Spatial Prediction Modeling of Geogenic Chromium in Groundwater Using Soft Computing Techniques.

Area of Science:

• Environmental Science
• Hydrogeology
• Computational Science

Background:

• Groundwater contamination by chromium (Cr) presents a significant human health risk.
• Underestimation of Cr contamination severity is likely due to insufficient well testing.
• Geogenic Cr poses a threat, necessitating accurate predictive tools for risk assessment.

Purpose of the Study:

• To estimate geogenic chromium concentrations in groundwater using various predictive models.
• To compare the performance of soft computing techniques (ANN, GEP, MARS, M5 Tree) against ensemble (RF) and linear (MLR) models.
• To identify populations at risk from Cr contamination in northeastern Iran.

Main Methods:

• Utilized a dataset of 676 groundwater Cr concentration measurements.
• Employed and compared multiple predictive models: artificial neural networks (ANN), random forest (RF), gene expression programming (GEP), multivariate adaptive regression splines (MARS), M5 Tree, and multiple linear regression (MLR).
• Evaluated models based on predictive accuracy using geological and geochemical parameters.

Main Results:

• Artificial neural networks (ANN) exhibited the highest predictive accuracy for geogenic Cr concentration.
• Random forest (RF) provided competitive results, followed by GEP and MARS.
• Multiple linear regression (MLR) showed the lowest accuracy, indicating limitations in modeling complex geochemical processes.
• The ANN model identified over 600,000 individuals at risk in central and western regions.

Conclusions:

• Advanced predictive models, particularly ANN, are effective for estimating groundwater Cr contamination.
• These models are valuable tools for groundwater quality management and risk assessment.
• Findings highlight the potential applicability of these methods in other regions facing similar groundwater contamination challenges.