Elsevier

Atmospheric Environment

Volume 45, Issue 35, November 2011, Pages 6267-6275
Atmospheric Environment

Assessing temporally and spatially resolved PM2.5 exposures for epidemiological studies using satellite aerosol optical depth measurements

https://doi.org/10.1016/j.atmosenv.2011.08.066Get rights and content

Abstract

Land use regression (LUR) models provide good estimates of spatially resolved long-term exposures, but are poor at capturing short term exposures. Satellite-derived Aerosol Optical Depth (AOD) measurements have the potential to provide spatio-temporally resolved predictions of both long and short term exposures, but previous studies have generally showed relatively low predictive power. Our objective was to extend our previous work on day-specific calibrations of AOD data using ground PM2.5 measurements by incorporating commonly used LUR variables and meteorological variables, thus benefiting from both the spatial resolution from the LUR models and the spatio-temporal resolution from the satellite models. Later we use spatial smoothing to predict PM2.5 concentrations for day/locations with missing AOD measures. We used mixed models with random slopes for day to calibrate AOD data for 2000–2008 across New-England with monitored PM2.5 measurements. We then used a generalized additive mixed model with spatial smoothing to estimate PM2.5 in location–day pairs with missing AOD, using regional measured PM2.5, AOD values in neighboring cells, and land use. Finally, local (100 m) land use terms were used to model the difference between grid cell prediction and monitored value to capture very local traffic particles. Out-of-sample ten-fold cross-validation was used to quantify the accuracy of our predictions. For days with available AOD data we found high out-of-sample R2 (mean out-of-sample R2 = 0.830, year to year variation 0.725–0.904). For days without AOD values, our model performance was also excellent (mean out-of-sample R2 = 0.810, year to year variation 0.692–0.887). Importantly, these R2 are for daily, rather than monthly or yearly, values. Our model allows one to assess short term and long-term human exposures in order to investigate both the acute and chronic effects of ambient particles, respectively.

Highlights

► We found high out-of-sample R2 (R2 = 0.830). ► For days without AOD values, we also found high R2 (R2 = 0.810). ► Importantly, these R2 are for daily, rather than monthly or yearly, values. ► Our model allows to assess both short- and long-term human exposures concurrently.

Introduction

Particular matter (PM), or aerosols, is the general term used for a mixture of solid particles and liquid droplets found in the atmosphere emanating from either natural (dust and volcanic ash) or anthropogenic aerosols (biomass burning smoke, industrial pollution) (Kaufman et al., 2002). A large body of literature has demonstrated the association between both short- and long-term exposures to ambient PM and adverse health effects. These effects include asthma (Lin et al., 2002), cardiovascular problems (Barnett et al., 2006, Le Tertre et al., 2002, Peters et al., 2001b, Schwartz and Morris, 1995, Wellenius et al., 2005, Zanobetti et al., 2000), respiratory infections (Baccarelli, 2009, Dominici et al., 2006, Katsouyanni et al., 1996, Schwartz, 1996, Sunyer and Basagana, 2001, Zanobetti et al., 2009), lung cancer and mortality (Dockery et al., 1993, Franklin et al., 2006, Pope et al., 2002, Schwartz, 1994). This association with ambient PM remains even after matching on other air pollutants and temperature in case–crossover analyses. However, most time series studies rely on a limited number of PM2.5 monitors in their study regions, which introduces exposure error, and likely biases the effect estimates downward (Zeger et al., 2000), and are unable to produce estimates in locations without monitoring.

Cohort studies that estimate the effects of air pollution on, for example, survival using nearest monitor as the exposure metric also suffer from exposure error (Laden et al., 2006, Pope et al., 1995, Miller et al., 2007, Peters et al., 2001a). Spatial interpolation/modeling methods (Maheswaran et al., 2005) can also be problematic due to limited data availability. The error for monitor-based studies can be substantial, with, for example, some subjects in the American Cancer Society (ACS) study residing more than 100 miles away from the monitor. A study that examined the effect of restricting the analysis of the ACS cohort to only subjects residing in the same county as the sulfate monitor found that the coefficient of long-term sulfate exposure doubled compared to the original study (Willis et al., 2003). More recently, Ostro and coworkers reported results from the California Teachers Cohort, and compared results matching people to monitors within 8 km to results matching people to monitors within 30 km. The slopes for sulfate also doubled when closer exposure was used (Ostro et al., 2006).

Simple kriging methods can produce more spatially resolved exposure, but fail to account for local emission sources, such as the presence of highways or stationary sources between two monitoring stations rendering such interpolation inappropriate. Exposure metrics from such studies thus ignore intra-urban variation in exposure and may be missing data in rural areas.

Land use (LU) regression models can do much better at capturing long-term differences in exposure between locations; however, since the LU terms are generally not time varying, their temporal resolution tends to be limited, and based on the sparse PM2.5 monitoring network (Aguilera et al., 2007, Briggs et al., 2000, Gryparis et al., 2009, Liu et al., 2007, Ryan et al., 2008, Yanosky et al., 2009). Hence, they can only assess long-term exposures which are adequate for chronic health effects studies, but not acute ones. Moreover, lack of monitors in exurban and rural locations, small towns, etc means participants from such locations are either excluded, or are assigned estimates from the LU regression that may have more error. And because of the siting strategies for monitors and their modest number, certain exposure scenarios may be under or unrepresented in calibrating the land use regression.

Due to its large spatial coverage and reliable repeated measurements, satellite remote sensing, provides another important tool for monitoring aerosols, particularly for areas and exposure scenarios where surface PM2.5 monitors are not available (Engel-Cox et al., 2004, Gupta et al., 2006, Koelemeijer et al., 2006, Liu et al., 2004). One important and common aerosol parameter retrieved from satellite sensors is the aerosol optical depth (AOD), which measures the light extinction by aerosol scattering and absorption in the atmospheric column. AOD is a function of the aerosol mass concentration, mass extinction efficiency, hygroscopic growth factor (a function of relative humidity), and effective scale height that is mainly determined by the vertical distribution of aerosols (Kaufman and Fraser, 1983). In principle, such data can yield an estimate of PM2.5 concentrations at any location.

Multiple studies published over the last decade have established quantitative relationships between satellite-derived AOD and PM2.5. These studies used various statistical methods from linear regression models (Chu et al., 2002, Hutchison, 2003, Schäfer et al., 2008) to more complex mixed models and generalized additive mixed models (GAM) (Liu et al., 2009, Paciorek et al., 2008). However, these generally report low to moderate predictive power, or lack detailed high resolution predictions across large space-time domains.

In a previous paper, we showed that day-specific calibrations of AOD data using ground PM2.5 measurements from a spatial monitoring network greatly improves the ability of AOD values to predict PM2.5 across Massachusetts (Lee et al., 2011).

An important limitation of using AOD data is that grid-specific values are missing on some days, mostly due to cloud or snow cover. In addition, AOD produces predictions for grid cells- not addresses, which may be sufficient for small area health studies, but results in less accuracy when individual addresses are available. In this paper we extend our previous work to incorporate LU regression and meteorological variables to predict PM2.5 concentrations for days when AOD measures are not available, and to provide the address specific predictions of LU regression. Specifically, we developed and validated models to predict daily PM2.5 at a 10 × 10 km resolution and at local addresses across New-England region for the years 2000–2008.

Section snippets

Study domain

The spatial domain of our study included the New-England region comprising the states of Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island and Vermont (Fig. 1).

AOD data

The Moderate Resolution Imaging Spectroradiometer (MODIS) is a relatively new instrument aboard the Earth Observing System (EOS) satellites (King et al., 1992, Salomonson et al., 1989). The MODIS sensors are located on polar orbiting and sun-synchronous Terra and Aqua satellites. The Terra satellite was launched on the

Results

Fig. 4 presents a scatter plot of the AOD–PM2.5 relationship before (Fig. 4a) and after (Fig. 4b) the stage 1 calibration showing the significant fit improvement gained by calibrating with our stage 1 model.

Fig. 5 presents a density plot exhibiting the daily variation of AOD slopes between 2000 and 2008 during the stage 1 calibrations, and shows there is considerable day to day variability in slope.

The first stage models all had very high out-of-sample fits for each year and the entire study

Discussion

In this paper we examined the relationship between PM2.5 ground measurements and MODIS AOD data in New-England during the period 2000–2008. One key finding of this study is that our novel prediction models perform significantly better than previous prediction models which assumed that the relationship between PM2.5 and MODIS AOD data remains constant over time, and better than LU regression alone. Another major key feature of our combination of LU regression with AOD data is the ability to

Conclusion

In summary, we have clearly demonstrated how AOD can be used reliably to predict daily PM2.5 mass concentrations. By including LU terms and later spatial smoothing our models perform much better than previous AOD–PM2.5 models. In addition, our model allows one to assess short term and long-term human exposures in order to investigate both the acute and effects of ambient particles, respectively.

Acknowledgments

Supported by the Harvard Environmental Protection Agency (EPA) Center Grant RD83479801, NIH grant ES012044 and the Environment and Health Fund (EHF) Israel. The authors also want to thank Dr. William L Ridgway, Science Systems and Applications, Inc. Climate and Radiation Branch, Code 613.2, NASA Goddard Space Flight Center, Greenbelt, MD 20771 and Steven J. Melly, department of environmental health, Harvard school of public health, Harvard University.

References (53)

  • A. Baccarelli

    Breathe deeply into your genes!: genetic variants and air pollution effects

    American Journal of Respiratory and Critical Care Medicine

    (2009)
  • A.G. Barnett et al.

    The effects of air pollution on hospitalizations for cardiovascular disease in elderly people in Australian and New Zealand cities

    Environmental Health Perspectives

    (2006)
  • D. Chu et al.

    Validation of MODIS aerosol optical depth retrieval over land

    Geophysical Research Letters

    (2002)
  • D. Dockery et al.

    An association between air pollution and mortality in six US cities

    The New England Journal of Medicine

    (1993)
  • F. Dominici et al.

    Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases

    JAMA

    (2006)
  • EPA

    USEPA National Emissions Inventory

    (2010)
  • M. Franklin et al.

    Association between PM2.5 and all-cause and specific-cause mortality in 27 US communities

    Journal of Exposure Science and Environmental Epidemiology

    (2006)
  • A. Gryparis et al.

    Measurement error caused by spatial misalignment in environmental epidemiology

    Biostatistics (Oxford, England)

    (2009)
  • C. Homer et al.

    Development of a 2001 national landcover database for the United States

    Photogrammetric Engineering and Remote Sensing

    (2004)
  • K. Katsouyanni et al.

    Short term effects of air pollution on health: a European approach using epidemiologic time series data: the APHEA protocol

    Journal of Epidemiology and Community Health

    (1996)
  • Y. Kaufman et al.

    Light extinction by aerosols during summer air pollution

    Journal of Applied Meteorology

    (1983)
  • Y. Kaufman et al.

    A satellite view of aerosols in the climate system

    Nature

    (2002)
  • M.D. King et al.

    Remote sensing of cloud, aerosol, and water vapor properties from the Moderate Resolution Imaging Spectrometer (MODIS)

    IEEE Transactions on Geoscience and Remote Sensing

    (1992)
  • F. Laden et al.

    Reduction in fine particulate air pollution and mortality: extended follow-up of the Harvard Six Cities study

    American Journal of Respiratory and Critical Care Medicine

    (2006)
  • A. Le Tertre et al.

    Short-term effects of particulate air pollution on cardiovascular diseases in eight European cities

    Journal of Epidemiology and Community Health

    (2002)
  • H. Lee et al.

    A novel calibration approach of MODIS AOD data to predict PM2.5 concentrations

    Atmospheric Chemistry and Physics Discussions

    (2011)
  • Cited by (322)

    View all citing articles on Scopus
    View full text