Agreement between MODIS hotspots and burned area perimeters in mediterranean fires

MODIS hotspots have been widely used as a proxy of fire occurrence, especially in those areas where forest fire statistics are scarce of not available. Even acknowledging the interest of hotspots to analyze fire patterns, they do not show all characteristics of fire regimes in a particular area, as the detection rates are very much influenced by fire length and size. This paper compares a time series of fire perimeter data generated from multitemporal analysis of Landsat-TM data and hotspots derived from the MODIS sensor (MOD14). The five study sites are located in Mediterranean areas of Spain, Portugal and central Chile. The time series includes several years for each study site, to tackle temporal trends in fire occurrence to hotspots relations. The comparison between hotspots and fire perimeters shows that the MODIS MOD14 product detects large fires (>500 ha) with high reliability (omission errors generally lower than 5%), but they do not properly detect small fires (<50 ha), which have omission errors close to 80%. Commission errors (false positives) are generally around 15% and mostly occur in agricultural areas with dry-hot soils in summer time. The confidence level assigned by the MOD14 product designers shows a clear relation with the detection level, although the intermediate confidence level also shows good detection levels. For the study sites, the average burned surface for each hotspot was calculated at 81 ha.