Prediction of monthly discharge volume by different artificial neural network algorithms in semi-arid regions

Prediction of monthly discharge volume is important for reservoir management and evaluation of drinking-water supplies. Also, it is very essential in arid and semi-arid regions due to the lack of observed data. This study compared four artificial neural network (ANN) algorithms to predict the monthly discharge volume from Idenak Watershed in Kohkiloye Boier Ahmad Province in southwestern Iran. These algorithms, including resilient backpropagation (ANN_RP), scaled conjugate gradient (ANN_SCG), variable learning rate (ANN_GDX), and Levenberg–Marquardt (ANN_LM), were applied to monthly discharge volume data. The transfer function employed was the tangent sigmoid, and input vectors were constructed in different ways during the algorithm development. The algorithms were trained and tested using a 36-year data record (432 monthly values) selected randomly. Comparison of the algorithms showed that ANN_SCG performed better than the other algorithms, where the values of R2 and root mean square errors during validation were 0.78 and 63 million cubic meters. Furthermore, the input vector consisting of precipitation [P(t)], antecedent precipitation [P(t − 1)], and antecedent monthly discharge volume with one time lag [V(t− 1)] was superior to the other input vectors for monthly discharge volume prediction. Generally, the proposed models are capable for prediction of monthly discharge volume in arid and semi-arid regions.