Evaluating groundwater salinity is crucial, particularly in arid and semi-arid regions where access to fresh water is essential for various needs. In this study, the GALDIT method, which uses expert judgment to score six key parameters to assess vulnerability to seawater intrusion (SWI), was utilized to evaluate the susceptibility of the Salmas and Urmia aquifers. However, depending on expert judgment may lead to subjectivity and potential bias in assessing vulnerability. To address these limitations, different artificial intelligence-based models were applied to enhance model performance. In the Urmia aquifer, the vulnerability index obtained by GALDIT varied from 3.8 to 6.8, and in the Salmas aquifer, it ranged from 4.3 to 7.8. The difference in the GALDIT vulnerability index ranges primarily reflects distinct hydrogeological conditions in each area. A qualitative comparison between the GALDIT index maps and the observed Electrical Conductivity (EC) distribution highlighted notable spatial mismatches, suggesting that the GALDIT index alone may not fully capture the spatial variability of salinity levels. To assess vulnerability using AI-based models, parameters of GALDIT index were used as input, while observed EC values, as effective salinity indicators, were used as the model outputs. The results showed that all AI-based models outperformed the conventional, expert-based GALDIT index in predicting aquifer vulnerability. While the GALDIT model demonstrated limited correspondence with observed salinity patterns, the AI-based models—especially BO-ELM and Integrated Ensemble—provided significantly improved predictions. These models achieved the highest accuracy in both Urmia and Salmas Plains, with DC values approaching 0.99–1.0 and substantially reduced RMSE, effectively capturing the complex spatial variability of salinity in the aquifers. Although the proposed framework is demonstrated using the Urmia and Salmas aquifers, the methodology is transferable and can be applied to other coastal aquifers with similar hydrogeological settings through site-specific recalibration, provided sufficient data are available.
Khajeh et al. (Fri,) studied this question.