Monitoring large areas for forest change using Landsat: Generalization across space, time and Landsat sensors

Landsat 7 ETM+ provides an opportunity to extend the area and frequency with which we are able to monitor the Earth's surface with fine spatial resolution data. To take advantage of this opportunity it is necessary to move beyond the traditional image-by-image approach to data analysis. A new approach to monitoring large areas is to extend the application of a trained image classifier to data beyond its original temporal, spatial, and sensor domains. A map of forest change in the Cascade Range of Oregon developed with methods based on such generalization shows accuracies comparable to a map produced with current state-of-the-art methods. A test of generalization across sensors to monitor forest change in the Rocky Mountains indicates that Landsat 7 ETM+ data can be combined with earlier Landsat 5 TM data without retraining the classifier. Methods based on generalization require less time and effort than conventional methods and as a result may allow monitoring of larger areas or more frequent monitoring at reduced cost. One key component to achieving this goal is the improved availability and affordability of Landsat 7 imagery. These results highlight the value of the existing Landsat archive and the importance for continuity in the Landsat Program. (C) 2001 Elsevier Science Inc. All rights reserved.