Gradient-Boosted Machine Learning Models for Tree Volume Estimation Using Forest Health Indicators

Abstract

Background and Objective: Accurate estimation of tree volume is essential for evaluating forest productivity, biomass accumulation, and carbon storage. This study aimed to develop a scalable and interpretable machine-learning framework for predicting tree volume using integrated forest health indicators.

Materials and Methods: A multi-index forest health dataset incorporating canopy, soil, and ecological variables was used to train and evaluate predictive models. Three machine-learning algorithms—Linear Regression, Random Forest, and Extreme Gradient Boosting (XGBoost)—were implemented and assessed using a 70/15/15 training, validation, and testing data split. Model interpretability was examined using SHapley Additive Explanations (SHAP) to identify the most influential predictors.

Results: Among the evaluated models, XGBoost demonstrated superior predictive performance on the independent test dataset, achieving a root mean square error (RMSE) of 2.143, a mean absolute error (MAE) of 1.602, and a coefficient of determination (R²) of 0.947. SHAP analysis indicated that canopy width, crown density, and soil fertility were the most significant contributors to tree volume estimation.

Conclusion: The findings highlight the effectiveness of gradient-boosted machine-learning models for accurate and interpretable tree volume prediction. The proposed approach provides a robust, data-driven framework with strong potential for large-scale forest monitoring, carbon accounting, and sustainable forest resource management.