Spectral signature analysis of false positive burned area detection from agricultural harvests using Sentinel-2 data

Abstract

Accurate mapping of burned area is of key importance for fire emissions modeling and post-fire rehabilitation planning. In this research, Sentinel-2 data were used to analyze the difference in spectral signature between burned area and false positives from agricultural harvests. 26 fires that were mapped in the field using Global Navigation Satellite System during 2017 and 2018 were analyzed over California and Utah, USA. Individual Sentinel-2 bands and a wide range of commonly used spectral indices for burned area were tested using a spectral separability index. The separability index assessed discrimination between the classes burned area and 1) unburned area and 2) area in agricultural land that were flagged as false positive from agricultural harvest. Separability values higher than one indicate good separation and the higher the values, the better the separation. For each class, we first determined the multitemporal difference, i.e. the absolute value of the pre-minus-post-change value. Second, we compared with other classes by using the spectral separability index. We found that for the burned-to-unburned comparison the near and shortwave infrared spectral regions and spectral indices that make use of these spectral regions, were the best discriminators (separability (M) values of approximately 2), corroborating findings from earlier works. For the burned-to-agricultural false positive comparison Sentinel-2 bands 4 and 5, corresponding with the Red and Red-edge spectral regions, were the best discriminator (M−values greater than 2). Consequently, spectral indices containing the Red band show a similar strong separability of agricultural false positives. The results from our research reveal an additional layer of information that could be exploited to minimize false positives in agricultural lands in space-borne burned area products.