California Desert Resource Inventory Using Multispectral Classification of Digitally Mosaicked Landsat Frames

Time and budget constraints have precluded the use of conventional mapping techniques for the U.S. Bureau of Land Management to produce a comprehensive resource inventory of the California deserts region as mandated by the California Desert Conservation Act. A Landsat mosaic at full resolution for ten scenes over the California deserts region was used to provide a continuous data base for mUltispectral thematic classification. Procedures for adjustment of brightness values between frames and the digital mosaicking of the Landsat frames to standard map projections were developed for this task and are discussed. The principle of transect sampling was adopted as a means to obtain a uniform classification throughout the entire desert while generating one set of classification statistics. The transects were sel'ected by the BLM science team to include all variable types in the landscape. The initial set of unsupervised statistical clusters was reduced by the BLM staff to 100 primary statistical clusters and applied to twelve small (512x512) test areas extracted from the Landsat mosaic. Subsequent to verification, classification was performed on the entire desert mosaic in lOxl o segments. Resource class assignment was aided by including, in a post Classification procedure, DMA/NCIC digital terrain elevation data (with derived measures of slope gradient and aspect) registered to the Landsat mosaic. The combination of local terrain variations and a global sampling strategy based on transects provided the framework for an *This paper presents one phase of research conducted at the Jet Propulsion Laboratory, California Institute of Technology, under NAS 7-100, sponsored by the National Aeronautics and Space Administration. for an accurate classification throughout the entire desert region.