An intelligent hybrid deep learning-machine learning model for monthly groundwater level prediction

Predicting groundwater level is necessary for environmental protection. Thus, our research suggests a hybrid artificial intelligence model that integrates the particle swarm optimization (PSO) algorithm, the coati optimization (COO) algorithm, the gated recurrent unit (GRU), and the adaptive neuro-fuzzy inference system (ANFIS) to address these challenges. The new model, called PSO-COO-GRU-ANFIS (PCGA), operates in several steps. First, the model uses PSO-COO to set ANFIS and GRU parameters. Then, the model uses GRU to extract hidden patterns from the data. In the final step, the extracted patterns are input into ANFIS to generate predictions. Specifically, the proposed model is used to precisely forecast monthly groundwater levels (GWLs) in the Ardabil Plain, Iran. The PCGA is also compared with several benchmark models, and its accuracy is evaluated using multiple evaluation metrics. Our findings show that the PCGA model achieves a mean absolute error (MAE) of 1.90 and a Nash-Sutcliffe efficiency (NSE) of 0.90. The PCGA model enhances the MAE and NSE of all other prediction models by 14-77% and 1-20%, respectively. The results also highlight the effectiveness of the PSO-COO algorithm and the GRU model in improving forecast precision, as the hybrid optimization approach significantly reduced error fluctuations during parameter tuning, and the GRU demonstrated strong capability in capturing long-term temporal dependencies in groundwater level data. Overall, the findings of the current study show that the PCGA model is a robust tool for forecasting monthly groundwater levels. The PCGA model demonstrates superior predictive capability compared to conventional and standalone models. By combining the strengths of optimization algorithms, deep learning, and fuzzy inference systems, the model effectively captures nonlinear and dynamic relationships in groundwater level (GWL) data.