Identifying Key Controlling Factors and Improving Pedotransfer Functions for Saturated Soil Hydraulic Conductivity Using Principal Component Analysis and Artificial Neural Networks

Saturated hydraulic conductivity ( K Sat ) is a critical soil property governing hydrological processes across scales, particularly at the soil surface, where it controls infiltration and water movement into deeper layers. Measuring K Sat in the field is challenging due to high variability, time‐intensive methods, costly equipment, and the need for skilled labor and large samples. Soil heterogeneity and land‐use effects further complicate extrapolation of K Sat values. This study aimed to (i) identify key‐driving factors that relate K Sat with hydro‐physical soil attributes and land use systems at a watershed scale, (ii) evaluate the performance of artificial neural networks (ANNs) in predicting K Sat by comparing them with multiple linear regression (MLR) models, and (iii) assess whether variables derived from principal component analysis (PCA) could improve model K Sat predictions. ANN and MLR models were developed for a watershed in southern Brazil, using an experimental grid of 178 points. Soil properties (bulk density, total porosity, micro‐ and macroporosity, organic carbon, particle‐size distribution, and K Sat ) were measured in the 0–20 cm layer, and land use was mapped from satellite imagery. PCA was applied to derive principal components (PCs), which were used as predictors in modeling. Results showed that incorporating PCs significantly improved ANN performance, with the best model (combining total porosity, soil texture, three PCs, and land use) achieving an R 2 of 0.70 and RMSE of 0.41 m h −1 , outperforming the corresponding MLR model. This indicates that K Sat variability is driven by both soil and land‐use factors and their orthogonal recombinations capturing the dataset's variance structure.