Improving European Black Pine Stem Volume Prediction Using Machine Learning Models with Easily Accessible Field Measurements

Abstract

Reliable prediction of tree stem volume is crucial for effective forest management and ecological assessment. Traditionally, regression models have been applied to estimate forest biometric variables, yet they often fall short when handling the complex, non-linear patterns typical of biological data, potentially introducing biases and errors. Tree stem volume, a critical metric in forest biometrics, is generally estimated through easily measured parameters such as diameter at breast height (d) and total tree height (h). This study investigates advanced machine learning (ML) techniques—Extreme Gradient Boosting (XGBoost), epsilon-Support Vector Regression (ε-SVR), and Random Forest regression (RFr)—to predict the stem volume of European black pine (Pinus nigra) on Mount Olympus, Greece, using basic field measurements. Machine learning (ML) approaches demonstrated substantial improvements in prediction accuracy compared to traditional non-linear regression-based models (RMs). Notably, XGBoost significantly enhanced predictive performance by reducing the Furnival index (FI) by as much as 42.3% (from 1.1859 to 0.1056) and 21.3% (from 0.1475 to 0.1161) in the test and fitting datasets, respectively, for the single-entry model. For the double-entry model, XGBoost achieved FI reductions of 40.5% (from 0.1136 to 0.0676) and 41.3% (from 0.1219 to 0.0715) in the test and fitting datasets, respectively. These findings highlight the potential of ML models to improve the accuracy of forest inventory predictions, thereby supporting more effective and data-driven forest management strategies.