Optimization of neural networks for precipitation analysis in a humid region to detect drought and wet year alarms

Abstract

In this study, annual precipitation was forecast by coding in MATLAB software environment based on a non-linear autoregressive neural network (NARNN), non-linear input-output (NIO) and NARNN with exogenous input (NARNNX). Historical precipitation data (27 precipitation gauge stations located in Gilan, Iran) were used as two 21 year sets from 1968 to 1988 and from 1989 to 2009 for calibration and testing of the networks, respectively. Results showed that the accuracy of the NARNNX was better than that of the NARNN and NIO, based on r values. However, performance of the networks was not satisfactory because the number of neurons in the hidden layer and the roles of training, validation and testing phases were lacking flexibility and change. Optimization of the number of neurons in the hidden layer and the determination of the best role among the different phases led to improvement of network accuracy. The r values were a0.73 only for five stations in the optimized NARNN and a0.74 only for those stations with optimized NIO.