Fine Extraction of Forest Burned Area by Using Fusion Visible Light UAV Image with Sentinel−2A image

We selected a forest fire which occurred in Xinglong Village, Majie Town, Yiliang County, Kunming City in April 2020 as our research case. First, we fused visible light image(R, G, B) derived from unmanned aerial vehicle(UAV) and Sentinel−2A multi-spectral imagery by using Gram−Schmidt(GS) and PC Spectral Sharping(PCSS) algorithms respectively. Second, 6 quantitative evaluation indicators were selected to evaluate the quality of the fusion results. Finally, fusion image was used to extract the forest burned area within the burned boundary based on Random Forest(RF) algorithm, then we compared the accuracies of extracting forest burned area with Sentinel−2A image only. The burned areas extracted from the 2 images were compared and analyzed with the appraisal personnel through field investigation, GPS coordinates and manual vectorization of forest burned area based on UAV images. The findings indicated that the forest burned area extraction accuracies of UAV and Sentinel−2A fusion image and Sentinel−2A image only were: producer's accuracy(96.14%, 95.18%), user's accuracy(97.79%, 96.57%), and Kappa coefficient(0.83, 0.76), respectively. Compared with the forest burned area derived from artificial map which combined field investigation with vectorization by appraises, the relative errors were −3.5% for fusion image and −6.2% Sentinel−2A image only. The results show that the fusion image of UAV and Sentinel−2A image for forest burned area extraction by RF has a higher accuracy than that of Sentinel−2A image only, and the details of the boundary between the burned area and the unburned area are more obvious. Our method can greatly improve the efficiency of judicial appraisal of the forest fire and get an accurate burned area, especially, increase the scientificity and objective of the judicial appraisal of the forest fire.