Vegetation detection through smoke-filled AVIRIS images: An assessment using MODIS band passes

Radiometrically calibrated, Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) images acquired during the Smoke, Clouds and Radiation in Brazil (SCAR-B) experiment were processed to simulate vegetation index (VI) imagery with the Moderate Resolution Imaging Spectroradiometer (MODIS) band passes. Data sets were extracted from tropical forested areas, burned fields, and shrub/grassland areas over both clear and variable smoke conditions with average aerosol optical thickness (AOT) values at 0.67 mu m of 0.14, 1.1, and 1.9, respectively. The atmospheric resistant VIs and various middle-infrared (MIR) derived VIs were then analyzed with respect to their ability to minimize atmospheric "smoke" contamination, The atmospheric resistant VIs utilized the blue band for correction of the red band, while the MIR-derived VIs used the MIR region (1.3 - 2.5 mu m) as a substitute for the red band since it is relatively transparent to smoke, yet remains sensitive to green vegetation. The performance of these indices were assessed and compared with the normalized difference vegetation index (NDVI) and the soil-adjusted vegetation index (SAVI). Over the tropical forests the NDVI and SAVI had high relative errors over all smoke-filled atmospheric conditions (50-80% error), while the atmospheric resistant VIs resulted in a 50-80% relative error only over thick levels of smoke. Over optically thin levels (AOT at 0.67 mu m < 1.1) they performed much better with a 20-40% relative error. The MIR-derived VIs, on the other hand, outperformed all other Vis over forested areas (less than or equal to 5% error). However, over burned fields with minimal amounts of green biomass the MIR-derived VIs had the highest levels of error due to smoke (> 40%), while all other indices had errors below 20%. In the shrub/grassland site, the atmospheric resistant indices behaved similarly with the MIR-derived indices, with both less sensitive to smoke than the NDVI and SAVI. We conclude that the MIR indices, particularly with MODIS band 7 (2.13 mu m), are useful in vegetation monitoring over forested areas during the burning season. However, they did not perform well in areas outside of forests such as burned areas and shrub/grassland.