Forecasting groundwater level changes using machine learning techniques in Tazerbo area, Al Kufra Basin, southeast Libya

Abstract

Due to the increasing demand for consumable water, groundwater management is critical, especially in arid and semi-arid regions. Effective management strategies are crucial for maintaining a sustainable water supply and promoting environmental health. Machine learning models can identify patterns and trends that help forecast fluctuations in groundwater levels. These predictions are essential for sustainable water resource management, thereby preventing water shortages. In this study, a machine learning model based on a time series neural network "Nonlinear Autoregressive Exogenous Neural Network (NARX-NN)" was made to predict the groundwater levels in Tazerbo, Al Kufra Basin, southeast Libya. The proposed model uses annual data of the groundwater levels for the last two decades (2004-2024) collected from 14 piezometric wells in the study area. The model was trained and validated using statistical performance metrics, including R², MSE, and RMSE, achieving high predictive accuracy across all wells. The model performed excellently during training and testing. Using NARX-NN, scenario-based forecasts for 2030 and 2040 were generated for 14 wells under two pumping rates 255,000 m³/day and 400,000 m³/day. At the current rate, groundwater is projected to decline by ~ 2 m by 2030 and ~ 1.6 m by 2040. Under higher pumping rates, drawdowns could exceed 50 m by 2030 and 2040. The results reveal spatially variable trends in groundwater decline, with significant drops projected in the northern and eastern zones under increased extraction. These findings offer valuable insights into sustainable groundwater management and long-term planning in water-stressed basins.