Dating Fire Events on End of Season Maps of Burnt Scars

Abstract

Forested areas cover circa 38% of Continental Portugal and the observed increasing trend in the extent and severity of wildfires points to the need for accurate and timely knowledge of the total burnt area. The official fire database is the one provided by the Portuguese Forest Service (DGRF) and is based on information supplied by fire and forest services. Since 1990, maps of burnt areas have been yearly produced based on information from Landsat-TM. A recent study for the period 1984-1989 has pointed out severe discrepancies between ground- and satellite-based data, raising the need to devise procedures aiming to correct such discrepancies. The present work represents a first attempt to assess the potential of using NOAA/AVHRR imagery to assigning dates to burnt scars on end of season maps as the ones derived from Landsat-TM. We begin by degrading to the AVHRR scale a MODIS-based end of season map of burnt scars. Degradation was simply performed by computing the fraction of burnt MODIS pixels inside each AVHRR pixel. Then, we built up a neuro-fuzzy model that assigns to each AVHRR pixel the "possibility" of representing a burnt area. The model uses as input pixel values of AVHRR channel 2 and was trained using a composite of minimum values of that channel and the corresponding fractions of burnt MODIS pixels of the degraded end of season map. The model was then applied to individual AVHRR images and further refined in order to eliminate errors associated to contamination by clouds and water bodies, geo-rectification problems and dark backgrounds. It is shown that the refined model underestimated by 11 % the total amount of burnt area as obtained from DGRF data and that the differences reduced to 1% when DGRF data were restricted to records greater than 100 ha. The model was then used to assign dates to burnt scars in the MODIS-based end of season map. Obtained results are quite encouraging since deviations (NOAA-MODIS) between -2 (-1) and +1 (0) days represent 85% (70%) of the total and may be attributed to differences in orbital times of passage of NOAA and TERRA/AQUA.