An Inter-Agency Approach for Determining Regional Land Cover in the American Southwest: The Southwest Regional Gap Analysis Project William G. Kepner1, David F. Bradford?, and Todd D. Sajwaf 1U.S. Environmental Protection Agency, Office of Research and Development, P.O. Box 93478, Las Vegas, NV 89193; 2University of Idaho, College of Natural Resources, P.O. Box 444332, Moscow, ID 83844 Abstract The Gap Analysis Program is a national inter-agency program that maps the distribution of plant communities and selected animal species and compares these distributions with land stewardship to identify "gaps" in biodiversity protection. GAP uses remote satellite imagery (Landsat 7) and Geographic Information System (GIS) technology to assemble and view large amounts of biological and land management data to identify areas where conservation efforts may not be sufficient to maintain diversity of living natural resources. Historically, GAP has been conducted by individual states. However, this has resulted in inconsistencies in mapped distributions of vegetation types and animal habitat across state lines because of differences in mapping and modeling protocols. This was further compounded from the lack of a national vegetation classification nomenclature. In response to these limitations, GAP embarked on a second-generation effort to conduct the program at a regional scale using 1) a vegetation classification scheme applicable across the U.S.; 2) ecoregional units as the basis for segmenting the landscape into manageable units; and 3) inter-agency investigator teams with land cover analysis and environmental protection expertise. The program's first formalized multi- state effort includes five Southwestern states (Arizona, Colorado, Nevada, New Mexico, and Utah), which comprise nearly one-fifth of the conterminous United States. GAP Program Overview: A "gap" is the lack of representation or under- representation of an element of biodiversity (plant community or animal species) in an area intended for its long-term maintenance. Gap analysis is a national program about keeping common species common by providing a geographic approach to map biological diversity. The GAP methodology is straightforward: 1) map the distributions of natural plant communities, 2) map predicted distributions of terrestrial vertebrate species, 3) map the degree of management for biodiversity maintenance, and 4) analyze the representation of vegetation and animal species distributions in the conservation network to identify "gaps" in long-term security. Introduction Land cover Mapping and First Generation GAP: While the first generation of western GAP projects was highly innovative for their time, there were unforeseen problems. As the various western GAP projects were completed and stitched together, the vegetation maps ex- hibited abrupt changes in their classification systems and community distributions at state boundaries. Animal species distribution maps, modeled largely from vegetation maps, also revealed abrupt changes at state boundaries. Three sources of these problems were identified: 1) separate vegetation classification systems for each state, 2) unique methodologies for constructing predictive maps of plant communities, and 3) state-by-state differences in habitat modeling protocols. Materials and Methods The National Vegetation Classification System (NVCS), developed by NatureServe, is the basis for plant community classification for the SWReGAP project. Based on the plant community characterization data collected in the field, each site is assigned an alliance, ecological system, and National Land Cover Data (NLCD) label (Table 1). To classify the vegetation of a 5-state region (Figure 1) requires thousands of training sites. Field crews select training sites opportunistically based onhomogeneity ofplantspecies composition, landform, and spectral characteristics. Three essential steps are performed at each site: plant community characterization, site delineation, and photographic documentation. Plant Community Characterization: Two basic types of information are collected for each training site: 1) ocular estimates of vegetative cover by life form and abiotic ground cover (e.g. rock fragments, bedrock, water) and 2) measurements Table 1. Modified NVCS for the SWReGAP Project Level Primary Basis for Classification Examples NLCD Coarse land use/land cover ¦Woody/Herbaceous Wetland Ecological System Aggregation of plant communities ISSSslf Dominant/diagnostic species of the uppermost stratum Poprf™ Tmiporarily- Table 1 Photographic Documentation: To document each training site, a digital photograph is collected as a reference should any questions arise regard- ing its alliance or ecological system labels. Once all of the training site polygons for a mapzone are collected, they are intersected through various digital datalayers. In order to and classification of the landscape setting (e.g. landform, topographic attributes). Site Delineation: A polygon delineating the training site is hand-digitized in the field utilizing satellite imagery, digital elevation models (DEMs), and digital raster graphs (DRGs) as guides. Field Training Sites (Total 92,100) further standardize methodologies throughout the five states, we used consistent datalayers for modeling. The geospatial data layers include Landsat 7 Enhanced Thematic Mapper Plus (ETM+) imagery acquired between 1999 and 2001 for 3 seasons (spring, summer, fall), digital elevation model data, and STATSGO soils data. Classification and Regression Trees: Classification trees recursively partition a data set into increasingly "pure" subsets based on a multitude of predictor variables. In the case of SWReGAP, the pure subsets are groups of field sites that belong to the same alliance or ecological system. The output of a classification tree is a set of decision rules. Accuracy Assessment: The final predictive vegetation maps were completed and subjected to various accuracy assessment procedures. Our methods include withholding a proportion (20%) of the training dataset to use in a con- ventional accuracy assessment and review of draft vegetation maps by regional experts. Southwest Digital Land Cover Map 125 Land Cover Classes Results was used to discriminate land cover types, while a minority of classes (e.g. urban classes, sand dunes, burn scars, etc.) were mapped using other techniques. Twenty mapping areas, each characterized by similar ecological and spectral characteristics, were modeled independently of one another. These mapping areas, which included a 4-km overlap, were subsequently mosaicked to create the regional dataset. An internal validation for modeled classes was performed on a withheld 20% of the sample data. While the modeling area encompassed these five Southwestern states (Arizona, Colorado, Nevada, New Mexico, and Utah), the actual GIS dataset can be downloaded as a subset of the 5-state region using state, county, Figure 2 depicts the digital land cover prod- uced for the SWReGAP project. More than 1.5 billion 30m pixels have been classified into 125 land cover classes to develop a seamless land cover map for the 5-state area. The information is available from the Utah State University server based at http://earth.gis.usu.edu/swgap/. Multi-season satellite imagery (Landsat ETM+) from 1999-2001 was used in conjunction with digital elevation model (DEM) derived datasets (e.g. elevation, landform, aspect, etc.) to model natural and semi-natural vegetation. For the majority of classes, a decision tree classifier TNC ecoregion, Bailey ecoregion, and SWReGAP mapping zone configurations. Each file contains a folder with the dataset in Arclnfo grid or ERDAS Imagine format as specified, FGDC (Federal Geospatial Data Committee) meta-datafile(s), and a .pdf document of land cover class descriptions. As an example, we demonstrate land cover for Clark County, Nevada in Figure 3. In this example, 39 land cover classes (Table 2) are displayed with mapping accuracies varying from 32 to 93% per class. The total map accuracy for Clark County was estimated at 75.3%. Future Directions _ The SWReGAP project has recently developed an online Nevada Geospatial Data Browser. It includes complete GIS coverages and meta-data for the entire state of Nevada. The intent of the Nevada Geospatial Data Browser is to 1) develop a central repository for the Nevada SWReGAP spatial data and 2) to provide a mechanism for public distribution of Nevada geospatial information to other researchers, public agencies, resource managers, non-governmental organizations, decision-makers, and user groups. This product will provide for long-term record keeping (archiving) and easy online public access. The coverages are available for download and the meta-data include important information relative to acquisition, location, processing level, file size and format. For More Information http://www.epa.gov/nerlesdl/land-sci/gap.htm http://earth.gis.usu.edu/swgap/ Partner Agencies: m uses e for a changing world UNIVERSITY £ NatureServe -] 1261eb05 db.SF orum-Kepner Notice: Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy Mention of trade names or commercial products does not constitute endorsement or recommendation by EPA for use. epascienceforum Collaborative Science for Environmental Solutions ------- |