Extrapolation of point data for resource mapping

Abstract

Since the early seventies, the usefulness of satellite imagery for producing vegetation maps has been well known. Even small scale imagery such as advanced very high resolution radiometer (AVHRR) has been applied to continental vegetation mapping. Much has been written about how remote sensing can provide this kind of data for a GIS, however, little attention has been placed on the role of GIS for supplying data to automate the image classification process. Large data bases such as the USDA SCS National Resources Inventory (NRI) now being constructed have a potential for filling this role. The NRI has a potential for contributing to the improvement of vegetation mapping in at least three ways: Automated classification of imagery using NRI sample points to create training sets, classification verification through use of NRI sample points as field verification points, and automated map generation directly from the NRI data base through use of Thiessen polygons. Classification accuracy of a TM scene classified using NRI data to produce training sets was fair (43.6% overall classification accuracy), considering the process was completely automated with no human judgement involved. Adding ancillary data and temporal image information should improve this figure. Classification accuracy was tested for training sets of various sized polygons constructed around the NRI points. It was found that the smallest polygon (3x3 box) around each point resulted in a training set that produced the most accurate classification. Land use/land cover m aps can be easily generated from the NRI data base using the Thiessen polygon process. They provide a good overview of the conditions for a large area such as a state, however, for a county sized map, there are too few NRI data points to allow a reasonable depiction of the true conditions. Thiessen polygon maps do not depict accurate boundaries due to the process used for polygon generation, however, with the large number of sample points in the NRI, they provide a quick approximation of the conditions and have a potential for displaying land use/land cover change as the NRI is updated every five years.