Machine Learning Approach to Biomass Estimation: Integrating Satellite and Ground Data in Sal Forests of Jharkhand

Accurately estimating Above Ground Biomass (AGB) is crucial for understanding forest carbon dynamics and improving ecological monitoring. This study refines AGB estimation in tropical Sal forests using machine learning (ML) techniques and vegetation indices (VIs) derived from Sentinel imagery. By analysing correlations among different VIs, both positive and negative correlations influenced by environmental factors and the intrinsic properties of the indices are identified. The study compares several ML algorithms, integrating Sentinel imagery with ground-based measurements for enhanced prediction accuracy. AGB values in the study area range from 10.531 to 254.930 Mg/ha, averaging 125.25 ± 61.12 Mg/ha. Among the tested ML models, XGBoost, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM) demonstrated strong performance, with R² values of 0.83, 0.81, 0.80, and 0.78, respectively. Random Forest (RF) and Decision Tree models performed moderately well, with R² values of 0.71 and 0.79, while the Generalized Additive Model (GAM) showed the lowest performance with an R² of 0.65. This scalable, non-destructive approach provides valuable insights for environmental management, offering critical support to researchers, policymakers, and forest management professionals in monitoring biomass and managing forest resources more effectively.