Comparing the potential of tree-based and area-based forest height metrics for aboveground biomass estimation in complex forest landscapes

• Comparative analysis of height metrics advances understanding of AGB estimation challenges. • Machine learning demonstrates potential for improving biomass prediction accuracy. • Regional variability underscores the need for adaptive approaches to height metric selection. Forest height metrics are vital for predicting aboveground biomass (AGB), a key indicator of carbon dynamics and sustainable forest management. This study explores the comparative performance of tree-based and area-based forest height metrics in AGB estimation using 755 field plots across Beijing, incorporating field-measured tree diameter at breast height and LiDAR-derived tree heights. Two modeling approaches—parametric mixed-effects models (MEM) and non-parametric machine learning (ML) algorithms—were applied to evaluate predictive accuracy. The results indicate the following: (1) Tree-based metrics that align more closely with forestry definitions demonstrate higher predictive accuracy than area-based metrics, particularly Lorey’s mean height and top height. (2) Among machine learning models, CatBoost, which incorporates Lorey’s mean height, achieve the highest accuracy ( R 2 = 0.688, relative RMSE = 41.85 %, MAE = 18.15 Mg/ha). (3) While area-based metrics are widely used in large-scale assessments due to their scalability, our results underscore the superior precision of tree-based metrics in AGB estimations, showing an 11.0 % to 23.1 % improvement of R 2 over the area-based metrics. (4) Regional variations across Beijing further highlight the need to tailor metric selection to specific landscape and modeling objectives. These findings provide critical insights for optimizing AGB estimation methods, supporting carbon stock assessments, forest monitoring, and management strategies.