Analysis of Soil Water Retention Data Using Artificial Neural Networks

Abstract

Many studies of water flow and solute transport in the vadose zone require estimates of the unsaturated soil hydraulic properties, including the soil water retention curve (WRC) describing the relationship between soil suction and water content. An artificial neural network (ANN) approach was developed to describe the WRC using observed data from several soils. The ANN approach was found to produce equally or more accurate descriptions of the retention data as compared to several analytical retention functions popularly used in the vadose zone hydrology literature. Given sufficient input data, the ANN approach was also found to closely describe the hysteretic behavior of a soil, including observed scanning wetting and drying curves.