Enhancement groundwater level prediction using hybrid ANN-HHO model: case study (Shabestar Plain in Iran)

Nowadays, in order to sustainably manage surface and groundwater resources, it is crucial to have knowledge of groundwater levels. The Shabestar Plain is one of the plains with water stress that was selected as the study area in this study. A question arises regarding how models with hybrid properties can boost the capabilities of metamodels in light of the increasing development of models and their combination with optimization algorithms. In order to try to find the answer, in this study, the Harris hawk optimization (HHO) algorithm was combined with an artificial neural network (ANN) model and was evaluated with the aim of simulating the monthly GWL of Shabestar Plain during the period 2001–2019. Training and testing of the developed models are accomplish using 80% and 20% of the monthly data set, respectively. In addition, a variety of statistical indicators were used to evaluate the developed models, including coefficient of determination (R2), Nash–Sutcliffe efficiency coefficient (NSE), root mean square error (RMSE), standard deviation ratio RMSE-observation (RSR), mean absolute error (MAE), uncertainty level (U95), and bias percentage (PBIAS). Analysis of the results shows that the hybrid model (ANN-HHO) produces accurate results. Based on testing phase, the ANN-HHO model shows a very good agreement with the measured data in scenario7 (R2 = 0.940, RMSE = 0.108(m), MAE = 0.081, NSE = 0.933, RSR = 0.256, U95 = 0.300, and PBIAS = 0.0009) and ANN (R2 = 0.798, RMSE = 0.232(m), MAE = 0.187, NSE = 0.697, RSR = 0.550, U95 = 0.602, and PBIAS = 0.004). Thus, ANN-HHO hybrid model is reasonable for predicting monthly groundwater levels. A hybrid model can provide water resources managers with a useful tool for future groundwater research and provide a reliable approach.