Predicting soil infiltration rate using Artificial Neural Network

The infiltration rate is an important parameter in soil, hydrological, ecological and agricultural studies. It plays the main role as the input parameter in models for water flow and solute transport in the vadose zone. In this study, Multilayer Artificial Neural Network "ANN" using the backpropagation algorithm was selected to estimate the steady infiltration rate covering different types of Libyan soils. The activation function was selected LOGSIG in the middle and exit layers. The input data were the percentage of sand, silt and clay, bulk density, saturated hydraulic conductivity and the volumetric water content in soil at -10 kPa. The performance of the ANN models was evaluated against a set of data that never seen by the model during the training phase. Multivariate linear regression model (MLR) based on the percentage of silt, saturated hydraulic conductivity and volumetric water content in soil at -10 kPa was also developed to determine infiltration rate for evaluation purpose. The results obtained in this study showed a good agreement between the measured data and the ANN simulated. The values of mean absolute error and root mean square error were slightly smaller in ANN steady infiltration rate model compared to the developed Multivariable linear regression model to estimate the infiltration rate. Although the results of these comparisons encourage the using ANN in practice, it would be valuable to have large local soil database from many different sites, in order to make a stronger assessment of the ANN models.