Enhancement of Groundwater-Level Prediction Using an Integrated Machine Learning Model Optimized by Whale Algorithm

The present study attempted to predict groundwater levels (GWL) obtained from precipitation and temperature data based on various temporal delays. The radial basis function (RBF) neural network–whale algorithm (WA) model, the multilayer perception (MLP–WA) model, and genetic programming (GP) were used to predict GWL. The objectives were: (1) to prepare robust hybrid ANN models; (2) to study the combination of ANN models and optimization algorithms; and (3) to study uncertainty related to the input parameters of the models, whereby three scenarios with different inputs were considered. The results showed that for the first scenario, in which the input data were just the average of the region temperature and three temporal delays of 3, 6, and 9 months were considered, the models based on the three simultaneous temperature inputs with mentioned delays had higher performance as compared to the inputs just belonging to temperature input. The MLP–WA model was the best model among all. For the test stage, the mean absolute error of the MLP–WA model decreased to 30% and from 31 to 38% as compared to the radial basis function–whale algorithm (RBF–WA) and GP models, respectively. The second scenario was the evaluation of the predicted GWL based on the precipitation data of 3, 6, and 9 months. The results showed that the three variations of precipitation data as simultaneous input improved the models’ performance. The third scenario was considered in which the data from average precipitation and temperature were simultaneously used. The best results were obtained when the precipitation and temperature data with delays of 3, 6, and 9 months were used as input.