Multi-Source Remote Sensing Data Fusion for Aboveground Biomass Estimation in Tropical Forests: Recent Advances, Challenges, and Future Trends

Accurate estimation of aboveground biomass (AGB) in tropical forests is critical for global carbon cycle research and climate policies such as REDD+, but single remote sensing sources face limitations: optical sensors suffer from saturation and cloud interference, SAR has saturation and terrain-induced distortions, and lidar (e.g., GEDI) provides high-precision vertical structures but sparse coverage. Multi-source data fusion - including data-level, feature-level, decision-level fusion, and spatio-temporal scaling - leverages the complementary strengths of optical, SAR, and LiDAR data to mitigate these issues, machine learning (e.g., RF, XGBoost) and deep learning (e.g., CNN, Transformer) to improve performance, and hybrid models to balance learning ability and robustness. Key challenges including data heterogeneity, geometric/radiometric inconsistencies, saturation in dense forests, cloud-induced optical data loss, and sparse GEDI footprints are addressed through validation frameworks (involving forest maps, NFI, airborne LiDAR) and uncertainty quantification methods (e.g., spatial cross-validation, Monte Carlo simulations). Regional case studies (Amazon, Congo Basin, Southeast Asia) show optimal strategies for specific environments, while future directions focus on collaborative GEDI with P-band SAR, dynamic monitoring through temporal fusion, physical based deep learning, federated/transfer learning, interpretable AI, and standardized REDD+ protocols. Based on a systematic review of AGB estimation studies in tropical forests in the past five years, this paper summarizes the development progress of multi-source remote sensing data fusion and advanced algorithms, identifies the existing technical bottlenecks, and proposes a future-oriented comprehensive optimization framework to improve the accuracy and robustness of AGB estimation.