Groundwater salinity modeling and mapping using machine learning approaches: a case study in Sidi Okba region,...

Area of Science:

• Hydrogeology
• Environmental Science
• Machine Learning Applications

Background:

• Groundwater salinization is complex, with limited data on controlling factors hindering accurate prediction and mapping.
• Accurate mapping of groundwater salinity (GWS) is crucial for water resource management, especially in arid and semi-arid regions.
• Existing methods often overlook the impact of input variable combinations on model accuracy.

Purpose of the Study:

• To model and map groundwater salinity (GWS) in the Mio-Pliocene aquifer, Sidi Okba region, Algeria, using machine learning.
• To identify critical factors influencing GWS and assess their impact on predictive model accuracy.
• To compare the performance of various machine learning models for GWS prediction.

Main Methods:

• Utilized a limited dataset of electrical conductivity (EC) measurements and digital elevation model (DEM) derivatives.
• Applied feature selection methods (RFE, FFS, BFS) to identify optimal input variable combinations.
• Trained and evaluated five machine learning models: Random Forest (RF), HyFIS, KNN, CRM, and SVM.

Main Results:

• The Random Forest (RF) model demonstrated superior performance in both training and testing phases (e.g., R=0.854 training, R=0.831 testing).
• Feature selection methods identified crucial, often overlooked, input variable combinations that significantly improved model accuracy.
• The generated GWS map revealed alarming salinization levels in low-elevation areas distant from water sources.

Conclusions:

• Machine learning, particularly the RF model, offers enhanced predictive performance for GWS with minimal input variables.
• The study highlights the importance of considering input variable interactions for accurate GWS modeling.
• Findings provide critical insights for managing groundwater resources and addressing salinization in the Sidi Okba region.