An Accuracy Assessment of 1992 Landsat-MSS Derived Land Cover for the Upper San Pedro Watershed (U.S./Mexico)
The utility of Digital Orthophoto Quadrangles (DOQs) in assessing the classification accuracy of land cover maps derived from Landsat MSS data was investigated. Initially, the suitability of DOQs in distinguishing between different land cover classes was assessed using high-resolution airborne color video data. A cross-tabulation of the analysts DOQ labels and the reference video label was produced and had an overall accuracy of over 92%. This indicated that the DOQ data could be used to identify and distinguish between the different land cover classes.
Three land cover maps for the upper San Pedro Watershed were available for accuracy assessment. These land cover maps were interpreted and generated from the 1973/4, 1986, and 1992, North American Landscape Characterization (NALC) project data sets, by Instituto del Medio Ambiente y el Desarrollo Sustentable del Estado de Sonora (IMADES), Hermosillo, Sonora. The Environmental Protection Agency (EPA) supplied Arizona Remote Sensing Center (ARSC) with approximately 60 DOQs for 1992. Most of the land cover classes were fairly well represented in the DOQs and covered between 24% and 41% in eight out of ten land cover classes. Only the Barren and Agriculture classes were poorly represented in the available DOQs covering 5.3% and 14.2% of the map area, respectively.
During the initial classification accuracy assessment using the DOQs, it became apparent that that the NALC derived land cover maps were not geometrically registered properly to the DOQs. Once the mis-registration problem was addressed, we proceeded to perform the classification accuracy assessment. A total of 457 sample points were used for the accuracy assessment. Allocation of sample points to land cover classes was through stratified (by land cover class area) random sampling, with a 20-sample minimum for the smallest classes. Map labels for the sample points were compared with reference DOQ labels and an error matrix generated. An overall classification accuracy of about 75% was obtained.
You will need Adobe Acrobat Reader, available as a free download, to view some of the files on this page. See EPA's PDF page to learn more about PDF, and for a link to the free Acrobat Reader.