Mapping forest biomass is key for promoting sustainable forest management and carbon budgets at both local and larger scales. We examined a machine learning (ML) model driven by domain-wide airborne LiDAR, spaceborne SAR, and optical data to estimate the forest biomass at a fine spatial resolution of 20 m × 20 m across Taiwan. Four ML algorithms were evaluated to determine the most effective biomass estimation method, and the effectiveness of 23 predictive variables derived from multisensor data was assessed. One of the machine learning models, extreme gradient boosting (XGBoost), yielded the best results compared to the other three algorithms (multilayer perceptron neural network (MLPNN), K-nearest neighbor (KNN), and random forest (RF)), with a coefficient of determination (R2) for the national forest inventory of 0.67, 0.71, 0.54, and 0.52 for broadleaf, conifer, mixed and bamboo forests, respectively. The feature importance values of XGBoost model indicated that the canopy height model (CHM) was the most influential factor in estimating the aboveground biomass (AGB). Our estimation revealed average dry weight AGB density values of bamboo, broadleaf, conifer, and mixed forests of 63.10, 201.94, 179.05, and 195.88 Mg ha−1, respectively, with an average of 189.39 Mg ha−1 across Taiwan. Total forest AGB in Taiwan is 403.99 million Mg, corresponding to a carbon stock of 190.57 million Mg C. This study not only contributes to the development of effective forest management strategies and climate change actions in this region but also highlights the importance of leveraging advanced technologies for sustainable forest management practices.
Nguyen et al. (Thu,) studied this question.