Bayesian Networks as a screening tool for exposure assessment

Abstract

A tiered approach to contamination exposure assessment is currently adopted in many countries. Increasing the site-specific information in exposure assessments is generally recommended when guideline values for contaminants in soil are exceeded. This work details a Bayesian Network (BN) approach to developing a site-specific environmental exposure assessment that focuses on the simple mapping and assessment of assumptions and the effect of new data on assessment outcomes. The BN approach was applied to a floodplain system in New South Wales, Australia, where site-specific information about elevated antimony (Sb) concentrations and distribution in soils was available. Guidelines for exposure assessment in Australia are used as a template for this site, although the approach is generic. The BN-based assessment used an iterative approach starting with limited soil Sb data (41 samples ranging from 0 to 18 mg kg⁻¹ Sb) and extending the model with more detailed Sb data (145 samples ranging from 0 to 40 mg kg⁻¹ Sb). The analyses identified dominant exposure pathways and assessed the sensitivity of these pathways to changes in assumptions and the level of site-specific information available. In particular, there was a 10.8% probability of exceeding the tolerable daily intake of Sb in the case study when the limited soil Sb data was used, which increased to 26.2% with the more detailed sampling regime. There was also a 47% decrease in the probability of overexposure to Sb when the dermal bioavailability of arsenic (a similar metalloid) was used as a surrogate measure instead of a default bioavailability of 100%. We conclude that the BN approach to soil exposure assessment has merit both in the context of Australian and international soil exposure assessments.