Prediction capability of different soil water retention curve models using artificial neural networks

Direct measurements of soil water retention curve (SWRC) are costly and time consuming. So far, less investigation has been carried out on the prediction capability of different models using artificial neural networks (ANNs). In this study in total 75 soil samples were collected from Guilan province, north of Iran. The basic soil properties namely sand, clay and bulk density were used as predictors and the parameters of ten SWRC models were forecasted by ANNs. The prediction capability of each model was examined based on three criteria in nine groups of samples: total, fine (clay and silty clay) and medium (clay loam, silt loam, silty clay loam and loam) textural groups and six soil texture classes. Overall, the Boltzman, Tani, Gardner, Campbell and van Genuchten models produced the best results. However, bimodal models (Durner, Seki and Dexter) established on non-uniform pore size distribution with two modes (peaks) in soils showed low prediction capability in this study. Therefore, further research is needed. Sensitivity analysis indicated that the residual and saturated water contents were largely dependent on clay content.