The increasing of forest fires highlights the importance of rapidly and accurately quantifying the burned areas, which is crucial evidence for determining the cause of fire and assigning responsibility. This study develops an operational tool based on the cloud-computing abilities of the PIE-Engine platform and Sentinel-2 satellite imagery. The burned areas are rapidly extracted by applying differential spectral indices such as dNDVI, dNBR and dNBR2 using the adaptive thresholds determined by the Otsu algorithm. Compared with field investigations, the accuracy of using a wildfire case in Liuhe Village, Hubei Province is high. The findings indicate that dNBR can achieve an area accuracy of 98.22 % and a pixel-level F1-score of 0.94, significantly higher than dNDVI. Although the accuracy of the random forest model is slightly high, the dNBR based method achieves an excellent balance between computational efficiency (about 2 s of processing time) and accuracy. In addition, we have developed a user-friendly web application that allows for custom parameter settings and visualizes results as a quantitative burned area map. This tool only requires a web browser, greatly reducing the technical barriers of remote sensing applications and providing a transparent, efficient, and accessible solution for supporting forest fire investigation and emergency response.
Mao et al. (Thu,) studied this question.