MODIS-Landsat fusion for large area 30Â m burned area mapping

- Method to map 30 m burned areas at regional scale
- Multi-temporal Landsat reflectance combined with MODIS active fire detections
- Semantic rule-based change detection
- Tested for 1.9 million km2 in the Western United States
- Comparison with MTBS burned area polygons

Fire products derived from coarse (500 m to 1 km) spatial resolution satellite data have become an important source of information for the fire science and applications communities. There is however a demand for moderate (30 m) spatial resolution burned area data sets, systematically generated at regional to global scale, that to date has been only partially met. This paper presents a methodology to fuse multi-temporal Landsat Enhanced Thematic Mapper plus (ETM +) data with Moderate Resolution Imaging Spectroradiometer (MODIS) active fire detections to map systematically burned areas at 30 m resolution. A multistage mapping approach is used with an initial per-pixel change detection based on spectral-rule based pre-classification of Landsat 30 m time series to identify candidate burned areas. The candidate burned area objects are then either retained or discarded by comparison with contemporaneous MODIS active fire detections. The methodology is demonstrated for 1.9 million km2 over the Western United States using all the Landsat 7 ETM + data and Terra MODIS active fire detections acquired in 2002. Systematic evaluation conducted by per-pixel and per-burned area object comparison with the burned area perimeters provided by the USGS Monitoring Trends in Burn Severity (MTBS) project is presented, and shows a good agreement in the identification of burning patterns but a likely underestimation of the total area burned. Future research to refine and further test the methodology is discussed.