Development of Wavelet-Kstar Algorithm Hybrid Model for the Monthly Precipitation Prediction (Case Study: Synoptic Station of Ahvaz)

Predicting hydrological parameters, especially rainfall, has played a very important role in water resources management and planning. Therefore, the development of methods giving accurate estimates has always been of interest to researchers. In this study, precipitation data from the Ahvaz synoptic station in the period of 2018-1961 were used to develop Kstar and Gene Expression Programming wavelet hybrid models (WKstar and WGEP). The performance of the applied models was evaluated using statistical indices, including the correlation coefficient (CC), Nash-Sutcliffe (NS), Kling–Gupta (KGE) and the Willmott Index (WI). Initially, the Kstar and GEP individual models were implemented, with a delay in precipitation input up to four months ago and month numbers. Results showed that both models achieved maximum accuracy with a time delay of one month (M1 senario), but their performance was very low and unacceptable. Regarding that both models with the M1 pattern have the best performance, so the M1 inputs decomposed by five different wavelet functions to detail and approximat subsets and were represented to the models. The results showed that the performance of wavelet hybrid models was significantly improved, so that the NS index increased from 0.139 to 0.607. In addition, the best performance of WKstar and WGEP hybrid models was obtained with the inputs of the Daubechies wavelet of order four and the decomposition level two. Also, there was no significant difference between the two developed hybrid models statistically, but using the violin plot it was found that the WKstar model is more suitable for predicting precipitation at the Ahvaz synoptic station.