Adaptive Gaussian-PSO XGBoost Model for Alpine Forests Aboveground Biomass Estimation Using Spaceborne PolSAR and LiDAR Data

Abstract

Accurate estimation of forest aboveground biomass (AGB) is fundamental to forest management and ecosystem monitoring. Natural forest ecosystems are an important guarantee to maintain the global ecological balance and carbon cycle, but the complex climate, dramatic topographic relief, and saturation effects make it difficult to achieve reasonable AGB estimation of alpine forests with commonly used optical data. In this study, spaceborne dual-polarimetric synthetic aperture radar and light detection and ranging data were combined to break through the limitation of optical data, and the information on the vertical structure inside the forests was extracted, to achieve high-precision forest AGB estimation and reveal the distribution pattern of forest AGB. An adaptive Gaussian-particle swarm algorithm XGBoost model (AGP-XGBOOST) was proposed to improve the forest AGB estimation, which adjusted the PSO through the built-in adaptive parameter of the Gaussian function to achieve the hyperparameter optimization for the XGBoost model. The proposed method was validated with the forest survey data, and classic machine-learning models were constructed for comparison. The comparative analysis was carried out using natural forests in the eastern Tibetan Plateau as an example, and the results showed that the proposed AGP-XGBOOST model consistently maintained the best performance across all models, and the AGB estimation errors caused by the combined data source decreased by 30.8%, 24.4%, and 10.1% compared to the independent data sources. In addition, the forest AGB mapping showed that the distribution pattern of forest AGB on the eastern Tibetan Plateau was significantly affected by terrain fluctuations.