Predicting soil hydraulic conductivity using random forest, SVM, and LSSVM models

Understanding the hydraulic properties of soil is essential to solve many management problems in agriculture and the environment. Water quality affects soil hydraulic conductivity. Soil hydraulic properties play an important role in nature's water cycle and are used as basic information in designing irrigation and drainage systems, hydrological issues, and soil quality assessment. In the current study, soil sampling is performed from different areas and its hydraulic conductivity was measured using the drop load method and then predicted using support vector machine (SVM) and least‐squares support vector machine (LSSVM) models. The model inputs were: soil texture (percentage of sand, silt, and clay particles), salinity (electrical conductivity), pH, sodium adsorption ratio, soil porosity, and bulk density and the output was soil hydraulic conductivity. Correlation coefficient, root mean square error (RMSE), mean bias error (MBE), and Nash–Sutcliffe efficiency (NSE) were used to evaluate the models and compare them. Based on evaluation criteria the best performance was obtained for random forest (RF) (R = 0.89, RMSE = 0.53, mean absolute error (MAE) = 0.54, and NSE = 0.72). Following RF, the SVM with (R = 0.69, RMSE = 1.32, MAE = 0.69, and NSE = 0.48) performed better than LSSVM model.