Abstract
The prediction of Field Capacity (FC) and Permanent Wilting Point (PWP) is essential for efficient soil management and agricultural sustainability, as in the Una River Watershed (URW), located in the state of São Paulo, Brazil, a region notable for its irrigated agriculture, which requires management tailored to its specific conditions. This study evaluates the efficiency of Pedotransfer Functions (PTFs) generated by Multiple Linear Regression (MLR) and Artificial Neural Networks (ANNs) in estimating FC and PWP from granulometric fractions and Organic Matter (OM) content. We analyzed 35 soil samples (0–20 cm), and the models were implemented in a Python environment. MLR stood out for its simplicity and efficiency, achieving R² values of up to 0.59 for FC and 0.87 for PWP, requiring fewer variables and lower computational costs. Meanwhile, ANNs reached R² values of up to 0.73 for FC but demanded greater computational complexity and more larger datasets. The results demonstrate that both methodologies provide adaptable solutions for estimating soil hydraulic parameters, and their integration into agricultural management systems is strategic for improving water-use efficiency and promoting sustainable practices.