Evaluation of artificial neural networks approach for simulation of advanced wetting front of loam soils from a point source

Shape (diameter and depth of wetted soil) and trend of advanced wetting front under a trickle is a function of soil properties, discharge and duration of applied water. With respect to numerous effective factors on shape and trend of advanced wetting front under drip irrigation, and capability of artificial neural networks, it seems that with collection of information for a relative extensive range of effective parameters, shape and trend of advanced wetting front in soil could be predicted. In this research, in 1383 from cropping soil of zayanderod river banks of isfahan for combining various variables (such as loam textured soil and applied discharges of 2, 4, 8 and 12 lit/hr with 48 lit of irrigation water), first shape and trend of advanced wetting front was measured with a physical model. Then, using Matlab ver 7 software, an artificial neural network named ANN-L was designed to predict shape and trend of advanced wetting front under a trickle.The results showed that for this loam soil and each of the four applied discharges, the ANN-L has the ability to predict trend of wetting front. In the designed network, RMSE was estimated as 0.2878 and the coefficient of determination was R2=0.995. Small RMSE and large R2 values of network shows proper match between trends of observed and predicted wetting fronts. The sensitivity analysis on parameters entered in ANN-L showed that with omission of irrigation water amount and time of irrigation, the performance of this neural network is weakened. The omission of. Physical properties of soil have lesser effect on neural network’s performance (p<0.05). With regard to the results, the error of ANN-L network is equal to 1% which is not important in applied cases. Thus, using ANN-L is recommended for prediction of trend of advanced wetting front under trickle irrigation in similar conditions.