Mapping above-ground biomass and canopy mean height in high mountainous forest areas with Sentinel-2 multi-spectral image based on machine learning algorithms

Forest mean canopy height (CMH) and aboveground biomass (AGB) are key indicators of forest ecosystem productivity. However, in high-altitude mountainous areas, complex topography and limited field data make accurate assessments challenging. This study was conducted in the Taohe National Nature Reserve, Gansu Province, China. The Sentinel-2 imagery and DEM were fused to estimate the AGB and CMH using four machine learning algorithms: random forest (RF), XGBoost, CatBoost, and multilayer perceptron. The stratified sampling was used to reduce the underestimation of high AGB values and overestimation of low AGB values in RF training. Results show that the RF model had the best estimation effect on AGB (R² = 0.6612), and the CatBoost model had the best estimation effect on CMH (R² = 0.7394). SHapley additive exPlanations (SHAP) analysis revealed that slope aspect was the greatest impact on CMH, whereas the Sentinel-2 blue band (B2) had the greatest impact on AGB. CMH plays an important role in improving the accuracy of AGB estimation. Based on the optimal model, AGB and CMH maps with a resolution of 10 m were generated, revealing a clear east–west biomass gradient. This study highlights the effectiveness of combining machine learning with SHAP-based interpretability for forest monitoring in complex mountainous environments.