The Implementation Of A Regional Multi-model Ensemble Forecast System And Its Preliminary Applications

The summer storm in southern China due to complex terrains, limited places and long duration, has always been a difficult forecasting issue. On the other hand, success of ensemble forecast in the field of global numerical weather prediction models propel that mesoscale & small-scale short-term ensemble forecast research to acquire a breakthrough. Therefore, this paper based on the results of previous studies, using multi-model ensemble forecasting system to simulate southern China heavy rain, analyses and discusses model differences or the differences between physical processes and issues such as the need for mesoscale ensemble forecast.A mesoscale multi-model ensemble forecasting system, with the characteristics of multi-mode, multi-initial values and multi-physical processes, is established based on ARPS and WRF, moreover, the program of dual BGM disturbance and different parameterized programs of physical processes are simultaneously utilized to make an analysis of a heavy rainfall in south China on June 19-20,2010.The heavy rainfall consisted of two rain belts distributed as "V", corresponding to the two rainfall periods. However, it was caused by the combined effect of 500hPa trough, the subtropical high,850hPa shear line, and the mesoscale vortex. At the same time, some important physical variables were very conducive to the formation and development of heavy rain. The evolution of these physical variables and precipitation has a good relationship. The convergence/divergence and the vertical circulation, caused by high & low level jet stream configuration, were conducive to storm formation and development.As the numerical simulation shows, the result reveals that the multi-model ensemble forecasting system could well simulate this precipitation process, which is highly sensitive to the initial value and physical processes.compared with a single deterministic prediction, it has apparent advantages of accurate and efficiency forecasting, at the moment, if this forecasting system is accepted. It also gets strong rainfall information 24-36 hours earlier than control system.According to the model and the cumulus convective parameterization schemes which members belong to, the multi-model ensemble forecasting were divided to sub-ensemble forecast to analyze. In addition, this kind of ensemble prediction is better than that based on a single model or a parameterized program of a single physical process. From the result of this process,it is obviously obtained that parameterized programs of physical processes have more opportunities to improve than models.Finally, members were divided to be conducted a cluster analysis according to their forecast results quality. In terms of storm rainfall prediction, physical processes and initial values are more sensitive to the disturbance of the low-level than that of high-level, and this may be one of the most important reasons why the forecasting results of some members are good while others are not.