A method for extracting burned areas from Landsat TM/ETM+ images by soft aggregation of multiple Spectral Indices and a region growing algorithm

Since fire is a major threat to forests and wooded areas in the Mediterranean environment of Southern Europe, systematic regional fire monitoring is a necessity. Satellite data constitute a unique cost-effective source of information on the occurrence of fire events and on the extent of the area burned. Our objective is to develop a (semi-)automated algorithm for mapping burned areas from medium spatial resolution (30 m) satellite data. In this article we present a multi-criteria approach based on Spectral Indices, soft computing techniques and a region growing algorithm; theoretically this approach relies on the convergence of partial evidence of burning provided by the indices. Our proposal features several innovative aspects: it is flexible in adapting to a variable number of indices and to missing data; it exploits positive and negative evidence (bipolar information) and it offers different criteria for aggregating partial evidence in order to derive the layers of candidate seeds and candidate region growing boundaries. The study was conducted on a set of Landsat TM images, acquired for the year 2003 over Southern Europe and pre-processed with the LEDAPS (Landsat Ecosystem Disturbance Adaptive Processing System) processing chain for deriving surface spectral reflectance ρi in the TM bands. The proposed method was applied to show its flexibility and the sensitivity of the accuracy of the resulting burned area maps to different aggregation criteria and thresholds for seed selection. Validation performed over an entire independent Landsat TM image shows the commission and omission errors to be below 21% and 3%, respectively.