Estimation of Soil Organic Carbon Using Artificial Neural Networks and Regression Models

Aim of study: This study aimed to estimate the soil organic carbon (SOC) content in red pine (Pinus brutia Ten.) plantation areas under semiarid climatic conditions using multiple linear regression (MLR) and artificial neural network (ANN) models. Area of study: The research was carried out within the boundaries of the Tokat-Niksar Ayvaz Forest Management Directorate, Türkiye. Material and method: The study material consisted of 92 disturbed soil samples collected from the topsoil (0-10 cm and 10-30 cm) at 46 sampling plots, along with vegetation and topographic indices. Various physical and chemical analyses were performed on the soil samples. In addition to soil properties, topographic and remote sensing indices were used as predictor variables. SOC estimations were carried out using MLR and ANN models. Main results: According to the findings, SOC content increased with higher wilting point (WP), elevation, and normalized difference vegetation index (NDVI), while it decreased with increasing lime content (CaCO₃), pH, and bulk density (BD). In the MLR model, WP, BD, pH, CaCO₃, soil moisture index (MSI), easterly exposure (EExp.), and valley depth (VDep.) were identified as significant variables, resulting in an R² value of 0.822. The ANN model showed higher predictive performance with the same variables, yielding R² = 0.937 and a low relative prediction error (4.3%). Research highlights: Both models were found suitable for SOC estimation; however, the ANN model represented complex and nonlinear environmental interactions more effectively.