Machine Learning Modeling Based on Microbial Community for Prediction of Natural Attenuation in Groundwater

Area of Science:

• Environmental Science
• Microbiology
• Data Science

Background:

• Natural attenuation is a key remediation strategy for contaminated groundwater.
• Evaluating attenuation potential is critical for achieving remediation goals within set timelines.
• Historic pesticide manufacturing sites often present complex contamination challenges.

Purpose of the Study:

• To develop and assess a machine learning approach for predicting natural attenuation of contaminants.
• To utilize microbiological data and contaminant attenuation rates for predictive modeling.
• To compare the performance of different machine learning models in predicting natural attenuation feasibility and rates.

Main Methods:

• Long-term monitoring of microbial communities and contaminants (BTEX, CAHs) at a pesticide manufacturing site.
• Development of a machine learning model using Random Forest Classification (RFC) followed by Random Forest Regression (RFR) or Artificial Neural Networks (ANNs).
• Model training and cross-validation using microbiological data and contaminant attenuation rates.

Main Results:

• RFC accurately predicted natural attenuation feasibility for both BTEX and CAHs, identifying key microbial genera.
• RFR was sufficient for BTEX but unreliable for CAHs attenuation rate prediction.
• ANN demonstrated superior performance in predicting attenuation rates for both BTEX and CAHs.
• A composite RFC and ANN model reduced mean absolute percentage errors.

Conclusions:

• Machine learning, particularly a combined RFC and ANN approach, shows significant potential for predicting natural attenuation.
• Integrating field microbial data with machine learning enhances the accuracy of natural attenuation predictions.
• This approach offers a promising tool for evaluating and managing contaminated groundwater sites.