Detecting soil information on the Konza prairie using high resolution satellite data

Abstract

Computer pattern recognition techniques were used to discriminate soil information from Landsat TM and French SPOT satellite data on the Konza prairie near Manhattan, Kansas. Digital Elevation Model (DEM) data were merged to Landsat TM and SPOT data to delineate soil mapping units within the study area. Soil mapping units from a digitized soil map were compared with a classified soil spectral map obtained from Landsat TM or SPOT, and DEM derived elevation, slope, and aspect data using an overall accuracy assessment. The overall accuracy of soil spectral classes from TM and SPOT data was improved after DEM data were merged. Higher separability of soil mapping units derived from Landsat TM and DEM data on upland positions of the study area was obtained using statistical divergence analysis. Ratioing, intensity transformation, and low frequency filtering were performed before supervised classification was used. A better average accuracy of soil mapping units was found from low frequency filtering. A higher overall accuracy derived from TM data in the dormant season was obtained compared with the accuracy in the growing season. This result indicates that satellite data acquired in dormant season are more useful for soil information detection. The overall accuracy from Landsat TM and SPOT data was not significantly different because of scale reduction of SPOT data. The overall accuracy from Landsat TM and SPOT data was about 55 to 57 %. This study demonstrates that high resolution Landsat TM and SPOT satellite data can be used to aid soil consociation delineation at the second order level in areas where the dominant land use is rangeland.