Tree stem mean diameter reduction factor prediction through advanced modeling approaches

Sustainable management of natural resources relies on accurate modelling of forest attributes to prevent degradation. This study explores advanced modelling techniques, including Artificial Neural Networks (ANN) and Support Vector Regression (SVR), for estimating the mean stem diameter reduction factor (taper) of standing fir trees (Abies x borisii-regis Matff.). These methods are compared against traditional non-linear regression model (NLR), developed using the Levenberg-Marquardt optimization algorithm. The ANN models employ cascade correlation, generalized regression, and Bayesian regularization back-propagation architectures, while the ε-SVR approach is assessed for its robustness. The results show that support vector regression (ε-SVR) achieved the lowest relative errors in model fitting, improving by 0.60% over cascade correlation and generalized regression and by 0.67% over Bayesian regularization. Regarding generalization ability, the ε-SVR model performed best, with a relative error of 4.90%, which was slightly lower than cascade correlation (by 0.1%), generalized regression (by 0.01%), and Bayesian regularization (by 0.04%). A comparative analysis between machine learning approaches and standard regression revealed that the ε-SVR model had the lowest mean error (0.0715), while the non-linear regression (NLR) model showed a higher mean error of 0.0955, which means 1.35 times greater. These findings highlight the strong capability of machine learning methods in accurately estimating and predicting the diameter reduction factor of trees, effectively capturing its non-linear behaviour compared to traditional regression models. Overall, this study underscores the potential of advanced machine learning techniques to enhance accuracy and adaptability in sustainable forest management.