized gap-analysis vertebrate models and multi-year monitoring data to: (1) describe new methods for accuracy assessment, (2) assess the agreement of gap-analysis vertebrate models of reptile and amphibian (herpetofauna) and mammal species richness with sampling-based models, and (3) determine the spatial correspondence between the nodes of highest richness for predicted and sampling-based models. Study Area Our study site was located within the 78,000-ha Savannah River Site (SRS), near Aiken, SC (Figs. 1 and 2). When the SRS was acquired in 1950 by the Department of Energy as a nuclear production facility, 67% was forested and 33% was agricultural. All forest stands, except those with limited access on the floodplain, had been logged (Workman and McLeod 1990). The site was closed to the public in 1951, and the USDA Forest Service planted pine seedlings on former crop and pastureland, beginning in 1952, as an initial forest restoration effort. By 1963, about 90% of the area was covered by young forests (Golley et al. 1965). Currently, 95% of the site is covered with pine and hardwood forests and wetland habitats (Gibbons et al. 1997). Methods We used multi-year monitoring data to determine vertebrate distributions, against which we compared gap-analysis predictive models. We assessed the accuracy (model agreement, omission and commission errors) of the gap-analysis predictive models by comparing them with a model based on monitoring data for individual species, by taxon, and for all species, and also by comparing the spatial correspondence of species richness. Comparisons between gap analysis and monitoring models were made with two different base maps to determine if there was a change in error rates when using different land-cover classifications. We used a SRS region map developed by Imm (1997) (Fig. 1) and the South Carolina gap-analysis land-cover classification map (Fig. 2) for the same landscape. Vertebrate sampling We trapped herpetofauna and small mammals at five randomly chosen replicates of each of the seven major SRS land-cover types (Workman and McLeod 1990). The locations of our sites were selected randomly, and trap lines began >30 m from an edge (generally a gravel road). Small mammals were sampled on each replicate during the fall season for five years (1999–2003) utilizing Sherman live traps, tomahawk traps, and pitfall/drift fence arrays (Fig. 3). Herpetofauna were sampled during the fall of three years (2001–2003) and two summer seasons (2002, 2003) using pitfall/drift fence arrays, funnel traps, coverboards, PVC pipes, and visual (incidental) captures (Fig. 3). 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 637 Mammals. Mammal traps were opened nightly and checked daily for three consecutive days, twice a season. Paired Sherman live traps were placed at 10-m intervals along three parallel 150-m trap lines that were placed 30 m apart, for a total of 96 Sherman traps per trapping period. Tomahawk box traps were placed to sample medium-sized carnivores and omnivores (a total of 12 tops/transect) at 30-m intervals along the outer-two Sherman trap lines. At 50-m intervals along the center trap line, pitfall/drift fence arrays were established between two Sherman lines to sample shrews and mice (a total of 3 arrays/transect, 4 buckets per array, resulting in 12 pitfall traps). Drift fences were constructed using 50-cm high aluminum flashing, and 19-liter plastic buckets served as pitfalls (Gibbons and Bennet 1974). Small mammals were identified to species and marked with individually numbered 0.635-cm Monel fish and small-mammal ear tags, and meso-mammals were marked with spray paint. Herpetofauna. Herpetofauna were sampled for nine consecutive days, twice per season. A double-ended funnel trap constructed from hardware cloth was placed in the middle of each segment of each pitfall-drift fence Figure 3. Within-replicate design of mammal and herpetofauna trapping within each of the 35 study areas (five replicates of seven land-cover types) at the Savannah River Site. Mammals were captured using Sherman traps (96), tomahawk traps (12), and three pitfall/drift-fence arrays (12 total buckets). Herpetofauna were captured using pitfall traps (12 buckets), funnel traps (9), PVC pipes (6), and coverboards (8). Visual captures were also recorded. 638 Southeastern Naturalist Vol. 6, No. 4 array, totaling nine funnel traps per site. Eight cover boards were placed at each site, and were set in two arrays. Cover boards were 61- x 122-cm sheets of tin or plywood, and each array consisted of two tin and two plywood cover boards. We used PVC pipes (6 per site) to sample herpetofauna species (mainly treefrogs) that escape pitfall traps and may not be commonly found using cover boards. Pipes were 1.5-m sections of opaque white 3.2-cm diameter PVC pipe inserted upright into the ground. Visual captures also were recorded. Amphibians and reptiles were identified and marked with visible implant fluorescent elastomer or were toe-clipped. Land-cover data SRS land cover. Landscape features within the Savannah River Site (e.g., topography, geomorphologial classification, soil classification) were used to develop an ecological classification GIS coverage in ArcView (Imm 1997). We modified Imm’s (1997) land-cover classification by grouping similar land-cover classes (e.g., southern mixed hardwood, pine hardwood, and pine-bay hardwood forest were all classified as mixed forest) into seven distinct vegetation types: bottomland hardwood, swamp-edge, mixed forest, hardwood slope, planted pine, Carolina bay, and sandhill (Fig. 1). Workman and McLeod (1990) describe the vegetation characteristics of these classes. Gap-analysis land cover. Between August and December 2001, 758 polygons were surveyed using a combination of remote sensing image interpretation and ground truthing to compile the dominant land covers of South Carolina. A 27-class raster habitat-based classification with a resolution of 30 m was produced from Landsat TM imagery dating from 1991–1993 (Schmidt et al. 2001). The Savannah River Site area was clipped from the South Carolina gap-analysis map. The SRS area included 22 of the 27 gap analysis land-cover classes (Fig. 2), but only 10 classes were applicable to our terrestrial-based study: swamp, bottomland/floodplain forest, closedcanopy evergreen forest/woodland, needle-leaved evergreen mixed forest/ woodland, pine woodland, dry deciduous forest/woodland, mesic deciduous forest/woodland, dry mixed forest/woodland, mesic mixed forest/woodland, and wet evergreen. These classes are described on the gap-analysis website: ftp://ftp.gap.uidaho.edu/products/South_Carolina/gis/landcover/grid/ scgapveg2.html. The gap-analysis habitat classification differed from the SRS classification because the different maps were developed for different uses. Neither classification is correct or incorrect, but they differ in their aggregation and delineation of land-cover categories. Therefore, we created a crosswalk table that converted gap-analysis classes into SRS classes (Appendix 1) to provide comparison between models based on the two classification systems using SRS land cover as the base map. SRS classes also were cross-walked into gap-analysis classes using the latter as the base map to determine how errors were affected by the classification schemes. 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 639 Models Monitoring-based models. This study focused on terrestrial species; therefore, our capture data apply to a 200-m swamp-edge buffer rather than the entire swamp class at SRS. Because gap analysis is more effective in modeling common, abundant species (Boone and Krohn 1999), and because general trapping techniques such as we employed are also more likely to sample common species, we limited the species we modeled to those which were commonly captured. We modeled mammal and reptile species with a minimum of 20 captures and amphibian species with a minimum of 35 captures. We set a criteria that captures within a given land-cover class must account for ≥5% of the captures for a species for it to be considered “present” in that land-cover class (Martin and McComb 2003). Trapping effort was equal in all replicates and land-cover classes. Thirty-one species met our criteria, including 10 reptiles, 13 amphibians, and 6 mammals (Table 1). We created a species-habitat matrix of presence/absence for each species in each land-cover class based on our monitoring data, with a value of “1” designating presence and a value of “0” designating absence of a species. This matrix was created using both the SRS and gap-analysis base maps. Gap-analysis-based models. Gap-analysis-generated habitat affinities for herpetofauna and mammals were developed primarily from literature review and obtained from the South Carolina Department of Natural Resources (Schmidt et al. 2001). These animal-habitat associations also were crosswalked into SRS land-cover classes. This information was used to build a matrix of species versus land-cover class for our 31 focal species. These species were predicted to be present (value of “1”) or absent (value of “0”) in each land-cover type. Commission and omission errors Agreement, commission, and omission error rates were calculated for individual species and averaged by taxon for each base map by area and land-cover class. Rates were then calculated within each land cover and across the Savannah River Site. An area or land-cover type was in spatial agreement between the gap analysis and monitoring-based models if both predicted the species to be either present or absent within that landcover type or area. Spatial correspondence. Composite raster species richness maps for amphibian, reptile, mammal, and combined taxa were produced by adding the individual species models to produce a composite map of overall monitoring- based richness, and predicted richness for the gap-analysis models. We compared the gap analysis predictive model to our monitoring-based model, using both the SRS and gap analysis land-cover classifications as our base maps to determine spatial correspondence of species richness. The monitoring-based richness model was subtracted from the gap-analysis predicted-richness model. A value of 0 occurred and was defined as spatial 640 Southeastern Naturalist Vol. 6, No. 4 correspondence when the number of species predicted to occur in a landcover class equaled the number of species that occurred based on field monitoring. Positive values occurred where gap analysis predicted species richness was greater than capture richness (commission errors), and negative values occurred where capture richness was greater than gap analysis predicted species richness (omission errors) (Allen et al. 2001a). One of the explicit focuses of gap analysis is not single-species models, but rather the identification of areas with potentially high species richness. Thus, we determined land-cover classes representing nodes of high species richness or “hotspots” (i.e., richness values ≥ 80% of the maximum possible richness; Allen et al. 2001a) for sample-based and gap-analysis models, using both the SRS and gap-analysis base maps. Table 1. Common herpetofauna and mammal species sampled in five replicates of seven landcover classes of the Savannah River Site, SC, for three and five years, respectively. Common name Scientific name Amphibians Southern Cricket Frog Acris gryllus Le Conte Marbled Salamander Ambystoma opacum Gravenhorst Mole Salamander Ambystoma talpoideum Holbrook Tiger Salamander Ambystoma tigrinum Green Southern Toad Bufo terrestris Bonnaterre Eastern Narrow-mouthed Toad Gastrophryne carolinensis Holbrook Green Treefrog Hyla cinerea Schneider Squirrel Treefrog Hyla squirrella Bosc Slimy Salamander Plethodon glutinosus Green (complex) Spring Peeper Pseudacris crucifer Wied-Neuwied Green Frog Rana clamitans Latreille Southern Leopard Frog Rana sphenocephala Harlan Eastern Spadefoot Toad Scaphiopus holbrooki Harlan Reptiles Green Anole Anolis carolinensis Voigt Six-lined Racerunner Cnemidophorus sexlineatus Linnaeus Black Racer Coluber constrictor Linnaeus Ringneck Snake Diadophis punctatus Linnaeus Five-lined Skink Eumeces fasciatus Linnaeus Broadhead Skink Eumeces laticeps Schneider Fence Lizard Sceloporus undulatus Bosc and Daudin Ground Skink Scincella lateralis Say Redbelly Snake Storeria occipitomaculata Storer Southeastern Crowned Snake Tantilla coronata Baird and Girard Mammals Southern short-tailed shrew Blarina carolinensis Bachman Least shrew Cryptotis parva Say Opossum Didelphis virginiana Kerr Eastern woodrat Neotoma floridana Ord Golden mouse Ochrotomys nuttali Harlan Cotton mouse Peromyscus gossypinus LeConte Raccoon Procyon lotor Linnaeus Southeastern shrew Sorex longirostris Bachman 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 641 Using this criteria, a land-cover class would be considered a “hotspot” if 80% of the species occurred or a “predicted hotspot” if 80% of the species were predicted to occur (i.e., 11 of the 13 amphibians, 8 of the 10 reptiles, 7 of the 8 mammals, or 25 of 31 total species). We then determined the correspondence between these actual and predicted nodes of highest richness for each taxon and combined taxa in each base map. Results Vertebrate sampling Models based on SRS land cover. We captured, marked, and released 23,105 individuals of 69 herpetofauna species and 1142 individuals of 16 mammal species. Species richness based on our sampling varied from 2–12 species per land-cover type for amphibians, 6–10 species for reptiles, 2–8 species for mammals, and 11–28 species for all taxa (Table 2). Gap-analysis- predicted species richness ranged from 5–13 species for amphibians, 2–10 species for reptiles, 3–8 species for mammals, and 10–31 for all taxa (Table 2). Models based on gap analysis land cover. Species richness based on our monitoring program varied from 2–11 species per land-cover type for amphibians, 7–10 species for reptiles, 2–7 species for mammals, and 11–28 species for all taxa (Table 3). Gap-analysis-predicted species richness varied from 5–13 species for amphibians, 2–10 species for reptiles, 3–8 species for mammals, and 9–30 species for all taxa (Table 3). Agreement: Commission and omission errors SRS land cover as a base map. Average amphibian accuracy was 68%, and commission errors occurred more often than omission errors in both area and land-cover comparisons (Table 4). Amphibians were most accurately predicted in Carolina bays (92%) and least accurately in sandhills (15%, due to commission error) (Table 2). Average reptile accuracy was 69%. Omission errors occurred more often than commission errors by both area and land cover. The highest reptile accuracy of gap-analysis-predicted models was found in hardwood slope and mixed forest (90%), and the lowest accuracy was in planted pine (40%, due to omission error). Average accuracy of gap-analysis mammal models was 64%, and omission and commission error rates were more or less equivalent by both area and land cover. The highest accuracy of gap-analysis-predicted mammal models was for bottomland hardwood (100%), and presence or absence was predicted poorly in sandhills (38%, due to commission error). Overall, individual-species models were on average accurate 67% of the time. Commission errors averaged 18% and omission errors averaged 14% for all species. The highest accuracy for all species predicted was bottomland hardwood (87%), while presence or absence of each species was predicted poorly in sandhills (39%, due to commission error). 642 Southeastern Naturalist Vol. 6, No. 4 Table 2. Species occurrence at the Savannah River Site (SRS) based on multi-year monitoring models and gap-analysis models, using SRS land cover as a base map. A “1” indicates species presence and a “0” indicates absence. The first number in each column shows presence or absence of a species based on monitoring data. The second number in each column shows the gap-analysis-predicted presence or absence of a species in each land-cover class. Agreement occurs when gap-analysis predictions correspond with capture data for presence or absence of a species. Omission errors occur when gap analysis predicts a species to be absent when it was actually present. Commission errors occur when gap analysis predicts a species to be present and it was absent based on capture data. Species BLH1 BAY HWS MIX PPI SDH SWA Amphibians Southern Cricket Frog 1/1 1/1 1/1 1/1 0/1 0/1 1/1 Green Frog 1/1 1/1 1/1 1/1 1/0 0/1 1/1 Southern Leopard Frog 1/1 1/1 1/1 1/1 1/0 0/1 1/1 Spring Peeper 1/1 1/1 0/1 0/1 0/0 0/1 1/1 Squirrel Treefrog 1/1 0/1 1/1 0/1 0/0 0/1 1/1 Green Treefrog 1/1 1/1 1/0 1/1 0/0 0/1 1/1 Southern Toad 1/1 1/1 1/1 1/1 1/1 0/1 1/1 Eastern Narrow-mouthed Toad 0/1 1/1 1/1 1/1 1/1 0/1 1/1 Eastern Spadefoot Toad 0/1 1/1 1/1 1/1 1/1 1/1 1/1 Marbled Salamander 1/1 1/1 0/1 1/1 0/0 0/1 1/1 Mole Salamander 1/1 1/1 0/0 0/1 1/0 0/1 0/1 Tiger Salamander 1/1 1/1 0/0 0/1 1/0 0/1 0/1 Amphibian totals 11/13 12/13 9/10 9/13 7/5 2/13 11/13 % amphibian agreement 85 92 77 69 54 15 85 % omission error 0 0 8 0 31 0 0 % commission error 15 8 15 31 15 85 15 Reptiles Green Anole 1/1 1/1 1/1 1/1 1/0 1/1 1/1 Six-Lined Racerunner 0/0 0/0 0/1 0/1 1/0 1/1 1/0 Five-Lined Skink 1/1 0/0 1/1 1/1 1/0 0/1 1/0 Broadhead Skink 1/1 0/1 1/1 1/1 1/0 0/1 1/1 Ground Skink 1/1 1/1 1/1 1/1 1/1 1/1 1/1 Fence Lizard 0/1 0/0 1/1 1/1 1/0 1/1 1/0 Black Racer 1/1 1/1 1/1 1/1 1/1 1/1 1/1 Ringneck Snake 1/1 1/0 1/1 1/1 0/0 1/1 1/1 Redbelly Snake 1/0 1/0 1/1 1/1 0/0 0/1 1/0 Southeastern Crowned Snake 0/0 1/0 1/1 1/1 1/0 1/1 1/0 Reptile totals 7/7 6/4 9/10 9/10 8/2 7/10 10/5 % reptile agreement 80 60 90 90 40 70 50 % omission error 10 30 0 0 60 0 50 % commission error 10 10 10 10 0 30 0 Mammals Southern short-tailed shrew 1/1 1/1 1/1 1/1 1/0 0/1 1/0 Southeastern shrew 1/1 1/1 1/1 1/1 1/1 0/1 1/1 Least shrew 0/0 1/0 1/0 1/1 1/0 0/1 1/0 Golden mouse 1/1 1/0 0/1 1/1 1/1 0/1 1/1 Cotton mouse 1/1 1/0 0/1 1/1 1/0 1/1 1/1 Woodrat 1/1 1/0 1/1 0/1 0/0 0/0 0/1 Opossum 1/1 1/1 1/1 1/1 1/1 0/1 1/1 Raccoon 1/1 1/1 1/1 1/1 1/0 1/1 1/1 Mammal totals 7/7 8/4 6/7 7/8 7/3 2/7 7/6 % mammal agreement 100 50 63 88 50 38 63 % omission error 0 50 13 0 50 0 25 % commission error 0 0 25 13 0 63 13 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 643 Gap-analysis land cover as a base map. Average amphibian accuracy was 55%, and commission errors occurred more than omission errors by both area and land cover (Table 4). Amphibian species were most accurately predicted in swamp-edge and bottomland/floodplain forest classes (85%), and were least accurate in pine woodland (15%, due to commission error) (Table 3). The Table 2, continued. Species BLH1 BAY HWS MIX PPI SDH SWA All species Vertebrate totals 25/27 26/21 24/27 25/31 22/10 11/30 28/24 % species agreement 87 71 77 81 48 39 68 % omission error 3 23 6 0 45 0 23 % commission error 10 6 16 19 6 61 10 1BLH = bottomland hardwood, BAY = Carolina bay, HWS = hardwood slope, MIX = mixed forest, PPI = planted pine, SDH = sandhill, SWA = swamp-edge. Figure 4. Spatial correspondence of species richness between gap-analysis-predicted occurrence and monitoring-based occurrence using Savannah River Site (SRS) land cover as the base map. Models of the presence of individual species were summed to get a total richness by land-cover class for predicted and actual occurrence. Captured richness was subtracted from predicted richness for each land-cover class. Negative values represent land-cover classes in which captured species richness was greater than predicted species richness. Positive values represent land-cover classes in which predicted species richness was greater than captured species richness. A zero would occur if gap analysis predicted the same number of species to be present that actually were based on capture data in that land-cover class. 644 Southeastern Naturalist Vol. 6, No. 4 Table 3. Species occurrence at the Savannah river Site (SRS) based on multi-year monitoring models and gap-analysis models, using gap-analysis land-cover as a base map. A “1” indicates species presence and a “0” indicates absence. The first number in each column shows the presence or absence of a species based on monitoring data. The second number in each column shows the predicted presence or absence of a species by gap analysis in each land-cover category. Agreement occurs when gap-analysis predictions correspond with capture data for presence or absence of a species. Omission errors occur when gap analysis predicts a species to be absent when it was actually present. Commission errors occur when gap analysis predicts a species to be present and it was absent based on capture data. Species S1 BF/F CCEF NEMF PW DDF MDF DMF MMF WE Amphibians Southern Cricket Frog 1/1 1/1 0/1 0/1 0/1 1/0 1/1 1/0 1/1 1/0 Green Frog 1/1 1/1 1/0 0/1 0/1 1/0 1/1 1/0 1/1 1/0 Southern Leopard Frog 1/1 1/1 1/0 0/1 0/1 1/0 1/1 1/0 1/1 1/0 Spring Peeper 1/1 1/1 0/0 0/1 0/1 0/0 0/1 0/0 0/1 1/1 Squirrel Treefrog 1/1 1/1 0/0 0/1 0/1 1/1 1/1 0/1 0/1 1/1 Green Treefrog 1/1 1/1 0/0 0/0 0/1 1/0 1/0 1/0 1/1 1/0 Southern Toad 1/1 1/1 1/1 0/1 0/1 1/1 1/1 1/1 1/1 1/0 Eastern Narrow-mouthed Toad 1/1 0/1 1/1 0/1 0/1 1/1 1/1 1/1 1/1 0/0 Eastern Spadefoot Toad 1/1 0/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 0/0 Marbled Salamander 1/1 1/1 0/0 0/1 0/1 0/1 0/1 1/1 1/1 1/0 Mole Salamander 0/1 1/1 1/0 0/1 0/1 0/0 0/0 0/1 0/1 1/1 Tiger Salamander 0/1 1/1 1/0 0/1 0/1 0/0 0/0 0/1 0/1 1/0 Slimy Salamander 1/1 1/1 0/1 1/1 1/1 1/1 1/1 1/1 1/1 1/0 Amphibian totals 11/13 11/13 7/5 2/12 2/13 9/6 9/10 9/8 9/13 11/3 % amphibian agreement 85 85 54 23 15 62 77 46 69 38 % omission error 0 0 31 0 0 31 8 31 0 62 % commission error 15 15 15 77 85 8 15 23 31 0 Reptiles Green Anole 1/1 1/1 1/0 1/1 1/1 1/1 1/1 1/0 1/1 1/0 Six-lined Racerunner 1/0 0/0 1/0 1/1 1/1 0/0 0/0 0/0 0/0 0/0 Five-lined Skink 1/0 1/1 1/0 0/1 0/1 1/0 1/1 1/0 1/1 1/0 Broadhead Skink 1/1 1/1 1/0 0/0 0/1 1/0 1/1 1/0 1/1 1/0 Ground Skink 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/0 Fence Lizard 1/0 0/0 1/0 1/1 1/1 1/1 1/0 1/1 1/0 0/1 Black Racer 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 1/1 Ringneck Snake 1/1 1/1 0/0 1/1 1/1 1/0 1/1 1/0 1/1 1/1 Redbelly Snake 1/0 1/0 0/0 0/1 0/1 1/0 1/1 1/0 1/1 1/0 Southeastern Crowned Snake 1/0 0/0 1/0 1/1 1/1 1/1 1/0 1/1 1/0 0/0 Reptile totals 10/5 7/6 8/2 7/9 7/10 9/5 9/7 9/4 9/7 7/3 % reptile agreement 50 90 40 80 70 60 80 50 80 40 % omission error 50 10 60 0 0 40 20 50 20 50 % commission error 0 0 0 20 30 0 0 0 0 10 Mammals Southern short-tailed shrew 1/0 1/1 1/0 0/0 0/1 1/0 1/1 1/0 1/1 1/0 Southeastern shrew 1/1 1/1 1/1 0/1 0/1 1/0 1/1 1/0 1/1 1/1 Least shrew 1/0 0/0 1/0 0/1 0/1 1/0 1/0 1/0 1/1 0/0 Golden mouse 1/1 1/1 1/1 0/1 0/1 0/1 0/1 1/1 1/1 1/0 Cotton mouse 1/1 1/1 1/0 1/1 1/1 0/1 0/1 1/1 1/1 1/0 Woodrat 0/1 1/1 0/0 0/0 0/0 1/1 1/1 0/1 0/1 1/0 Opossum 1/1 1/1 1/1 0/1 0/1 1/1 1/1 1/1 1/1 1/1 Raccoon 1/1 1/1 1/0 1/1 1/1 1/1 1/1 0/1 0/1 1/1 Mammal totals 7/6 7/7 7/3 2/6 2/7 6/5 6/7 6/5 6/8 7/3 % mammal agreement 63 100 50 50 38 38 63 38 75 50 % omission error 25 0 50 0 0 38 13 38 0 50 % commission error 13 0 0 50 63 25 25 25 25 0 All Species Vertebrate totals 28/24 25/26 22/10 11/27 11/30 24/16 24/24 24/17 24/28 25/9 % agreement 68 90 48 48 39 55 74 45 74 42 % omission error 23 3 45 0 0 35 13 39 6 55 % commission error 10 6 6 52 61 10 13 16 19 3 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 645 Table 3, continued. 1S = swamp-edge, BF/F = bottomland floodplain forest, CCEF = closed canopy evergreen forest/ woodland, NEMF = needle-leaved evergreen mixed forest/woodland, PW = pine woodland, DDF = dry deciduous forest/woodland, MDF = mesic deciduous forest/woodland, DMF = dry mixed forest/woodland, MMF = mesic mixed forest/woodland, WE = wet evergreen. Table 4. Agreement, omission, and commission errors for each species in two categories: number of land-cover classes and area of land-cover classes. Omission errors occur when gap analysis predicts a species to be absent when it was present based on monitoring models, and commission errors occur when gap analysis predicts a species to be present when it was absent based on monitoring models. Percent agreement, omission, and commission were calculated by dividing the respective number of land-cover classes or area by the total # of land-cover classes or area in each base map. The first number in each column represents the percent agreement, omission or commission using the Savannah River Site (SRS) land-cover classification. The second number represents the same calculations for the gap analysis land-cover classification. % OE = percent omission error, % CE = percent commission error. Landcover Area % landcover % area Common name agreement % OE % CE agreement % OE % CE Southern Cricket Frog 71/40 0/30 29/30 39/23 0/2 61/75 Green Frog 71/40 14/40 14/20 39/23 36/56 25/21 Southern Leopard Frog 71/40 14/40 14/20 39/23 36/56 25/21 Spring Peeper 57/60 0/0 43/40 57/72 0/0 43/28 Squirrel Treefrog 57/60 0/0 43/40 61/78 0/0 39/22 Sreen Treefrog 71/50 14/40 14/10 70/92 5/8 25/0 Southern Toad 86/70 0/10 14/20 75/79 0/0 25/21 Eastern Narrow-mouthed Toad 71/70 0/0 29/30 60/68 0/0 40/32 Eastern Spadefoot Toad 86/90 0/0 14/10 85/89 0/0 15/11 Marbled Salamander 71/50 0/10 29/40 70/71 0/0 30/29 Mole Salamander 43/40 14/10 43/50 21/19 36/54 43/27 Tiger Salamander 43/30 14/20 43/50 21/19 36/54 43/27 Slimy Salamander 86/80 0/10 14/10 64/46 0/0 36/54 Green Anole 86/70 14/30 0/0 64/46 36/54 0/0 Six-lined Racerunner 43/80 29/20 29/0 41/40 41/60 18/0 Five-lined Skink 57/30 29/50 14/20 34/17 41/61 25/21 Broadhead Skink 57/50 14/40 29/10 38/44 36/56 26/0 Ground Skink 100/90 0/10 0/0 100/100 0/0 0/0 Fence Lizard 57/50 29/40 14/10 44/34 41/66 15/0 Black Racer 100/100 0/0 0/0 100/100 0/0 0/0 Ringneck Snake 86/80 14/20 0/0 99/98 1/2 0/0 Redbelly Snake 43/30 43/50 14/20 54/60 21/18 25/21 Southeastern Crowned Snake 57/60 43/40 0/0 58/34 42/66 0/0 Southern shorttail shrew 57/40 29/50 14/10 34/39 41/61 25/0 Southeastern shrew 86/60 0/20 14/20 75/77 0/2 25/21 Least shrew 29/30 57/50 14/20 28/11 47/67 25/21 Golden mouse 57/50 14/10 29/40 69/71 1/0 30/29 Cotton mouse 57/60 29/20 14/20 58/38 37/54 5/8 Eastern woodrat 57/60 14/10 29/30 81/94 1/0 18/6 Opossum 86/80 0/0 14/20 75/79 0/0 25/21 Raccoon 86/70 14/10 0/20 64/46 36/54 0/0 Average amphibian agreement 68/55 6/16 26/28 54/54 12/18 35/28 Average reptile agreement 69/64 21/30 10/6 63/57 26/38 11/4 Average mammal agreement 64/56 20/21 16/23 61/57 21/30 19/13 Average agreement by species 67/58 14/22 18/20 59/47 19/23 23/30 646 Southeastern Naturalist Vol. 6, No. 4 highest reptile accuracy of gap analysis-predicted models was found in bottomland floodplain (90%), and the lowest accuracy was in wet evergreen and 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 647 closed-canopy evergreen forest classes (40%). Average reptile accuracy was 64%, and there were more omission errors than commission errors by both area and land cover. Average mammal accuracy was 56%, and omission errors were higher than commission errors by area comparison and lower by land Figure 5 (opposite page, upper figure). Spatial correspondence between nodes of highest richness (hotspots) predicted by gap analysis and those hotspots that occurred based on monitoring data using the Savannah River Site (SRS) land-cover classification as a base map. For a land-cover class to be considered a node of high richness, a minimum 25 of the 31 vertebrate species (top 20%) was required to be present. Figure 6 (opposite page, lower figure). Spatial correspondence of species richness between gap-analysis-predicted occurrence and occurrence derived from monitoring data, using the gap-analysis land-cover as the base map. Models of the presence of individual species were summed to get a total richness by land-cover class. Monitoring-based richness models were subtracted from predicted richness for each land-cover class. Negative values represent land-cover classes in which captured species richness was greater than predicted species richness. Positive values represent land-cover classes in which predicted species richness was greater than captured species richness. A zero would occur if gap analysis predicted the same number of species to be present as were detected with multi-year monitoring. Figure 7. Spatial correspondence between nodes of highest richness (hotspots) predicted by gap analysis and those derived from multi-year monitoring, using the gap analysis land cover. For a land-cover class to be considered a node of high richness, a minimum of 25 of the 31 vertebrate species (top 20%) was required. 648 Southeastern Naturalist Vol. 6, No. 4 cover. Mammal species were predicted with 100% accuracy in bottomland/ floodplain, and were predicted least accurately in three land covers (38% in pine woodland, dry deciduous forest, and mesic deciduous forest). Individual models were on average accurate 58% of the time. For all species, commission errors averaged 20%, and omission errors averaged 22%. Accuracy ranged from 90% in bottomland floodplain forest to 39% in pine woodland. Spatial correspondence SRS land cover as a base map. Predicted amphibian species richness was higher than monitoring-based richness in all but the planted pine land-cover class. These predictions ranged from a single commission error in Carolina bay and hardwood slope classes to 11 more species predicted to occur than were captured in the sandhill class. Additionally, there was no spatial correspondence between the number of species captured and those predicted by gap analysis for amphibians. However, three monitoring-based hotspots corresponded to predicted hotspots (bottomland hardwood, Carolina bay, and swamp-edge), with 85% overlap in predicted and actual species composition (Table 2). For reptiles, correspondence occurred in the bottomland hardwood class (7 species), with an 80% agreement of species composition (i.e., Fence Lizard was predicted to occur, but did not based on monitoring, and the Redbelly Snake was captured but not predicted to occur in gap-analysis models). Reptile species richness ranged from three more species predicted in the sandhill landcover class to 6 more species sampled than predicted in the planted pine class. Two monitoring-based reptile hotspots corresponded to predicted hotspots (hardwood slope and mixed forest), with 90% agreement in predicted and monitoring-based species composition. Mammal richness corresponded in the bottomland hardwood class with all 8 species predicted to occur correctly. The sandhill class differed the most, with 5 more species predicted to occur than were actually sampled. Four more species were documented during monitoring than were predicted in gap-analysis models in the Carolina bay class. Bottomland hardwood and mixed forest classes corresponded as mammal hotspots to those predicted by gap analysis. The remaining classes were either predicted to be a hotspot (sandhill and hardwood slope) or were a monitoring-based hotspot (Carolina bay, planted pine, and swamp-edge), but did not correspond (Table 2). There was no spatial correspondence between monitoring-based and gap-analysis-predicted models for all taxa (Fig. 4). Differences ranged from 19 more species predicted than were documented in the sandhill land-cover class to 12 more species occurring than were predicted in the planted pine class. All 31 species were predicted to occur in the mixed forest class, while 25 actually occurred based on monitoring, qualifying this land cover as a monitoring-based and predicted hotspot. Predicted and monitoring-based hotspot correspondence also occurred in the 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 649 bottomland hardwood class. Gap analysis predicted hardwood slope and sandhill classes as vertebrate hotspots, whereas Carolina bay and swampedge classes were identified as monitoring-based hotspots. The planted pine class did not qualify as a hotspot based on monitoring nor was it predicted to be one by gap analysis (Fig. 5). Gap-analysis land cover as a base map. There was no correspondence of amphibian species richness between monitoring-based models and those predicted by gap analysis. Differences ranged from 11 and 10 more species predicted than sampled in pine woodland and needle-leaved evergreen mixed forest/woodland, respectively, to 8 more species documented than predicted in wet evergreen. Two monitoring-based hotspots corresponded to predicted hotspots for amphibians (swamp-edge and bottomland floodplain forest), with an 85% agreement in predicted and actual species composition (Table 3). There was no correspondence of reptile species richness between the monitoring-based models and those predicted by gap analysis. Predicted reptile richness was higher than the monitoring-based richness in only 2 classes (by 2 species in needle-leaved evergreen mixed forest/woodland and by 3 species in dry deciduous forest). These two classes were predicted as hotspots, but did not correspond to the 6 monitoring-based hotspots. Monitoring- based richness ranged from 6 more species occurring than predicted (closed canopy evergreen forest/woodland) to only 1 more species occurring than predicted (bottomland floodplain forest). Monitoring-based mammal richness corresponded to predicted richness in the bottomland floodplain/forest class, with all species predicted correctly. Five more species were predicted to occur in pine woodland than were recorded based on monitoring, while 4 more species occurred than were predicted in the closed canopy evergreen forest and wet evergreen classes. The bottomland floodplain class was the only hotspot that was predicted correctly based on monitoring. Vertebrate species had spatial correspondence between predicted and actual vertebrate occurrence in one land-cover class (mesic deciduous forest; Fig. 6), with a 74% agreement in species composition. Differences ranged from 19 more species predicted to occur in the pine woodland class to 16 more species occurring based on monitoring than predicted by gap analysis in the wet evergreen class. Although gap analysis predicted four land-cover classes as hotspots and three were identified based on monitoring, there was only correspondence in the bottomland/floodplain class (Fig. 7). Discussion The best assessment of a model’s accuracy is validation with an independent data set. Therefore, we compared South Carolina gap-analysis models with models based on multi-year monitoring. Instead of assigning a priori 650 Southeastern Naturalist Vol. 6, No. 4 or likelihood of occurrence ranks to reduce commission errors based on the likelihood-of-occurrence, we only modeled species that were commonly captured. While likelihood-of-occurrence ranks methods may be useful in assessing models when not much is known about the data used to inform the model (e.g., the length of survey, size of study area, reliability of data collected by multiple investigators), it was more appropriate in our study to examine a subset of common species. Our results are based on some key assumptions. One assumption is that ≥20 captures for mammals and reptiles and ≥35 captures for amphibians was enough to build valid models. The second assumption is that modeling a species as present in a land-cover class where it was captured is valid only if ≥5% of the total captures for that species were in that land-cover class. Both assumptions are attempts to remove uncommon or transient animals from our models and ensure minimum data for model building. However, failure to detect a species on a site may be due to trapping difficulty, natural rarity, or spatial or temporal variability in habitat use rather than the absence of the animal. We tested our assumptions by performing sensitivity analyses that varyied the rules used. For both assumptions, we found that our minimum rules provided a reasonable balance between omission and commission error rates and number of species modeled (Figs. 8 and 9). Additionally, the assessed accuracy of species models was not related to the Figure 8. Impact of varying the required minimum number of individuals captured (x-axis) on omission and commission error rates (y-axis). The solid darker gray bars represent omission error rates, and the hatched bars represent commission error rates. Twenty captures allowed for a compromise between error rate and number of species modeled. 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 651 number of individuals sampled for species captured ≥20 times (p > 0.05). Another source of error for our monitoring models is differences in detection probabilities for species in different land covers; we did not test for this. However, our models and gap-analysis models are based on the presence or absence of a species, rather than its abundance, and thus this source of error should have been minimal for most species. Classification schemes developed for different purposes and/or at different scales aggregate within and among land-cover classes differently. Thus, converting between classification systems can increase the commission and omission errors of the models. There were several land-cover types that were not clearly delineated in the South Carolina gap analysis, which may have led to failure of animal-habitat associations to correctly predict species occurrence. For example, South Carolina gap analysis could not reliably separate the land-cover categories of swamp and bottomland hardwood based on their techniques or decision rules (Schmidt et al. 2001). Also, none of the 194 Carolina bays (786 ha) known to occur in the SRS area were present on the South Carolina gap-analysis map; therefore, we could not include that class in the gap-analysis-based model. Conversely, the SRS seven-class land-cover classification scheme was simpler than the gap analysis ten-class land-cover classification. Figure 9. Impact of varying the minimum percent of captures required (x-axis) for a species to be included in a land-cover class, on omission and commission error rates (y-axis). The gray bars represent omission error rates, and the hatched bars represent commission error rates. A 5% capture criteria balances omission and commission error rates. 652 Southeastern Naturalist Vol. 6, No. 4 Omission and commission error rates may vary depending on whether one calculates error rates based on land-cover classes or the area of agreement. Species with the same land-cover class agreement rates may differ in the area of agreement, so different calculations may be required based on research or management objectives. A land cover that is large in size relative to the landscape can influence error rates more than a small land cover, but the small land cover may be more ecologically important. The sandhill and planted pine land covers comprise almost two thirds of the SRS landscape, so their high error rates (61% and 51%, respectively) contributed greatly to reduce overall accuracy within our study area. Average model agreement for the two base land-cover maps was similar for both commission and omission error rates. Error rates were similar for mammals, and omission errors were higher than commission errors for reptiles. For amphibians, commission errors were higher than omission errors, perhaps due to differences in detectability in this taxon. The vertebrate species most accurately modeled in both land-cover maps and methods of assessment were either habitat generalists (e.g., Coluber Constrictor L. [Black Racer] and Didelphis virginiana Kerr [opossum]) or specialists (e.g., Neotoma floridana Ord [eastern woodrat]). Vertebrate species with lowest model-agreement rates may interact with finer-scale variables difficult to identify with remote sensing or may have lower detection rates (e.g., Ambystoma talpoideum Holbrook [Mole Salamander] and Cryptotis parva Say [least shrew]). Because gap analysis is a tool for predicting vertebrate distributions for use in conservation planning, Edwards et al. (1996) argues that commission error is preferred over omission error. High omission error could possibly lead to the exclusion of species from conservation plans. Incorporating uncertainty into gapanalysis models would enhance their applicability. One of the explicit focuses of gap analysis is to identify areas of potential high biodiversity, so we determined which land-cover classes contained these hotspots for each base map. There was some spatial correspondence when conducting analyses for each taxon separately as well as vertebrates as a whole, but correspondence was limited to three or fewer classes. Where there was spatial correspondence for within-taxon analyses, species identities often differed. This is important if managers are concerned with conservation of a particular species, as opposed to species richness. We do not present our results as a definitive accuracy assessment of South Carolina gap analysis or gap analyses in general. The resolution of vertebrate models in our analyses (30-m minimum mapping unit) is far higher than recommended for final gap analysis products (EPA’s EMAP hexagon; a 635-km2 hexagonal grid; White et al. 1992). Additionally, the five-year span of our data is still relatively limited. Therefore, our accuracy assessment of South Carolina gap-analysis models is presented to demonstrate our methods, and is likely to underestimate the accuracy of lower resolution models. By providing analyses based on both land2007 J.A. LaBram, A.E. Peck, and C.R. Allen 653 cover classifications utilized by managers in the region of our study area, we demonstrate how the utilization of different classifications affects the assessment of accuracy in animal-distribution modeling. Despite the disparate classification systems used by the SRS maps and gap-analysis maps, and the different uses for which these maps were created, error rates were similar in our comparisons, though the sources of error differed. One way to improve vertebrate models is to determine the sources of errors. Two possible sources are erroneous habitat-association models, or species models that are too simplistic. For the former problem, monitoring and sampling programs can provide information with enough spatial and temporal breadth to refine habitat-association models. In the latter case, models can be improved utilizing current knowledge that blends landscape ecology and population viability. Inclusion of landscape metrics may improve species models and give the user more confidence in management decisions based on output of the models. For example, Allen et al. (2001b) incorporated minimum critical-area criteria into species models to reduce commission errors arising from modeling an animal as present in a patch too small or disconnected to support a viable population. Most likely, commission errors propagate from a combination of these sources. Explicit consideration of uncertainty in gap-analysis models would be a great improvement. Our methods do provide an accuracy assessment of gap-analysis models. Further refinement of the vertebratemodeling process and investigation of sources of model error will improve the accuracy of predictive models critical for the application of gap analyses results to conservation decision making. Acknowledgments The South Carolina Cooperative Fish and Wildlife Research Unit is jointly supported by a cooperative agreement among the United States Geological Survey, the South Carolina Department of Natural Resources, Clemson University, and the Wildlife Management Institute. The Nebraska Cooperative Fish and Wildlife Research Unit is jointly supported by a cooperative agreement between the United States Geological Survey, the Nebraska Game and Parks Commission, the University of Nebraska-Lincoln, the United States Fish and Wildlife Service and the Wildlife Management Institute. Funding was provided by the Department of Energy-Savannah River Operations Office through the US Forest Service Savannah River under Interagency Agreement DE-IA09-00SR22188. We would like to thank L. Moore, W. Jarvis, and D. Imm for facilitating this project. J. Bock, S. Brobst, J. Cassell, J. LaPointe, Q. Lupton, J. Oldroyd, B. Roberts, B. Schlachter, D. Schwalm, and B. Timm aided in data collection. A. Garmestani, V. Egger, and B. Weeks reviewed an earlier draft of this manuscript. Literature Cited Allen C.R., L.G Pearlstine, D.P. Wojcik, and W.M. Kitchens. 2001a. The spatial distribution of diversity between disparate taxa: Spatial correspondence between mammals and ants across south Florida, USA. Landscape Ecology 16:453–464. 654 Southeastern Naturalist Vol. 6, No. 4 Allen, C.R., L.G. Pearlstine, and W.M. Kitchens. 2001b. Modeling viable mammal populations in gap analyses. Biological Conservation 99:135–144. Boone, R.B., and W.B. Krohn. 1999. Modeling the occurrence of bird species: Are the errors predictable? Ecological Applications 9:835–848. Boone, R.B., and W.B. Krohn. 2000. Predicting broad-scale occurrences of vertebrates in patchy landscapes. Landscape Ecology 15:63–74. Edwards, T.C., Jr., E.T. Deshler., D. Foster, and G.G. Moisen. 1996. Adequacy of wildlife habitat relation models for estimating spatial distributions of terrestrial vertebrates. Conservation Biology 10:263–270. Fielding, A.H. 2002. What are the appropriate characteristics of an accuracy measure? Pp. 271–280, In J.M. Scott, P.J. Heglund, M.L. Morrison, J.B. Haufler, M.G. Raphael, W.A. Wall, and F.B. Samson (Eds.). Predicting Species Occurrences. Island Press, Washington, DC. Gibbons, J.W., and D.H. Bennet. 1974. Determination of anuran terrestrial activity patterns by a drift-fence method. Copeia 1:236–243. Gibbons, J.W., V.J. Burke, J.E. Lovich, R.D. Semlitsch, T.D. Tuberville, J.R. Bodie, J.L. Greene, P.H. Niewiarowski, H.H. Whiteman, D.E. Scott, J.H.K. Pechmann, C.R. Harrison, S.H. Bennett, J.D. Krenz, M.S. Mills, K.A. Buhlmann, J.R. Lee, R.A. Seigel, A.D. Tucker, T.M. Mills, T. Lamb, M.E. Dorcas, J.D. Congdon, M.H. Smith, D.H. Nelson, M.B. Dietsch, H.G. Hanlin, J.A. Ott, and D.J. Karapatakis. 1997. Perceptions of species abundance, distribution, and diversity: Lessons from four decades of sampling on a government managed reserve. Environmental Management 21:259–268. Golley F.B., J.B. Gentry, L.D. Caldwell, and L.B. Davenport, Jr. 1965. Number and variety of small mammals on the AEC Savannah River Plant. Journal of Mammalogy 46:1–18. Imm, D. 1997. ArcView classification of the land covers of the Savannah River Site. USDA Forest Service. Savannah River Institute, New Ellenton, SC. Jennings, M.D. 2000. Gap analysis: Concepts, methods, and recent results. Landscape Ecology 15:5–20. Kiester ,A.R., J.M. Scott, B. Csuti, R.F. Noss, B. Butterfield, K. Sahr, and D. White. 1996. Conservation prioritization using gap data. Conservation Biology 10:1332–1342. Margules, C.R., and R.L. Pressey. 2000. Systematic conservation planning. Nature 405:243–253. Martin, K.J., and B.C. McComb. 2003. Amphibian habitat associations at patch and landscape scales in the central Oregon coast range. Journal of Wildlife Management 67:672–683. Scott, J.M., B. Csuti, J.D. Jacobi, and J.E. Estes. 1987. Species richness: A geographic approach to protecting future biological diversity. BioScience 37:782– 788. Scott, J.M., F. Davis, B. Csuti, R. Noss, B. Butterfield, C. Groves, H. Anderson, S. Caicco, F. Derchia, T.C. Edwards, Jr., J. Ulliman, and R.G. Wright. 1993. Gap analysis: A geographical approach to protection of biological diversity. Wildlife Monograph 123. Schaefer, S.M., and W.B. Krohn. 2002. Predicting vertebrate occurrences from species habitat associations: Improving the interpretation of commission error rates. Pp. 271–280, In J.M. Scott, P.J. Heglund, M.L. Morrison, J.B. Haufler, M.G. Raphael, W.A. Wall, and F.B. Samson (Eds.). Predicting Species Occurrences. Island Press, Washington, DC. 2007 J.A. LaBram, A.E. Peck, and C.R. Allen 655 Schmidt ,E.V., D. Otis, C. Aulbach, F. Tian, J.D. Scurry, Y. Alger, F.G.F. Smith, D. Fairey, and D. Gordon. 2001. The South Carolina gap-analysis project: Final report. South Carolina Cooperative Fish and Wildlife Research Unit, USGS. Clemson, SC. White, D., A.J. Kimerling, and W.S. Overton. 1992. Cartographic and geometric components of a global sampling design for environmental monitoring. Cartography and Geographic Information Systems 19:5–22. Workman, S.W., and K.W. McLeod. 1990. Vegetation of the Savannah River Site: Major community types. Savannah River Ecology Laboratory. Publication SRONERP- 19, National Environmental Research Park Program. Savannah River Ecology Laboratory, Aiken, SC. 137 pp. 656 Southeastern Naturalist Vol. 6, No. 4 Appendix 1. Cross-walk table used for conversion between Savannah River Site (SRS) and gap-analysis land-cover classifications. Gap land cover SRS land cover Pocosin/bay Bay Bottomland/floodplain forest Bottomland hardwood Wet evergreen Bottomland hardwood Dry deciduous forest/woodland Hardwood slope Mesic deciduous forest/woodland Hardwood slope Dry mixed forest/woodland Mixed Mesic mixed forest/woodland Mixed Closed canopy evergreen forest/woodland Planted pine Needle-leaved evergreen mixed forest/woodland Sandhill Pine woodland Sandhill Swamp Swamp Freshwater Water Marsh/emergent wetland Not applicable Wet scrub/shrub thicket Not applicable Dry scrub/shrub thicket Not applicable Sandy bare soil Not applicable Open canopy/recently cleared forest Not applicable Aquatic vegetation Not applicable Grassland/pasture Not applicable Cultivated land Not applicable Urban development Not applicable Urban residential Not applicable