During the last decade, soil contamination with volatile organic contaminants (VOC) received special attention because of their potential to cause indoor air problems. Moreover, research has shown that people spend 64% to 94% of there time indoors; therefore, the indoor air quality is of a primary importance for exposure to VOC. Human health risks to VOC—in cases of soil contamination—are often dominated by the exposure route ‘inhalation of indoor air’. Exposure is often a result of vapour transport from the soil or groundwater to the indoor air of the building. Within human health risk assessments, a variety of algorithms
are available that calculate transfer of soil gas to the indoor air. These algorithms suffer from a relatively high uncertainty due to a lack of representation of spatial and temporal variability. For such an application, these algorithms need to be further verified empirically against field observations so that they can be sufficiently reliable for regulatory purposes.
This paper presents the accuracy for seven algorithms by using observed and predicted soil and indoor air concentrations from three sites, where the groundwater had been contaminated with aromatic and chlorinated VOC.
The algorithms for vapour intrusion that are frequently used in European countries were included in this study and were Vlier–Humaan (Flanders), CSoil (Netherlands), VolaSoil (Netherlands), Johnson & Ettinger (USA), Risc (United Kingdom), and the dilution factor (DF) algorithms from Sweden and Norway. Three sites were investigated in more detail and samples were taken synoptically from the groundwater, soil and indoor air on four occasions. On the petroleum sites, the aromatic hydrocarbons benzene, toluene, ethylbenzene and xylenes were analysed and, on the dry cleaning sites, the chlorinated hydrocarbons tetrachloroethylene, trichloroethylene and cis 1,2-dichloroethene. To increase spatial resolution, measurements in groundwater and soil air were taken in three different zones at each site, in the close proximity of or in the building. During sampling, several relevant soil properties were measured like the bulk density, water and air filled porosity, soil temperature and depth of the groundwater. Also, building properties like the dimensions of the building and the quality of the floor were registered. The seven algorithms were applied to compare that observed with the predicted concentrations in soil and indoor air. The groundwater concentrations were used as a source contamination. The results from the algorithms were compared by using performance criteria to assess the accuracy of each algorithm.
All calculations are presented in a box plot that contains the predicted soil or indoor air versus the observed concentrations. Results from the applied criteria are presented for each algorithm.
Differences between predictions and observations were up to three orders of magnitude and can be partially related to the amount of parameters included in each algorithm and the mathematical concept used. For example, the inclusion or exclusion of a capillary fringe or temperature correction for the Henry constant: it is not clear why all algorithms tend to over-predict the soil air concentration. The prediction mostly starts with the calculation of a soil air concentration related to the Henry constant, followed by diffusive and/or convective transport to the soil surface and zone of influence around the building foundation. Further research is needed to investigate the over-predictions and the use of the Henry constant to calculate
the soil air concentration should be reviewed.
The algorithms with the highest accuracy were the Johnson and Ettinger and the Vlier–Humaan algorithms. The DF algorithms from Sweden and Norway resulted in higher over- and underpredictions than others. Results for the indoor air showed that all the algorithms calculate high and low concentrations in the indoor air when compared to observations. The algorithms with the highest accuracy were JEM, Vlier–Humaan and CSoil. The DF algorithm from Norway calculated concentrations that were frequently higher than observed concentrations and the Swedish DF algorithm showed frequent higher and lower concentration than observed. The conservatism of the most accurate algorithms is sufficient for regulatory purposes, and they can trigger an integrated programme of field observations (monitoring) or/and modelling.
The dataset used for this paper was derived from three sites with groundwater contamination and further verification of these algorithms should be done for other sites that have a vadose zone contamination.