Hybrid emotional neural networks and novel multi-model stacking algorithms for multi-lake water level fluctuation modeling

Abstract Lakes play a crucial role in the water cycle, and accurately modeling their water level fluctuation is vital for managing water resources, ecosystems, and flood control. The current study aims to develop a novel emotional artificial neural network optimized by genetic algorithm (GA-EANN), weighted average model stacking (WAE) and neural network-based stacking (NNE) that combines the outputs of emotional artificial neural network (EANN) and artificial neural network (ANN) for modeling the water level of four lakes of Rift Valley Lakes Basin (RVLB) in Ethiopia. Ten years of daily hydrological and weather data were used for model training and testing. The efficiencies of the employed algorithms were compared using graphical and statistical metrics such as root mean square error (RMSE), mean absolute error (MAE), Nash Sutcliffe Efficiency (NSE) and coefficient of determination (R 2 ). The modeling result demonstrated that the NNE method provided the best prediction accuracy at Hawassa Lake (RMSE = 0.056 m, NSE = 0.972 and R 2 = 0.972), Langano Lake (RMSE = 0.059 m, NSE = 0.978 and R 2 = 0.978) and Abiyata lake (RMSE = 0.052 m, NSE = 0.987and R 2 = 0.987) in the testing set. At Ziway Lake, the hybrid GA-EANN model surpassed other models with R 2 value of 0.976 in testing set. Generally, the NNE and optimized hybrid modeling (GA-EANN) significantly improved the performance of water level prediction.