Investigating the role of ENSO in groundwater temporal variability across Abu Dhabi Emirate, United Arab Emirates using machine learning algorithms

Accurate prediction of groundwater levels is crucial for managing groundwater resources efficiently. The complex aquifer heterogeneity and groundwater abstraction variation present challenges to have accurate groundwater level models over Abu Dhabi emirate, United Arab Emirates. In the present study, two data-driven models are employed, which are the Long Short-Term Memory (LSTM) and the Random Forest (RF) to develop a model for the prediction of monthly groundwater level in the Abu Dhabi Emirate. The incorporated data in the models are precipitation, terrestrial water storage, soil moisture, evapotranspiration, and the El Niño-Southern Oscillation (ENSO) 3.4 index. The groundwater monitoring wells data are obtained for 263 monitoring wells distributed over Abu Dhabi emirate for the period 2000–2023 in a monthly temporal scale. The models' performance was assessed using the Nash-Sutcliffe efficiency (NSE), root mean square error (RMSE) the coefficient of determination (R 2 ) and Percent bias (PBIAS). An optimization technique was also applied to address the impact of the lags on enhancing the groundwater level model. The LSTM model outperformed the RF model during the testing period, achieving R 2 = 0.79, NSE = 0.70, RMSE = 0.38 m and PBIAS = 0.2% with a 3-month lag. The global sensitivity analysis was applied to understand the importance of each parameter and its influence on the models’ output. This study highlights the potential use of data-driven models for the prediction of groundwater level which could aid water managers to monitor the groundwater resources at a regional scale. The developed model can serve as an alternative approach for predicting groundwater level change over the Abu Dhabi Emirate. • Groundwater level predictions using LSTM and RF models achieved satisfactory accuracy across Abu Dhabi Emirate. • Incorporating time lags improved the models' accuracy for groundwater level predictions. • Global sensitivity analysis approach assessed the impact of input data on the model performance.