Uncertainties in air quality models degrade the skill of their predictions. A way to improve the forecasting results is to constrain the model with observations. This can be done in multiple ways, including modifying the initial conditions of the model (generally referred to as data assimilation) or by constraining the elements forcing the model such as emissions (typically called inverse modeling). Data assimilation and inverse modeling are also useful when skillful estimates of air quality or emissions are needed for other purposes, such as performing health impact assessments or designing policies to reduce emissions.
Through our research we study multiple aspects of data assimilation and inverse modeling, including the development of new methods, assessing the impact of observational data sets not previously assimilated, and demonstrating their use in near-real-time to speed-up their transition to operations.
The figures on the top-row show forecasting results valid for the same day and time (May 25th 00 UTC) from forecasts initialized on multiple days. As the initilization is closer to the forecasted date the forecast improves it's performance, but on the last day there is a large improvement due to data assimilation of aerosol optical depth (AOD) from a sensor (GOCI) on board of a geostationary satellite.The comparisson to AOD on the bottom panels is performed to independent data not yet assimilated at the time. These forecasts were performed in the framework of the KORUS-AQ campaign (flight tracks for that day shown on the bottom-right panel), and represented the first near-real time forecasts to assimilate geostationary AOD data.