Data from: Evaluation of parametric and nonparametric machine-learning techniques for prediction of saturated and near-saturated hydraulic conductivity

Parametric and nonparametric supervised machine learning techniques were used to estimate saturated and near saturated hydraulic conductivities (Ks, K10) from easily measurable soil properties including name of pedological horizon (HOR), soil texture (sand, silt & clay), organic matter (OM), bulk density (BD) and water contents (θpF1, θpF2, θpF3 and, θpF4.2) measured at four different matric heads (-10, -100, -1000, and -15848 cm). Using a stepwise linear model (SWLM) and the Lasso regression as parametric methods with 316 data in training and 135 data in testing phase, four pedotransfer functions (PTFs) were obtained in which water contents for both methods play an important role compared to other variables. SWLM showed better performance than Lasso in the testing phase for log(Ks) and log(K10) prediction with RMSE of 0.666 and 0.551 cm d-1 and R2 of 0.26 and 0.65. Nonparametric supervised machine learning methods trained and tested with similar data set significantly improved the accuracy of Ks prediction with R2 of 0.52, 0.36 and 0.53 for Gaussian regression process (GPR), support vector machine (SVM) and Ensemble (ENS) method in the testing stage. These methods also described 74.9, 66.7 and 72.5% of the variation of log(K10). Bootstrapping method validated the strong performance of nonparametric techniques. Feature selection capability of GPR determined that instead of using a model with all predictors, HOR, silt, θpF1 and θpF3 are sufficient for the prediction of log(Ks) or log(K10), HOR, silt, and OM can predict as accurate as the comprehensive model with all variables.