Mapping and Hydrologic Attribution of Temporary Wetlands Using Recurrent Landsat Imagery

In areas with a high density of ephemeral wetlands, traditional mapping protocols may underestimate occurrence of wetlands when single-date base-imagery is utilized. In the Pleistocene Sand Dunes Ecoregion in Oklahoma, National Wetland Inventory (NWI) maps created using base-imagery from a dry year omitted large numbers of ephemeral wetlands. To improve the likelihood of capturing inundated depressions, we classified water pixels from 51 Landsat images (3 images per year: pre/early, peak, and late/post growing season) from 1994 to 2011. Several image classification methods were tested but decision tree analysis with training pixels from multi-season imagery provided the greatest accuracy. Accuracy was determined through manual comparison of two Landsat images with concurrent aerial imagery (Kappa =0.96 and 0.93 for the two images). Wetland polygons were created from water/non-water rasters and given hydroperiod designations based on the number of inundated periods. Landsat-derived wetland maps identified 3156 wetland units, 718 more than the original 1980s NWI, with only 33.9 % agreement between the two maps. Finally, one meter LiDAR data were combined with classified Landsat images to determine the volume of water in wetlands during each image date. These wetland maps can assist with estimating the availability of inundated habitat during wet, dry, and average rainfall periods.