Comparison of level I land cover classification accuracy for MSS and AVHRR data

Abstract

The capabilities of the Advanced Very-High-Resolution Radiometer (AVHRR) for land-cover mapping were investigated by comparing the accuracy of land-cover information for the Washington, D.C. area derived from NOAA-7 AVHRR data with that from LANDSAT Multispectral Scanner Subsystem (MSS)s data. Unsupervised level I land-cover classifications were performed for MSS and AVHRR data sets collected on 11 July 1981. A detailed accuracy assessment was conducted based on ground data delineated on 12 U.S. Geological Survey 7·5 min series topographic maps. These results produced overall land-cover classification accuracies of 71·9 and 76·8 per cent for AVHRR and MSS, respectively. While the accuracies for predominant categories were similar for both sensors, land-cover discrimination for less commonly occurring and/or spatially heterogeneous categories was improved with the MSS data set. The AVHRR, however, performed as well as or better than the MSS in classifying large homogeneous areas. The application of AVHRR data with its lower processing cost and more frequent worldwide coverage appears promising for regional land-cover mapping.