Abstract Accurate estimation of forest aboveground biomass (AGB) is critical for quantifying carbon stocks, assessing ecosystem productivity, and informing climate mitigation strategies. Initially reliant on destructive sampling and allometric equations, AGB estimation has evolved with the proliferation of satellite missions, including Landsat, Sentinel, and GEDI. Modern AGB assessments leverage multispectral, radar (SAR), and LiDAR data—each with distinct capabilities and limitations. Optical sensors provide vegetation indices like NDVI and IPVI but suffer from saturation in dense forests. SAR offers cloud penetration and structural insights yet struggles with noise and moisture sensitivity. LiDAR provides precise vertical structure measurements, excelling in high-biomass regions but is limited by cost and coverage. ML algorithms such as Random Forest (RF), Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Gradient Boosting Machines (GBM) enhance biomass predictions by capturing complex non-linear relationships within multi-source geospatial data. Recent studies report R² values ranging from 0.70 to 0.86 using integrated datasets, highlighting the superiority of ensemble and hybrid approaches. A temporal analysis reveals a shift from linear regressions to deep learning models like CNN-LSTM, enabled by cloud computing platforms (e.g., Google Earth Engine) and increased data accessibility. The co-evolution of sensor technology and ML techniques has significantly improved AGB estimation accuracy and scalability. This review underscores the potential of integrated geospatial-ML frameworks to support sustainable forest management and global carbon accounting, establishing a strong foundation for future AI-driven environmental monitoring systems.
Kumar et al. (Fri,) studied this question.