Improving the performance of the BAIM index for burnt area mapping using MODIS data

Abstract

The analysis of fire effects and its influence on the structure, composition and functioning of terrestrial ecosystems (Michalek et al., 2000) at regional and global scales relies on the feasibility of obtaining accurate estimations on burnt areas. Remote sensing has proven to be an adequate tool for this objective and many different methodologies have been proposed in order to obtain maps of burnt area on different scales. Most methodologies are based on the use of spectral indices, some of which have been traditionally applied to estimate vegetation conditions, such as the NDVI, whereas others have been specifically designed for burnt area mapping (Kasischke and French, 1996; Barbosa et al., 1996; Martín and Chuvieco, 1998; Trigg and Flasse, 2001; Chuvieco et al., 2005). Among the latter is the Burnt Area Index (BAI) which was originally designed for burnt area mapping using NOAA-AVHRR data (Martín, 1998). BAI was recently adapted to MODIS data (BAIM) increasing its discrimination ability by including information from the MODIS SWIR bands (Martín et al., 2005). In spite of the good performance showed by the BAIM in discriminating recently burned areas in different environments (Martin et al., 2005; http://www.geogra.uah.es/aql/), some confusions have been reported regarding specific land covers. This paper proposes an improvement of BAIM by implementing two new factors in order to increase the discrimination capability and to reduce reported confusions between burnt and unburnt covers (especially those containing water bodies).

The design of the Improved BAIM index (IBAIM) is based on the statistical analysis of sample pixels obtained from MODIS images. The samples where selected using images from two different years (2001 and 2003) and include the most representative landcovers in the Iberian Peninsula as well as burned areas. The accuracy of the new index was tested using official fire perimeters obtained with GPS and also by visual interpretation of high resolution images.