Tracking Major Sources of Water Contamination Using Machine Learning.

Area of Science:

• Environmental microbiology
• Water quality assessment
• Computational hydrology

Background:

• Current microbial source tracking methods using grab samples are insufficient for understanding contamination dynamics over space and time.
• Predictive modeling of microbial contamination sources can significantly aid watershed management strategies.

Purpose of the Study:

• To evaluate the efficacy of various machine learning models in predicting major microbial contamination sources within a watershed.
• To identify key environmental variables influencing microbial source determination.

Main Methods:

• Six machine learning models (KNN, Naïve Bayes, SVM, NN, Random Forest, XGBoost) were developed.
• Models utilized land cover, weather, and hydrologic data to predict microbial sources (human vs. non-human).
• Model performance was assessed using accuracy and Receiver Operating Characteristic (ROC) Area Under the Curve (AUC).

Main Results:

• All tested models successfully predicted microbial sources with average accuracies from 69% (Naïve Bayes) to 88% (XGBoost).
• XGBoost demonstrated the highest performance (AUC = 0.88), followed by Random Forest (AUC = 0.84).
• Precipitation and temperature were identified as the most influential factors in predicting microbial sources via Random Forest analysis.

Conclusions:

• Machine learning models, especially XGBoost, offer a powerful tool for predicting dominant microbial contamination sources.
• These models effectively leverage relationships between microbial contaminants, daily weather, and land cover data.
• The findings provide valuable insights for improved watershed management and water quality monitoring.