Data Assimilation Research Team Events / Media

The torrential rains result from the short duration extreme rainfall system is of most critical for the disaster prevention. However, the limited predictability is the essence of the short duration extreme rainfall system due to the multi-scale interaction, fast evolution and strong nonlinearity. The assimilation of the radar observation with rapid, continuous update cycle is a key to level up the predictability of such a system.

The continuous rapid update cycle is able to capture and keep convective-scale structure and avoid the model spin-up problems. However, many challenges were faced in the continuous update cycle data assimilation. For example, the limited-area model systems in general suffer a deficiency to effectively represent the large-scale features and are unavoidable to experience the obvious large-scale forecast errors. In particular, the domain size is restricted due to the compromise of increasing model resolution and limited computer resources. Furthermore, the model errors are ease to accumulate over the sparse observation area, especially as the data assimilation system configured as a continuous cycle mode.

In this study, a multi-scale blending scheme using a low-pass spatial filter (Hsiao et al. 2015) was applied to a continuous cyclic radar data assimilation system. The blending scheme combines the global model analysis and the convective scale model forecast. It is expected the blended field takes the advantage from the global large scale environment and the convective scale perturbations. The scheme was applied to the hourly updated 3DVAR based radar data assimilation system. In addition, it also applied to re-center the ensemble mean of the cyclic LETKF radar data assimilation system. Case studies show that the blending scheme is able to correct the bias of the large scale monsoon flow from the global model and keep the convective rainfall structure from the convective scale radar data assimilation system. The results also show that the performance of quantitative precipitation forecasts from both the 3DVAR and LETKF radar data assimilation system improved significantly as applying the blending scheme. The more detailed sensitivity on the blending scheme also discussed in this study.

Keywords: blending scheme, radar data assimilation, continuous cycling