Burned area detection using Sentinel-1 SAR data: A case study of Kangaroo Island, South Australia

Abstract

Burned-area maps are useful in disaster management and in response to bushfire events. In this paper, we explored the capabilities of synthetic aperture radar (SAR) Sentinel-1 in detecting and mapping the bushfire-affected areas. Fires in Kangaroo Island, Australia, in 2019–20 known as the “Black Summer” were selected as a case study. We applied a random forest method to the Sentinel-1 image classification to detect the burned areas over Kangaroo Island. Radar burn difference (RBD), radar burn ratio (RBR), and delta modified radar vegetation index (ΔRVI) were calculated and imported as inputs to the random forest classifier. An independent reference map was generated using the difference normalize burn ratio (dNBR) and Sentinel-2 images and was used as the ground truth to evaluate the accuracy of the SAR-based burned-area detection map. Our results show that the SAR-based burned area detection map outperforms the MODIS MCD64. The feature importance in the random forest method indicates that RBDVH is the most important index (importance value of 0.35) followed by RBDVV (0.20), ΔRVI (0.18), RBRVH (0.17), RBRVV (0.10). The random forest method's precision, accuracy and kappa index were 94%, 94%, 0.87, respectively, while corresponding metrics for the MODIS MCD64 products were 92%, 91%, 0.83, respectively.