STUDY OF CORN AND SOYBEAN LANDSAT MSS DATA CLASSIFICATION PERFORMANCE AS A FUNCTION OF SCENE CHARACTERISTICS
Abstract
In order to fully utilize remote sensing to inventory crop production, it is very important to identify the factors that affect the accuracy of Landsat classifications. The objective of this study was to investigate the effect of several scene characteristics involving crops, soils, and weather variables on the accuracy of Landsat classifications of corn and soybeans. Segments sampling the U.S. Corn Belt was classified using a Gaussian maximum likelihood classifier on data from two key acquisition periods. "Wall-to-wall" field observations of all cover types were used for training and accuracy assessment. In order to investigate the effect of scene characteristics on classification performance, a total of 29 variables was defined and estimated for each segment. Classification accuracies were generally high with an average overall test field accuracy of 84.6%, and Landsat proportion estimates were well related to ground truth proportions with coefficients of determination greater than .90 for the bias corrected Landsat proportions for both corn and soybeans. Overall accuracies were higher and presented lower segment-to-segment variability than both corn and soybean accuracies. Corn accuracies were almost as high as overall accuracies. Multilinear regression indicated that several groups of scene characteristics are important to explain the variability in the classification accuracy measures and that six independent variables were enough to explain most of the variability in these measures. Field size had a strong effect on crop classification accuracy and small fields tended to present low accuracies even when the effect of mixed pixels was eliminated. Other scene characteristics variables such as proportions of corn and soybeans, crop diversity index, proportion of all field crops together, drainage, slope, soil order, long-term average soybean yield, maximum yield, relative position of the segment in the Corn Belt, weather factor and development stage variables appeared to be the most important variables accounting for variability in classification accuracy.
Degree
Ph.D.
Subject Area
Agronomy
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.