METHODOLOGY FOR THE DETECTION OF OBSERVABLE PHENOMENA IN TIME SERIES OF DAILY SATELLITE IMAGES: APPLICATION TO BURNED AREAS

摘要

We have developed a pixel level methodology for observable phenomena detection in time series of daily satellite images. It consists of several steps: pre-processing, construction of the statistical variables space for a Bayesian classifier, algorithm design, obtaining probability maps and postprocessing. It has been designed a software tool to apply this methodology to detect burned areas in North American boreal forests using AVHRR images of 0.05 ° (～ 5 km). The results are comparable to those of other recently published studies that used higher spatial resolution images.