Development of Ambient Air Quality Population-Weighted Metrics for Use in Time-Series Health Studies
Posted on: Saturday, 10 May 2008, 03:00 CDT
By Ivy, Diane Mulholland, James A; Russell, Armistead G
ABSTRACT A robust methodology was developed to compute population- weighted daily measures of ambient air pollution for use in time- series studies of acute health effects. Ambient data, including criteria pollutants and four fine particulate matter (PM) components, from monitors located in the 20-county metropolitan Atlanta area over the time period of 1999-2004 were normalized, spatially resolved using inverse distance-square weighting to Census tracts, denormalized using descriptive spatial models, and population-weighted. Error associated with applying this procedure with fewer than the maximum number of observations was also calculated. In addition to providing more representative measures of ambient air pollution for the health study population than provided by a central monitor alone and dampening effects of measurement error and local source impacts, results were used to evaluate spatial variability and to identify air pollutants for which ambient concentrations are poorly characterized. The decrease in correlation of daily monitor observations with daily population-weighted average values with increasing distance of the monitor from the urban center was much greater for primary pollutants than for secondary pollutants. Of the criteria pollutant gases, sulfur dioxide observations were least representative because of the failure of ambient networks to capture the spatial variability of this pollutant for which concentrations are dominated by point source impacts. Daily fluctuations in PM of particles less than 10 mum in aerodynamic diameter (PM^sub 10^) mass were less well characterized than PM of particles less than 2.5 mum in aerodynamic diameter (PM^sub 2.5^) mass because of a smaller number of PM^sub 10^ monitors with daily observations. Of the PM^sub 2.5^ components, the carbon fractions were less well spatially characterized than sulfate and nitrate both because of primary emissions of elemental and organic carbon and because of differences in measurement techniques used to assess these carbon fractions.
(ProQuest: ... denotes formulae omitted.)
INTRODUCTION
In numerous epidemiologic investigations, ambient air pollution has been associated with acute respiratory and cardiovascular health outcomes.1-7 Many of these studies utilize existing health and air quality monitoring databases because of their large sample size and relative low cost. However, these two databases are not directly linkable because air quality monitors are point measurements, often temporally integrated, whereas health outcomes respond to temporally varying concentrations over a large spatial area. Different methods for spatial interpolation and modeling have been used to overcome this, but there is much discrepancy on the appropriate method to use to provide the most representative results.
Spatial interpolation techniques are used to increase spatial coverage of ambient air pollution measurements, which are often defined by spatially sparse air monitoring networks. The simplest method for representing air pollution levels is to use air pollution data from a central monitor. 8,9 The central monitor approach has been extended to nearest monitor methods by defining sub-regions around the monitors and evaluating them individually for the entire study area.10 Spatial averaging methods have also been used.11-13 Marshall et al.14 developed a metric by weighting data by the inverse distance-square to each Census tract and then population- weighted these values. Kriging and universal kriging methods have been used in epidemiologic studies, providing both a concentration estimate and an uncertainty estimate.15-17 These methods require dense monitoring networks to be representative of areas with different land uses, and application in health studies may lead to overrepresentation of rural sites in population exposure estimates.
Modeling and proximity to source methods have become more common in estimating ambient air pollution exposure. Distance-to-roadway modeling has become increasingly popular in traffic-related health studies on particulate matter and nitrous oxides18-20; Hoek et al.21 have extended distance-to-roadway modeling by incorporating measurements into the estimate of exposure to roadway emissions.21 However, these methods are not applicable to secondary pollutants and nontraffic-related primary pollutants. Land-use regression models have also been developed to improve spatial interpolation of air pollution.22,23 The development of these models, however, requires a dense monitoring network for calibration, and, similar to distanceto- source methods, land-use regression models are difficult to apply to secondary pollutants and typically do not use direct measurements in their estimates. Jerrett et al.24 have suggested that integrated meteorological emission models would provide better spatial coverage in estimates. Tong et al.25 have evaluated the use of the CMAQ (Community Multiscale Air Quality) model for spatial coverage of ozone and found that CMAQ dampened temporal variability and exaggerated spatial variability between urban and rural areas.
A retrospective study of the relationship between acute health effects and ambient air pollution is being conducted in the 20- county Atlanta metropolitan region. 6,7 This study takes advantage of fine particle composition data measured at several locations in the study area since 1999. The study includes over 1 million emergency department visits each year for respiratory and cardiovascular illnesses. For this population, only zip code of residence is known. Here, we present results of the development of a procedure for computing daily populationweighted metrics of ambient air pollution for use in this health study. Our objectives were to: (1) compute daily population-weighted ambient air pollution concentrations using data from available monitors, minimizing bias associated with the use of different measurement methods; (2) compute spatially resolved daily estimates of ambient air pollution and provide daily estimates of spatial variance; (3) maximize completeness by computing population-weighted values when station data are missing and providing a measure of error associated with computation with data missing; and (4) assess the representativeness of the air pollution metrics. Here, our focus is on characterizing the larger scale distribution of ambient pollution in the Atlanta metropolitan area using data from the ambient monitoring system of the U.S. Environmental Protection Agency (EPA) that is designed to assess compliance with health-based ambient air quality standards. The small-scale variation of ambient air pollution, such as that associated with distance to roadway, is not of interest in this study, which supports a time-series study of emergency department visits in which only zip codes of residence are known about the study population.
MATERIALS AND METHODS
Ambient air quality data for the 1999-2004 6-yr time period were analyzed from the Southeastern Aerosol Research and Characterization (SEARCH) network,26 EPA's Air Quality System (AQS), and the Assessment of Spatial Aerosol Composition in Atlanta (ASACA) network.27 Data from two SEARCH monitors were used: Jefferson Street, Atlanta (urban) and Yorkville (rural), located approximately 60 km west of Atlanta. The AQS sites used included one Species and Trends Network (STN)28 monitor-South Dekalb, located 15 km east of downtown Atlanta near the intersection of two major highways. Pollutants analyzed are as follows: nitrogen dioxide (NO2), nitrogen oxides (NOx), carbon monoxide (CO), ozone (O3), sulfur dioxide (SO2), particulate matter of particles less than 10 [mu]m in aerodynamic diameter (PM10), particulate matter of particles less than 2.5 [mu]m in aerodynamic diameter (PM2.5), and PM2.5 components elemental carbon (EC), organic carbon (OC), nitrate (NO3 -), and sulfate (SO4 2-). Monitors were selected that are located in the 20- county metropolitan Atlanta area (Figure 1). Hourly gas data were used to compute daily 1-hr maxima (NO2, NOx, CO, SO2) and daily 8- hr maxima (O3); 24-hr average PM measures were used (PM10, PM2.5, EC, OC, NO3 -, SO4 2-). Year 2000 Census data were used for 660 Census tracts located within the 20-county metropolitan Atlanta area.
Data from 5 SO^sub 2^ monitors, 6 NO2 and NO^sub x^ monitors, 5 CO monitors, 5 O3 monitors, 9 PM^sub 10^ monitors, 11 PM^sub 2.5^ monitors, and 6 PM^sub 2.5^ composition monitors were used in this study (Figure 1). In some cases, differences in the measurement methods used by the different networks resulted in measurement bias. Slightly higher SO^sub 2^ values were obtained from the SEARCH monitors than the AQS monitors because of less SO^sub 2^ loss by condensation in the SEARCH sampling line. The SEARCH monitors have slightly lower NO^sub x^ measurements than the AQS monitors; for the SEARCH monitors, independent measurements of NO2 and NO are summed, whereas the AQS measurement of total NO^sub x^ includes some NOy (total reactive nitrogen oxides) species. The SEARCH PM^sub 10^ data used in this study were obtained by summing a filter-based Federal Reference Method (FRM) PM^sub 2.5^ measurement and a PMcoarse (PM^sub 10^ - PM^sub 2.5^) measurement obtained using a dichotomous sampler. The daily AQS PM^sub 10^ data at Georgia Tech, on the other hand, were obtained using a semicontinuous method. AQS PM^sub 10^ measurements taken every sixth day were obtained by FRM measurement. The SEARCH and AQS PM^sub 2.5^ data used were FRM measurements; the ASACA PM^sub 2.5^ measurements at Fort McPherson, South Dekalb, and Tucker were tapered element oscillating microbalance (TEOM) measurements. For PM^sub 2.5^ EC/OC components, the SEARCH network measurement method was thermal optical reflectance (TOR), whereas the ASACA and AQS measurements were by thermal optical transmittance (TOT). The TOR method yields higher EC and lower OC than the TOT method.29 Derivation of Population-Weighted Metrics
Air quality data, particularly primary pollutant data, tend to have a lognormal distribution.30 To evaluate this, power transformations and Hinkley dlambda statistics were calculated for each pollutant at each monitor forlambda-0.5, 0, 0.5, and 1.31 Overall, an optimal power transformation of lambda = 0 was found, indicating that the pollutant distributions are best described by a lognormal function.
In addition to being log-transformed, data were normalized:
... (1)
Here, i,k is the normalized value of the pollutant at monitor i for day k, [mu]i is the mean of ln(Ci,k) values for a year at monitor i, and iota is the standard deviation of ln(Ci,k) values for year at monitor i. Thus, the distribution of i has an annual mean of 0 and an annual standard deviation of 1. These normalized values were inverse distancesquare weighted to the 660 Census tracts as follows.
... (2)
Here, Vj,k is the interpolated normalized value for each day k at each Census tract j, and Dij is the distance from monitor i to Census tract j. Normalized values, as opposed to actual concentrations, were used to produce a smoother interpolated surface and increase the robustness of the metric when monitor data are missing. That is, without normalization, interpolation would result in average concentrations "floating" to regions where no monitors are located. In the case of a limited monitoring network of pollutants with concentrations that are much higher near the urban center than in surrounding rural areas (e.g., vehicular emission pollutants), direct interpolation would lead to unrealistic spatial distributions. The interpolation method used here is based entirely on the ambient monitor data and does not require the use of artificial boundary conditions. Moreover, without normalization the impact of missing data on these interpolations might be such that the results are only useful if data are available from all monitors. Such a reduction in completeness of the dataset might substantially decrease the power of a time-series health study.
The normalized value at each Census tract was then converted back to a concentration using descriptive models of the means and standard deviations as a function of distance from the urban center.
... (3)
Here, [mu]i is the modeled mean of ln(Cj,k) values for the year at Census tract j and Dij is the modeled standard deviation of ln(Cj,k) values for the year at Census tract j. Logistic and linear functions were used to model the annual means and standard deviations, respectively, providing a smooth spatial surface in which local source impacts and biases due to differences in measurement methods are minimized. This procedure allows for daily anisotropic pollutant fields, but the annual average pollutant fields (means and standard deviations) are assumed to be isotropic (i.e., dependent on radial distance only). This assumption has been assessed in previous work.32 The impact of prevailing wind direction on annual pollutant fields in the southeast is less pronounced than in other regions of the United States because of a relatively stagnant air mass, particularly in summer. After ambient air pollutant levels measured at an urban monitor and a rural monitor are described in the next section, spatial model fits of annual means and standard deviations are presented.
The final step was to population-weight the concentrations in each Census tract, resulting in an overall concentration to represent the study area for each day.
... (4)
Here, Pj is the population of Census tract j, and Ck is the population-weighted concentration on day k. In a large population health study of the type being conducted in the 20-county Atlanta metropolitan region, populationweighted metrics are likely to be more representative of exposure of the study population to ambient air pollution than data from a central monitor.
Urban and Rural Monitor Pollution Levels
Median and the 25-75% quartile range concentrations are given in Table 1 for daily measures of ambient pollutants from 1999 to 2004 at the Jefferson Street, Atlanta (urban) and at the Yorkville, Georgia (rural) SEARCH sites. Also shown is the urban-rural Pearson R2 value. Primary pollutant concentrations are much higher at the urban site than the rural site. NO^sub x^ concentrations are approximately 10 times higher, CO concentrations are 3-5 times higher, and SO^sub 2^ and PM^sub 2.5^ EC concentrations are about twice as high, on average, at the urban site than at the rural site. Urban-rural SO^sub 2^ concentrations are less correlated than the other primary pollutant gases, consistent with the semivariogram analysis of Wade et al.32 Concentrations of secondary pollutants, on the other hand, are more uniform across the study area and are more correlated than concentrations of primary pollutants. Urban-to- rural average ratios for O3 and PM^sub 2.5^ SO4 2- and NO3 - components are 0.85, 1.05, and 1.14, respectively; urban-to-rural R2 values for these secondary pollutants range from 0.635 to 0.844. PM^sub 10^ mass, PM^sub 2.5^ mass, and PM^sub 2.5^ OC, which have both primary and secondary components, have urban-torural ratios ranging from 1.3 to 1.4 and have R2 values ranging from 0.350 to 0.725.
Spatial Distribution of Monitor Annual Means
For the retrospective health study in Atlanta, without detailed information about the location of subjects within the study area and with a limited number of monitors available for characterizing daily fluctuations in ambient pollutant levels, daily ambient pollutant metrics representative of the population were desired, as well as estimates of the uncertainty in the daily fluctuations on the basis of spatial variability. The approach already described, which provides daily population-weighted concentrations as well as daily spatial distributions, requires that spatial means and standard deviations be modeled. Analysis of data on population and vehicle miles traveled (VMT) provides an indication of how pollutant concentrations may vary spatially. Data from the 2000 Census, shown in Figure 2a, indicate an exponential decrease in population density with respect to distance from the urban center for the 20-county Atlanta metropolitan area, defined here as the intersection of the north-south and east-west interstate highways (see Figure 3). The distribution of VMT might provide a better indicator of the spatial distribution of pollutants dominated by mobile source emissions, such as CO, NO2/NO^sub x^, and PM^sub 2.5^ EC. A plot of VMT density, generated using the Atlanta Regional Commission's 13- county traffic demand model (calibrated for year 2000), which provides VMT estimates for over 38,000 roadway links,33 shows a decrease in the log of VMT density with increase in distance from the urban center fitted with a logistic function (Figure 2b). VMT is high within the perimeter highway, which has a radius of approximately 20 km and then decreases with increasing distance from the city center.
Descriptive models of annual mean and standard deviation were fit for each year and for each pollutant. Two examples are shown in Figure 3: 1-hr maximum NO2 (Figure 3, a, c, and e) and 24-hr average PM^sub 2.5^ mass (Figure 3, b, d, and f), both in 2004. Although the number of ambient monitors is limited (6 for NO2 and 12 for PM^sub 2.5^), the graphs suggest that the annual distributions of mean and standard deviation can be modeled with radial symmetry, consistent with previous work demonstrating the isotropic nature of correlations of pollutant measurements between pairs of monitors in the metropolitan Atlanta region.32 Similar models were obtained in each of 6 yr (1999-2004) for each of 11 pollutants.
Results for all pollutants averaged over all years are shown in Figure 4. As expected, mean CO and NO2/NO^sub x^ concentrations were best fit by logistic functions, whereas other pollutant concentrations were best fit by exponential functions. PM^sub 2.5^ EC was best fit by an exponential function, although the limited number of monitors and the difference in the measurement methods (TOR versus TOT) makes this assessment more subjective. Mean values of primary pollutants decreased much more than mean values of secondary pollutants with distance from the urban center, except for the primary pollutant SO^sub 2^. Sources of SO^sub 2^ are dominated by a few point sources of coal combustion emissions that are not located in the urban center.
The plots of standard deviation (Figure 4, b and d) demonstrate how annual variation differs over space. For the pollutants studied, annual variation is driven in large part by seasonal variation rather than day-to-day variation (e.g., day-of-week variation). Annual variation of SO^sub 2^ was greatest over the entire study area because of sporadic plume fumigation events. Variation in CO decreased most dramatically with distance from the urban center. CO from the oxidation of organics such as isoprene is a large contributor in rural areas, whereas CO from mobile source emissions is a large contributor in the urban center. The latter has a strong seasonal variation because of high emissions in winter from vehicle cold starts. NO2 has a greater seasonal variation in rural areas than in the urban center, whereas NO has greater seasonal variation in the urban center. The four PM^sub 2.5^ components studied have greater variation than total PM^sub 2.5^ because these components have different seasonal profiles. EC, OC, and NO3 - are highest in winter, whereas SO4 2- is highest in summer. RESULTS
Population-Weighted Pollutant Concentrations
Calculated population-weighted concentrations and their variation (Figure 5) are found to be highly correlated (r > 0.83) with data from the central monitor (Jefferson Street). However, population- weighted concentrations of primary pollutants are much lower than the central monitor data because of decreasing levels of these pollutants with increasing distance from the urban center.
In addition to calculating daily population-weighted pollutant concentrations, we have calculated daily population-weighted spatial variation as follows.
... (5)
Here, Sk is the daily population-weighted spatial variance of a pollutant and C[mu] k is the average concentration on day k of all Cj,k. In a large population time-series health study in which population-weighted pollutant concentrations are used for exposure, population-weighted spatial variance is a measure of uncertainty in the exposure variable. Uncertainty in the exposure variable can lead to a bias to the null and a widening of the confidence interval in the estimation of health risk ratios.34 Normalization of the spatial variation to the temporal (day-to-day) variation of population- weighted values indicates that the spatial variations of vehicle emission pollutants, that is NO2, NO^sub x^, CO, and PM^sub 2.5^ EC, are high relative to their temporal variations (Figure 6). Because temporal variation provides the power with which to observe an association in a time-series health study, high values of the ratio of spatial variation to temporal variation would translate to low power in risk assessment. In a time-series study of the relationship of air pollution and acute health effects, these results suggest that using spatially resolved measures for these pollutants might provide better indicators of exposure.
Evaluation of Model Performance
The normalized bias (NBias) between monitor data (yi) and calculated concentration values (xi) at the Census tract nearest the monitor is calculated using:
... (6)
The monitor data and calculated values are highly correlated, as expected, with R2 values of 0.94 or greater for all pollutant measures. Lack of perfect correlation is due to the standard deviations not being perfectly modeled and the monitors not being located exactly at the zip code centroids. Bias is introduced by the smoothing of the mean and standard deviation profiles over space; results are shown in Figure 7. In many cases, this bias is desirable because it corrects for bias in measurement method or for local source impacts. A few examples are noted. The positive bias associated with the SEARCH NO^sub x^ calculation is due to different sampling protocol. The SEARCH monitors at Jefferson Street (A) and Yorkville (T) have negative biases for NO^sub x^ because they measure less NO^sub x^ than the AQS monitors that likely measure other NO^sub x^ in addition to NO and NO2. The SEARCH monitors have positive biases for SO^sub 2^ because they have less loss in sampling. The SEARCH monitors have a negative OC bias and positive EC bias because of the different temperature set points used by the TOR and TOT methods. Local source impacts are also observed. The South Dekalb monitor (I) is located near a major roadway, resulting in positive biases for NO^sub x^ and EC. Fire Station No. 8 (C) is located near a rail yard and a roadway with heavy diesel traffic; it has positive PM^sub 10^ and PM^sub 2.5^ biases.
To evaluate model performance in predicting the spatial distribution of daily pollutant levels, the correlation of monitor observations and model predictions calculated without using data from that monitor are shown as a function of distance to the urban center in Figure 8. As distance from the urban center increases, the number of monitors decreases and the variability between monitors increases, resulting in decreasing predictive capability. For pollutants that are predominantly secondary in nature (i.e., formed in the atmosphere), such as O3 and PM^sub 2.5^ total and SO4 2- and NO3 - component masses, high correlations (r > 0.8) are obtained even for sites within 65 km of the urban center. On the other hand, pollutants strongly associated with mobile sources, such as NO2/ NO^sub x^, CO, and PM^sub 2.5^ EC, are not well predicted at rural sites, with R values between 0.3 and 0.4 for the Yorkville site located approximately 64 km from the urban center. The ability to predict the SO^sub 2^ concentrations is particularly poor. Major sources of SO^sub 2^ in the Atlanta area are coal combustion point sources, in particular a coal-fired power plant located 11.5 km northwest of the urban center. When a plume from this plant impacts the Atlanta area, its width is narrow resulting in a spatially heterogeneous pollutant field that is not well characterized by the ambient monitors. The correlation of observations and predictions for PM^sub 2.5^ OC, which has significant primary and secondary components, is intermediate.
DISCUSSION
Assessment of Monitor Representativeness
To evaluate how representative of the study population the daily fluctuations of ambient air pollution at each monitor are, monthly correlations between the populationweighted metric and each monitor were calculated. Results, calculated as the average of the monthly Pearson R2 values, indicate that data from stations closest to the urban center are most representative of (i.e., most correlated with) the population-weighted ambient level (Table 2). For primary pollutants, such as NO^sub x^, CO, SO^sub 2^, and PM^sub 2.5^ EC, the correlations of the population-weighted values and the monitors greater than 50 km from the urban center are much lower than those for secondary pollutants, such as O3 and PM^sub 2.5^ SO4 2- and NO3 - .
The low correlation of SO^sub 2^ measurements between the Jefferson Street (JS) monitor and monitors located nearby (Georgia Tech [GT] is 1.5 km from JS; Confederate Avenue [CA] is 8.3 km from JS) is because of the spatial heterogeneity of coal combustion plume impacts. The population-weighted CO values are less correlated with data from the Roswell Road (RR) monitor than data from either South Dekalb (SD) or Dekalb Tech (DT) despite RR being located nearer the urban center, likely because of nearby roadway emission impacts at RR. Finally, lower PM mass correlations are observed for the Fire Station #8 (FS8) PM^sub 2.5^ and PM^sub 10^ monitors, possibly because of nearby rail yard and roadway emission impacts at FS8.
These results provide a relative measure of the representativeness of ambient air quality monitors. Work is ongoing to convert these correlation values to error estimates and quantitatively assess the impact of this error on health risk assessment in terms of a bias to the null and widening of the confidence interval of risk ratio estimates.
Assessment of Completeness and Error Associated with Missing Data
To maximize completeness of the dataset, it was desirable to compute the population-weighted average on days when data from some monitors were not available. This introduces error, but not bias, relative to the calculation using all monitor data available. To quantify this error, we used the method of data withholding to calculate the normalized root mean square error (NRMSE) associated with using data from different numbers of monitors:
... (7)
Here, yk are the daily population-weighted values calculated using data from all monitors, and xk are these values calculated with data withheld. In Figure 9, results are shown for each pollutant as a function of the percentage of completeness of days when data are missing from 1999 to 2004. The total number of monitors and the percentage of total days with missing data are also given for each pollutant. For example, data were available from each of five SO^sub 2^ monitors located in the study area (three centrally located) on 1897 of 2192 days (85%) during the 6-yr study period (Figure 7a). Population-weighted averages were calculated using available data with a NRMSE of 0.11 on 97 days of the 335 days with missing data (29%), with a NRMSE of 0.12 on 44 days (42% cumulative), with a NRMSE of 0.20 on 102 days (73% cumulative), and so on. On 7 days, or 2% of the 335 days with missing data, data from only one of the three centrally located SO^sub 2^ monitors were available and population-weighted averages were calculated with a NRMSE of 0.41.
For pollutant gases, the largest error is associated with calculating population-weighted SO^sub 2^ concentrations on days when not all of the five SO^sub 2^ monitors report data, consistent with the finding of Wade et al.32 that the spatial distribution of ambient SO^sub 2^ in Atlanta is more poorly characterized than the other criteria pollutants. On the other hand, the lowest error is associated with calculating population-weighted O3 concentrations on days when not all of the five O3 monitors report data, except for some winter days when data from only the Yorkville monitor were available. For PM^sub 2.5^ components, the largest NRMSE is associated with EC and OC.
For use in time-series health studies, there is a tradeoff between maximizing exposure data completeness and minimizing exposure variable error. In health models that use a 3-day moving average of a daily pollutant measure, which is the a priori model used in the Atlanta studies,6,7 1 missing day of a pollutant measure results in a loss of 3 days from the epidemiologic analysis, decreasing the statistical power to detect associations. On the other hand, the addition of error to the exposure measurement can result in loss of statistical power as well. Current work is quantitatively assessing these impacts on the risk ratio estimates. CONCLUSIONS
A method for calculating population-weighted concentrations of ambient air pollution using data available from standard monitoring networks was developed and applied to the 20-county Atlanta metropolitan area. The methodology results in a high correlation between monitor data and modeled estimates for the Census tract where the monitor is located (R2 > 0.94), but allows for bias to dampen effects of measurement differences and local source impacts. This procedure allows for maximum completion of datasets for use in time-series health studies, with errors calculated for estimates performed with incomplete monitor data. In addition, the procedure allows for an assessment of the representativeness of ambient air pollutant monitors in a study area. Results are being used in ongoing investigations of the relationship between ambient air pollution and acute health effects in Atlanta.
ACKNOWLEDGMENTS
This work was supported by subcontracts from Emory University under grants from EPA (R82921301, R83096001, R82897602, and RD83107601) and the National Institute of Environmental Health Sciences (R01ES11294). The authors also thank researchers at the Southern Company and at Atmospheric Research and Analysis, Inc., for assistance in using the ASACA data and the SEARCH data, respectively.
IMPLICATIONS
A methodology for computing population-weighted metrics of ambient air pollution, including gas pollutants and PM mass and composition, from standard monitoring networks was developed and applied to the 20-county Atlanta metropolitan area. Measurement bias associated with differences in sampling protocol and impacts of local sources were dampened, with spatially resolved results highly correlated with observations. Population-weighted values were calculated to maximize completeness and minimize error due to missing data. Measurements of primary pollutants were shown to be representative of a much smaller area than secondary pollutants. Results are being used to investigate relationships between ambient air pollution and acute health effects.
REFERENCES
1. Air Quality Criteria for Particulate Matter; EPA/600/P-99/ 002bB; U.S. Environmental Protection Agency; Office of Research and Development; National Center for Environmental Assessment: Research Triangle Park, NC 2001.
2. Dockery, D.W.; Pope, C.A. Acute Respiratory Effects of Particulate Air Pollution; Ann. Rev Public Health 1994, 15, 107- 132.
3. Committee of the Environmental and Occupational Health Assembly of the American Thoracic Society. Health Effects of Outdoor Air Pollution, Part 2; Am. J. Respir. Crit. Care Med. 1996, 153, 477- 498.
4. Samet, J.M.; Zeger, S.L.; Dominici, F.; Curriero, F.; Coursac, I.; Dockery, D.W.; Schwartz, J.; Zanobetti, A. The National Morbidity, Mortality, and Ambient Air Pollution Study Part II: Morbidity, Mortality, and Ambient Air Pollution in the United States; Research Report 94; Health Effects Institute: Cambridge, MA, 2000.
5. Brook, R.D.; Franklin, B.; Cascio, W.; Hong, Y.; Howard, G.; Lipsett, M.; Luepker, R.; Mittleman, M.; Samet, J.; Smith, S.C., Jr. Air Pollution and Cardiovascular Disease: a Statement for Health Care Professionals from the Expert Panel on Population and Prevention Science of the American Heart Association; Circulation 2004, 109, 2655-2671.
6. Metzger, K.; Tolbert, P.; Klein, M.; Peel, J.; Flanders, W.D.; Todd, K.; Mulholland, J.; Ryan, P.B.; Frumkin, H. Ambient Air Pollution and Cardiovascular Emergency Department Visits; Epidemiol. 2004, 15, 46-56.
7. Peel, J.; Tolbert, P.; Klein, M.; Metzger, K.; Flanders, W.D.; Todd, K.; Mulholland, J.; Ryan, P.B.; Frumkin, H. Ambient Air Pollution and Respiratory Emergency Department Visits; Epidemiol. 2005, 16, 164- 174.
8. Hernandez-Garduno, E.; Perez-Neria, J.; Paccagnella, A.M.; Munguia- Castro, M.; Catalan-Vazquez, M.; Rojas-Ramos, M. Air Pollution and Respiratory Health in Mexico City; J. Occup. Environ. Med. 1997, 39, 299-307.
9. von Klot, S.; Wolke, G.; Tuch, T.; Heinrich, J.; Dockery, D.W.; Schwartz, J.; Kreyling, W.G.; Wichmann, H.E.; Peters, A. Increased Asthma Medication Use in Association with Ambient Fine and Ultra- fine Particles; Eur. Respir. J. 2002, 20, 691-702.
10. Linn, W.S.; Szlachcic, Y.; Gong, H., Jr.; Kinney, P.L.; Berhane, K.T. Air Pollution and Daily Hospital Admissions in Metropolitan Los Angeles; Environ. Health Perspect. 2000, 108, 427- 434.
11. Katsouyanni, K.; Zmirou, D.; Spix, C.; Sunyer, J.; Schouten, J.P.; Ponka, A.; Anderson, H.R.; Le Moullec, Y.; Wojtyniak, B.; Vigotti, M.A.; Bacharova, L. Short-Term Effects of Air Pollution on Health: a European Approach Using Epidemiological Time-Series Data; Eur. Respir. J. 1995, 8, 1030-1039.
12. Burnett, R.T.; Cakmak, S.; Raizenne, M.E.; Stieb, D.; Vincent, R.; Krewski, D.; Brook, J.R.; Philips, O.; Ozkaynak, H. The Associations between Ambient Carbon Monoxide Levels and Daily Mortality in Toronto, Canada; J. Air & Waste Manage. Assoc. 1998, 48, 689-700.
13. von Klot, S.; Peters, A.; Aalto, P.; Bellander, T.; Berglind, N.; D'Ippoliti, D.; Elosua, R.; Hormann, A.; Kulmala, M.; Lank, T.; Lowel, H.; Pekkanen, J.; Picciotto, S.; Sunyer, J.; Forastiere, F. Ambient Air Pollution is Associated with Increased Risk of Hospital Cardiac Readmissions of Myocardial Infarction Survivors in Five European Cities; Circulation 2005, 112, 3073-3079.
14. Marshall, J.D.; Riley, W.J.; McKone, T.E.; Nazaroff, W.W. Intake Fraction of Primary Pollutants: Motor Vehicle Emissions in the South Coast Air Basin; Atmos. Environ. 2003, 37, 3455-3468.
15. Mulholland, J.A.; Butler, A.J.; Wilkinson, J.G.; Russell, A.G.; Tolbert, P.E. Temporal and Spatial Distributions of Ozone in Atlanta: Regulatory and Epidemiological Implications; J. Air & Waste Manage. Assoc. 1998, 48, 418-426.
16. Tolbert, P.E.; Mulholland, J.A.; Macintosh, D.L.; Xu, F.; Daniels, D.; Devine, O.J.; Carlin, B.P.; Klein, M.; Dorley, J.; Butler, A.J.; Nordenberg, D.F.; Frumkin, H.; Ryan, P.B.; White, M.C. Air Quality and Pediatric Emergency Room Visits for Asthma in Atlanta, Georgia; Am. J. Epidemiol. 2000, 151, 798-810.
17. Buzzelli, M.; Jerrett, M.; Burnett, R.; Finklestein, N. Spatiotemporal Perspectives on Air Pollution and Environmental Justice in Hamilton, Canada, 1985-1996; Ann. Assoc. Am. Geograph. 2003, 93, 557-573.
18. English, P.; Neutra, R.; Scalf, R.; Sullivan, M.; Waller, L.; Zhu, L. Examining Associations between Childhood Asthma and Traffic Flow Using a Geographic Information System; Environ. Health Perspect. 1999, 107, 761-767.
19. Wilkinson, P.; Elliott, P.; Grundy, C.; Shaddick, G.; Thakrar, B.; Walls, P.; Falconer, S. Case-Control Study of Hospital Admission with Asthma in Children Aged 5-14 Years: Relation with Road Traffic in North West London; Thorax 1999, 54, 1070-1074.
20. Buckeridge, D.L.; Glazier, R.; Harvey, B.J.; Escobar, M.; Amrhein, C.; Frank, J. Effect of Motor Vehicle Emissions on Respiratory Health in an Urban Area; Environ. Health Perspect. 2002, 110, 293-300.
21. Hoek, G.; Brunekreef, B.; Goldbohm, S.; Fischer, P.; van den Brandt, P.A. Associations between Mortality and Indicators of Traffic-Related Air Pollution in the Netherlands: a Cohort Study; Lancet 2002, 360, 1203-1209.
22. Briggs, D.J.; Collins, S.; Elliott, P.; Fischer, P.; Kingham, S.; Lebret, E.; Pryl, K.; Van Reeuwijk, H.; Smallbone, K.; Van der Veen, A. Mapping Urban Air Pollution Using GIS: a Regression-Based Approach; Int. J. Geograph. Info. Sci. 1997, 11, 699-718.
23. Gilbert, N.L.; Goldberg, M.S.; Beckerman, B.; Brook, J.R.; Jerrett, M. Assessing Spatial Variability of Ambient Nitrogen Dioxide in Montreal, Canada, with a Land-Use Regression Model; J. Air & Waste Manage. Assoc. 2005, 55, 1059-1063.
24. Jerrett, M.; Arain, A.; Kanaroglou, P.; Beckerman, B.; Potoglou, D.; Sahsuvaroglu, T.; Morrison, J.; Giovis, C. A Review and Evaluation of Intraurban Air Pollution Exposure Models; J. Expos. Anal. Environ. Epidemiol. 2005, 15, 185-204.
25. Tong, D.Q.; Mauzerall, D.L. Spatial Variability of Summertime Tropospheric Ozone over the Continental United States: Implications of an Evaluation of the CMAQ Model; Atmos. Environ. 2006, 40, 3041- 3056.
26. Hansen, D.A.; Edgerton, E.; Hartsell, B.; Jansen, J.; Burge, H.; Koutrakis, P.; Rogers, C.; Suh, H.; Chow, J.; Zielinska, B.; McMurry, P.; Mulholland, J.; Russell, A.; Rasmussen, R. Air Quality Measurements for the Aerosol Research and Inhalation Epidemiology Study; J. Air & Waste Manage. Assoc. 2006, 56, 1445-1458.
27. Butler, A.J.; Andrew, M.S.; Russell, A.G. Daily Sampling of PM^sub 2.5^ in Atlanta: Results of the First Year of the Assessment of Spatial Aerosol Composition in Atlanta Study; J. Geophys. Res. Atmos. 2003, 108, 8415.
28. Implementation Plan: PM^sub 2.5^ Monitoring Program; U.S. Environmental Protection Agency; available at http://www.epa.gov/ ttn/amtic/files/ ambient/pm25/pmplan3.pdf (accessed 2007).
29. Chow, J.C.; Watson, J.G.; Crow, D.; Lowenthal, D.H.; Merrifield, T. Comparison of IMPROVE and NIOSH Carbon Measurements; Aerosol Sci. Technol. 2001, 34, 23-34.
30. Ott, W.R. A Physical Explanation of the Lognormality of Pollutant Concentrations; J. Air & Waste Manage. Assoc. 1990, 40, 1378-1383.
31. Wilks, D.S. Statistical Methods in the Atmospheric Sciences; Elsevier Science and Technology: San Diego, 2005; Vol. 91, pp 21- 63.
32. Wade, K.S.; Mulholland, J.A.; Marmur, A.; Russell, A.G.; Hartsell, B.; Edgerton, E.; Klein, M.; Waller, L.; Peel, J.L.; Tolbert, P.E. Instrument Error and Spatial Variability of Ambient Air Pollution in Atlanta, Georgia; J. Air & Waste Manage. Assoc. 2006, 56, 876-888.
33. Atlanta's Traffic Model; Atlanta Regional Commission; available at http://atlantaregional.com/html/357_ENU_HTML.htm (accessed 2007). 34. Zeger, S.L.; Thomas, D.; Dominici, F.; Samet, J.M.; Schwartz, J.; Dockery, D.; Cohen, A. Exposure Measurement Error in Time-Series Studies of Air Pollution: Concepts and Consequences; Environ. Health Perspect. 2000, 108, 419-426.
Diane Ivy, James A. Mulholland, and Armistead G. Russell
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA
About the Authors
Diane Ivy is a master's student, Jim Mulholland is a professor, and Ted Russell is the Georgia Power Distinguished Professor of Environmental Engineering in the School of Civil and Environmental Engineering at the Georgia Institute of Technology. Please address correspondence to: Jim Mulholland, School of Civil and Environmental Engineering, Georgia Institute of Technology, 311 Ferst Drive, Atlanta, GA 30332-0512; phone: 1-404-894-1695; fax: 1-404- 894- 8266; e-mail: jim.mulholland@ce.gatech.edu.
Copyright Air and Waste Management Association May 2008
(c) 2008 Journal of the Air & Waste Management Association. Provided by ProQuest Information and Learning. All rights Reserved.
Source: Journal of the Air & Waste Management Association
User Comments (0)
RSS Feeds