Applicability of Multi-date Land Cover Mapping using Landsat-5 TM Imagery in the Northeastern US

Abstract

In many situations, multi-date image classification improves classification accuracies. However, with improved accuracies comes increased image processing time and effort. This work investigates the circumstances under which multi-date image classification is significantly better than single-date classification using Landsat-5 TM imagery for southeastern New Hampshire. Multiple Landsat images were processed for every three years from 1986 to 2010 and classified using an object-based image analysis approach (OBIA) with a classification and regression tree (CART) technique. Two maps were created for each of the mapping years, one using a single image, and another using multiple images from that year. The multi-date classification process generally performed better than the single-date process. However, the significance of the improvement was primarily dependent on the accuracy of the single-date map. Therefore, if the accuracy of the single-date classification is acceptable, it may not be necessary to perform the multi-date classification.