Modelling the Isotope composition of groundwater using hydrochemical properties in eastern Saudi Arabia: Implementation of innovative data intelligence techniques

Al-Qatif coastal region in eastern Saudi Arabia is an arid region with limited surface water resources and vulnerable to seawater intrusion. The study focused on modelling and prediction of the isotope composition (δ¹⁸O and δ²H) of coastal groundwater using Artificial Intelligence (AI) models utilizing readily available groundwater hydrochemical dataset. The study aims to understand the geochemical evolution of groundwater and the impact of seawater intrusion on arid coastal environment. Eight AI algorithms (KNN, SVR, RF, ET, Bag, AdaBt, GRB, and CAT) and stacking ensemble models were developed to predict the δ¹ ⁸O and δ²H isotopes of the groundwater using a dataset of physicochemical parameters, ions and elements, and isotopes from 47 wells. The study shows that the stacking ensemble models outperformed individual algorithms. The optimum model for δ¹ ⁸O (O_M1) was achieved with R² of 0.9858, MAE of 0.0440, and Pearson correlation of 0.9941. for δ²H the optimal model (H_M1) was achieved with R² of 0.9317, MAE of 0.5334, and Pearson correlation of 0.9658. The significant relationship between hydrochemical parameters and isotopic composition indicate that the variation in groundwater chemistry is mostly associated with mixing processes, primarily driven by seawater intrusion in the coastal region. The study demonstrates the potential of AI-based models to predict the isotopic signature and groundwater dynamics in similar coastal arid environments. • Predictive models of δ 18 O and δ 2 H isotopes of coastal groundwater were developed. • Groundwater's physicochemical properties and major ions utilized. • The stacking ensemble models outperform traditional machine learning techniques. • The best models achieve an accuracy of 0.98 for δ 18 O and 0.93 for δ 2 H. • Calcium, chloride, and bromide are the main influences.