Development of a high-resolution (400 m) operational air quality early warning system for Delhi, India through integrated chemical data assimilation

Managing air quality levels in the National Capital Region (NCR), especially Delhi, India has emerged as a complicated task. It is now a matter of top priority to develop meaningful policy options. Short-term air quality forecasts can provide timely information about forthcoming air pollution episodes that the decision-makers can use to implement temporary emission control measures and reduce public exposure to extreme air pollution events. Although India has developed air quality forecasting systems for NCR, it was challenging to predict acute air pollution episodes during which hourly PM_2.5 concentrations exceed 300 µg/m³. In this perspective, a very high-resolution (400 m) operational air quality prediction system has been developed to predict extreme air pollution events in Delhi and issue timely warnings. This modeling framework consists of a high-resolution fully coupled state-of-the-science Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) and three-dimensional Variational (3DVAR) framework of the community Gridpoint Statistical Interpolation (GSI) system. The system assimilates satellite aerosol optical depth (AOD) retrievals at 10 km resolution, real-time crop residue burring at 1km resolution, surface PM_2.5 data from 43 air quality monitoring stations, and uses high-resolution dynamical emissions (400 m) from various anthropogenic sources. The chemical data assimilation is further integrated with dynamical downscaling to obtain improved chemical conditions for the 400 m resolution domain. This paper summarizes the performance of the model forecasts for the winter season 2019-2020 and the evaluation of the model against the observations. Here, we demonstrate that the assimilation of chemical data in a coupled weather-air quality model improved the overall accuracy of PM_2.5  forecasts in New Delhi by about 70 % during the winter season 2019-2020. Results show that the skill score for the poor (AQI 200-300), very-poor (AQI 300-400) and sever pollution (AQI 400-500) days is relatively promising for the hit rate with a value of 0.74 for (very-poor). This indicates that the model has reasonable predictive accuracy for air quality events. False Alarm rate (0.19), missing rate (0.32) are low, and the probability of detection is relatively high (0.67), indicating that the performance of the real-time forecast is better for both very poor events and no-very poor events.