Improving automated detection of land cover change for large areas using Landsat data

Detection of land cover change using multi- temporal data is crucial for many Earth Science issues. We examine the potential of several alternative approaches for automated detection of land cover change using Landsat data. Specifically we test their efficacy in consistently detecting change for large areas by examining their performance for Eastern Paraguay. We used Landsat imagery with limited training data solely from 10 km squares located at one degree intersections. Traditional classifiers showed a substantial decrease in accuracy when extrapolating from one scene to multiple other scenes, whereas decision tree and support vector machine classifiers maintained accuracy.