An automatic monitoring method for disturbance caused by deforestation and forest fires

Timely and accurate detection and assessment of forest disturbances are crucial for early warning and risk prevention of forest disasters. This study presents an automated forest disturbance monitoring algorithm with three main steps: (1) extracting features that capture the remote sensing response of forest fires and deforestation and creating threshold maps of features; (2) using the threshold maps to extract potential forest disturbance pixels as training samples to train a one-class support vector machine (OC-SVM) and detect forest disturbance regions; and (3) applying diagnostic rules to diagnose the disturbance type as either burned or deforested areas. We validated the method using 10 global cases (five forest fires and five deforestation events). The results show that Sentinel-2 bands 6, 7, 8 and 8A, the mean textures of bands 6, 7, 8 and 8A and the difference vegetation index (DVI) are effective for disturbance detection. The overall detection accuracy of the 10 cases is 98.46% on average, with an average F1 score of 0.923 and a type diagnosis accuracy of 100%. This approach enables high-precision, automated detection and classification of forest disturbances and can be extended to monitor other disturbance types, offering a valuable tool for forest disaster response and management.