Above-ground biomass estimation in tropical dry deciduous forests using multi-sensor data from ALOS-2 and sentinel-2 with advanced machine learning approaches

Abstract

Accurate estimation of above-ground biomass (AGB) is essential for sustainable forest resource management and carbon stock assessment. This study focuses on the Pembi Forest, Telangana, India, where synthetic aperture radar (SAR) data from ALOS-2 PALSAR-2 L-band and multispectral data from Sentinel-2 were integrated with biophysical information to develop AGB prediction models. Predictor variables included SAR backscatter and polarization-derived features, Sentinel-2 spectral indices and band combinations, principal components, leaf area index (LAI), and global canopy height. Light Gradient Boosting Machine (LGBM), Support Vector Regression (SVR), Extreme Gradient Boosting (XGB), Artificial Neural Network (ANN), and a weighted voting ensemble were evaluated. The ensemble model achieved the highest predictive performance, with R² = 0.67, RMSE = 24.35 t ha⁻¹ and MAE = 14.97 t ha⁻¹. Spatial AGB distribution maps showed prominent variation in the study area, providing valuable information for resource management and conservation planning. The findings highlight the potential of combining SAR and multispectral remote sensing data with machine learning techniques to improve accuracy of AGB estimation, surpassing the performance of conventional single-model methods and advancing the understanding of forest biomass dynamics.