Statistical downscaling of coarse-resolution fine particulate matter predictions over the contiguous United States: model development, evaluation, and implication in health impact assessment.

Fine particulate matter (PM2.5) predictions at a high spatial resolution (i.e., neighborhood scale) are critically needed to better understand the health impacts of air pollution, especially at neighborhood scales. This work develops a statistical downscaling approach to predict PM2.5 at a 1-km grid resolution over the contiguous United States (CONUS) under baseline and future energy transition scenarios and estimate health benefits utilizing the Environmental Benefits Mapping and Analysis Program (BenMAP). To this end, we incorporate the satellite-based high-resolution aerosol optical depth (AOD), land use data, and PM2.5 composition predicted by the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) at 36-km into daily multi-linear regressions over different climate regions of the CONUS. Compared to the WRF-Chem baseline predictions in 2008-2012, 1-km PM2.5 estimates enhance the accuracy by increasing the yearly correlation coefficients from ~0.4 to ~0.8 and reducing normalized mean errors from ~47 % to ~27 %. Future 1-km PM2.5 is projected by combining the baseline 5-yr (2008-2012) monthly-averaged training coefficients with high-resolution statistically improved projected AOD and PM2.5 subsets from WRF-Chem. BenMAP with WRF-Chem predictions under future energy scenarios shows an average of 2478 fewer deaths per year in 2050 in New York City and Boston due to PM2.5, while the downscaled PM2.5 shows less PM2.5 reduction and about half the health benefit of the WRF-Chem projections. The downscaling approach is more computationally efficient than running the 3-D air quality model with a 1-km spatial grid resolution. This work uniquely combines WRF-Chem outputs and statistical downscaling to provide high-resolution and high-fidelity PM2.5 predictions.