Uncertainties in Determining Targeting Observation of the Heavy Rainfall Caused by A Southwest Vortex in China in June 2008

Targeting observation is to locate an area around a certain weather system in which adding more observations may have much more benefit to the forecast accuracy than from anywhere else. Previous works on targeting observation were mainly based on single model result. However, from ensemble forecast point of view, different members with equally possible initial conditions may have different targeting areas, which makes the operational implementation of targeting observation a difficult problem. Besides, recent works also demonstrated that targeting observation may depend on the method utilized to determine the targeting area. This work is to explore the uncertainties in determining the targeting area in a southwest vortex case in China in June 2008. The targeting area of this event was determined through two different ways. One is through direct data assimilation method by exploring the impact of assimilating synthetic observation at different locations around the vortex at the initial time on the 24-h rainfall via OSSE (Observing System Simulation experiment) with WRF-3DVar. The member which performed the most close to the NCEP analysis in the ensemble was used as the truth. Synthetic wind profiler observations extracted from the truth were assimilated into one member that did not have a decent simulation in terms of the 24-h precipitation and stream field (referred to as bad members). We explored the targeting area using different bad members with different magnitudes and patterns of initial error. Results show that using a different bad member produced apparently different targeting areas. Some bad members were even not sensitive to data assimilation thus no targeting area. This result indicates large uncertainties in the locations of targeting area in the ensemble with standard observational and analysis error. This uncertainty was further confirmed when we used a different method to obtain the targeting area through calculating the correlation between the 24-h total precipitation and the initial wind and temperature. We found that the maximum correlation didn't match the targeting area determined by the OSSE method. These results show that the location of targeting area is closely associated with the uncertainties of both the initial condition and the methods used to produce the targeting area, which suggests that the effectiveness of targeting observation needs careful evaluation especially in the context of ensemble forecast.