An alternative approach to allometric above‐ground woody biomass models using symbolic regression and terrestrial laser scanning
- Author
- Tom Verhelst (UGent) , Louise Terryn (UGent) , Mathias Disney, Joanne Nightingale, Niall Origo (UGent) , Maxime Van Haeverbeke (UGent) , Hans Verbeeck (UGent) and Kim Calders (UGent)
- Organization
- Project
- Abstract
- The accurate quantification of forest above-ground biomass (AGB) is crucial for monitoring forest functioning and carbon stocks. Forest AGB maps are commonly derived from earth observation products and used in many downstream applications, such as global vegetation and carbon cycle models or the monitoring of forest primary production. Forest AGB maps rely on the accuracy of AGB field estimates as calibration, which are typically generated through allometric AGB models. Several key issues exist in many current allometric models which can cause strong prediction bias and unreliable estimates of prediction uncertainty intervals. Reference datasets are small, sparse and lack large trees. Furthermore, common modelling approaches can fail to reflect the observed size-to-mass scaling across larger tree size ranges, and underlying modelling assumptions are often violated. We used a terrestrial laser scanning (TLS) derived plot-level, temperate tree inventory dataset to construct size-to-volume allometric models in a Bayesian framework. We developed the models over three stages, starting with a conventional linear functional form after a log–log transformation of the data. Model performance was then improved over the next two stages using symbolic regression, an algorithm that selects suitable mathematical equations to describe relations in a dataset. In stage two, this was done by adjusting the model functional form. In stage three, we incorporated an expression for the observed heteroskedasticity into the allometric model. Overall model accuracy consistently improved around 5% after the model adjustments and posterior distributions better reflected the data. The introduction of the adjusted functional forms reduced prediction bias across the entire predictor range. By modelling heteroskedasticity, the volume prediction uncertainty for large trees specifically was reduced by up to 10%–15%. Practical implication: We provide a new workflow to construct tree allometric AGB models, which serves as a template to improve the estimation of tree volume, AGB or carbon stocks. Following the workflow, we suggest pathways to improve model performance, present three ready-to-use allometric models for Acer pseudoplatanus in southern Great Britain and discuss their use cases based on different tree structural metrics. Our results also demonstrate the advantages of Bayesian models, which naturally resolve several issues in traditional allometric models.
- Keywords
- Q-ForestLab, allometric models, Bayesian modelling, heteroskedasticity, symbolic regression, TLS, TREE, TRANSFORMATION, EQUATIONS, WOODLAND, HEIGHT, BIAS
Downloads
-
Verhelst et al. - 2026 - An alternative approach to allometric above-ground woody biomass models using symbolic regression an.pdf
- full text (Published version)
- |
- open access
- |
- |
- 3.50 MB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01M0S8Z2DGWW98F88C4GYFKQXK
- MLA
- Verhelst, Tom, et al. “An Alternative Approach to Allometric Above‐ground Woody Biomass Models Using Symbolic Regression and Terrestrial Laser Scanning.” ECOLOGICAL SOLUTIONS AND EVIDENCE, vol. 7, no. 3, 2026, doi:10.1002/2688-8319.70310.
- APA
- Verhelst, T., Terryn, L., Disney, M., Nightingale, J., Origo, N., Van Haeverbeke, M., … Calders, K. (2026). An alternative approach to allometric above‐ground woody biomass models using symbolic regression and terrestrial laser scanning. ECOLOGICAL SOLUTIONS AND EVIDENCE, 7(3). https://doi.org/10.1002/2688-8319.70310
- Chicago author-date
- Verhelst, Tom, Louise Terryn, Mathias Disney, Joanne Nightingale, Niall Origo, Maxime Van Haeverbeke, Hans Verbeeck, and Kim Calders. 2026. “An Alternative Approach to Allometric Above‐ground Woody Biomass Models Using Symbolic Regression and Terrestrial Laser Scanning.” ECOLOGICAL SOLUTIONS AND EVIDENCE 7 (3). https://doi.org/10.1002/2688-8319.70310.
- Chicago author-date (all authors)
- Verhelst, Tom, Louise Terryn, Mathias Disney, Joanne Nightingale, Niall Origo, Maxime Van Haeverbeke, Hans Verbeeck, and Kim Calders. 2026. “An Alternative Approach to Allometric Above‐ground Woody Biomass Models Using Symbolic Regression and Terrestrial Laser Scanning.” ECOLOGICAL SOLUTIONS AND EVIDENCE 7 (3). doi:10.1002/2688-8319.70310.
- Vancouver
- 1.Verhelst T, Terryn L, Disney M, Nightingale J, Origo N, Van Haeverbeke M, et al. An alternative approach to allometric above‐ground woody biomass models using symbolic regression and terrestrial laser scanning. ECOLOGICAL SOLUTIONS AND EVIDENCE. 2026;7(3).
- IEEE
- [1]T. Verhelst et al., “An alternative approach to allometric above‐ground woody biomass models using symbolic regression and terrestrial laser scanning,” ECOLOGICAL SOLUTIONS AND EVIDENCE, vol. 7, no. 3, 2026.
@article{01M0S8Z2DGWW98F88C4GYFKQXK,
abstract = {{The accurate quantification of forest above-ground biomass (AGB) is crucial for monitoring forest functioning and carbon stocks. Forest AGB maps are commonly derived from earth observation products and used in many downstream applications, such as global vegetation and carbon cycle models or the monitoring of forest primary production. Forest AGB maps rely on the accuracy of AGB field estimates as calibration, which are typically generated through allometric AGB models.
Several key issues exist in many current allometric models which can cause strong prediction bias and unreliable estimates of prediction uncertainty intervals. Reference datasets are small, sparse and lack large trees. Furthermore, common modelling approaches can fail to reflect the observed size-to-mass scaling across larger tree size ranges, and underlying modelling assumptions are often violated.
We used a terrestrial laser scanning (TLS) derived plot-level, temperate tree inventory dataset to construct size-to-volume allometric models in a Bayesian framework. We developed the models over three stages, starting with a conventional linear functional form after a log–log transformation of the data. Model performance was then improved over the next two stages using symbolic regression, an algorithm that selects suitable mathematical equations to describe relations in a dataset. In stage two, this was done by adjusting the model functional form. In stage three, we incorporated an expression for the observed heteroskedasticity into the allometric model.
Overall model accuracy consistently improved around 5% after the model adjustments and posterior distributions better reflected the data. The introduction of the adjusted functional forms reduced prediction bias across the entire predictor range. By modelling heteroskedasticity, the volume prediction uncertainty for large trees specifically was reduced by up to 10%–15%.
Practical implication: We provide a new workflow to construct tree allometric AGB models, which serves as a template to improve the estimation of tree volume, AGB or carbon stocks. Following the workflow, we suggest pathways to improve model performance, present three ready-to-use allometric models for Acer pseudoplatanus in southern Great Britain and discuss their use cases based on different tree structural metrics. Our results also demonstrate the advantages of Bayesian models, which naturally resolve several issues in traditional allometric models.}},
articleno = {{e70310}},
author = {{Verhelst, Tom and Terryn, Louise and Disney, Mathias and Nightingale, Joanne and Origo, Niall and Van Haeverbeke, Maxime and Verbeeck, Hans and Calders, Kim}},
issn = {{2688-8319}},
journal = {{ECOLOGICAL SOLUTIONS AND EVIDENCE}},
keywords = {{Q-ForestLab,allometric models,Bayesian modelling,heteroskedasticity,symbolic regression,TLS,TREE,TRANSFORMATION,EQUATIONS,WOODLAND,HEIGHT,BIAS}},
language = {{eng}},
number = {{3}},
pages = {{18}},
title = {{An alternative approach to allometric above‐ground woody biomass models using symbolic regression and terrestrial laser scanning}},
url = {{http://doi.org/10.1002/2688-8319.70310}},
volume = {{7}},
year = {{2026}},
}
- Altmetric
- View in Altmetric