Timely delineation of burned areas is essential for post-fire assessment and recovery planning. In this study, we mapped wildfire impacts around Javanrud, Iran, using Sentinel-2 Level-2A imagery acquired before and after a summer 2022 event (pre-fire: 2022-08-16; post-fire: 2022-08-26). For each epoch, we computed the NDVI and NBR indices and derived their bi-temporal differences (dNDVI and dNBR). A Random Forest (RF) classifier was trained on a feature stack comprising three key VNIR/SWIR bands (B4, B8, B12) along with NDVI, NBR, dNDVI, and dNBR to distinguish burned from unburned pixels within an Area of Interest (AOI) defined as the burn polygon buffered by 5 km. Training labels were obtained from the provided burn shapefile (positive samples) and stratified random samples within the buffer zone (negative samples), with clouds and shadows masked out.<br />On held-out tiles, the RF model achieved an overall accuracy (OA) of 0.96, F1-score of 0.90, and AUC of 0.997, with a burned-area estimate of 13.07 ha. Permutation-based feature importance identified dNBR and SWIR–NIR features as dominant predictors, offering a physically consistent explanation of model behavior. The proposed workflow—combining simple spectral differencing with a lightweight supervised learner—provides a fast, reproducible, and cost-effective solution for post-fire mapping in heterogeneous, mountainous terrain.