Investigating the differences between receptor and dispersion modeling for concentration prediction and health risk assessment of volatile organic compounds from petrochemical industrial complexes

Abstract

Receptor and dispersion models both provide important information to help understand the emissions of volatile organic compounds (VOCs) and develop effective management strategies. In this study, differences between the predicted concentrations of two models and the associated impacts on the estimated health risks due to different theories behind two models were investigated. Two petrochemical industrial complexes in Kaohsiung city of southern Taiwan were selected as the sites for this comparison. Although the study compares the approaches by applying the methods to this specific area, the results are expected to be adopted for other areas or industries. Ninety-nine VOC concentrations at eight monitoring sites were analyzed, with the effects of diurnal temperature and seasonal humidity variations being considered. The Chemical Mass Balance (CMB) receptor model was used for source apportionment, while the Industrial Source Complex (ISC) dispersion model was used to predict the VOC concentrations at receptor sites. In the results of receptor modeling, 54% ± 11% and 49% ± 20% of the monitored concentrations were contributed by process emissions in two complexes, whereas the numbers increased to 78% ± 41% and 64% ± 44% in the results of dispersion modeling. Significant differences were observed between two model predictions (p < 0.05). The receptor model was more reproducible given the smaller variances of its results. The effect of seasonal humidity variation on two model predictions was not negligible. Similar findings were observed given that the cancer and non-cancer risks estimated by the receptor model were lower but more reproducible. The adverse health risks estimated by the dispersion model exceeded and were 75.3%–132.4% of the values estimated by using the monitored data, whereas the percentages were lowered to the range from 27.4% to 53.8% when the prediction was performed by using the receptor model. As the results of different models could be significantly different and affect the final health risk assessment, it is important to carefully choose an appropriate model for prediction and to evaluate by monitoring to avoid providing false information for appropriate management.