Testing artificial neural networks for estimating relationships between diameter at breast height and stump diameter in Quercus cerris

OpenAlex İngilizce

ABSTRACT Background: Accurate prediction of diameter at breast height (DBH) from stump diameter measurements is essential for sustainable forest management, particularly in post-harvest assessments, illegal logging investigations, and carbon accounting protocols in Quercus cerris stands. Traditional allometric models often fail to capture the complex nonlinear relationships inherent in tree growth patterns, necessitating advanced computational approaches. This study aims to develop and validate artificial neural network (ANN) models for predicting DBH from stump diameter in Q. cerris stands across diverse ecological conditions in Turkish forests. Results: We conducted an extensive field survey across 103 systematically distributed sample plots encompassing 1,077 individual Q. cerris trees within the Bursa Regional Directorate of Forestry, Türkiye. The study area represented diverse site conditions including five site quality classes, stand age, and varying stand densities. Fourteen traditional regression models were compared against five distinct ANN architectures with varying activation functions. The ANN model employing hyperbolic tangent sigmoid activation function between input and hidden layers coupled with pure-linear function between hidden and output layers (ANN-2) with 8 neurons demonstrated superior predictive performance (SSE = 567.07, RMSE = 2.290, MSE = 4.797, R²adj = 0.9484). Among conventional models, quadratic regression (Model 4) exhibited competitive performance but showed evidence of heteroscedasticity and model assumption violations. Conclusion: This research demonstrates the superior capability of artificial neural networks in modeling complex DBH-stump diameter relationships in Q. cerris compared to traditional statistical approaches. The developed ANN models provide forest managers with robust, accurate tools for retrospective forest assessments, illegal logging quantification, and sustainable management planning.