Detection of biomass burning smoke in satellite images using texture analysis

Classification results using texture analysis is presented for forest fire smoke from satellite remote sensing data. Texture analysis is carried out for normalized difference images calculated from visible and thermal infrared images of the Indonesian forest fire in 1997. Smoke regions are identified by assuming threshold values for the resulting texture feature as well as for radiances in the original and difference images. It is found that when the thresholds are chosen appropriately for GMS visible and infrared spin scan radiometer, 94% pixels exhibit agreement between the classification results using the texture analysis and the supervised Euclidean classification. Agreement is found for 96% pixels in mutual verification using the VISSR image and a concurrent NOAA advanced very high resolution radiometer image. A correlation coefficient of 0.91 is obtained between the results from the two sensors in the variation of the number of smoke pixels accumulated for 12 days in September 1997. Additionally, it is confirmed that as the threshold value of the texture feature is increased, the variation range of the aerosol optical thickness is also increased. As a whole, this study indicates that texture analysis provides quite reasonable results in the smoke detection when appropriately combined with the spectral information.