Forecasting furrow irrigation infiltration using artificial neural networks

An artificial neural network (ANN) was developed for estimating the infiltrated water volume (Z) under furrow irrigation. A feed-forward neural network using back-propagation training algorithm was developed for the prediction. Four variables were used as input parameters; inflow rate (Qo), furrow length (L), waterfront advance time at the end of the furrow (TL) and infiltration opportunity time (To). The Z was the one node in the output layer. The data used to develop the ANN model were taken from published experiments. The ANN model predicted Z over a wide range of the input variables with statistical analysis indicating that it can successfully predict Z with a high degree of accuracy. Performance evaluation criteria indicated that the ANN model was better than the two-point method using a volume balance model. Using testing and validation data sets to compare the ANN model with the two-point method shows that the two-point method had a mean coefficient of determination (R2) value that was about 3.6% less accurate than that from the ANN model. Also, the mean root mean square error (RMSE) value of 0.0135 m3 m−1 for the two-point method was almost double that of mean values for the ANN model. The relative errors of computed Z values for the ANN model were mostly around ±10%. Therefore, the ANN model is applicable to other soils and to different furrow irrigation hydraulics.