The Temperature Effect on the Estimation of Basin Outflow by Perceptron and Convolutional Neural Networks with Wavelet Analysis

Estimation of basin outflow under the influence of various parameters is a complex process and in case of basin information lack, the analytical models are not applicable. On the other hand, Artificial Intelligence models do not require basin information, and some, such as the convolutional neural network (CNN), have recently been widely used in hydrology. The purpose of this study was to investigate the performance of CNN in estimating the outflow in terms of temperature, precipitation and inflow to the basin. In this study, CNN, combination of CNN with wavelet analysis (WCNN), and perceptron neural network (MLP) were used to evaluate the effect of temperature on the outflow within the Ghaleh Shahrokh Chelgard basin from 1992 to 2015. Each model was run 20 times and the mean values ​​of correlation coefficient(R), root mean square error (RMSE) and Nash Sutcliffe coefficient (NS) were calculated. Lags of one, two and three-month of temperature and rainfall data are also included as input data. Wavelet analysis was used for noise reduction and the results showed that CNN with a lag of three months had ,  and  equal to 14.20 (m3/s), 0.922 and 0.772, respectively. In contrast, the WCNNT3 method with Daubechies wavelet, which is a combination of CNN and wavelet analysis, ​​with level four performance and resolution two (WCNNT3-db42) had indices respectively equal to 9.45 (m3/s), 0.945 and 0.863. Accordingly, CNN performed better than MLP, and the WCNN method with wavelet analysis showed better performance than CNN.