Optimizing soil water retention predictions using a subtractive clustering-based ANFIS model

Soil water content indicators, particularly field capacity (FC) and permanent wilting point (PWP), are fundamental parameters influencing irrigation management, water retention capacity, infiltration, drainage, hydraulic conductivity, plant water stress, and solute dynamics. Their direct measurement is often costly. To address this, pedotransfer functions (PTFs) have emerged as efficient, rapid, and economical alternatives for estimating these characteristics. This study evaluates the capability of multiple linear regression, classification and regression trees (C&RT), artificial neural networks (ANN), k-nearest neighbor (k-NN), and adaptive neuro-fuzzy inference system (ANFIS), for estimating soil FC and PWP. A total of 240 soil samples were collected and split into training and testing datasets. Key input variables such as clay, silt, sand, and bulk density were considered, while FC and PWP served as the outputs. Sensitivity analysis and statistical validation were used to assess model performance, primarily based on the determination coefficient (R2) and normalized root mean square error (NRMSE). Among the models tested, ANFIS (for FC; R2 = 0.77, NRMSE = 0.138 and PWP; R2 = 0.79, NRMSE = 0.149) demonstrated superior predictive accuracy, making it the most suitable approach for estimating soil water content in diverse soil textures. Consequently, the integration of ANFIS into the development of soil PTFs is recommended for soils in northwestern Iran.