Annual Rainfall Forecasting Using Hybrid Artificial Intelligence Model: Integration of Multilayer Perceptron with Whale Optimization Algorithm

Rainfall, as one of the key components of hydrological cycle, plays an undeniable role for accurate modelling of other hydrological components. Therefore, a precise forecasting of annual rainfall is of the high importance. In this regard, several studies have been tried to predict annual rainfall of different climate zones using machine learning and soft computing algorithms. This study investigates the application of an innovative hybrid method, namely Multilayer Perceptron-Whale Optimization Algorithm (MLP-WOA) to predict annual rainfall comparatively to the ordinary Multilayer Perceptron models (MLP). The models were developed by using 3-Input variables of annual rainfall at lag1, 2 and 3 corresponding to P_t-1, P_t-2 and P_t-3, respectively of two synoptic stations of Senegal (Fatick and Goudiry) in the time period of 1933–2013. 75% of the dataset were utilized for training and the other 25% for testing the studied models Accurateness of the mentioned models was examined using root mean squared error, correlation coefficient, and KlingGupta efficiency. Results showed that MLP-WOA3 and MLP3 using both P_t-1, P_t-2 and P_t-3 as inputs presented the most accurate forecasting in Fatick and Goudiry stations, respectively. In Fatick station, MLP-WOA3 decreased the RMSE value of MLP3 by 18.3% and increased the R and KGE values by 3.0% and 130%, respectively in testing period. But, in Goudiry station, MLP-WOA3 increased the RMSE value of MLP3 by 3.9% and increased the R and KGE values by 10.2% and 91% in testing period. Therefore, it can be realized that the MLP-WOA3 could not able to reduce the RMSE value of correspondent MLP model in Goudiry station. The conclusive results indicated that MLP-WOA slightly improved the accuracy of correspondent MLP models and may be recommended for annual rainfall forecasting.