Assimilating AOD retrievals from GOCI and VIIRS to forecast surface PM2.5 episodes over Eastern China

In this study, Geostationary Ocean Color Imager (GOCI) AOD and Visible Infrared Imaging Radiometer Suite (VIIRS) AOD data were assimilated to forecast surface PM2.5 concentrations over Eastern China, by using the three-dimensional variational (3DAVR) data assimilation (DA) system, to compare DA impacts by assimilating AOD retrievals from these two types of satellites. Three experiments were conducted, including a CONTROL without the AOD assimilation, and GOCIDA and VIIRSDA with the assimilation of AOD retrievals from GOCI and VIIRS, respectively. By utilizing the Weather Research and Forecasting with Chemistry (WRF/Chem) model, 48h forecasts were initialized at each 06 UTC from 19 November to 06 December 2013. These forecasts were evaluated with 248 ground-based measurements from the air quality monitoring network across 67 China cities. The results show that overall the CONTROL underestimated surface PM2.5 concentrations, especially over Jing-Jin-Ji (JJJ) region and Yangtze River Delta (YRD) region. Both the GOCIDA and VIIRSDA produced higher surface PM2.5 concentrations mainly over Eastern China, which fits well with the PM2.5 measurements at these eastern sites, with more than 8% error reductions (ER). Moreover, compared to CONTROL, GOCIDA reduced 14.0% and 6.4% error on JJJ region and YRD region, respectively, while VIIRSDA reduced respectively 2.0% and 13.4% error over the corresponding areas. During the heavy polluted period, VIIRSDA improved all sites within YRD region, and GOCIDA enhanced 84% sites. Meanwhile, GOCIDA improved 84% sites on JJJ region, while VIIRSDA did not affect that region. These geographic distinctions might result from spatial dissimilarity between GOCI AOD and VIIRS AOD at time intervals. Moreover, the larger increment produced by AOD DA under stable meteorological conditions could lead to a longer duration (e.g., 1-2 days, > 2 days) of AOD DA impacts. Even though with AOD DA, surface PM2.5 concentrations were still underestimated clearly over heavy polluted periods. And 3% sites performed worse, where low PM2.5 values were observed and CONTROL performed well. With this study, the results indicate that AOD DA can partially improve the accuracy of PM2.5 forecastS. And the obvious geographic differences on forecasts emphasize the potential and importance of combining AOD retrievals from GOO and VIIRS into data assimilation.