Compositing criteria for burned area assessment using multitemporal low resolution satellite data

Five different multitemporal image compositing criteria were compared for the specific aim of monitoring burned areas using AVHRR-GAC 5 km data. Two of them were based on a classical maximum value technique: maximum value composite of NDVI (MVCN) and of surface temperature (MVCTS). Three others were based on a minimum value technique: minimum value composite of Channel 1 (M(i)VCB1), of Channel 2 (M(i)VCB2) and of albedo (M(1)VCA). The compositing period used was 1 week, and four output channels suitable for identifying burned areas were generated: AVHRR Channel 2 (Ch2), normalized difference vegetation index (NDVI), surface temperature (T-S), and global environment monitoring index (GEMI). The different compositing criteria were assessed based on their ability to retain burned areas and also on their choice of close to nadir pixels. The separability od burned and unburned areas for each of the compositing criteria was assessed for each of the individual output channels. The criterion that showed best results in terms of separability was M(i)VCA followed very closely by M(1)VCB2; the MVCN criterion clearly showed the least merit of all. An overall analysis of the satellite zenith angles showed that the M(i)VCB2 criterion selected the smallest angles or closer to nadir pixels followed by M(i)VCA, MVCN, MVCTs, and M(i)VCB1. Although impossible to choose a single best compositing criteria in absolute terms, M(i)VCA proved to be the best compromise because it chose near to nadir pixels and performed best in terms of separability between burned and unburned spectral signatures. Future work will determine how to combine different channels (e.g., GEMI, Ch2, T-s, NDVI) for burned area detection, once the M(i)VCA compositing has been performed. (C) Elsevier Science Inc., 1998.