Grantee Research Project Results

During year 3, we have accomplished most of the objectives that are critical to achieve our final goals. Our major findings and their significances are summarized below.

 

Air quality modeling could potentially improve exposure estimates for use in epidemiological studies. We investigated this application of air quality modeling by estimating location-specific (point) and spatially aggregated (county level) exposure concentrations of particulate matter with an aerodynamic diameter less than or equal to 2.5 µm (PM_2.5) and ozone (O₃) for the eastern United States in 2002 using the Community Multi-scale Air Quality (CMAQ) modeling system and a traditional approach using ambient monitors. The monitoring approach produced estimates for 370 and 454 counties for PM_2.5 and O₃, respectively. Modeled estimates included 1,861 counties, covering 50% more population. The population uncovered by monitors differed from those near monitors (e.g., urbanicity, race, education, age, unemployment, income, modeled pollutant levels). CMAQ overestimated O₃ (annual normalized mean bias = 4.30%), while modeled PM_2.5 had an annual normalized mean bias of -2.09%, although bias varied seasonally, from 32% in November to -27% in July. Epidemiology may benefit from air quality modeling, with improved spatial and temporal resolution and the ability to study populations far from monitors that may differ from those near monitors. However, model performance varied by measure of performance, season, and location. Thus, the appropriateness of using such modeled exposures in health studies depends on the pollutant and metric of concern, acceptable level of uncertainty, population of interest, study design, and other factors.

 

In the context of source-apportionment using the CMAQ model, during this project year, we focus on two important tasks: (1) the model evaluation of CMAQ and the Comprehensive Air Quality Model with Extensions (CAMx), and (2) the application of CAMx with the particle source apportionment technology (PSAT) for source apportionment and comparison with results from CMAQ with the Brute force method (BFM). Model evaluation shows that CMAQ performs better for max 1- and 8-h O₃, and the wet deposition fluxes of NO₃^-, SO₄^2-, and NH₄⁺ and both models perform similarly for 24-h average PM_2.5 and its components. For source apportionment in January 2002, the three most important sources of PM_2.5 domainwide are coal combustion, biomass burning, and other mobile sources by CAMx/PSAT and biomass burning, miscellaneous area sources, and coal combustion by CMAQ/BFM. The two methods agree that the top three contributors in July are coal combustion, miscellaneous area sources, and industrial processes, but differing considerably in the magnitude of contributions from coal combustion. For primary PM species, the source contributions from CMAQ/BFM and CAMx/PSAT are very similar, as the concentrations of these species are linearly related to emissions. However, large discrepancies exist in the source apportionment of secondary PM species due to the fact that CMAQ/BFM accounts for the interactions of secondary species through photochemistry and aerosol formation processes, whereas CAMx/PSAT does not account for these interactions.

 

The evaluation of physically based computer models for air quality applications is crucial to assist in control strategy selection. Selecting the wrong control strategy has costly economic and social consequences. The objective comparison of mean and variances of modeled air pollution concentrations with the ones obtained from observed field data is the common approach for assessment of model performance. One drawback of this strategy is that it fails to calibrate properly the tails of the modeled air pollution distribution, and improving the ability of these numerical models to characterize high pollution events is of critical interest for air quality management. In this work, we introduced an innovative framework to assess model performance, not only based on the two first moments of models and field data, but also on their entire distribution. Our approach also compares the spatial dependence and variability in both models and data. More specifically, we estimate the spatial quantile functions for both models and data, and we apply a nonlinear monotonic regression approach on the quantile functions taking into account the spatial dependence to compare the density functions of numerical models and field data. We use a Bayesian approach for estimation and fitting to characterize uncertainties in data and statistical models. We assess the performance of the EPA CMAQ model to characterize ozone ambient concentrations. Our approach shows a 75% reduction in the root of mean square error (RMSE) compared to the default approach based on the two moments of models and data.

 

To relate adverse health effects to PM_2.5 sources and to develop effective control strategy for PM_2.5, source apportionment studies for exposure of PMPM_2.5 are needed. Limitations of existing source apportionment studies were identified. The objectives of this work were to: (1) demonstrate and evaluate a methodology for apportioning human exposure to PM_2.5 emission sources; (2) quantify the variability of exposure apportionment for different sub-groups of the population; (3) identify seasonal differences in source contributions to total PM_2.5 exposure; and (4) analyze effects of infiltration factors on source apportionment. The Stochastic Human Exposure and Dose Simulation Model for PM_2.5 (SHEDS-PM) was used to apportion non-ambient PM sources and to determine the contribution of ambient exposure to total exposure. Two receptor modeling approaches were used to apportion ambient concentrations to sources: chemical mass balance (CMB) and Positive Matrix Factorization (PMF). U.S. EPA Speciation Trends Network (STN) speciated PM_2.5 fixed site monitoring data for two counties in New York City (Bronx and Queens) were used to characterize the composition of ambient PM_2.5 over space and the study periods are April, July, October, and December 2002. Secondary aerosols and motor vehicles contribute most among ambient sources. Smoking contributes most among non-ambient sources to the exposure for smokers and non-smokers living with smokers. Home cooking contributes most among non-ambient for non-smokers that do not live with smokers. Seasonal variability in ambient source apportionment is affected by air exchange rate and secondary PM formation. Sensitivity analysis on infiltration factors indicates that source-specific infiltration factors lead to significant differences in source apportionment. The exposure apportionment method was further demonstrated based on case studies for exposure modeling domains in North Carolina along the I-40 corridor and for Harris County, Texas.

 

We introduced a modeling framework for estimating the acute effects of personal exposure to ambient air pollution in a time series design. First, a spatial hierarchical model was used to relate Census tract-level daily ambient concentrations and simulated exposures for a subset of the study period. The complete exposure time series is then imputed for risk estimation. Modeling exposure via a statistical model reduces the computational burden associated with simulating personal exposures considerably. This allows us to consider personal exposures at a finer spatial resolution to improve exposure assessment and for a longer study period. The proposed approach is applied to an analysis of fine particulate matter of less than 2.5 µm in aerodynamic diameter (PM_2.5) and daily mortality in the New York City metropolitan area during the period 2001 to 2005. Personal PM_2.5 exposures were simulated from the Stochastic Human Exposure and Dose Simulation (SHEDS). Accounting for exposure uncertainty, the authors estimated a 2.32% (95% posterior interval: 0.68, 3.94) increase in mortality per a 10 µg/m³ increase in personal exposure to PM_2.5 from outdoor sources on the previous day. The corresponding estimates per a 10 µg/m³ increase in PM_2.5 ambient concentration was 1.13% (95% confidence interval: 0.27, 2.00). The risks of mortality associated with PM_2.5 also were higher during the summer months.

 

We studied the impact of speciated PM_2.5 exposure on premature mortality, using daily mortality data at the county level nationwide for years 1999-2005, and monitoring data. Overall, a 10-μg/m3 increase in PM_2.5 at lag 1 is associated with 0.25 (95 percent interval -0.08, 0.58) in all-cause mortality. We conducted similar analysis for the PM components; the organic carbon and the silicon were the components that appeared to have a significant impact on mortality. We also evaluated the impact of different metrics to characterize exposure on the estimated risk, using air quality model (CMAQ), monitoring data, and combining CMAQ with monitoring data, by introducing a new statistical framework to fuse data. To characterize the multipollutant effect, we introduced joint models using a basis representation that allowed for nonlinear interactions between pollutants. We found that the metric for PM_2.5 can have a substantial impact on the results. For example, using monitoring data in North Carolina we found a 1.5% increase in risk for a 10 μg/m³ increase in PM_2.5 (z-score 2.7) and when using the CMA-monitoring combined surfaces the estimated increase is 3.2% (z-score 6.0). Also, the use of the general interaction between PM_2.5 and ozone shows dramatic interactions that are not observed in the linear main effects analysis.

http://www4.stat.ncsu.edu/~fuentes Exit

http://environment.yale.edu/profile/bell Exit

http://www4.ncsu.edu/~frey/ Exit

http://www.meas.ncsu.edu/faculty/zhang/zhang.html Exit