Abstract

This paper examines the accuracy of Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) generated 72 hr fine particulate matter (PM2.5) forecasts in Delhi during the crop residue burning season of October-November 2017 with respect to assimilation of the Moderate Resolution I... Show moreThis paper examines the accuracy of Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) generated 72 hr fine particulate matter (PM2.5) forecasts in Delhi during the crop residue burning season of October-November 2017 with respect to assimilation of the Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth (AOD) retrievals, persistent fire emission assumption, and aerosol-radiation interactions. The assimilation significantly pushes the model AOD and PM2.5 toward the observations with the largest changes below 5 km altitude in the fire source regions (northeastern Pakistan, Punjab, and Haryana) as well as the receptor New Delhi. WRF-Chem forecast with MODIS AOD assimilation, aerosol-radiation feedback turned on, and real-time fire emissions reduce the mean bias by 88-195 mu g/m(3) (70-86%) with the largest improvement during the peak air pollution episode of 6-13 November 2017. Aerosol-radiation feedback contributes similar to 21%, similar to 25%, and similar to 24% to reduction in mean bias of the first, second, and third days of PM2.5 forecast. Persistence fire emission assumption is found to work really well, as the accuracy of PM2.5 forecasts driven by persistent fire emissions was only 6% lower compared to those driven by real fire emissions. Aerosol-radiation feedback extends the benefits of assimilating satellite AOD beyond PM2.5 forecasts to surface temperature forecast with a reduction in the mean bias of 0.9-1.5 degrees C (17-30%). These results demonstrate that air quality forecasting can benefit substantially from satellite AOD observations particularly in developing countries that lack resources to rapidly build dense air quality monitoring networks. Show less