Modelling Vertical Infiltration in an Unsaturated Porous Media Using Neural Network Architecture

Abstract

The vadose zone is the dynamic area that controls the flux of water between the surface and groundwater and partitions rainfall into infiltration, runoff, groundwater recharge and Evapotranspiration. Richards’ equation is the key relationship for addressing basic soil hydrological processes within the vadose zone. In this study, black box neural network models and white numerical models which are derived from system knowledge to understand and model the unsaturated flow process are integrated using feed forward neural network architecture. The method uses the capabilities of neural network to learn differential data having physical values. The vertical infiltration through the soil represented by Richards's equation, the moisture content and pressure head in the soil profile are formulated in this study. The flow equation is subjected to initial and boundary conditions that signify the applied water (head) during rainfall and irrigation application time at the top boundary of the soil profile. The simulated neural network model is trained and validated using the analytical solution of Richards’ equation. The neural network algorithm allows us to obtain fast solutions of Richards’ equation starting from randomly sampled data sets predicting the soil moisture content and pressure head in the wetting profile of the vadose zone. The algorithm allows us to achieve good approximation results without wasting memory space and computational time and therefore reducing the complexity of the problem due to the parallel structure of the network.