Research Article | Open Access

Volume 2026 |Article ID 100120 | https://doi.org/10.1016/j.plaphe.2025.100120

Nondestructive individual tree aboveground biomass estimation using a hierarchical Bayesian approach in combination with individual tree competition indices

Zengrui Zhang,¹ Yuting Zhao,¹ Zhen Zhen,¹ Yinghui Zhao ,¹ Jun Li,² and Yuan Zhou³

¹School of Forestry, Key Laboratory of Sustainable Forest Ecosystem Management-Ministry of Education, Northeast Forestry University, Harbin, 150040, PR China
²Heilongjiang Geological Science Institute, Harbin, 150040, PR China
³Ecosystem Big Data Research and Development Center, School of Forestry, Northeast Forestry University, Harbin, 150040, PR China

Abstract

Ecological variables like aboveground biomass (AGB) are often spatially autocorrelated, and AGB prediction may be underestimated if spatially correlations are ignored in remote sensing–based models. Thus, incorporating spatial correlations into AGB prediction models is crucial for accurate AGB estimation, especially in natural secondary forests with complex structures and intense competition. Terrestrial laser scanning (TLS) enables fine-scale and nondestructive measurements of individual trees while reconstructing the complete spatial structure and competitive relationships of the forest. Consequently, the utilization of TLS data for developing an individual tree AGB model that considers competition permits nondestructive AGB estimation at both the tree and plot levels. Focusing on 13 natural secondary forest sample plots located in northeast China, this study combined UAV and TLS LiDAR data to explore the applicability of the hierarchical Bayesian spatial approach (INLA-SPDEs) to nondestructive individual tree AGB estimation in natural secondary mixed forests. The analyses also considered the effect of the individual tree competition indices. This study used the INLA-SPDEs method to construct four models (a base model, a Bayesian spatial model, a hierarchical Bayesian model, and a hierarchical Bayesian spatial model) to estimate individual tree AGB. The results showed that relative to the base model (R2 = 0.836), the model fitting accuracy of the models incorporating random effects were improved, while the hierarchical Bayesian spatial model that included two random effects had the best estimation results (R2 was increased by 13.52 %, and the RMSE was decreased by 53.34 %). The results of the study indicate that the INLA-SPDE method that considers spatial autocorrelation is both efficient and robust for biomass estimation. Integrating Bayesian, spatial correlation, and individual tree competition factors allowed us to implement effective AGB estimation for complex forest ecosystems with significant hierarchical structures. The results thus provide strong support for spatial modeling and the analysis of ecological processes.