Ponente
Descripción
A data assimilation method combined with machine learning has been developed and applied to adjust anthropogenic emissions and improve forecasting accuracy in Chinese unified atmospheric chemistry environment (CUACE) model. This is an attempt to combine data assimilation and machine learning. Nudging method was used to create the database of nudging gain matrixes, using simulations of CUACE and the ground-based observations. Then these data are employed to train a machine learning model using extremely random trees method (ExRT), and to store the relations between nudging gain matrixes and simulations in the trees. Observations was used again to find out the proper nudging gain matrixes using the trained machine learning model.The Nudging-ExRT emission inventories was calculated by the original emission inventories and the Nudging-ExRT gain matrixes. During its application for the operational CUACE five days forecasts in China in March 2021, the data assimilation of anthropogenic emissions had a good performance in most of the periods for PM2.5 and O3, and the optimization did not weaken over time. As for PM2.5, the hourly average spatial correlation coefficient (R) increased from 0.39 to 0.45 and the root mean square error (RMSE) decreased from 43.59 µg/m3 to 40.71 µg/m3. As for O3, R increased from 0.08 to 0.11, RMSE decreased from 63.66 µg/m3 to 56.70 µg/m3. This simplicity, efficiently and extensibility framework of Nudging-ExRT method has been proved to be a good way to adjust anthropogenic emissions in CUACE and still remains much to be done in the future.
