Predicted Vertebrate Species Distributions and Richness
Suddenly, as rare things will, it vanished.
- R. Browning
3.1 Background
All species range maps are predictions about the occurrence of those species within a particular area (Csuti 1994). Traditionally, predicted distributions of species have been derived from sample collections made at individual points or in grids (Scott et al. 1993). This approach typically relies on the location of specimens, and includes limited information on the ecological conditions that favor the presence of the species. Habitat features, such as vegetation, also have been used in conservation and management to predict species presence (Verner et al. 1986, Morrison et al. 1992) and can enhance traditional approaches despite some limitations (Scott et al. 1993). In this chapter, we describe vertebrate species distributions predicted using both point locality records and habitat conditions.
The purpose of the vertebrate species maps developed for gap analysis is to provide more precise information about the current distribution of individual native species within their general ranges. With this information, better estimates can be made about the actual amount of habitat area and the nature of its configuration. Gap analysis uses the predicted distributions of native vertebrate species to evaluate their conservation status relative to existing land management (Scott et al. 1993). Previous to this effort there were no maps available, digital or otherwise, showing the likely present-day distribution of species, by habitat, across their ranges in Wyoming. Because of this, ordinary species (i.e., those not threatened with extinction or not managed as game animals) are generally not given sufficient consideration in land-use decisions. As incremental loss of habitat occurs, the decline of such species can, and does, result in an accelerating increase in numbers of threatened or endangered species. Creating a consistent spatial framework for storing, retrieving, analyzing, and updating our knowledge about the status of each vertebrate species is one of the most necessary and basic elements for preventing further erosion of biological resources.
Besides gap analysis, the maps of vertebrate species distributions described in this chapter may be used to answer a wide variety of management, planning, and research questions relating to individual species or groups of species. In addition to the maps, great utility may be found in the consolidated species locality records and literature that are assembled into databases used to produce the maps.
3.2 Methods
The modeling approach used to predict vertebrate distributions in Wyoming included five steps. First, criteria were developed to choose which species would be included in the current analyses. Second, the distributional limits of each species were defined by recording the species' presence or absence within the Environmental Protection Agency's (EPA) hexagon grid system for Wyoming (White et al. 1992). Third, we developed a Wildlife-Habitat Relationships (WHR) database which defined the affinities of terrestrial vertebrate species to habitat features including land cover types, riparian/aquatic habitats, and elevation. Fourth, the hexagon and WHR databases were used in a GIS-modeling process which assigned species to habitat polygons based on their known or expected occurrence within hexagons and their association to habitat features. Finally, hardcopy maps of predicted species distributions were reviewed by over 60 acknowledged experts including state and federal biologists, university professors, and Audubon Society members.
3.2.1 Criteria for Including Species in WY-GAP
There are over 600 terrestrial vertebrate species known to occur in Wyoming (Wyoming Natural Diversity Database 1994). Many of these species are rare or accidental migratory birds which have been documented within the state only a few times. We developed the following set of criteria to include species in our analysis. Species were included if they were:
1) year-round, summer, or winter resident as defined by Oakleaf et al. (1992),
2) neotropical migratory bird as defined by Oakleaf et al. (1992),
3) migratory shorebird or waterfowl as defined by Oakleaf et al. (1992),
4) exotic game species as defined by Wyoming Natural Diversity Database (1994),
5) species or sub-species of management concern (listed as endangered, threatened, candidate, sensitive, or TNC State Rank of 2),
6) sub-species recognized as the only representative of its species in Wyoming,
7) all amphibian and reptile species and subspecies in Wyoming as listed by Baxter and Stone (1985).
Wyoming-specific field guides and atlases, in addition to the opinion of experts, were used to decide whether a species met these criteria. In particular, "accidental" or "rare" migrant birds, and exotic non-game mammals and birds were not included (Wyoming Natural Diversity Database 1994, Dorn and Dorn 1990, and Oakleaf et al. 1992). Some species, like the house mouse (Mus muculus) and the Norway rat (Rattus norvegicus) are not uncommon in Wyoming, but we did not include them in our analysis because they are non-native species. The taxonomy and nomenclature used to describe species was adopted from TNC and selected as a standard by the National GAP (Wilson and Reeder 1993, AOU Committee on Classification and Nomenclature 1983, Collins 1990, Frost 1985).
3.2.2 Occurrence of Species within Hexagons
Counties and latilongs are common units used to document the general location of species. Wyoming consists of 23 counties (average size = 10,950 km2) and 28 latilong blocks (average size = 9,004 km2). Using either of these geographic units to make species predictions would have overestimated distributions of species in cases where a species' range extended only partly into a county or latilong. To reduce this problem, we mapped the distributional limits of species using smaller, hexagon units (635 km2) which are part of a global hexagonal grid system developed by the EPA (White et al. 1992). Advantages to using the hexagon grid include its equal area sampling structure, its independence from political and administrative boundaries (resulting in more consistent mapping of animal distributions), and its hierarchical structure which can facilitate increasing or decreasing grid densities in future analyses (White et al. 1992).
Species were recorded within each of the 436 hexagons for Wyoming using 1 of 7 definitions (Table 3.1). We adopted the first 3 definitions of species occurrence from the Biodiversity Research Consortium (Master et al. 1995), which is a complementary effort coordinated by EPA's Habitat/Biodiversity Program whose objective is to identify areas of the country where risks to biodiversity are greatest. The remaining 4 definitions (Table 3.1) were developed to enhance the species-hexagon database and are shown as part of the vertebrate species maps (Merrill et al. 1996b). We used only the data classified in the first 4 categories to conduct our gap analysis. Statement of probabilities in these descriptors were used as guidelines to subjectively qualify the occurrence of a species within a hexagon consistent with the descriptions in Table 3.1. At this time, they do not represent a quantified analysis of the probability of occurrence. Future refinements to the database may allow a quantified probability statement of species occurrence.
Three primary sources of information were used to document the occurrence (or expected occurrence) of a species within a hexagon: (1) species locality records, (2) published range maps, and (3) the opinions of experts. Species locality records (i.e., recorded occurrences of observed, trapped, or killed individuals) were obtained from 16 existing wildlife databases collected from state and federal agencies, conservation groups, museums, and outdoor science schools in Wyoming (Table 3.2). Fifteen of the species databases were non-spatial, tabular databases which included Public Land Survey System (PLSS) descriptions or coordinates for the location of observed species. PLSS locational descriptions were converted to latitude-longitude coordinates for import into Arc/Info using a fortran program called TR-LL (Morgan and McNellis 1965). Hexagons encompassing locality records with a date > 1950 were coded as Confirmed, while those populated with locality records < 1950 were coded as Historical. Historical hexagons that were immediately adjacent to other hexagons coded as Confirmed, Probable, or Possible, were initially included within a species' current distribution. In cases where the historical hexagon was geographically isolated from a species' contiguous range, the hexagon was initially excluded from the species' current distribution, but was not removed from the species-hexagon range maps. Later, when expert reviewers examined the maps (see below), they were given the chance to modify historical records as necessary.
CONFIRMED (C) The species is confidently assumed (> 95% certain) or known to occur in the hexagon. Information sources confirming occurrence within a hexagon included species locality records and expert opinion.
PREDICTED (PR) The species is predicted to occur in the hexagon based on the "fact-pattern" (i.e., presence of suitable habitat or conditions and historical record and/or presence in adjacent hexagons[s]); at least 80% certain that the species occurs in the hexagon. Information sources used to document a species within a hexagon included expert opinion only.
POSSIBLE (PO) The species possibly or potentially occurs in the hexagon; its estimated likelihood of occurrence in the hexagon is thought to be between 80% and 10% (or less for extremely rare species where suitable habitat or conditions may be present). Information sources used to document a species as Possible within a hexagon included expert opinion and published range maps.
HISTORICAL (H) The species is confidently assumed (> 95% certain) or known to have occurred in the
(Included) hexagon prior to 1950. The historical presence within the hexagon was included as part of the species' current distribution. Information sources used to document a species as historical (included) within a hexagon included species locality records and expert opinion.
HISTORICAL (Hx) (excluded) The species is confidently assumed (> 95% certain) or known to have occurred in the hexagon prior to 1950. The historical presence within the hexagon was not included as part of the species' current distribution. Information sources used to document a species as historical (excluded) within a hexagon included species locality records and expert opinion.
QUESTIONABLE (?) (excluded) The occurrence of the species within a hexagon was still in question after having been reviewed by experts. Hexagons coded as questionable were not included as part of the species' current distribution. Information sources used to document a species as questionable within a hexagon included expert opinion only.
EXCLUDED (X) The documented occurrence of a species was excluded by expert review after once having been coded as confirmed, predicted, or possible. Information sources used to document a species as excluded within a hexagon included expert opinion only.
Range maps published by Clark and Stromberg (1987) and Baxter and Stone (1985) also were used to document the occurrence of species within hexagons for mammal and herptile species. Wyoming-specific range maps for birds did not exist. For mammals and herptiles, the geographic range of each species was manually transferred from paper maps to the computerized hexagon grid using a mouse to select the hexagons which overlapped with range map polygons. Hexagons populated in this manner were coded as Possible.
Table 3.2. Databases used to document species occurrence within hexagons.
| Database | Source | No. of Records | Date of Acquisition | Wildlife Observation System* | Wyoming Game & Fish Department | 666,567 | 5/92 |
| Element occurrence Database* | Wyoming Natural Diversity Database | 2,880 | 7/94 |
| Vertebrate Museum Database | Museum Databases | 4,389 | |
| Wildlife Observation Database | Grand Teton National Park | 6,668 | 3/92 |
| Devils Tower Fauna Database | Devils Tower National Monument | 199 | 4/92 |
| Green River Sage Lek Database | BLM -Green River Resource Area | 128 | 9/92 |
| Green River Raptor Database | BLM -Green River Resource Area | 1,577 | 9/92 |
| Lander Raptor Database | BLM -Lander Resource Area | 162 | 3/92 |
| Kemmerer Raptor Database | BLM -Kemmerer Resource Area | 125 | 2/92 |
| Cody Raptor Database | BLM -Cody Resource Area | 1,060 | 7/92 |
| Cody Nongame Bird Database | BLM -Cody Resource Area | 225 | 7/92 |
| Grizzly Bear Database | NPS -Interagency Study Team | 9,338 | 3/92 |
| M.A.P.S database | Teton Science School | 332 | 10/92 |
| Amphibian Survey Database | Teton Science School | 35 | 10/92 |
| Wind River Wildlife Database | U.S. Fish & Wildlife Service | 2,775 | 3/93 |
| Great Divide RA Raptor Database | BLM - Great Divide Resource Area | 3,266 | 1993 |
* Includes additional records from 1994 or 1995 for specific areas and/or taxonomic groups
Species-hexagon range maps developed from locality records and published range maps were reviewed by over 60 acknowledged experts consisting of federal and state biologists, university professors, and Audubon Society members (Appendix 3.1). Reviewers were asked to check, and if necessary, correct the hexagon occurrences that were based on questionable locality records or range maps. Reviewers were also given the opportunity to add animal occurrences within hexagons using the definitions in Table 3.1. The 1994 review of the species-hexagon range maps represented the first of two distinct map reviews.
Maps of species richness within hexagons were derived by totaling the number of species documented/expected to occur within hexagons and do not reflect species distributions modeled using habitat associations. For this analysis, we used only species occurrences which qualified as one of the first four definitions in Table 3.1. The five categories of species richness identified in the maps were determined using an equal-interval classification.
In developing the database for species distributions for Wyoming, we did not differentiate between breeding and winter ranges for bird species. Seasonal information for birds existed only by latilong blocks and interpolation of breeding ranges to the hexagon level within these larger units would have represented an unreasonable refinement of scale. The refinement to seasonal ranges also would have complicated the review process beyond reasonable time demands of the reviewers since most bird reviewers reviewed all 291 bird distribution maps. Further, the
conservation of bird species must consider the maintenance of habitat throughout the year (Csuti 1996). Future refinements to the bird distribution maps should separate breeding and wintering ranges and incorporate new information on seasonal habitat use by individual bird species.
3.2.3 Wildlife-Habitat Relationships
Once species were documented within the appropriate hexagons, we assigned species to spatially-explicit polygons of mapped habitat. We use the term habitat to represent areas characterized by several environmental features, specifically land cover, elevation, and the presence of riparian/aquatic features. WHR databases for Wyoming that existed at the initiation of this project contained information that was too general to predict species within the land cover types we mapped. For this reason, we compiled detailed WHR information and entered it into the Biological Conservation Database (BCD) developed and maintained by TNC. Vertebrate characterization abstracts within the BCD were used to document: (1) the associations of individual species to habitats, (2) sources of information which defined species-habitat associations, and (3) reviewer's notes on special habitat requirements which may limit the species' distribution within Wyoming.
Information used to complete the vertebrate characterization abstracts came from existing WHR databases, published and unpublished literature, and individuals having expert knowledge of a particular species. The majority of the WHR information was provided by the Colorado Division of Wildlife (Schrupp and Cade 1990) who developed a tabular database from an existing WHR publication (U.S. Forest Service 1981). In addition, we used WHR information from the UT-GAP and regional species guides to check and supplement WHRs defined by Colorado. We also completed an extensive literature review on habitat associations for 103 species of concern (i.e., federally listed as an endangered, threatened, or candidate species, USFS sensitive species, WGFD priority species, or species having a TNC state rank of 2) in Wyoming (Garber 1995) and on Wyoming species that were not recorded in the Colorado database. Lastly, information on species-habitat associations was recorded from expert reviewers who reviewed the species-habitat associations as part of the second review of the species distribution maps (see section 3.2.5). WHR information compiled from these three sources was input into the BCD and also Arc/Info as three separate species-habitat "matrices" and linked to the 3 GIS habitat layers described below to model species distributions.
Land Cover Matrix
Many of the documented associations between species and land cover types were derived from the Colorado database. A crosswalk between similar land cover types was developed to facilitate the transfer of information from the Colorado database to Arc/Info (Merrill et al. 1996b). Some of Colorado's WHR information was too specific, and in other cases, too general to be matched to Wyoming's land cover types. As a result, we did not include any of the Colorado habitats in our database that could not be confidently matched with Wyoming land cover types. The crosswalk did match land cover/habitat types from the Colorado database to 39
of the 41 land cover types mapped for Wyoming. One of the missing types (greasewood) was matched from UT-GAP's WHR database, and the other missing type, burned conifer, was added where appropriate to species' associations through literature and expert review.
Riparian/Aquatic Feature Matrix
Riparian areas are defined as lands adjacent to streams and rivers where vegetation is strongly influenced by the presence of water. In the arid west, riparian areas can constitute less than 1% of landscape (Chaney et al. 1991), yet their importance to the distribution of vertebrate species is far out of proportion to the area they represent (Gerhart and Olsen 1982; Szaro and Jackle 1985; Szaro and Belfit 1986, 1987; Finch 1989). Because riparian areas are often small and linear by nature they are difficult to map at the scale at which the land cover is produced (Csuti 1994), and as a result GAP has adopted a 40-ha MMU standard for delineating riparian and other wetland features in the land cover map (Jennings 1993). Although this is a significant reduction from the 100-ha unit used in mapping upland land cover types, many small riparian and aquatic features still are not distinguished from upland cover types. In order to better predict the distributions of species associated with riparian and aquatic areas, we modeled riparian areas by creating buffers around hydrographic (surface water) features. A similar approach was taken by the Idaho GAP (ID-GAP) and UT-GAP (Scott et al. 1993, Edwards et al. 1995). This approach, refined by varying the width of the buffer according to stream order, allowed us to approximate the location and amount of area in riparian vegetation zones. Unlike other riparian mapping approaches, such as aerial videography, it did not allow us to determine the vegetative composition or structure within the buffer. Another major limitation with our approach is that it did not identify wetlands associated with groundwater, which constitutes a significant proportion of total wetland habitat.
The riparian/aquatic model was developed in four steps. First, hydrographic features (streams, lakes, ponds, reservoirs) were extracted from USGS 1:100,000 scale digital line graphs (DLGs). Second, streams from the DLGs were then ordered using the automated Strahler stream ordering method developed by the USGS (Lanfear 1990). Third, buffer widths for each of the resultant seven stream orders and wide rivers (rivers represented by two shorelines in the DLGs) were determined by overlaying hydrographic features on a Landsat TM image of the southeast corner of the state (Path 34 / Row 31, 17 June 1991). Widths of the riparian vegetation were measured at approximately 1-km intervals along every perennial stream within the extent of the TM scene. Buffer widths were averaged by order (Table 3.3) and values rounded to the nearest 10 m were used for the buffer widths.
To refine predicted distributions of vertebrate species associated with riparian areas, the final step in developing the riparian model was to assign land cover types to the buffered areas. An initial attempt to classify land cover types within the buffered areas from spectral characteristics of Landsat images was not completed because sufficient ancillary data on riparian vegetation were not available and the field reconnaissance required for this interpretation required a time commitment beyond the scope of this project. The approach we used was to interpret riparian vegetation characteristics based on the land cover map (Chapter 2). Where a buffer intersected a polygon with a primary riparian cover type (cover type with largest area within the polygon) or secondary riparian cover type (cover type with second largest area within the polygon), that riparian cover type was assigned to the buffer. If there were no riparian cover types associated with the land cover polygon, the buffer segment of the polygon was designated as "unclassified riparian". We note that the riparian classification associated with the 2-ha MMU riparian map is limited because of the low resolution of the land cover map from which it was derived.
Table 3.3. Mean, standard deviation, and sample size (n) of riparian buffer widths measured on TM imagery for the southeastern portion of Wyoming.
| Stream Order | Mean | Standard deviation | n | Buffer Width (m) |
| 1 | 38.9 | 9.33 | 222 | 40 |
| 2 | 40.2 | 6.19 | 137 | 40 |
| 3 | 59.6 | 7.86 | 8 | 60 |
| 4 | 91.3 | 10.26 | 87 | 90 |
| 5 | 121.3 | 10.50 | 62 | 120 |
| 6 | 148.6 | 11.46 | 66 | 150 |
| 7 | 210.0 | 13.19 | 90 | 210 |
| Wide Rivers | 305.7 | 42.72 | 90 | 300 |
| Lakes/reservoirs/ponds | n/a | n/a | n/a | 90 |
Following the development of the riparian model, it was incorporated with the main land cover map to be used in the prediction of species distributions. We combined information on the presence of riparian/aquatic features from the land cover map and the riparian/aquatic model to develop a matrix which recorded the presence or absence of species within riparian and aquatic features (Appendix 3.2). Species associated with any of the mapped riparian habitats (forest-, shrub-, and grass-dominated riparian) in the land cover map were also assigned to modeled riparian types in which the riparian vegetation was unclassified. Our reviewers agreed that despite the fact that the majority of the modeled riparian was unclassified, associating species to the unclassified riparian was still likely to portray a more accurate representation of the species distribution than the riparian types in the land cover map alone, and this was confirmed in our accuracy assessment of riparian species (see section 3.4).
Because of the limitations of the riparian/aquatic model, discussed in detail in Appendix 3.3., we emphasize that its sole purpose is to improve the predicted distributions of vertebrate species, and it should not be considered a "stand alone" map of riparian/aquatic areas in Wyoming.
Elevation Matrix
The third habitat characteristic used to refine species distributions was elevation. The elevational gradient in Wyoming ranges from approximately 973 to 4185 m and introduces climatic zonation which often limits the distribution of vertebrate species. Elevational ranges used by vertebrate species were obtained from the Colorado database or literature sources and
summarized within the vertebrate characterization abstracts. In cases where there were no specific literature sources documenting species-elevation associations for Wyoming, sources from other states within the region (CO, MT, ID, UT) were used. In these cases, we adjusted the elevational range documented in the literature to similar ranges in Wyoming using the treeline elevation as a reference for adjustment. The rate of decline of the treeline between 40o N and 55o N latitude is approximately 100 m elevation per degree of latitude (Peet 1988, Driese et al. in press). For instance, sources of minimum and maximum elevation ranges from Colorado, usually Armstrong (1972) or Bailey and Niedrach (1965), were each reduced by 400 m for Wyoming species because the difference in the mean latitudes of Colorado (39o N) and Wyoming (43o N) was 4 degrees.
The species-elevation matrix was used in conjunction with a GIS layer of contoured elevation to restrict species distributions. The elevation layer was derived from a Digital Elevation Model (DEM) of 90-meter resolution and was produced with a contour interval of 150 m, chosen because it corresponded closely to values given for elevational ranges of species reported by Clark and Stromberg (1987) and other literature sources.
3.2.4 GIS Modeling of Species Habitat and Distributions
The GIS layers of hexagons, land cover, elevation, and riparian/aquatic areas were combined in a GIS overlay process to develop a composite "habitat layer" for predicting species distributions. In addition, we produced a similar layer excluding the modeled riparian/aquatic areas (but still including mapped riparian and aquatic features from the land cover map) to assess the effect that modeled riparian areas might have on predicted species distributions (see section 3.4 Accuracy Assessment). In the union process "sliver" polygons 0.2 ha were eliminated to remove small, insignificant polygons and to simplify the composite layer. Species occurrence was predicted in habitat polygons if: (1) species occurrence was documented in the hexagon, (2) suitable land cover was present, and (3) the land cover was within the documented elevational range for the species. Both the primary (land cover occupying the largest proportion of the area of each polygon) and secondary (land cover occupying the second largest proportion of the area of each polygon) types were used to place a species in a polygon of associated habitat. For reporting purposes, we summarized the area of a species' predicted distribution based on primary and secondary habitat types separately in Merrill et al. 1996b, but our analysis in Chapter 5 does not differentiate between the two designations and reflects the largest extent of the species' range.
Our modeling process sometimes resulted in species distributions which ended abruptly at the edge of hexagons, even when suitable habitat was present outside of the hexagon where species occurrence was not documented. To mitigate this problem, species distributions were extrapolated beyond the hexagon boundaries into immediately adjacent polygons of suitable habitat.
3.2.5 Expert Review of Species Distribution Maps
We conducted a second review of vertebrate species distribution maps in 1995. In this review, participants (Appendix 3.1) were asked to review both the WHR information used to predict species distributions and an 11 x 7.5-in color map of each species distribution. Initial attempts to have the reviewers provide an accuracy rating for each map were abandoned because it resulted in excessive demands on the reviewers' time. Upon completion of the expert review, suggested changes were incorporated into the databases.
3.2.6 Edgematching Species Distributions with Adjacent States
WY-GAP species-habitat associations were checked for consistency with UT-GAP species-habitat associations when we incorporated WHRs from both states into our species database. Comparison of associations between WY-GAP and CO-GAP were not possible at the time that the Colorado WHR was crosswalked to Wyoming land cover types, because the land cover classification for CO-GAP had not yet been developed. Since that time, spatial edge-matching of land cover types has been completed for Utah and Colorado. We expect that there will be some discrepancies in the distributions of species due to the different geographic units used by each state to define species ranges (e.g. latilong blocks, counties, hexagons).
3.3 Results
Distributions of 445 terrestrial vertebrate species were predicted including 291 birds, 116 mammals, 26 reptiles, and 12 amphibians. Of the 445 species, 370 species (83%) had an association with riparian/aquatic habitats, and 291 species (65%) had specific minimum and maximum elevational limits, documented in literature or by the reviewers (Appendix 3.2). A listing of WHRs, source references, habitat area summaries, and statewide distribution maps for each species are included in an atlas that is separate from this report (Merrill et al. 1996b). However, we give an example of this information in Appendix 3.3 of this report.
Total species richness within hexagons ranged from 113 to 333 with a mean of 179 (Fig. 3.1). Species richness appeared bimodal reflecting the low species richness of basins and high species richness of mountainous areas in the state. Hexagons containing the highest diversity of terrestrial vertebrate species were located near Jackson Hole (297, 297, and 303 species), Casper (333 species), and Buffalo (326 species) (Fig. 3.2).
Fig. 3.1. Frequency distribution of total vertebrate species
richness within 436 equal area hexagons located in Wyoming.
(figure not available in html)
Figure 3.2. Predicted distribution of total vertebrate species richness within hexagons across Wyoming.
Avian species richness ranged from 48 to 257 per hexagon (Fig. 3.3) with the highest species occurring in hexagons around Jackson (218, 219, and 225), Buffalo (249), and Casper (257) (Fig. 3.4).
Figure 3.4. Predicted distribution of species richness of birds within hexagons across Wyoming.
Mammalian species richness ranged from 49 to 75 species (Fig. 3.3) with the highest richness occurring in the mountainous regions and the lowest richness in the basins (Fig 3.5).
Figure 3.5 Predicted distribution of species richness of mammals within hexagons across Wyoming.
Only 3 to 7 amphibian species occurred per hexagon (Fig. 3.3) across Wyoming with the most diverse areas occurring near the towns of Laramie (7) and Douglas (7) (Fig. 3.6).
Figure 3.6 Predicted distribution of species richness of amphibians within hexagons across Wyoming.
Reptilian species richness ranged from 1 to 18 species (Fig 3.3), and was greatest in the eastern Platte river valley (15-18) and scattered hexagons near the Black Hills region (15) (Fig 3.7).
Figure 3.7 Predicted distribution of species richness of reptiles within hexagons across Wyoming.
3.4 Accuracy Assessment
Properly designed, long-term field surveys provide the best source of independent data to assess our predicted vertebrate distributions. The large size of Wyoming, the high number of vertebrate species in this analysis, and the spatial-temporal problems associated with interpolating animal ranges from survey records are all difficult to address with limited personnel, funds, and perhaps most importantly, time (Csuti 1994). We chose to follow an approach used by UT-GAP (Edwards et al. 1995), based on comparison with existing species checklists, to assess our predicted vertebrate distributions.
3.4.1 Methods
We compared lists of predicted species to checklists of terrestrial vertebrate species developed for 2 national parks/monuments, 2 wildlife refuges, 2 national forests-grasslands, 1 national recreation area, and a bird observation checklist developed for Jackson Hole which encompassed Grand Teton National Park (Fig. 3.8, Table 3.4). The species checklists compiled for all the areas were derived from published and unpublished reports that were not used directly in developing the WY-GAP databases. Of the 8 test areas, only 3 of them (Devils Tower National Monument, Yellowstone National Park, and the Bighorn National Recreation Area) had complete checklists for all 4 taxonomic groups. The other areas had checklists for either birds or mammals.
Table 3.4. Location, checklist source, size (ha), elevation range
(m), and predominant habitats of the 8 areas used to assess the
accuracy of predicted distributions of vertebrate species within
Wyoming.
(table not available in html; see figure above for location)
Number of omission errors (No), defined as the number of species not included on our list of predicted species, but present on the area's corresponding field checklist, and number of commission errors (Nc), defined as the number of predicted species included on our list, but not contained on the area's corresponding field checklist, were tabulated for all 8 areas. The accuracy of our predictions of species occurrences was derived by dividing the number of species which matched both lists (Nm) by the total number (Nt) of species contained on both lists. To determine the influence of the modeling strategies on the accuracy of species distributions, we conducted the accuracy assessment based on results generated both with and without inclusion of modeled riparian/aquatic areas and with and without the inclusion of species distributed within "Possible" hexagons.
3.4.2 Results
When species predictions were based on modeled riparian areas, our accuracy averaged 79.5% across sites and taxa (Table 3.5). The exclusion of modeled riparian areas generally had little to no effect on accuracy of predicting reptiles and mammals, but reduced the accuracy of predicting the occurrence of birds and amphibians at some sites by 10 - 30%. The reduction in accuracy was the result of species, such as waterfowl, shorebirds and riparian- or water-dependent birds and amphibians, which were omitted for one of two reasons. One third of these cases were species associated with cover types that were not mapped within the 40 ha MMU of the land cover map, and were represented only by modeled riparian within these sites. The remaining cases occurred when species were not recorded within the hexagons encompassing the assessment sites. The species were recorded in hexagons adjacent to the accuracy assessment site, and their habitat was extended into the site along corridors of modeled riparian because of the "smoothing" process applied in the habitat modeling procedure (see Section 3.2.4).
Errors of omission averaged 12.2% (0 - 36.6%) for all taxonomic groups, and were often high for birds (Table 3.5), indicating that our models tended to under-predict the presence of bird species. Of the 206 total bird omissions, only two birds, the blue grouse (Dendragapus obscurus) and northern saw-whet owl (Aegolius acadicus) were omitted because of an apparently erroneous restriction in elevation. Three birds, the long-billed dowitcher (Limnodromus scolopaceus), rosy finch (Leucosticte arctoa), and white-winged crossbill (Loxia leucoptera), were omitted because none of their associated land cover types were mapped within the areas. The remaining 201 (96%) omission errors were the result of no recorded occurrence of the species within any of the hexagons encompassing the accuracy assessment area(s).
Table 3.5 Commission errors, omission errors, matches, and percent
accuracy of predictged species occurrences.
(this table not available in html)
The highest omission error occurred for birds in the Bighorn Canyon National Recreation Area (BCNRA). The northern portion of BCNRA extends into Montana and contains additional habitat types not present in the Wyoming portion of the site, which may affect bird species composition (Anderson et al. 1987). The species checklist used in this comparison was compiled for both the Wyoming and Montana portions of BCNRA and it was not possible to determine which species were present only in the Wyoming portion of the BCNRA from the species check list. Errors of omission were also high for the Bighorn National Forest area, probably due to problems in interpreting the actual boundaries of the area used to compile the checklist, which extended beyond the official boundary of the National Forest.
Errors of commission averaged 8.3% (0 - 34.8%) for all taxonomic groups and were highest for mammals (Table 3.5), indicating that our models tended to over-predict the presence of mammal species. Most of the commission errors for mammals were the result of over-predictions of bat, rodent and rabbit/hare species. For example, of the 40 predicted to be present, but not on the checklists, 31 species were either bats, rodents or rabbits/hares. In particular, at Devils Tower National Monument, which had the highest commission error of the four accuracy assessment sites for mammals, 22 of the 23 committed species were within these taxa. Over-predicted distributions of bat, rodent, and rabbit/hare species were related to a lack of point locality data used to define range extent. Lack of information resulted in the inclusion of many hexagons labeled as "Possible" in the distributions of these taxa because published range maps showed these species widely distributed across large portions of the state. The remaining nine commission errors included species such as the wolverine, marten, lynx and black bear. These species were incorrectly predicted to occur at National Elk Refuge or DTNM because their habitat existed within the hexagons encompassing these sites, though the species had never actually been documented within the boundaries of the sites (or had been extirpated from the sites).
Exclusion of Possible hexagons in predicting species distributions generally reduced the number of commission errors for species with an uncommon or unknown distribution, but significantly increased the omission errors for widely-distributed and common species (e.g., thirteen-lined ground squirrel, Nuttall's cottontail, and striped skunk). Exclusion of Possible hexagons increased the accuracy rating of mammals at two sites, but it also greatly reduced the accuracy of mammals at the other two sites (Table 3.6). The exclusion of Possible reduced the accuracy of our predictions of amphibian and reptilian species by an average of 37%. However, there was little substantive effect on the accuracy of bird predictions.
Table 3.6. Comparison of accuracy assessment results across 8 areas, with (P) and without (NP) the use of the "possible" designation of species occurrence within hexagons which were used to develop species distribution maps. Dashed lines indicate that a checklist for the taxonomic group was not available for that area.
We did not find strong evidence that error rates decreased with increasing size of the assessment area (Figure 3.9; not available in html) as suggested by UT-GAP (Edwards et al. 1996). The number of assessment sites available to us was low and incorporation of Wyoming's results with results from other state gap analysis projects may provide a better analysis of these patterns.
3.5 Limitations and Discussion
Successful assessment of the protection status of species through gap analysis requires accurate mapping of species distributions. The goal set by National GAP is to produce maps that predict species occurrences with an overall accuracy of 80% or higher (Csuti 1994). Our average accuracy (79.5 %) fell just at or below this level. With one exception, accuracy ratings of individual sites were within the range reported by UT-GAP (Edwards et al. 1996). The exception was the Bighorn Canyon National Recreation Area which included areas outside Wyoming that were not modeled. UT-GAP reported accuracy rates that, on average, were highest among birds and mammals while we found our accuracy was highest for amphibians and lowest for mammals. Part of the GAP effort is to determine for which species landscape-scale modeling efforts are least likely to apply and, therefore, would be inappropriate (Scott et al. 1996). In mapping and reviewing species distributions in Wyoming, we identified species for which data were insufficient for modeling purposes and found several important factors that may contribute to potential errors in these maps that should be recognized when using them. Modeling species distribution was a two step process, and errors were introduced when mapping species ranges within hexagons, as well as when modeling species distributions using habitat associations.
3.5.1 Species Distributions Within Hexagons
Limits to a species' range were determined by defining the presence of a species within hexagons using locality records. For many species there were an inadequate number of locality records to confidently determine its range. For example, sightings of the fisher (Martes pennanti) were uncommon and often questioned by our reviewers resulting in limited data for describing the overall range of the fisher. In particular, there was a dearth of information for many bat species and some small mammals which was most likely due to their inconspicuous and/or nocturnal behavior. In one instance, we did not have sufficient new data to map the distribution of the three, recently-recognized species of rosy finch (Leucosticte tephrocotis, L. atrata, L. australis) because existing locality records for the rosy finch did not differentiate between these new species.
To compensate for the lack of locality records for amphibians, reptiles, and mammals, we used existing range maps from Baxter and Stone (1985) and Clark and Stromberg (1987) to assign the presence of a species in a hexagon and labeled these hexagons as Possible. In contrast, range maps did not exist for birds and we relied solely on point locality records and expert opinion to determine ranges of birds. During the review process, we found that the reviewers of the maps were hesitant to extrapolate the range of birds far beyond known occurrences or to contract the ranges of amphibians, reptiles and mammals from published range maps. As a result, the number of hexagons designated as Possible is much lower for birds than for herptiles and mammals and maps of bird distributions are more fragmented. These differences may affect future management area evaluations. For example, Freitag et al. (1996) found that in evaluating the existing conservation reserve network in the Transvaal region of South Africa, the current system represented 66% of the hypothetical sites necessary to represent all species in the reserve system when based on point locality records, but only 38-54% when based on range maps. Which data source provides the most accurate representation of a species distribution is unknown since both types of data have their limitations (Freitag et al. 1995). Nonetheless, our accuracy assessment indicated that the inclusion of Possible hexagons increased the overall accuracy of the mammal and herptile distribution maps, and their exclusion had little effect on the accuracy of the bird distribution maps.
Distributions of some species were identified by reviewers as problematic due to possible misidentification in locality records where species' ranges overlap. Species with a high probability of misidentification included cottontail species (Sylvilagus floridanus, S. nuttallii, and S. audubonii); the least weasel (Mustela nivalis) and the ermine (M. erminea); the gray fox (Urocyon cinereoargenteus ocythous) and the swift fox (Vulpes velox velox); the Yuma myotis (Myotis yumanensis); the California myotis (Myotis californicus); the grasshopper sparrow (Ammodramus savannarum) and savannah sparrow (Passerculus sandichensis); and many of the empidonax flycatchers. Thus, the mapped distributions of these species should be used with some caution.
The point locality data, and the reviewers themselves, may have introduced biases into the distribution maps due to opportunistic rather than systematic sampling (i.e., uneven sampling). The location of species locality records collected in the field are undoubtedly influenced by population densities and existing transportation routes. The areas of highest diversity of birds (Fig. 3.4) were centered on the cities of Casper, Jackson and Buffalo, where there are active Audubon Society chapters. Members from these chapters also participated in the review of our bird distribution maps. Likewise, the lack of reviewers for the Thunder Basin National Grassland may, in part, have contributed to the low bird diversity in this area (Fig 3.4). Thus, areas of high or low species richness may be an artifact of mere data collection intensity or effort. Locality records are also likely biased against species with nocturnal behavior (e.g. bats, rubber boa [Charina bottae]), inconspicuous habits or small size. While we are confident that the review process reduced the omission errors in the species distribution maps, we must acknowledge the potential biases associated with "overconfidence of experts" (Fischhoff et al. 1981, Suter et al. 1987).
3.5.2 Habitat Associations and Species Mapping
Within hexagons, the reliability of predicting species distributions based primarily on vegetation that is mapped on a "coarse scale" has been questioned (Short and Hestbeck 1995, but see Davis 1996, Edwards 1996, Scott et al. 1996). Indeed, working with remotely sensed data limited our ability to map micro-habitats (e.g., caves, cliffs) and small "pocket" habitats such as juniper, aspen, or bitterbrush shrub which occur in narrow strips along ridges or within canyons. As a result, species could be under or overestimated. For example, the distribution of cedar waxwings (Bombycilla cedrorum) whose habitat includes "open aspen stands", may have been under-estimated due to our inability to map many of the smaller, interspersed stands of aspen in foothill environments. We compensated to some degree for this problem by using both primary and secondary land cover types to make species predictions.
In contrast, we mapped the distribution of other micro-habitat specialists by assigning them to broad land cover types, based on the assumption that certain land cover polygons contain the micro-habitat features of importance. For example, the distribution of the cliff chipmunk (Tamias dorsalis utahensis) and the canyon mouse (Peromyscus crinitus doutii) were predicted using juniper cover, even though these species are limited to rock outcrops that are usually encompassed by juniper habitats. As a result the distribution of these species are over-estimated. Our use of small geographic units such as the hexagon minimized the extent of over-estimation for micro-habitat specialists with restricted ranges, such as the canyon wren (Catherpes mexicanus), and the chimney swift (Chaetura pelagica), but it was difficult to minimize over-prediction for micro-habitat specialists with broad ranges. Many species of bats have broad geographic ranges, but may actually be limited within these extents because of special roosting requirements, features such as caves, abandoned mine shafts and buildings that could not be mapped at the scale of our land cover map. We have documented most of these micro-habitat mapping problems (Merrill et al. 1996b) and data users should be cognizant of these limitations.
The ability to predict species occurrences from generalized land cover types has also been questioned because associations between species occurrence and vegetation type are not always tight. Factors other than vegetation, such as climate or small scale features such as subcanopy vegetation, tree or snag density, or even spatial arrangement of a number of cover types may be required for reliable predictions (Short and Hestbeck 1995, Flather et al. 1996). Because topographic relief in Wyoming landscapes is a dominant feature that influences climate, we included elevation in our models of species distributions. We also included hydrologic and associated riparian features in our modeling efforts because in the arid west many species are associated with these features and often dependent on them (Finch 1989, Szaro and Belfit 1986, 1987, Szaro and Jackle 1985). Addition of a GIS layer depicting soil types might further improve predictions of fossorial species such as the Wyoming pocket gopher (Thomomys clusius) and the olive-backed pocket mouse (Perognathus fasciatus). Soil types and other more detailed features could not be included in our models because these features are not mapped across the entire state and the vast majority of species have not been studied in sufficient detail to determine their association with such fine-scale features or habitat configurations (Scott et al. 1996). In fact, we found that for many of the 445 species we modeled, habitat relationships have been described only very generally. In some extreme cases, the best habitat description for forest bird species was "associated with coniferous forests". We had to assign these species to all seven coniferous types resulting in generalized and potentially overestimated species distributions. However, even when species predictions are based on more detailed information, usually at finer scales, observed error rates have been equally variable and high (Block et al. 1994, Hollander et al. 1994, Timothy and Stauffer 1991, Raphael and Marcot 1986, Dedon et al. 1986).
3.6 Summary and Conclusions
Gap analysis procedures should not be regarded as a substitute for detailed biological inventories on species distributions (Scott et al. 1993). Rather they are a methodology for organizing existing data into static maps that represent dynamic distributions (Edwards et al. 1996). Uncertainty exists in the current predictions of species due to incomplete information, data biases, map resolution, habitat models, and dynamics of species populations. To date, there have been only a few efforts to quantify the effects of the uncertainty in the data used to map species distributions and its effect on the interpretation of the program's results (Stoms et al. 1992, Dean et al. 1996, Kohley in prep). Nonetheless, the gap distribution maps represent the most up-to-date compilation and review of species distributions in Wyoming.
Although species check lists provide a preliminary assessment of our ability to map species distributions, species lists usually are not completely independent sources of information that provide reliable accuracy assessments. For example, in Wyoming, data used in the species check lists were not directly used in determining species ranges, but past observations on which the lists were based are likely to have been incorporated into state-wide databases (although we could not identify them) and published range maps. Also, several of the species check lists were partially developed by map reviewers. We recommend that error assessment of vertebrate databases, including both statistical assessments of modeling approaches as well as field validations, become a priority of GAP now that a number of state gap databases are completed. Even with these additional assessment efforts, we suspect that the basic lack of information on ranges and habitat associations of many species will hinder even the best modeling capabilities. In the immediate future, we believe one of the most important contributions of WY-GAP is to provide a management framework for designing further field surveys and research projects toward improving our understanding of species distributions in Wyoming.