An advanced approach for drinking water quality indexing and health risk assessment supported by machine learning modelling in Siwa Oasis, Egypt

Abstract

Study region

Siwa Oasis is located very far (800 km) from the main water resources (Nile River) of Egypt and the people in the study area mainly rely on groundwater for all purposes

Study focus

The deterioration of drinking water quality and the accumulation of potentially toxic elements (PTEs) in water at high levels in arid regions such as Siwa Oasis in Egypt can pose significant risks to humans and living organisms. The methodology of study involved performing geochemical modeling, contamination source detection, and optimizing a new model using machine learning model for prediction of integrated weight water quality index (IWQI), Health risk indices (HI and HQ) regarding oral and dermal exposure to potentially toxic elements (PTEs).

New hydrological insights for the region

The key findings of this research showed that the Nubian sandstone aquifer (NSSA) is characterized mainly by mixed Ca-Mg-Cl/SO₄ fresh water type and influenced by silicate weathering. The nitrates sources fell between atmospheric inputs in the case of NSSA, soil nitrogen in Tertiary carbonate aquifer (TCA), springs, and drains, while sewage water strongly affects the lakes. The IWQI values demonstrated that water resources in the deep aquifer (NSSA) is appropriate for drinking with ranking of quality range from medium to excellent quality (IWQI < 150). The shallow aquifer (TCA) is suitable for drinking in the south east of the Oasis only with intermediate quality ranking (100 < IWQI < 150), while the poor water quality needs further treatment in the western side of Siwa Oasis. The non-carcinogenic risks evaluation revealed the vulnerability of child and adult to oral exposure of PTEs in the west and center of the investigated area. The feed forward back propagation neural network (FFBP-NN) model was a powerful tool for predicting IWQI and HI, where the relationship between the actual and predicted value had R² greater than 0.95 and mean square error (MSE) range from 5.4E-05–0.66, root mean square error (RMSE) between 0.006 and 0.81, and relative square error (RSE) between 0.001 and 2.4 E-05.