Environ. Sci. Technol. 2005, 39, 7817-7825
Effects of Wind and Air Trajectory Directions on Atmospheric Concentrations of Persistent Organic Pollutants near the Great Lakes WILLIAM D. HAFNER AND RONALD A. HITES* School of Public and Environmental Affairs, Indiana University, Bloomington, Indiana 47405
Three different regression models involving air temperature, time, and either wind direction or parametric or nonparametric air trajectory direction were used with concentrations of four representative persistent organic pollutants to quantitate the atmospheric transport of these compounds to the Great Lakes. The local wind and parametric trajectory models predicted an optimal source direction for each compound, whereas the nonparametric trajectory model was based on a hypothesized source region. All three regressions were used to calculate the factor by which the partial pressures of each compound measured at five sampling sites increased when the air came from a particular source direction. Dieldrin, chlordane, polychlorinated biphenyl, and polycyclic aromatic hydrocarbon partial pressures were used with each of these regressions, and the correlation coefficients (r2) were evaluated for each model, for each compound, and for each regression term. In general, with the exception of polycyclic aromatic hydrocarbons at some sites, the explanatory powers of the regressions were not improved by the inclusion of any of these directional terms.
Introduction The atmospheric concentrations of semivolatile, persistent organic pollutants (POPs) vary as a function of several factors, the most significant of which is atmospheric temperature at the sampling site (for a review, see ref 1). Other factors are related to the strengths of the compounds’ sources. These factors include (a) how much of a given pollutant is available at a given source and the rate it enters the atmosphere (sometimes called an emission rate of that source), (b) the temperature of the emission source (sources are generally less active at lower temperatures), (c) the distance between the sources and the sampling site, and (d) the change of the emission rate as a function of time (older sources are generally less active than newer ones). For a given compound at a given sampling site, the temperature and time (factor d) effects have been modeled with a simple regression
ln(P) ) a0 +
a1 + a2t T
(1)
where P is the atmospheric partial pressure of the compound * Corresponding author e-mail:
[email protected]. 10.1021/es0502223 CCC: $30.25 Published on Web 09/14/2005
2005 American Chemical Society
(in femtoatmospheres), T is the average daily temperature (in K) measured at the site during sample collection, and t is the time (in relative Julian days) when the sample was collected. This model does not include factors a-c above, all of which depend on linking the source strength to the measured concentration. Given that this linkage depends on the movement of the compounds through the atmosphere, these factors all have a directional component. There are several ways of including this directional component: Sometimes the wind direction at the sampling site is included as another term in these regression models (2-6). The wind direction approach is convenient; it can be easily measured at the sampling site and converted into a meaningful parameter. In fact, regression models using wind direction have been used to classify sources of PAHs and PCBs in the Great Lakes region (3-6). Unfortunately, wind direction at a given sampling site can be highly variable, and this is especially true near the Great Lakes. Large bodies of water can cause surface heating variations, which can lead to diurnal wind direction variations (6). For this reason, wind direction does not necessarily correlate with the direction of long-range transport and is generally thought to be helpful only in locating local sources. Another method is the potential source contribution function (PSCF) model, which we have used to indicate sources of POPs to the Great Lakes (7). The PSCF approach is a probabilistic method that uses backward air trajectories to produce maps indicating where potential source regions might exist. As useful as these maps have been in furthering our understanding of atmospheric transport of POPs to the Great Lakes, there is an unfortunate tendency to overinterpret this sort of qualitative, visual data. In particular, PSCF maps provide no quantitative indication of the source strength and cannot be used in a regression model. Still another approach of including a directional information model is based on backward air trajectories. Even though these trajectories are calculated using a grid of wind directions and speeds (8), trajectories give a better history of the movement of a sampled air mass over several days. Using backward air trajectories, high concentrations of chlorinated pesticides measured in Ontario have been linked to atmospheric transport from the southern United States (9). In addition, PCB and PAH concentrations have also been correlated to backward air trajectory directions (10, 11). One problem with trajectory modeling is the difficulty of classifying the trajectory direction, which prevents their use in regression models. James and Hites tried to overcome this problem through the use of a nonparametric regression model (12) in which the trajectories were sorted into those that arrived at the sampling site from the north versus those from the south (the presumed source of the pesticide toxaphene). All of these approaches have their problems. For example, does the wind direction at the sampling site during the time that the samples were taken have anything to do with the long-range transport of these compounds from distant sources? How does one include multidimensional data such as backward air trajectories in these models in anything other than a qualitative manner? To answer these and other questions, this paper has two goals: First, we will present and evaluate three methods for including directional information in the classic regression model. The first method is based on local wind direction, the second uses the same model but replaces wind direction with the average backward trajectory direction of air coming to the sampling site, and the third method is a nonparametric VOL. 39, NO. 20, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7817
FIGURE 1. Locations of the five United States IADN sampling sites. regression based on the average back-trajectory direction and the source regions of the PSCF maps. To date there has been no coordinated, direct comparison of these approaches using a self-consistent data set. Second, using these three methods, we will quantitatively evaluate the directions and source strengths of dieldrin, chlordane, polychlorinated biphenyls (PCBs), and some polycyclic aromatic hydrocarbons (PAHs) coming to the Great Lakes through the atmosphere. We will make these evaluations using the very large data set obtained by the Integrated Atmospheric Deposition Network (IADN). IADN was created by Annex 15 of the 1987 revision to the Great Lakes Water Quality Agreement. IADN is a long-term program for measuring POPs in the atmosphere of the Great Lakes (7) and a joint program of the United States and Canadian governments. In the U. S., the network consists of five fully equipped stations at Brule River, WI, Eagle Harbor, MI, Sleeping Bear Dunes, MI, Chicago, IL, and Sturgeon Point, NY (Figure 1). Measurements of numerous POPs have been made every 12 days since ∼1992 at each of these sites.
Experimental Methods Sampling and Chemical Methodology. A brief summary of the sampling and analytical methods is presented here, but full details are available elsewhere (13-16). With a modified Anderson high-volume air sampler (General Metal Works), an atmospheric sample was taken every 12 days beginning at 9:00 a.m. and running for 24 h such that ∼820 m3 of air was pulled through the sampler. Compounds present in the atmospheric vapor phase were collected on XAD-2 resin (Sigma). POPs measured in the gas phase include 18 PAHs, 22 organochlorine pesticides, and over 100 PCB congeners and congener groups. The atmospheric particle phase was also collected on a filter located before the XAD-2 resin, but
