Prediction of groundwater level variations using deep learning methods and GMS numerical model

One of the key elements of the hydrogeological cycle and a variable used by many water resource operating models is the variation in groundwater level (GWL). One of the biggest obstacles to the drawdown analysis and GWL forecasts is the absence of accurate and complete data. The application of diverse numerical models has been regarded as a reliable approach in recent years. Such models are able to determine the GWL for any given region by utilising a wide variety of statistics, data, and field measurements such as pumping experiments, geophysics, soil and land use maps, topography and slope data, a plethora of boundary conditions, and the application of complex equations. Artificial intelligence-based models need significantly less information. The purpose of this research is to predict the changes of GWL of Shazand plain by using the PSO-ANN, ACA-ANN hybrid methods and deep learning methods LSTM, LS-SVM, and ORELM and comparing with GMS numerical model. The model's accuracy is evaluated using a two-stage validation and verification process. Then Taylor's diagram was used to select the best model. Results show that ORELM with R, Nash, RMSE and NRMSE values equal to 0.977, 0.955, 0.512 and 0.058 respectively was the best performance in the test stage. After that is the PSO-ANN model. Using the Taylor diagram is another certain way to guarantee that you've picked the best possible model. The research results show that there is a link between the ORELM and the place that is most central to the reference point. Since the GMS model is complex and requires a large amount of data and a time-consuming calibration and validation process, the ORELM model can be utilised with certainty to predict the GWL across the entire plain. This research suggests that instead of using numerical models with a complex and time-consuming structure, deep learning methods with the least required data and with high accuracy should be used to forecast the groundwater level.