Mapping forest fire impact from LANDSAT-TM imagery

We address the problem of estimating fire impact in Mediterranean forests based on Landsat-TM imagery, for which we have used a forest fire in Andalucia (S. Spain) as a case of study. We processed two LANDSAT-TM scenes, acquired before and after the date of the fire. The post-fire scene was segmented and a table of segment statistics was submitted to a hierarchical model-based agglomerative clustering. The boundary defined by the classification closely matches the boundary defined by helicopter. We also produced an image classification of the pre-fire image based on image segmentation and canonical analysis and studied the trajectories of burnt and not-burnt centroids in the Kauth-Thomas plane. The trajectories indicated the existence of a differential response to fire and to phenologic change. We modeled the post-fire conditions as if the region had not been burnt and defined an Index of Fire Impact as the difference between the actual and the modeled second Kauth-Thomas component. The Index of Fire Impact is significantly related to field estimates, but with an scatter that introduces uncertainity in the inversion.