Forecasting Groundwater Levels using a Hybrid of Support Vector Regression and Particle Swarm Optimization

Forecasting the groundwater level is crucial to managing water resources supply sustainably. In this study, a simulation–optimization hybrid model was developed to forecast groundwater levels in aquifers. The model uses the PSO (Particle Swarm Optimization) algorithm to optimize SVR (Support Vector Regression) parameters to predict groundwater levels. The groundwater level of the Zanjan aquifer in Iran was forecasted and compared to the results of Bayesian and SVR models. In the first approach, the aquifers hydrograph was extracted using the Thiessen method, and then the time series of the hydrograph was used in training and testing the model. In the second approach, the time series data from each well was trained and tested separately. In other words, for 35 observation wells, 35 predictions were made. Aquifer’s hydrograph was evaluated using the forecasted groundwater level in the wells. The results showed that the SVR-PSO hybrid model performed better than other models in terms of Root Mean Square Error (RMSE) and coefficient of determination (\({R}^{2}\)) in both approaches. In the first approach, the SVR-PSO hybrid model forecasted the groundwater level in the next month with a training RMSE of 0.118 m and testing RMSE of 0.221 m. In the second approach, using the SVR-PSO hybrid model, the RMSE error was reduced in 88.57% of the wells compared to other models, and more reliable results were achieved. Based on the performance, the SVR-PSO hybrid model can be used as a tool for decision support and management of similar aquifers.