A methodological framework for decomposing the error budget of ALS-based above ground biomass estimation using a 3D forest reconstruction model
Science of Remote Sensing · 2026-12 · 14 · 100540
DOI: 10.1016/j.srs.2026.100540 ↗Kurz gefasst
Fachartikel in der Zeitschrift Science of Remote Sensing, 2026-12. Verfasst von Junliu Yang und 4 weiteren Autor:innen. Die Arbeit nennt 96 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
: Accurate quantification of tropical forest carbon stocks is essential for understanding global carbon cycle dynamics and evaluating climate mitigation strategies, yet substantial methodological hurdles persist in how to use airborne laser scanning observations for above-ground biomass mapping. Current area-based approaches rely on statistical relationships between ALS metrics and field-measured datasets, but systematic uncertainties from field measurement errors, allometric variability, wood density heterogeneity, and forest structural complexity, remain poorly evaluated. We propose to address this limitation by using forest structure model simulations to systematically decompose the error budget for ALS-based AGB prediction at the Paracou experimental forest in French Guiana. Using virtual forest scenarios, we conducted a sensitivity analysis of prediction error to individual error sources in power-law models relating AGB to ALS-derived mean canopy height. At 1 ha resolution, the standard approach, relying on pantropical allometries and species-specific wood density data, achieved 10.3% rRMSE. This was nearly twice as large as when all controllable error sources were eliminated (rRMSE = 5.71%). Within this framework, hierarchical error decomposition identified wood density and allometric relationships as the two largest controllable error sources in the 3D model, while co-registration between field and ALS data contributed comparatively little. Stem density and quadratic mean diameter explained 95% of the remaining forest type bias. However, comprehensive evaluation of complex lidar metrics (height quantiles, gap fraction, vertical complexity index, leaf area density profiles) at 1 ha resolution revealed negligible improvement over baseline MCH, indicating a fundamental limitation: under the structural conditions at Paracou canopy-centric metrics characterize vertical structure but show limited capacity to capture the below-canopy stem attributes that drive residual error. Increasing spatial resolution did not solve this problem, as models fitted at finer resolutions than 0.25 ha and subsequently aggregated performed worse than models fitted directly at coarse resolution. This study identified wood density variability as a major error source within the AGB-MCH model fitting framework at the site and found stem density detection as the fundamental challenge limiting further accuracy improvements. The forest structure framework demonstrates a systematic approach to decomposing confounded error sources in ALS-based biomass estimation, with applications for error budget analysis in forest monitoring systems.
Zitieren
APA
Yang, J., Fischer, F. J., Mirabel, A., Chave, J., & Vincent, G. (2026). A methodological framework for decomposing the error budget of ALS-based above ground biomass estimation using a 3D forest reconstruction model. Science of Remote Sensing, 14, 100540. https://doi.org/10.1016/j.srs.2026.100540
Vancouver
Yang J, Fischer FJ, Mirabel A, Chave J, Vincent G. A methodological framework for decomposing the error budget of ALS-based above ground biomass estimation using a 3D forest reconstruction model. Science of Remote Sensing. 2026;14:100540. doi:10.1016/j.srs.2026.100540
BibTeX
@article{Yang2026,
author = {Yang, Junliu and Fischer, Fabian Jörg and Mirabel, Ariane and Chave, Jérôme and Vincent, Grégoire},
title = {A methodological framework for decomposing the error budget of ALS-based above ground biomass estimation using a 3D forest reconstruction model},
journal = {Science of Remote Sensing},
year = {2026},
volume = {14},
pages = {100540},
doi = {10.1016/j.srs.2026.100540}
}
EUVIMED