Groundwater Level Forecasting using a Deep Learning Convolutional Neural Network (CNN) Model: Seidan-Farooq Study Area

Groundwater, as one of the vital resources for meeting human, agricultural, industrial, and environmental needs, plays a key role in water resource management. Given the increasing challenges in water resource management, accurate forecasting of groundwater levels is crucial for the sustainable utilization of these resources. In this study, a deep learning model using Convolutional Neural Networks (CNN) is applied to forecast groundwater levels in the Seidan-Farooq area within the Tashk Bakhtegan Maharloo basin, Fars province. The proposed model uses meteorological data, including precipitation, temperature, evapotranspiration, and historical groundwater level data as inputs. The results show that incorporating past groundwater levels (with a one-month lag) into the model has increased forecasting accuracy. In the first scenario, the model achieved a Nash-Sutcliffe efficiency (NSE) of 0.99 and a coefficient of determination (R²) of 0.98 using meteorological parameters. In the second scenario, where historical groundwater data was added, the RMSE decreased from 0.49 to 0.35 meters, and the model's stability index (PI) increased from 0.6 to 0.8. Additionally, error analysis and model accuracy on a seasonal and monthly basis indicate that the second scenario outperforms the first. These results demonstrate the model's high capability in simulating temporal changes and seasonal fluctuations in groundwater levels. This approach can serve as an effective tool for managing groundwater resources in regions facing a scarcity of observational data.