By incorporating both species-specific variability and spatial random effects, the method significantly improves prediction accuracy while remaining fully nondestructive. The findings demonstrate that LiDAR-derived tree attributes, combined with Bayesian spatial modeling, offer a powerful framework for quantifying biomass in complex forest ecosystems and strengthening carbon accounting efforts.
Natural secondary forests are a dominant component of northeastern China’s landscapes and an important terrestrial carbon reservoir. Their dense, mixed-species structures create strong competition among adjacent trees, influencing growth patterns and biomass distribution. Accurately estimating aboveground biomass (AGB) is critical for understanding forest carbon storage.Traditional tree-level AGB estimation methods rely on destructive sampling and often ignore spatial autocorrelation, leading to systematic underestimation. Meanwhile, LiDAR technologies—including UAV laser scanning and terrestrial laser scanning—now provide precise, nondestructive measurements of tree height, diameter, and crown structure. Yet few models incorporate these detailed structural attributes together with competition metrics and spatial variability. Due to these challenges, there is a strong need for advanced modeling techniques capable of capturing ecological complexity in biomass estimation.
A study (DOI: 10.1016/j.plaphe.2025.100120) published in Plant Phenomics on 17 October 2025 by Yinghui Zhao’s team, Northeast Forestry University, provides a robust foundation for nondestructive, high-precision biomass estimation at tree and stand levels.
Using a combined UAV–TLS LiDAR workflow, the study first delineated individual tree crowns and extracted tree-level structural parameters before developing aboveground biomass (AGB) models that incorporated competition indices and hierarchical Bayesian spatial components. Tree segmentation accuracy was assessed across 13 sample plots, and competition was quantified using both distance-independent metrics derived from LiDAR-measured diameter and height distributions, and distance-dependent indices calculated within radii ranging from 2.5 to 15 m. Stepwise regression and multicollinearity tests were then applied to identify optimal predictors for AGB modeling, after which four models—a base regression model, a hierarchical Bayesian model, a Bayesian spatial model, and a hierarchical Bayesian spatial model—were constructed using the INLA-SPDE framework to test the influence of species-level and spatial random effects. Corresponding results showed strong segmentation performance, with an average F-score of 0.88 and successful detection of 856 out of 971 reference trees. LiDAR-derived DBH and height estimates closely matched field measurements (R² = 0.985 and 0.926), validating their use in model development. Analysis of competition metrics revealed skewed distributions for height-related indices and identified a rapid increase in competitive intensity up to a radius of 10 m, which was therefore selected as the optimal spatial window for defining competing trees. Model comparisons demonstrated clear performance gains from incorporating random effects: relative to the base model, adding species effects improved R² by 11.84% and reduced RMSE by 42.06%, while adding spatial effects improved R² by 4.67% and reduced RMSE by 21.98%. The hierarchical Bayesian spatial model, which integrated both effects, delivered the highest accuracy (R² = 0.935; RMSE = 98.29 kg), narrower standard errors, and the most robust parameter estimates, particularly for DBH-related predictors, confirming the superiority of this integrated modeling strategy.
Improved AGB prediction supports accurate carbon stock assessments, ecosystem monitoring, and forest restoration planning. The method also enables deeper ecological insights by integrating competition dynamics, spatial patterns, and species-level variability into biomass models. As global carbon markets and climate policies increasingly depend on reliable forest carbon estimates, the integration of LiDAR data and Bayesian spatial modeling offers a scalable and transparent solution for national forest inventories, REDD+ programs, and long-term ecological research. The framework also supports future expansion into multispecies forests and multi-sensor data environments.
###
References
DOI
10.1016/j.plaphe.2025.100120
Original Source URl
https://doi.org/10.1016/j.plaphe.2025.100120
Funding information
This work was supported by the Key Project of National Key Research and Development Plan [2023YFF1304003]; National Natural Science Foundation of China [32071677]; and National Forestry and Grassland Data Center-Heilongjiang platform [2005DKA32200-OH].
About Plant Phenomics
Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.