Soil Water Infiltration Rate and Soil Infiltration Model Parameters Prediction Using Artificial Neural Network and Support Vector Machine

Water infiltration into the soil is one of important hydrological parameters. This study was conducted for Phillip and Horton parameters and final water infiltration rate prediction using artificial neural network and support vector machine (SVM). The soil water infiltration was measured in 100 points of Abarkouh city landscape (Yazd province) with double ring method. The samples from 0-30 cm of soil surface were analyzed for bulk density, texture, organic matter, sodium adsorption ratio, porosity, geometric mean particle diameter and geometric standard deviation of soil particle. The multilayer perceptron neural network (MLP) with 4 different scenarios with 3, 5, 7 and 9 inputs and SVM with 9 inputs were analyzed for infiltration parameters and final water infiltration rate prediction. The results showed that the network with 9 inputs had the greatest R2 and the lowest error in Phillip and Horton parameters prediction. The study of prediction ability of ANN for Horton and Phillip parameters showed that the greatest capability was related to final infiltration rate to net design as 9-5-1 with R2 equal to 0.84. The sensitivity analysis showed that the designed nets had greater sensitivity to soil sodium adsorption ratio and the organic matter than other 7 parameters. The SVM model had good ability to water infiltration rate prediction based on basic soil properties. SVM model had better ability in soil water infiltration rate prediction compared to ANN model.