Abstract: Forest fires are a critical disturbance factor affecting ecosystem balance, timely and accurate monitoring of burned area extent and assessment of vegetation recovery essential for ecological restoration are important. In this study, we investigated the "3·30" forest fire that occurred on Yaji Mountain, Beijing, in 2019 using multi-temporal Sentinel-2 imagery. We comprehensively applied differential index thresholding methods (dNDVI, dBAI, and dNBR) and supervised classification approaches (Maximum Likelihood Classification (MLC), Support Vector Machine (SVM), and Random Forest (RF)) to extract burned areas, and employed the Enhanced Vegetation Index (EVI) to evaluate vegetation recovery dynamics during 2018―2025. The results showed that the Random Forest algorithm achieved the highest extraction accuracy, with an overall accuracy of 89.72% and a Kappa coefficient of 0.79. The total burned area in the study region was 91.1 hm2, of which low-severity burns accounted for 80.8%, while moderate-severity burns accounted for 14.8%，while high-severity burns accounted for 4.4% and were mainly distributed around the scenic area and within dense Pinus tabuliformis forests, indicating notable ecological risks. Time-series EVI analysis revealed that post-fire vegetation exhibited a rapid recovery trend; by 2021, EVI values in most areas had recovered to pre-fire levels, and the recovery process entered a relatively stable stage after 2022. However, this recovery mainly reflected the increased coverage of understory shrubs and grasses, whereas the long-term recovery of tree communities requires continuous monitoring. This study demonstrated the effectiveness of the Random Forest method for burned area extraction in complex mountainous environments, and the findings provide a scientific basis for post-fire ecological restoration and sustainable forest management.

Key words: burned area, vegetation index, random forest, vegetation restoration