Abstract
The Minnesota State Planning Agency has developed a geographically referenced Land Management Information System which is being used extensively for planning purposes. Land use categories in the system were originally coded from aerial photographs; this method is inefficient for updating the large-area data base. Landsat data and many computer-assisted techniques are available to analyze the classification system and to update the land use data base. The data derived from a Landsat analysis could be used to supplement the existing data base and to complement detailed interpretations of aerial photographs.
This study had as its primary objective an evaluation of computer manipulation, classification, and accuracy assessment techniques for use in updating land use data in the Land Management Information System. Four approaches to statistical computer manipulation (polygons selected from cathode ray tube displays, unsupervised clustering, polygons selected from aerial photographs and data extracted from the existing land use data base) were attempted. The resulting statistics were applied to the image data by three pattern-recognition algorithms: minimum distance to the mean, maximum likelihood, and canonical analysis with minimum distance to the mean. Twelve output images were compared to photo interpreted samples, ground-verified samples, and the current land use data base for accuracy assessment.
The results of this study indicate that for a reconnaissance inventory, statistical computer manipulation via polygons selected from aerial photographs applied with the canonical analysis and minimum distance algorithm is the most accurate and efficient approach. Crosstabulation with the accuracy samples indicated classification accuracies between 20 to 40 percent. These accuracy levels could probably be increased with the availability of appropriate seasonal coverage and the collection of more timely multidate supporting data.
Date of this Version
1981