MA2 Amphibian Research and Monitoring
Ballroom F, Level 4
8:00 AM - 12:00 PM, Monday, 10 November 2003
Chair: Rostad, Colleen ,
Co-chair: Battaglin, William ,
(013) Modeling Wetland Probability in Yellowstone National Park.
Wright, C.1, 1 Montana State University, Bozeman, MT, U.S.
ABSTRACT- Surveys of amphibian wetland habitat are an important component of the USGS Amphibian Research and Monitoring Initiative and typically rely on the National Wetland Inventory (NWI) to identify wetlands. However, NWI maps are often unavailable in digital format, appear to be biased toward errors of omission, and represent a static snapshot of wetland extent. As a method for acquiring multi-temporal information over large areas, satellite remote sensing addresses shortcomings of the NWI. In Yellowstone National Park, the probability of wetland occurrence was estimated across the landscape using a combination of Landsat multispectral imagery (30 m resolution) and ancillary GIS data (landform, cover, and habitat types, soil attributes, hydrologic wetness index, etc). This approach combines temporally dynamic spectral variables with static environmental attributes related to wetland presence. Wetland probability was modeled using a hierarchical classification tree methodology. At the coarsest level, the probability of occurrence of palustrine wetland (the predominant type in Yellowstone National Park) was mapped. Pixels identified as palustrine wetland were then assigned probabilities of membership in five palustrine wetland classes (emergent, forested, scrub-shrub, aquatic bed, and unconsolidated shore). At the finest level, palustrine emergent and palustrine forested pixels were classified by their water regime (saturated, seasonally or intermittently flooded). Results from year 2000 models indicated that palustrine wetland was identified with 91% accuracy. The five palustrine wetland types were discriminated with an overall accuracy of 81%. Comparison of wetland probability maps for a dry year (1988) and a wet year (1997) was informative for understanding the potential magnitude of spatiotemporal variability in Yellowstone wetlands. Wetland probability maps were used to plan 2003 amphibian surveys in Yellowstone National Park, and as a component of landscape-scale amphibian habitat models under development by USGS.
Key words: Remote Sensing, Wetlands, Classification Trees, Amphibian Habitat