Site-specific updating and aggregation of Bayesian belief network models for multiple experts

Area of Science:

• Environmental Science
• Risk Assessment
• Bayesian Statistics

Background:

• Combining multiple expert opinions is crucial for complex environmental problems like hazardous chemical cleanup.
• Bayesian Belief Networks (BBNs) offer a framework for expert knowledge integration but require robust methods for model aggregation.
• Assessing groundwater contamination, specifically the occurrence of reductive dechlorination of trichloroethene (TCE), necessitates reliable risk models.

Purpose of the Study:

• To develop and apply a novel method for combining multiple expert opinions encoded in BBN models.
• To tailor aggregate risk models to site-specific conditions using observed evidence.
• To improve the prediction accuracy of groundwater contamination cleanup feasibility assessments.

Main Methods:

• Utilized Bayes Rule to update individual expert BBN models with observed evidence.
• Computed posterior probability weights for each expert model based on evidence consistency.
• Aggregated 21 expert BBN model predictions using weighted averaging for a groundwater contamination case study.

Main Results:

• The developed method effectively weights expert models, giving more influence to those consistent with site data.
• The aggregate BBN model prediction differed significantly from simple model averaging.
• Demonstrated the method's applicability using both a pedagogical example and a real-world groundwater contamination scenario involving TCE.

Conclusions:

• The proposed Bayesian weighting method provides a more accurate and site-specific approach to aggregating expert BBN models.
• This technique enhances the reliability of risk assessments for hazardous chemical cleanup in contaminated groundwater.
• The findings support the use of evidence-based weighting for expert models in environmental risk management.