Alonso-Canas, Itziar; Chuvieco, Emilio (2015). Global burned area mapping from ENVISAT-MERIS and MODIS active fire data. REMOTE SENSING OF ENVIRONMENT, 163, 140-152.

Abstract
We present the development of a global burned area (BA) algorithm based on MERIS imagery along with the assessment of the global BA results for three years (2006-2008). This work was developed within the Fire Disturbance project under the European Space Agency Climate Change Initiative programme, which aimed to generate long-term BA information for climate models. Our algorithm combined temporal series of MERIS reflectances with thermal information from MODIS HS (hotspot) product. The algorithm included two-steps. Firstly, cumulative distribution functions were computed to discriminate the most clearly burned pixels using regionally-oriented near infrared reflectance thresholds. Secondly, a contextual criterion improved the spatial detection of the burned patches from the seed pixels. BA estimates for the three target years range from 3.6 to 3.8 million km(2). Results were validated from a statistically designed sample of fire perimeters generated from Landsat multi-temporal imagery. Intercomparison with existing BA products was also carried out. Results from global validation datasets provided an average overall accuracy higher than 0.95. The accuracy of the BA category was lower than the accuracy of the unburned one. Within the former, average omission and commission errors were lower for areas where the proportion of burned area was higher (for pixel-based error matrices, commission error (CE) was 0.52 and omission error (OE) was 0.51), than for those areas with very low BA proportion (CE = 0.60 and OE = 0.74). In terms of total BA estimation, errors were generally well balanced, with a tendency towards underestimation.(34%). Relevant impact of temporal reporting accuracy was found for these validation datasets. Intercomparison with other existing BA datasets pointed out similar spatial and temporal trends, with high correlation with GFED v4 BA estimations for the three years (r(2) > 0.974). (C) 2015 Elsevier Inc All rights reserved.

DOI:
10.1016/j.rse.2015.03.011

ISSN:
0034-4257