Chemical Data Assimilation to Improve Short-Term PM2.5 Predictions over the United States

State and local air quality forecasters across the United States use air quality forecasts from the National Air Quality Forecasting Capability (NAQFC) from the National Oceanic and Atmospheric Administration (NOAA) as one of the key tools to protect the public from adverse air pollution, by distributing timely information about air pollution episodes. This project funded by the National Aeronautics and Space Administration (NASA) aims to enhance the decision-making process by improving the accuracy of NAQFC short-term predictions of ground-level particulate matter of less than 2.5 µm in diameter (PM 2.5 ) by exploiting NASA Earth Science Data with chemical data assimilation. The NAQFC is based on the Community Multiscale Air Quality (CMAQ) model. To improve the initialization of PM 2.5 in CMAQ, we developed a new capability in the community Gridpoint Statistical Interpolation (GSI) system to assimilate Terra/Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth (AOD) retrievals in CMAQ. Specifically, we developed new capabilities within GSI to read/write CMAQ data, a forward operator that calculates AOD at 550 nm from CMAQ aerosol chemical composition, and an adjoint of the forward operator that translates the changes in AOD to aerosol chemical composition. A generalized background error covariance program called “GEN_BE” has been extended to calculate background error covariance using CMAQ output. The background error variances are generated using a combination of both emissions and meteorological perturbations to better capture sources of uncertainties in PM 2.5 simulations. The newly developed CMAQ-GSI system is used to perform daily 24-h PM 2.5 forecasts with and without data assimilation from 15 July to 14 August 2014, and the resulting forecasts are compared against AirNOW PM 2.5 measurements at 550 stations across the U. S. We find that the assimilation of MODIS AOD retrievals improves initialization of the CMAQ model, resulting in an
improved correlation and reduced bias. However, we notice a large bias in nighttime PM 2.5 simulations which is primarily associated with a very shallow boundary layer in the model. The developments and results will be discussed in detail during the presentation.