PREDICTED ANIMAL DISTRIBUTIONS AND SPECIES RICHNESS

TABLE OF CONTENTS

1. INTRODUCTION

2. LANDCOVER CLASSIFICATION AND MAPPING

3. PREDICTED ANIMAL DISTRIBUTIONS AND SPECIES RICHNESS

3.1. Introduction

3.2. Methods

3.2.1. Mammals

3.2.2. Birds

3.2.3. Reptiles

3.2.4. Amphibians

3.3. Results

3.4. Accuracy Assessment

3.5. Limitations and Discussion

4. LAND STEWARDSHIP

5. ANALYSIS BASED ON STEWARDSHIP AND MANAGEMENT STATUS

6. CONCLUSIONS AND MANAGEMENT IMPLICATIONS

7. DATA USE AND AVAILABILITY

8. LITERATURE CITED

9. GLOSSARY

10. GLOSSARY OF ACRONYMS

11. APPENDICES AND MAPS

3.1. Introduction

All species range maps predict occurrence within a particular area (Csuti 1994). Traditionally, predicted occurrences of most species begin with samples from collections made at point locations. Most species range maps are small-scale (e.g., >1:10,000,000) and derived primarily from point data for field guides. The purpose of Gap Analysis vertebrate species maps are to provide more precise information about current distribution of individual native species within their general ranges. With these data, better estimates can be made about configuration and actual amounts of habitat.

Gap Analysis maps are produced at a scale of 1:100,000 and are intended for applications at landscape or "beta" scale (homogeneous areas generally covering 1,000 to 1,000,000 hectares (ha) and made up of more than one kind of natural community). Applications of these data to site or stand-level analyses (site - a microhabitat, generally 10 to 100 m²; stand - a single habitat type, generally 0.1 to 1,000 ha; Whittaker 1977, also see Stoms and Estes 1993) are likely to be compromised by finer-grained patterns of environmental heterogeneity that are resolved at those levels.

Gap Analysis uses predicted distributions of native vertebrate species to evaluate their conservation status relative to existing land management (Scott et al. 1993). However, maps of vertebrate species distributions 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 maps, great utility may be found in consolidating specimen collection records and literature into databases used to produce maps.

Previous to this effort no maps were available, digital or otherwise, showing likely present-day distribution of species by habitat type across their ranges. Because of this, ordinary species (i.e., those not threatened with extinction or managed as game animals) are generally not given sufficient consideration in land-use decisions in context with large geographic regions. Decline of these species due to incremental habitat loss can, and does, result in one threatened or endangered "surprise" after another. Frequently, records that do exist for these "ordinary" species are truncated by state boundaries. Simply creating a consistent spatial framework for storing, retrieving, manipulating, analyzing, and updating status of each vertebrate species is a necessary and basic element for preventing further erosion of biological resources.

3.2. Methods

U.S.G.S. Biological Resources Division's Gap Analysis Program (BRD-GAP) has standards for generating predicted terrestrial distribution. Breeding habitat patches outside breeding distribution (but contiguous with habitats within breeding distribution) are included while all habitats outside of geographic range are excluded (Figure 3.1.). Tomlin (1990) refers to this procedure as a zonal operator which can be performed by a Geographic Information Systems (GIS). Geographic Resource Analysis Support System (GRASS4.1) does not include a module to perform zonal operations. Generating models by combining existing GRASS4.1 modules with custom programming proved to be time consuming (each model can run 16-30 hours on a SUN 690 server with 320 megabytes of RAM) and created maps that required large amounts of hard disk space. Therefore, topologies between all contiguous vegetation polygons (units for breeding habitat) and counties of Arkansas (units for breeding distribution) were obtained by running GRASS4.1 module r.stats and output was loaded into INFORMIX where model generation occurred. In this manner, GRASS4.1 determined spatial relationships between vegetation polygons and counties, while INFORMIX performed linkages between terrestrial vertebrate species, appropriate vegetation types, county of occurrence and vegetation polygons. Terrestrial vertebrate models were generated by INFORMIX Structured Query Language (SQL) queries that obtained all appropriate vegetation totally within a county of occurrence or contiguous with a county of occurrence. Maps were generated quickly (all 322 models ran in 23 hours) by this methodology because topological relationships were only determined once. Disk storage requirements of resultant files were minimal as they were only database tables. All other analysis, including accuracy assessment, was performed using this two step methodology of topological determination by GRASS4.1 and database linkage by INFORMIX. Queries were used to generate maps by unloading SQL outputs to files and were used as inputs for GRASS4.1 module r.reclass. These reclass maps were viewed on a digital display and later directed to a printer for hardcopy production (Catanzaro and Smith 1995).

In order to keep Arkansas Gap Analysis Program (AR-GAP) models simple, most individual species received little special attention. However, species requiring special habitats, such as caves, riparian areas, or water bodies, received additional modeling. Locations of caves within Arkansas were

Figure 3.1. Schematic representation of the steps necessary to model terrestrial vertebrates for NBS GAP.

identified using U.S.G.S. Geographic Names Information System (GNIS). Non-flowing aquatic features were generated by patching water defined by TM sensor into doubly-lined drainages mapped in 1992 U.S.G.S. Topographically Integrated Geographic Encoding and Reference (TIGER) digital

dataset. Standing water was obtained from 30 m vegetation map and waterbodies were separated into large bodies (greater than 100 ha) and small bodies (less than 100 ha). Perennial streams and annual streams, defined by 1992 TIGER files, were rasterized at 30 m. Flowing water layers were patched into standing water layers to produce a final water layer map. Methodology for model generation of species requiring water habitats duplicated procedures used for terrestrial species.

3.2.1. Mammals:

Sixty-seven mammals were included in AR-GAP (Appendix 11.10.) following taxonomy of Banks et al. (1987). County of occurrence range maps were generated by combining museum specimen locations obtained from Dr. Rick McDaniel (Arkansas State University, Jonesboro) and Dr. Bob Wiley (University of Arkansas, Monticello) with distributions shown in Sealander and Heidt (1990). Natural history information contained in Sealander and Heidt (1990) was used for habitat association matrix data.

3.2.2. Birds:

A complete list of AR-GAP birds was developed with following criterion: species must have at least 5 sightings in Arkansas, or more than 1 sighting or breeding record within Arkansas since 1980. Species clearly out of their range were excluded. These criteria reduced a total of 381 avian species that have been reported in Arkansas (James and Neal 1986; James et al. 1994) to 144 species (Appendix 11.10.). Avian range maps were compiled using three datasets. County level source data from Arkansas Birds (James and Neal 1986), checklists developed by Neal and Mlodinow (1988) and Hyatt and Moren (1990) were stored in INFORMIX. A matrix of bird species and counties was reviewed by the Gap Avian Committee (Appendix 11.11.) and any appropriate additions or changes were made. The Gap Avian Sub-Committee (Appendix 11.11.) reviewed this augmented dataset and changes made were stored in INFORMIX. Avian habitat association matrix follows Hamel (1992) and was loaded into INFORMIX after vegetation types were cross-walked into Arkansas's Natural Vegetation Classification (Figure 3.2.). Modifications were made for those species occurring in agricultural and/or urban areas, species inhabiting water bodies or riparian areas, and species not included by Hamel (1992). A dynamic linkage between INFORMIX and GRASS 4.1 was created and maps were generated for each species, reviewed by the Gap Avian Sub-Committee and corrected as needed.

3.2.3. Reptiles & Amphibians:

A complete list of reptiles and amphibians was generated with assistance from Dr. Stan E. Trauth (Arkansas State University, Jonesboro) and followed Banks et al. (1987) taxonomy. One-hundred and ten reptiles and amphibians were included in AR-GAP (Appendix 11.10.). County of occurrence range maps and habitat association matrix data were obtained from Conant and Collins (1991).

3.3. Results

A map of each terrestrial vertebrate in Arkansas was generated by AR-GAP (e.g. Figure 3.3.; see List of Maps). Each map followed a generalized format with breeding distribution shown alongside breeding habitat and final predictive model was displayed next to a legend for all three maps. In order to simplify explanation of data produced by AR-GAP, examples in both this chapter and

chapter 5 will be focused on Ovenbirds (Seiurus aurocapillus), a common neotropical migrant found throughout much of Arkansas. Ovenbirds in Arkansas were limited to Interior Highlands (upper left, Figure 3.3.) as shown by James & Neal (1986), Neal & Mlodinow (1988) and Hyatt & Moran (1990).

Throughout AR-GAP data collection, it rapidly became evident that researching gaps in our

Figure 3.2. A graphical representation of the hierarchical relationships between those vegetation units that were able to be discriminated by remote sensing techniques followed by the Arkansas Gap Analysis project (hierarchy from Foti et al. (1994)). The bold groupings are the vegetation units used to create the Wildlife-Habitat Relationship models (groupings based on Hamel (1992)).

chapter 5 will be focused on Ovenbirds (Seiurus aurocapillus), a common neotropical migrant found throughout much of Arkansas. Ovenbirds in Arkansas were limited to Interior Highlands (upper left, Figure 3.3.) as shown by James & Neal (1986), Neal & Mlodinow (1988) and Hyatt & Moran (1990).

Throughout AR-GAP data collection, it rapidly became evident that researching gaps in our

protective network of managed lands uncovered significant gaps in knowledge about distribution of terrestrial vertebrates, as well as their natural history (Smith and Catanzaro 1996). Two categories, Predicted by Gap Avian Committee and Predicted by Gap Avian Sub-Committee, were added in an attempt to rectify informational gaps (upper left, Figure 3.3.). Ovenbirds were predicted to occur in

Figure 3.3. Example of three datasets used to generate predicted distributions of breeding terrestrial vertebrates.

protective network of managed lands uncovered significant gaps in knowledge about distribution of terrestrial vertebrates, as well as their natural history (Smith and Catanzaro 1996). Two categories, Predicted by Gap Avian Committee and Predicted by Gap Avian Sub-Committee, were added in an attempt to rectify informational gaps (upper left, Figure 3.3.). Ovenbirds were predicted to occur in

three Arkansas counties. Potential habitat used by breeding Ovenbirds was distributed evenly across Arkansas (upper right, Figure 3.3.). Habitats of higher value to Ovenbirds (mixed pine-hardwoods and oak-hickory) were limited principally to Boston Mountains (see Figure 5.2 for a map of ecoregions). Significant amounts of marginal Ovenbird habitat (elm-ash-cottonwood, loblolly pine-shortleaf pine, and oak-gum-cypress) were found throughout Arkansas. Ovenbird predictive model (lower left, Figure 3.3.) was a combination of breeding distribution and breeding habitats. Habitats within counties occupied by Ovenbirds were highlighted, while habitats outside their breeding distribution were not. Exceptions occur only when contiguous vegetation polygons crossed over into counties where Ovenbirds were not thought to breed (southern Arkansas).

Breeding distribution maps were translated from counties to Environmental Monitoring Assessment Program (EMAP) 635 km² hexagons (White et. al 1992) for further analysis. Range of biodiversity was quite narrow when displayed in this manner. Only 70 species separated the most specious hexagon from the least (Figure 3.4.). Least specious hexagon included 200 species and most specious hexagon included 270 species (62% and 83% of Arkansas's terrestrial vertebrate diversity respectively). Less specious hexagons were located in Mississippi Alluvial Basin where landscapes have been highly modified for agricultural purposes. Hexagons with high diversity of terrestrial vertebrates were at the confluence of Arkansas Valley, Ouachita Mountain, and Mississippi Alluvial Basin ecoregions. Other hexagons with high biodiversity were in western Arkansas and parts of Ouachita Mountains. Avian hexagon biodiversity ranged from 80 to 126 species (56% to 88% of Arkansas avian diversity) (Figure 3.5.). Hexagons with high avian biodiversity were in central, northwest, and southwest Arkansas. Hexagons with high avian biodiversity were found along borders of two or more ecoregions. Mammalian biodiversity per hexagon ranged from 48 species to 62 species (72% to 93% of Arkansas mammalian diversity) (Figure 3.5.). Hexagons with high concentrations of mammalian biodiversity were found in northwest Arkansas while relatively low concentrations of mammals were found in southern regions. Amphibian hexagon biodiversity ranged from 18 species to 33 species (40% to 73% of Arkansas amphibian diversity) (Figure 3.5.). Hexagons with high concentrations of amphibian biodiversity were scattered throughout Interior Highlands while low concentrations of amphibian biodiversity were found in Mississippi Alluvial Basin. Reptilian hexagon biodiversity ranged from 41 species to 60 species (64% to 94% Arkansas reptilian diversity) (Figure 3.5.). A concentration of reptiles was found in central Arkansas. Mississippi Alluvial Basin was not depauperated of reptiles as it was for other taxonomic groups.

Terrestrial vertebrate diversity map (Figure 3.6.) totals all 322 individual predictive models of breeding birds, mammals, reptiles and amphibians. Vegetation polygon with highest biodiversity in Arkansas was oak-gum-cypress and provided habitat for 214 species (64% Arkansas biodiversity). Other areas of high terrestrial biodiversity were bottomland hardwood regions in central and southwest. Low areas of biodiversity were agricultural areas. Major centers of avian biodiversity were upland oak-hickory forests (Figure 3.7.). Vegetation polygon with highest avian diversity provided habitat for 107 bird species (74% of Arkansas avian biodiversity). There was no major center of mammalian diversity (Figure 3.7.). Vegetation polygon with highest mammalian biodiversity in Arkansas provided habitat for 39 species (58% of Arkansas mammalian biodiversity). Major centers of amphibian biodiversity were oak-gum-cypress areas (Figure 3.7.). The polygon with highest amphibian biodiversity provided habitat for 38 species (84% of Arkansas amphibian biodiversity). Most areas in Arkansas had similar numerical biodiversity of amphibians, due to even spatial distributions of water bodies. Major centers of reptilian biodiversity were oak-gum-cypress (Figure 3.7.). The polygon with highest reptile biodiversity in Arkansas provided habitat for 57

species (90% of Arkansas reptilian biodiversity).

3.4. Accuracy Assessment

Other GAP projects have obtained species lists from well studied management areas and compared these lists to GAP predicted lists. Very few areas in Arkansas have been well studied, therefore data from North American Breeding Bird Surveys (BBS) were used to test predicted avian distributions.

Figure 3.4. Distribution of terrestrial vertebrates.

Figure 3.5. Hexagon diversity for birds, mammals, reptiles, amphibians.

Figure 3.6. Terrestrial vertebrate diversity (clump).

BBS data were collected during an annual late spring/early summer survey which started in 1966. Fifty census points were located 0.8 km (0.5 miles) apart on a 39 km (24.5 mile) roads. Each point was sampled by volunteers for three minutes and all species and number of birds seen and heard within 0.402 km (0.25 miles) of census points were recorded (see Erskine 1978; Robbins et al. 1986 for full methodology). Effective geographic resolution BBS was a section, an aggregation of ten

Figure 3.7. Diversity for birds, mammals, reptiles and amphibians.

BBS census points, thus each BBS routes had five sections. There were 32 Arkansas BBS routes (160 sections) run during 1980-1992. While this dataset was obtained in digital format, it was not referenced to a digital map. A digital base map was generated to render this dataset as a GIS (Catanzaro and Smith 1996). Base map generation for BBS data involved isolating roads representing BBS routes on 1:100,000 DLGs. Each section was uniquely attributed and a 0.8 km buffer was applied to represent area sampled by each section. Final base map was related back to BBS data by linking each section to its respective BBS data stored in INFORMIX.

Topological relationships between BBS and completed vegetation map was determined using GRASS4.1 module r.stats. A series of SQLs determined relationships between species predicted presence or absence and BBS data. Errors of commission and omissions were generated and results were classified into four categories: PPV - species predicted to occur within a BBS section and confirmed by BBS data (positive predictive value); NPV - species predicted to be absent from all vegetation polygons within a BBS section and not found by BBS data (negative predictive value); Omission error - species predicted to be absent from all vegetation polygons within a BBS section and found by BBS data; Commission error - species predicted to occur within a BBS section and not found to by BBS data (Appendix 11.12.).

Predicted distributions of mammalian, reptilian and amphibian species were not tested. While digital datasets of museum specimens for these taxonomic groups were obtained, a lack of funds and time precluded such analysis.

Upon inspection of accuracy assessment data, it was evident that there was a significant difference between errors of omission and commission. For example, Ovenbird commission error was 38% and omission error was 3% while Louisiana Waterthrush commission error was 24% and omission error was 1% (Appendix 11.12.). Avian models tended to overpredict rather than underpredict (Smith and Catanzaro 1996). There could be two reasons for this phenomenon. First, there were a total of 2,596 predictive records in breeding distribution database (i.e. counties attributed as Predicted by Gap Avian Committee or Predicted by Gap Avian Sub-Committee). Using predictive data to build a predictive model could have introduced bias. These predictive records accounted for 15% of 16,750 county distributional records and it was quite possible that commission errors may be higher in those counties a species was originally predicted to occur in. It should be noted however, that 86% of 2,596 predictive records were for species that were thought to be statewide breeders. These predictions filled information gaps associated with common species, as opposed to range extensions. Secondly, there were several large polygons within Arkansas. The largest natural vegetation polygon, shortleaf pine, had an area of 411,874 ha and touched thirteen counties. The second largest natural vegetation polygon, white oak, northern red oak - shortleaf pine _ hickory, had an area of 272,397 ha and touched nine counties. If a species was found to occur in only one of these counties, GAP methodology dictated that the entire vegetation polygon must be included in the predictive model. It is hypothesized that large polygons will cause an overprediction to occur (see landcover section for further analysis).

3.5. Limitations and Discussion

All maps, including AR-GAP products are predictions of ground based features. AR-GAP vertebrate modeling used simple models and their performance tested. These simple models should be viewed as a starting points for discussion of Arkansas terrestrial vertebrate distribution.

Vegetation structure (as opposed to composition) is an important factor determining avian biodiversity (Cody 1985), temperature and moisture regimes are important predictors of reptile and amphibian distribution (Gans 1976, 1977, 1982; Duellman and Trueb 1986), soil types predict fossorial rodents (Churchfield 1990), and caves are necessary to most bats (Gaisler 1979: Kunz 1982). General variables are difficult to correctly identify from orbiting commercial satellite sensors and using such imagery to discern structural differences among vegetation types, sub-canopy composition or structure is exceedingly difficult (Gonzales 1994; Hepinstall et. al 1996). While TM sensor does have a thermal component, spatial resolution of this spectral band is currently 120 m, a substantial decrease from 30 m limits its' usefulness. Crist & Cicone (1984) developed an index related to vegetation wetness through a mathematical transformation of TM data. This index does not identify soil moisture, and for forested areas is highly dependent upon season and vegetation mixes. Identification of soil types by aerial photographs is available (Soil Survey Division Staff 1993), but satellite identification has met with only limited success (Sabins 1978) and depend upon quality data inputs. Recognizing limitations of spaceborne platforms, AR-GAP attempts to predict distributions of terrestrial vertebrates solely on occurrence of appropriate vegetation types.

Some species of reptiles, amphibians, mammals, and birds are microhabitat specialists, living in habitats which can not be identified by small scale maps (e.g. > 1:100,000). While mappable surrogate habitats may be used, results can be unsatisfactory due to oversimplification of a species requirements. Contrastingly, some species are habitat generalists and their models may be uninformative. Raccoons (Procyon lotor) are widely distributed in Arkansas (all 75 counties) and are found in a variety of habitats including forested, urban and agricultural areas. As a result, AR-GAP methodology predicted raccoons to occur in 97% of Arkansas. Uninformative results will also be generated when information gaps occur. While some taxa (e.g. game, endangered and/or threatened, or scientifically interesting species) have been thoroughly studied, allowing development of complicated models, most biodiversity has not been studied sufficiently to generate even generalized models. A combination of these inherent difficulties interacting with the strength of relationships defined in the wildlife-habitat matrix will ultimately determine model performance.

Irrespective of these difficulties in general model techniques, programmatic restraints in modeling 322 terrestrial vertebrates have given rise two specific issues particular to AR-GAP and need to be addressed. In order to conclude AR-GAP in a timely matter, terrestrial vertebrate modeling was completed long before a smoothed 100 ha vegetation map was finalized. Effects of this choice are unclear, this unsmoothed map was used as the input for a GRASS4.1 module to find contiguous areas (r.clump). Results of r.clump may be biased for map categories that are one cell wide or map category features that are diagonal in nature which are not considered contiguous. Unsmoothed data as an input layer in r.clump more than likely improperly inflated numbers of contiguous polygons and may have affected model performance.

Use of BBS data to test AR-GAP vertebrate models had several limitations, but free data, large sample size (160 BBS sections) with each section of equal area are considerable strengths. BBS data only cover breeding birds (a little over half of terrestrial vertebrates included in AR-GAP). Other limitations include: BBS was not designed to sample biodiversity per se, it was designed to sample population trends; not every bird species included in AR-GAP had been recorded by BBS; and BBS may be biased against species that breed before or after survey times, species that not active during times surveyed or species difficult to locate. Model accuracies must be interpreted cautiously and