Groundwater is a vital resource for communities in arid regions, such as Saudi Arabia, where seawater intrusion poses a significant threat. To determine groundwater quality with traditional methods is time-consuming and costly process. In this study, we developed a machine learning method as a tool for predicting seawater intrusion in coastal areas using groundwater chemical parameters. We analysed 99 water sample data from the Red Sea and 42 from eastern coast of Saudi Arabia for 13 chemical parameters and conducted two experiments with and without chloride parameter. Unlike previous studies that developed complex machine learning methods, we focus on lightweight machine learning algorithms to model seawater intrusion for possibility of sensor deployment in future. The machine learning algorithms assessed and compared are namely decision trees, ensemble machine learning methods (Random Forest and Gradient Boosting), support vector machines, multilayer perceptrons, and ridge regression. With the chloride parameter included, ridge regression emerged as the best model, with an R2 of 99.8% and the lowest Akaike and Bayesian information criteria values. Without chloride data, multilayer perceptron achieved a marginally higher R2 of 98.7% but required longer training times. Notably, a high accuracy was maintained without chloride parameter, demonstrating the capacity of machine learning for augment monitoring. This research underscores the effectiveness of ridge regression and neural networks in predicting seawater intrusion, enabling the accurate mapping of vulnerable areas to inform adaptive management strategies for groundwater preservation. The regional scope of this study offers valuable insights for other arid coastal regions that rely on groundwater and threatened by seawater intrusion.
Read full abstract