Empirically‐Derived Building‐Specific Learning Models for Predicting Occupant Vapor Intrusion Exposures

ABSTRACT Vapor intrusion (VI) is a chemical exposure pathway that refers to the migration of hazardous vapors from a subsurface vapor source, such as contaminated soil or groundwater, through the soil and into an overlying building or structure. Release and transport of these compounds through the VI pathway can result in health risks that are typically estimated by collecting and analyzing samples, representing the results as an average concentration over time, and then comparing these measurements to risk screening criteria established by regulatory agencies. These criteria determine which mitigation option, if any, is to be implemented, as well as the specified response timing. While spatial variability and temporal dynamics are well‐documented, the most common VI risk assessment approaches consist of samples collected via random timing driven by practitioner convenience and scheduling constraints. Previous studies have demonstrated that randomly timed sample results do not typically represent conservative conditions and are susceptible to underestimation and false negative results. More recent studies demonstrated that samples collected during upward vapor flux conditions driven by advective transport can result in improved risk assessments and a reduction in the potential for false negative conclusions. The current study builds upon these findings by generating empirical building‐specific correlations between barometric pressure patterns and exposure concentration measurements during upward flux conditions. These correlations are then used to predict past occupant exposure profiles by applying the building‐specific relationships to historical barometric pressure records. Three examples are described, and considerations for future VI investigations are presented. For each of these, a linear relationship between the barometric pressure drop rate and indoor concentration was observed. Subsequently, the most rapid drop in historic barometric pressure corresponded with a prediction of the highest indoor air concentration. This study represents the first time this type of building‐specific learning model has been developed and used to estimate occupant exposures. This novel exposure prediction approach has the potential to improve VI risk assessments and mitigation design and to expedite risk management decisions.