Predicting bacterial transport through saturated porous media using an automated machine learning model

Area of Science:

• Environmental microbiology
• Hydrogeology
• Data science

Background:

• Escherichia coli (E. coli) from manure-amended soil can contaminate groundwater.
• Predicting bacterial transport is key to mitigating microbiological risks.

Purpose of the Study:

• To develop machine learning models for predicting E. coli transport in saturated porous media.
• To identify key variables influencing bacterial subsurface movement.

Main Methods:

• Compiled 377 datasets from 61 studies on E. coli transport.
• Trained six machine learning algorithms using eight input variables.
• Evaluated model performance based on bacterial retention scenarios.

Main Results:

• Machine learning models effectively predicted E. coli transport, despite low individual variable correlations.
• Gradient Boosting Machine and Extreme Gradient Boosting showed superior performance.
• Pore water velocity, ionic strength, median grain size, and column length were key predictors.

Conclusions:

• Machine learning offers a valuable tool for assessing E. coli subsurface transport risk.
• Data-driven approaches are feasible for predicting various environmental contaminant transport.