Multiscale Assimilation of Radar and AWS Data for Typhoon Haikui's Record-Breaking Landfalling Precipitation Forecast

As the low-pressure trough associated with the remnants of Typhoon Haikui interacted with the southwesterly monsoon, the Pearl River Delta witnessed a record-breaking extreme precipitation event, leading to four fatalities and dozens injuries. With continuous efforts to develop state-of-art modeling frameworks and to deploy ground-based instruments in recent years, especially the high-density automatic weather stations (AWS) and radar data, opportunities for an improved prediction of the landfalling TCs have emerged.

In our study of the event, we assimilated radar data and AWS surface data using a multiscale hybrid ensemble-variational (EnVar) assimilation approach. First, the multistep strategy is employed to assimilate AWS surface data and radar data in different steps. Then, to better represent multiscale uncertainties in the background error covariance (BEC), we also explored how to introduce multiscale ensemble BEC in the hybrid EnVar method. Two strategies of generating multiscale ensemble BEC are designed, using multi-resolution ensemble BEC or filtering ensemble BEC into different scales at different assimilation steps of AWS surface data and radar data.

Compared to using the same ensemble BEC in all steps, both strategies properly obtained increments at different scales from AWS and radar data, and thus improved forecast skills of heavy rainfall. Filtering ensemble BEC into different scales in different steps demonstrated better performance, with the highest forecast skills of accumulated precipitation and lowest forecast errors of surface variables. Further diagnosis revealed that the improvement of precipitation forecast skill was attributable to the stronger cold pools and gust winds near the surface as well as deeper saturated water vapor layers within the convections. More details will be shown in the conference.