Geo-hydroclimatological-based estimation of sediment yield by the artificial neural network

Abstract

An artificial neural network (ANN) model is proposed for the estimation of sediment yield in Lake Urmia sub-basins. The number of model parameters were extended as far as possible to all geometric, geological and hydroclimatological parameters of the sub-basin. Also, various ANN structures, learning rules, and transfer functions were examined. The examinations show that extended delta and hyperbolic tangent were the best functions for the proposed ANN model. The best structure for the ANN model is a triangle with two hidden layers, containing five neurons in its first and three neurons in its second hidden layer. The comparison between the proposed and regional analysis models showed a notable increase in the accuracy by using the proposed model. Mean absolute error and the maximum absolute error of the estimation reduced to 2.5% and 3% of those regional analysis models, respectively, and therefore ANN model is recommended for sediment yield estimation.