Performance Evaluation of Artificial Neural Networks conjunct with Genetic Algorithm for Estimation of Soil Infiltration Rate (Case Study: Khoda afarin Region of East Azerbaijan Province)

Infiltration plays a pivotal role in the hydrologic cycle by effectively acting to partition water into surface and subsurface components. Direct measurement of infiltration rate is expensive and work and time consuming. Artificial Neural Networks (ANNs), Gene Expression Programing (GEP) and hybrid of ANN and Genetic Algorithm (ANN-GA) can be used for estimation of soil infiltration rate as an indirect methods. The main objective of this research was to develope an infiltration rate model in Khoda afarin region based on the collected data (88 double ring infiltration) and some soil properties. The Pierson correlation revealed among the soil properties, sand and silt contents, porosity and organic matter have the most correlation with the infiltration rate. Determination Coefficient (R2) and Normalized Root Mean Square Error (NRMSE) were calculated to be 0.88 and 7.9%, respectively for the ANN method and 0.75 and 11.3% for the GEP method. Both ANN and GEP methods perform poorly, in extrapolating the minimum and maximum amount of infiltration rate. The hybrid model of ANN-GA was the best model in terms of statistical indices including R2 (0.93) and RMSE (6.1%). This model comprised of 4 neurons (sand, silt, porosity percentage and OM) in input layer and 5 neurons using sigmoidal tangent functions in the hidden layer and linear activation functions in the output layer. The results indicated that the neural-genetics algorithm can be used to optimize weight parameter of artificial neural network. Overall the hybrid ANN-GA model showed better performance than the other models, so that the R2 and NRMSE for the hybrid model were 0.93 and 6.1% respectively. Therefore it is suggested as a powerful tool for estimating infiltration rate.