Predictive Modeling of Volume and Biomass in Pinus pseudostrobus Using Machine Learning and Allometric Approaches
This study aims to evaluate the effectiveness of machine learning algorithms in predicting key forest metrics—stem volume, root system volume, and organ biomass (including leaves, branches, stem, and root)—for Pinus pseudostrobus var. Lindley, based on morphological measurements from the same trees....
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-01-01
|
| Series: | Forest Science and Technology |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/21580103.2025.2456295 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study aims to evaluate the effectiveness of machine learning algorithms in predicting key forest metrics—stem volume, root system volume, and organ biomass (including leaves, branches, stem, and root)—for Pinus pseudostrobus var. Lindley, based on morphological measurements from the same trees. The novelty of this study lies in applying five machine learning algorithms—Random Forest, Neural Networks, Gradient Boosting Machines, Support Vector Machines (SVM), and k-Nearest Neighbors (k-NN)—to predict these metrics, using data from the destructive analysis of 98 individual trees aged from eight months to five years. For comparison, we also applied univariate allometric models, adjusted with nonlinear least squares and quantile regression. The results indicate that Random Forest, k-NN, and SVM outperformed the other algorithms, demonstrating superior predictive accuracy for both biomass and volume. A key innovation of this study is its demonstration of how machine learning, with its ability to model complex, nonlinear relationships, can serve as a powerful tool for forest management. Quantile regression, combined with nonlinear least squares, proves most effective when the relationships are well-defined, allowing for tailored parameter adjustments that enhance predictions, particularly in the presence of heteroscedasticity. |
|---|---|
| ISSN: | 2158-0103 2158-0715 |