CHANGE DETECTION FOR MONITORING FOREST DEFOLIATION

Monitoring of environmental conditions such as forest defoliation by insects over large areas is facilitated by automated approaches to change detection using remotely sensed data. This study evaluated four change detection techniques using multispectral, multitemporal SPOT data for identifying changes in hardwood forest defoliation caused by gypsy moth, Lymantria dispar L. The change detection techniques considered were principal components analysis, image differencing, spectral-temporal (layered temporal) change classification, and post-classification change differencing. The study area comprised approximately 148 square kilometres in Warren and Shenandoah Counties, Virginia. Reference information of defoliation were aerial sketch maps developed by the U.S. Forest Service. Results indicate that defoliation may be best determined by image differencing and principal components analysis. A pair-wise test of significance determined that the four techniques resulted in significantly different accuracies at a 95 percent probability level. Principal components and image differencing analyses are generally more complex than post-classification because data no longer represent actual sensor data values, and classification involves identifying change, rather than cover, classes. These techniques are simpler than post-classification approaches, which require independent classification prior to change detection.