Among air pollutants, particulate matter with a diameter of 2.5 microns or less (PM2.5) is associated with some of the most serious health concerns. This is because these tiny particles can penetrate deep into the lungs and other body organs, thereby increasing the risk for many health conditions including respiratory illnesses, cardiovascular disease, cancer, and birth defects.

Various networks of ground-based sensors routinely measure the abundance of PM2.5 (μg/m3) but the spatial coverage is rather sparse because of the costs involved in operating a sensor network. Several studies have sought to overcome this limitation by using satellite-derived Aerosol Optical Depth (AOD) with regression and/or numerical models to estimate ground-level PM2.5 within the Earth’s boundary layer. Zhang et al. (2009) presented a comprehensive study for the 10 EPA regions across the United States using multi-linear regression between the PM2.5 abundance observed by the EPA and the Moderate Resolution Imaging Spectroradiometer (MODIS) AOD and a set of meteorological parameters. The best correlations of PM2.5 with AOD were observed for the eastern states in summer and fall, with EPA region 4 having a correlation coefficient of more than 0.6. The poorest correlations were observed for the southwestern states, with EPA region 9 having a correlation coefficient of approximately 0.2. Weber et al. (2010) extended the study of Zhang et al. (2009) for five EPA monitoring sites in the Baltimore/Washington DC Metro area by considering AOD from MODIS, the Multi-Angle Imaging Spectroradiometer (MISR), and the Geostationary Operational Environmental Satellite (GOES). The PM2.5 estimates of Zhang et al. (2011) and Weber et al. (2010) are made available through the Infusing satellite Data into Environmental Applications (IDEA) website (http://www.star.nesdis.noaa.gov/smcd/spb/aq/).

In an elegant study Van Donkelaar et al. (2006) presented a global estimate of the long-term average PM2.5 concentrations between 2001-2006 using both satellite observations of AOD from MODIS and a global chemical transport model to estimate η=PM2.5/AOD. The 3D chemical transport model used was GEOS-Chem. Van Donkelaar et al. (2006) found significant spatial agreement with North American PM2.5 measurements (correlation coefficient of 0.77) and with non-coincident measurements elsewhere (correlation coefficient of 0.83).

In this study we have used a proprietary machine learning approach to estimate η=PM2.5/AOD entirely from observations. We used PM2.5 observations from the United States, Europe, Africa, Australia and Asia to create a comprehensive training dataset spanning more than a decade. We then used this training dataset to estimate η as a function of the satellite AOD at multiple wavelengths and all the associated parameters that are available with the AOD (such as the angstrom exponent, scattering angle, cloud masks, surface reflectivity, and viewing geometry) and the meteorological analyses. Fifty independent trainings were performed using this training dataset, for each of these fifty trainings there was a random selection of 66% of the data for use in the training, with 34% of the data left out. The statistics shown in Table 1 and Figure 1 is the mean solution for these fifty independent trainings. Very careful attention is paid to ensure that the PM2.5 observations and satellite observations are coincident in space and time to within a great circle separation of 0.02˚ (approximately 2 km) and a time window of 30 minutes. This is done for the standard and Deep Blue retrieval algorithms of MODIS Terra and Aqua. This can be thought of as the global fully non-linear multivariate extension to the pioneering work of Zhang et al. (2009).

Van Donkelaar, A., Martin, R. V. & Park, R. J. (2006) Estimating ground-level pm2.5 using aerosol optical depth determined from satellite remote sensing. J Geophys Res, 111.

Weber, S. A., Engel-Cox, J. A., Hoff, R. M., Prados, A. I. & Zhang, H. (2010) An improved method for estimating surface fine particle concentrations using seasonally adjusted satellite aerosol optical depth. Journal of the Air & Waste Management Association, 60, 574-585.

Zhang, H., Hoff, R. M. & Engel-Cox, J. A. (2009) The relation between moderate resolution imaging spectroradiometer (MODIS) aerosol optical depth and pm2.5over the united states: A geographical comparison by u.S. Environmental protection agency regions. Journal of the Air & Waste Management Association, 59, 1358-1369.

Zhang, H., Lyapustin, A., Wang, Y., Kondragunta, S., Laszlo, I., Ciren, P. & Hoff, R. M. (2011) A multi-angle aerosol optical depth retrieval algorithm for geostationary satellite data over the united states. Atmospheric Chemistry and Physics, 11, 11977-11991.