Dien, N.T.; Hirai, Y.; Sakai, S. (2015) Proceedings of 35th International Symposium on Halogenated Persistent Organic Pollutions, pp. 19481
Introduction
Polybrominated diphenyl ethers (PBDEs), including penta-BDEs and octa-BDEs, were listed as persistent organic pollutants (POPs) in the Stockholm Convention in 2009. In Japan, penta-BDEs and octa-BDEs were phased out in 1990 and 1999; deca-BDE is still in use, but its domestic demand is gradually decreasing (from 10,000 t in FY1990 to 990 t in FY2011).1 To examine the effectiveness of reduction policies, environmental levels of PBDEs have been monitored at multiple stations in Japan. The Japan Ministry of Environment (JMOE) has performed annual surveys of atmospheric PBDEs (2009−2012) present in bulk concentrations.2 JMOE data showed spatial variation regarding different sampling sites (Figure 1) and revealed the dominance of BDE-209 (87−97%), followed by BDE-47 (1−7%); then BDE-99 and BDE-183 (1−2% for each) (Figure 2). Thus, there is a question as to which factors may be relevant to spatial variation of atmospheric PBDE concentrations. Meteorological conditions (i.e., ambient temperature, rainfall, relative humidity, and wind speed) are considered important factors influencing atmospheric concentrations of PBDEs. Socioeconomic factors related to population density also influenced the distribution of the various PBDEs among regions.3 Therefore, a multiple regression model combining environmental and anthropogenic factors is needed to address PBDE trends. On the other hand, JMOE data also showed variations in the limit of detection (LOD) values by year (Table 1). Especially, for BDE-183, more than 1/3 of the samples were non-detectable (nd). If these "nd" values are completely ignored, important information may be lost, and hence, inappropriate conclusions could be drawn. In this case, consistent estimates can be obtained by the method proposed by Tobin.4 This approach is usually called the "Tobit" model and is a special case of the more general censored regression model. Our motivation in this study was first to determine, using multiple linear regression of measurements, which factors were predominantly influencing atmospheric PBDEs. Secondly, we use the Tobit model to deal with nd values and to explore how such treatment can affect the results of regression models.