Continuous rainfall-runoff simulation by artificial neural networks based on efficient input variables selection using partial mutual information (PMI) algorithm

Knowledge on the natural ability of basins is one of fundamental needs to optimal utilization of runoff. Thus, rainfall-runoff simulation in basins is of utmost importance. Continuous simulation of rainfall-runoff in Maroun basin performed using Artificial Neural Networks (ANNs) in order to evaluate the ability and accuracy of ANN for runoff estimation. Considering the fact that the number of rainy days per year less than sunny days, so runoff is caused by two different mechanisms. In continuous rainfall time and a few days later, runoff mainly is from high discharge and low base time. But on most days when there is no rainfall, baseflow has low discharge and long base time .Thus, in this research a double criterion model of rainfall-runoff includes model on rainy days and non rainy days were examined. Also efficient input variables on runoff in the Maroun basin are determined using the partial mutual information (PMI) algorithm. Comparison of statistical criteria between the single criterion model and double criterion model indicated that the double criterion model were more accurate. Therefore, the Nash-Sutcliff coefficient of single criterion model and double criterion model for test stage of network were 0.86 and 0.94 respectively.