Remote sensing of wetlands in Yellowstone National Park

Abstract

As part of the Amphibian Research and Monitoring Initiative, satellite remote sensing was used to identify potential wetland amphibian habitat in Yellowstone National Park. Landsat Thematic Mapper imagery was combined with ancillary predictors of wetland occurrence including habitat type, cover type, landform type, bedrock geology, soil attributes, terrain measures, and climate data. Classification trees were used to predict the likelihood of palustrine wetland occurrence across the Yellowstone landscape. Wetland maps generated by this study are intended to address shortcomings of the National Wetland Inventory in Yellowstone; namely, errors-of-omission and the temporally invariant nature of the inventory. Relative importance of Thematic Mapper imagery, image texture information, terrain measures, and thematic spatial data were assessed by comparing the accuracy of classification trees trained with different subsets of predictors. In general, classification trees using all available predictors exhibited the greatest accuracy. Classification tree structure was similar in models generated with satellite imagery from different years. Also, similar accuracy rates were found across years. The results indicate that the method could be applied to annual wetland monitoring. Average producer's accuracy for the palustrine wetland class was approximately 0.92. Five palustrine wetland classes were discriminated with an average overall accuracy of approximately 0.83. A wetland map derived from a 1 August, 2003 Thematic Mapper image was field-verified in late-summer and early-autumn of 2003. The classification tree methodology appears to be sensitive to within-season wetland condition and yearly wetland variability. Satellite remote sensing located wetlands not mapped by the National Wetland Inventory and also identified upland sites erroneously classified as wetland. Methods developed by this study are general enough to be applied in other physiographic settings and should prove to be useful to amphibian conservation efforts over large geographic extents.