Developing High Resolution Particulate Matter Surfaces for use in Community Health Assessments

Urban decision makers have growing needs for high spatial resolution data on urban air particulate matter concentrations. Data are needed to better quantify existing health burdens at the neighborhood scale, to identify and prioritize exposure reduction strategies for pollution hot spots, to track progress in achieving air quality-related health improvement goals, and, increasingly, to assess health co-benefits of longer- term carbon mitigation strategies. Newly available retrievals from NASA MODIS satellite remote sensing provide opportunities to construct high resolution PM_2.5 spatial fields for intra-urban public health assessments, as well as serve as a launching pad for further downscaling using available and emerging low-cost sensors in conjunction with land use regression. The objective of this study was to construct gridded PM_2.5 spatial fields for New York City based on 1x1 km MAIAC satellite-based aerosol optical depth retrievals, and to further downscale 100x100 m using a hi-density urban network of in-situ sensor and land use data.

We first used a random forest algorithm to predict PM_2.5 concentrations at 21 regulatory monitoring stations in and around New York City in 2015 based on a set of potential explanatory variables that included two satellite AOD products (AQUA and TERRA), meteorological data, vegetation density, land use density, elevation, and traffic density. We then augmented the model using data from the New York City Community Air Survey (NYCCAS), a network of street-levels monitors measuring 2-week average PM_2.5 every quarter at appx 100 sites. We compared the PM_2.5 output data generated by these two models, as well as with PM_2.5 estimates produced by the Global Burden of Disease Study. We then combined the PM data with baseline disease and mortality rates to estimate health impacts on a 100 m grid over NYC.

Predictions R² in the range of 0.73-0.85 were achieved for PM_2.5 in NYC. Predicted PM_2.5 based on the full model are depicted graphically in the attached figure. Preliminary results suggest that it is possible to develop very high-resolution estimates of ground-level PM2.5 using a combination of remote sensing, in-situ sensors, and land use variables, revealing important features that are relevant to source identification, human exposure assessment, and health impact assessment.