Comparing 30 Meter Imagery From Landsat 5 and 7 for Crop Area Estimation

摘要

This paper compares digital, multi-spectral imagery from the Landsat 5 Thematic Mapper with that of the Landsat 7 Enhanced Thematic Mapper when used to discriminate crop types for area estimation. Comparisons are done for different types of crop areas in several states, using images that are only one day apart. The overlaps between adjacent paths in several major crops states are used to define the analysis areas. A simple non-parametric paired sample sign test is used to determine statistical significance of differences. Standardized techniques as used in the Agency's Cropland Data Layer Project are used for image processing, including a modified supervised pattern recognition /clustering approach for cover type signatures and maximum likelihood for categorization. Ground truth data consist of 211 one and two square mile areas selected in a stratified random sample and visited in June of the corresponding crop year. Sampling rates for the ground data range from one in 30 to one in 166 depending on the state and land use stratum.