A multi-source data fusion approach for high-resolution mapping of forest structure and carbon storage

Accurate estimation of forest carbon storage is essential for implementing nature-based climate solutions and achieving carbon-neutrality goals, particularly in China, which has the world's largest afforestation efforts. Although remote sensing offers powerful tools, accurately quantifying carbon in heterogeneous forest landscapes remains challenging due to the complex vertical structure and the difficulty of scaling plot-level measurements. This study presents a multi-source data fusion framework that integrates unmanned aerial vehicle LiDAR, airborne LiDAR, Sentinel-1/2 satellite imagery, and ground inventory data to produce a high-resolution map of forest carbon storage in the Tongshan District of Xuzhou City. The key lies in stratifying forests into single- and multi-layer types using canopy entropy (CE)—a metric derived from LiDAR point clouds that quantifies structural complexity. Separate random forest models were then developed for each forest type, with input features optimized using SHAP analysis. This stratified modeling approach significantly improves prediction accuracy (RMSE%: 6.23% for single-layer and 8.61% for multi-layer forests) compared to a unified model (RMSE% = 10.23%). The resulting carbon map estimates the total forest carbon storage in Tongshan at 1.30 Tg, revealing substantial carbon sequestration potential in rehabilitated mining areas associated with ecological restoration projects. This method demonstrates that incorporating structural stratification based on CE can effectively reduce estimation errors caused by forest heterogeneity, providing robust technical support for regional high-precision carbon monitoring.