Machine learning approaches for groundwater vulnerability assessment in arid environments: Enhancing DRASTIC with ANN and Random Forest

This study proposes Artificial Intelligence methods, namely, Artificial Neural Networks (ANN) and Random Forest (RF), for developing groundwater vulnerability maps in arid regions while minimizing data requirements. While the DRASTIC approach is widely used for assessing intrinsic groundwater vulnerability, its predefined weights and ratings are controversial due to their dependence on expert judgment. Using importance analysis with an RF Regressor on data from Qatar, as a case study representing an arid environment, the study revealed that soil media and groundwater recharge have negligible effects on vulnerability in arid regions. Both ANN and RF models showed good agreement with the original DRASTIC vulnerability map when using only five of the seven DRASTIC parameters, with correlation coefficients more than 0.8. Statistical analysis confirmed both models have good reliability, though the RF model demonstrated a slightly better performance with lower Mean Absolute Error values (2.9 for training, 3.2 for validation) compared to the ANN model (3.6 for training, 3.7 for validation). The study shows that groundwater vulnerability assessment in arid environments with DRASTIC using RF is more time efficient and accurate compared to the ANN.