Improving infiltration prediction by point-based PTFs for semi-arid soils in southern of Iran

Soil water infiltration plays an important role in the hydrological cycle. Accurate prediction of infiltration requires for evaluating irrigation, drainage, run-off and watershed management. The direct infiltration measurement at the field large scale is time-consuming, costly and difficult. Therefore, the aim of this study was to derive and evaluate soil water infiltration models for a semi-arid region in Marvdasht plain, southern of Iran. The infiltration data were measured in 72 points with 3 replications. In each point, the basic soil properties were measured and used as an input data. The multiple linear regression (MLR), support vector machines (SVMs) and feed-forward multilayer perceptron artificial neural networks (ANNs) model were developed to estimate cumulative infiltration data at time intervals including 5, 10, 20, 45, 150, 210 and 270 min. The results of this study indicated that the Kostiakov–Lewis model provided the most accurate predictions in comparison with Kostiakov, USDA-NRCS, Philip, Horton and Green-Ampt water infiltration models. Moreover, the results of the derived MLR-, SVM- and ANN-based PTFs models at different interval times showed that the SVMs-based PTFs model yielded in the accurate prediction of water infiltration ( R 2 of 0.496–0.720 and RMSE of 0.800–15.108). It was concluded that the SVMs model could be an applicable and reliable method for predicting cumulative soil water infiltration. The accurate prediction of water infiltration will be helpful for designing an efficient irrigation system, crop water requirement and solute transport in semi-arid regions.