Elsevier

Environmental Modelling & Software

Volume 109, November 2018, Pages 182-190
Environmental Modelling & Software

Application and evaluation of two model fusion approaches to obtain ambient air pollutant concentrations at a fine spatial resolution (250m) in Atlanta

https://doi.org/10.1016/j.envsoft.2018.06.008Get rights and content

Highlights

  • Novel model fusion methods that estimate 24-hr PM2.5and 1-hr max NOx and CO at a 250m resolution are developed and applied.

  • Estimates show steep spatial gradients near roadways while retaining comprehensive emissions and secondary formation.

  • Modeled PM2.5, NOx, and CO results in Atlanta show high temporal and spatial correlations with available observations.

  • Methods can be applied to other locations, pollutants, and model inputs.

Abstract

Epidemiologic studies rely on accurately characterizing spatiotemporal variation in air pollutant concentrations. This work presents two model fusion approaches that use publicly available chemical transport simulations, dispersion model simulations, and observations to estimate air pollutant concentrations at a neighborhood-level spatial resolution while incorporating comprehensive chemistry and emissions sources. The first method is additive and the alternative method is multiplicative. These approaches are applied to Atlanta, GA at a 250 m grid resolution to obtain daily 24-hr averaged PM2.5 and 1-hr max CO and NOx concentrations during the years 2003–2008 for use in health studies. The modeled concentrations provide comprehensive estimates with steep spatial gradients near roadways, secondary formation and loss, and effects of regional sources that can influence daily variation in ambient pollutant concentrations. Results show high temporal and spatial correlation and low biases across monitors, providing accurate pollutant concentration estimates for epidemiologic analyses.

Introduction

Air pollution has been linked to adverse health effects, including cardiorespiratory morbidity and mortality (Brunekreef and Holgate, 2002; Delfino et al., 2005; Pope et al., 2009) and adverse birth outcomes (Chang et al., 2012; Darrow et al., 2011). A limitation of many population-based health studies is the inability to estimate steep spatial gradients in intraurban pollutant exposures driven by local emission sources, like vehicle traffic on roadways. Simulation studies show that an inability to accurately capture spatial variability in heterogeneous pollutants can bias risk ratio estimates in epidemiologic studies (Goldman et al., 2011), and errors in exposure misclassification due to spatial variability assumptions are especially important for epidemiologic studies of long-term pollutant exposures (Wilson et al., 2005).

Reliance on air quality observational data for exposure analyses can introduce exposure measurement error because the lack of a dense monitoring network can lead to unobserved spatial variation in pollutant concentrations (Gryparis et al., 2009; Sarnat et al., 2010; Wilson et al., 2005). Air quality modeling addresses this problem by creating spatially and temporally resolved concentration fields constructed from simulating emissions, chemistry, physics, and meteorology impacting pollutants. However, specific model results are often limited by either spatial resolution or an inability to capture complex chemistry and a vast array of emissions sources. Two types of emissions-based models are commonly used: chemical transport models and dispersion models (e.g., Gaussian plume and variants). Eulerian grid-based chemical transport models simulate the transport and chemical transformation of pollutants emitted from thousands of emissions sources over a large spatial domain. However, because all emissions are evenly distributed within one computational grid (often 36 km2 but seldom below 1 km2), these models do not simulate local effects of individual sources, which can result in steep spatial gradients that occur on scales < 1 km. Vehicles on roadways are an example of a local source that can drive steep spatial gradients in pollutants (Weijers et al., 2004). 59.5 million people lived within 500 m of heavily trafficked roads in 2010, and PM2.5 mass and component concentrations can double close to road sources (Beevers et al., 2013; Rowangould, 2013; Zhu et al., 2002). These high concentrations on roadways are not discernable using low resolution grids, leading to the need for additional exposure assessment methods to accurately characterize pollutant gradients in urban areas. Dispersion models can capture these gradients by using plume, puff, or particle representations but often do not take into account non-linear chemical reactions that contribute to the formation of major pollutants, like PM2.5. They are also not used to derive daily estimates over large spatial domains (1000s of km's) due to model parameter limitations. Additionally, all types of models are subject to biases from model parameters and inputs. Therefore, neither chemical transport models nor dispersion models alone can estimate temporally and spatially resolved air pollutant concentrations with comprehensive emissions precursor information and chemistry. Reducing exposure misclassification by improving spatial and temporal resolution of air pollutant concentration estimates, reducing model biases, including emissions from all sources, and simulating chemistry in best-estimate simulations of concentrations is critical to minimizing error in epidemiologic studies.

Different methods have been utilized to reduce error and improve spatial resolution of air pollutant concentration estimates at unmonitored locations. Land-use regression (LUR) variable models have proven effective tools for fine-scale modeling by incorporating landscape characteristics, such as elevation and distance to roadway, with data from monitors and/or dispersion models (Marshall et al., 2008; Michanowicz et al., 2016). However, LUR models are specific to one study area with particular land-use characteristics. Other modeling approaches utilize various techniques to combine observations and/or different model outputs to represent intraurban air pollution. One method uses linear combinations of wavelet basis functions to blend data from monitors (Crooks and Isakov, 2013), the photochemical model CMAQ (Community Multiscale Air Quality model) (Byun and Schere, 2006), and the plume dispersion model AERMOD (AMS/EPA Regulatory Model) (Cimorelli et al., 2005) while another method nests the local dispersion model ADMS-Urban (McHugh et al., 1997) in the regional photochemical model CAMx (Comprehensive Air Quality Model with Extensions) (ENVIRON International Corporation, 2014; Stocker et al., 2014). “Hybrid” methods add fine-scale dispersion model outputs to broader-scale estimates of pollutant concentrations, often referred to as background concentrations. Previous hybrid studies have used observations from central monitoring stations after subtracting out concentrations due to local emissions for estimates of urban background (Stein et al., 2007). Other work utilized chemical transport models, like CMAQ, run without local emissions to determine urban background (Dionisio et al., 2013; Stein et al., 2007). An advanced method for calculating urban background was developed using space-time ordinary kriging to combine monitoring data, CMAQ, and CMAQ with zeroed out emissions (Arunachalam et al., 2014). These background estimates are added to model outputs from dispersion models like AERMOD and RLINE (Research LINE model) (Snyder et al., 2013) and/or Lagrangian models like HYSPLIT (Hybrid Single Particle Lagrangian Integrated Trajectory model) that characterize primary concentrations from stationary and roadway sources (Chang et al., 2017; Dionisio et al., 2013; Stein et al., 2007). Care must be taken with these hybrid methods to prevent double-counting of emissions in the dispersion models and background calculations. Here, a method is developed for fusing multi-model and observational data to estimate total pollutant concentrations from local and regional sources at a fine spatial resolution without needing to estimate the urban background a priori with a separate chemical transport model run without direct emissions of the pollutants of interest. Removing the need to simulate background concentrations using multiple photochemical air quality model runs but instead calculating urban background empirically can save computational time and prevent the methods from missing the chemistry and secondary formation or loss associated with local sources that chemical transport models cannot simulate if those emissions are zeroed out. We present the development of two novel, computationally efficient model fusion approaches to estimate air pollutant concentrations at a fine spatial resolution with comprehensive emissions and chemistry and their applications to the Atlanta, GA region.

Each model fusion method developed in this work uses mathematical combinations of outputs from a chemical transport model that provides chemistry and local and regional sources and a dispersion model that provides fine spatial resolution simulations of inert pollutants from a local source, along with limited observations as available. One method is an additive approach and the other is a multiplicative approach. Both methods are applied to Atlanta, GA using simulations from the chemical transport model CMAQ and RLINE. RLINE is a steady-state dispersion model that simulates near surface releases of primary and chemically inert pollutants from line-sources, like vehicles on roadways (Snyder et al., 2013), and with 1.6 billion vehicles worldwide, the association between mobile source emissions and disease development is of particular concern (HEI Panel on the Health Effects of Traffic-Related Air Pollution, 2010). Emissions from road transportation are estimated to be the largest source contribution to premature deaths due to PM2.5 pollution in the United States (Caiazzo et al., 2013), and previous studies have shown that resolving steep spatial gradients in pollutant concentration near roadways is beneficial to epidemiologic studies (Chang et al., 2017), which is why RLINE was chosen for the applications of these model fusion approaches. Results of the model fusion applications include a time series of daily concentration estimates of 1-hr CO and NOx and 24-average PM2.5 at a 250 m grid resolution across Atlanta, GA during the years 2003–2008. Although the applications of these methods focus on vehicles as local sources, the methods can be extended to other local, non-roadway facilities. Overall, the inclusion of comprehensive regional and local sources and chemistry with fine-scale spatial gradients can provide comprehensive estimates of pollutant concentrations, reducing biases in exposure estimates used for epidemiologic analyses. Results are currently being used in spatiotemporal epidemiologic analyses of birth outcomes associated with air pollution and city planning and environmental justice studies investigating the relationships between air pollution, health, socio-economic factors, and infrastructure characteristics (Davis et al., 2017).

Section snippets

Methods

This work describes the development of two novel methods that fuse data from multiple models and observations to create comprehensive estimates of air pollutant concentrations at a fine spatial resolution. The two model fusion approaches combine results from a chemical transport model (CTM) and dispersion model (DISP) to obtain pollutant (PM2.5, NOx, and CO) concentration estimates. The model fusion approaches are applicable to pollutants whose small scale spatial variation is captured by the

Results

The additive and multiplicative model fusion methods were applied to obtain daily estimates of 24-hr average PM2.5 and 1-hr maximum NOx and CO for Atlanta, GA during the years 2003–2008. The general mathematical equations for the model fusion methods (eq. (1) and eq. (2)) can be rewritten to apply to the specific inputs used in this paper and are expressed in equations (3), (4).PM2.5(x,d)=[(OBSCMAQ(l,d)OBSRLINE¯coarse(l,y))interpolated]+OBSRLINE(x,y)Gas(x,d)=[(OBSCMAQ(l,d)OBSRLINE¯coarse(l

Discussion

Capturing steep spatial gradients in pollutant concentrations near roadways is critical for exposure assessment in epidemiologic studies. Monitoring stations are too sparse to accurately capture spatial gradients in pollutant concentration and can miss very high concentrations near roadways. Some air quality models can estimate fine spatial resolutions (dispersion models) but do not simulate atmospheric chemistry processes or incorporate regional emissions sources. Other models, like chemical

Acknowledgements

This publication was developed under Assistance Agreement No. EPA834799 awarded by the U.S. Environmental Protection Agency to Emory University and Georgia Institute of Technology. It has not been formally reviewed by EPA. The views expressed in this document are solely those of the authors and do not necessarily reflect those of the Agency. EPA does not endorse any products or commercial services mentioned in this publication. This paper was also made possible by the ARCS Foundation.

References (46)

  • J.G. Wilson et al.

    A review of intraurban variations in particulate air pollution: implications for epidemiological research

    Atmos. Environ.

    (2005)
  • H. Yu et al.

    Cross-comparison and evaluation of air pollution field estimation methods

    Atmos. Environ.

    (2018)
  • X.X. Zhai et al.

    Spatial and temporal source apportionment of PM2.5 in Georgia, 2002 to 2013

    Atmos. Environ.

    (2017)
  • X.X. Zhai et al.

    Calibrating R-LINE model results with observational data to develop annual mobile source air pollutant fields at find spatial resolution: application in Atlanta

    Atmos. Environ.

    (2016)
  • S. Arunachalam et al.

    A method for estimating urban background concentrations in support of hybrid air pollution modeling for environmental health studies

    Int. J. Environ. Res. Publ. Health

    (2014)
  • S.D. Beevers et al.

    Air pollution dispersion models for human exposure predictions in London

    J. Expo. Sci. Environ. Epidemiol.

    (2013)
  • D. Byun et al.

    Review of the governing equations, computational algorithms, and other components of the models-3 Community Multiscale Air Quality (CMAQ) modeling system

    Appl. Mech. Rev.

    (2006)
  • H.H. Chang et al.

    Time-to-Event analysis of fine particle air pollution and preterm birth: results from North Carolina, 2001-2005

    Am. J. Epidemiol.

    (2012)
  • S.Y. Chang et al.

    Finely resolved on-road PM2.5 and estimated premature mortality in central North Carolina

    Risk Anal.

    (2017)
  • A.J. Cimorelli et al.

    AERMOD: a dispersion model for industrial source applications. Part I: general model formulation and boundary layer characterization

    J. Appl. Meteorol.

    (2005)
  • X.C. Cong et al.

    On-road measurements of pollutant concentration profiles inside Yangkou tunnel, Qingdao, China

    Environ. Geochem. Health

    (2016)
  • J. Crooks et al.

    A wavelet-based approach to blending observations with deterministic computer models to resolve the intraurban air pollution field A wavelet-based approach to blending observations with deterministic computer models to resolve the intraurbanair pollution field

    J. Air Waste Manag. Assoc.

    (2013)
  • D. D'Onofrio

    Atlanta Roadside Emissions Exposure Study - Methodology & Project Overview

    (2016)
  • Cited by (0)

    View full text