论文部分内容阅读
针对2006年5月24～25日一次暴雨过程,通过一系列初值扰动试验探讨实际业务中建立集合预报系统的方法.运用45 km的WRF模式构建一个11个成员的集合预报系统来比较分析不同的扰动方案、扰动的空间结构和扰动振幅对集合预报的影响,结果表明:(1)初值扰动的空间结构对暴雨集合预报的离散度影响很关键,而扰动振幅的影响却居次要地位.具有动力学结构的孵化扰动明显优于随机扰动.(2)集合预报比单一控制预报提供了更有价值的预报信息.例如在该个例控制预报中漏报的湖北监利强降水中心,在集合预报中有20%的概率,并且实况被包含在集合预报的预报范围之中.集合平均预报也明显优于控制预报.例如矫正了在控制预报中明显虚报的鄂东北的大暴雨中心,且集合平均预报的暴雨中心落在实际观测暴雨中心的附近.(3)集合离散度较好地反映了实际降水过程的可预报性.例如应用孵化扰动,其离散度的空间结构同降水预报误差的空间分布大致对应.“,”A Mesoscale Ensemble Prediction System (MEPS) is under development at Wuhan Heavy Rain Institute of CMA aiming for heavy rain forecasting. The initial results from the experiments of perturbing atmospheric initial conditions (ICs) are reported. With a 45 km-WRF model (NCAR ARW),a 11-member ensemble was integrated into 36 h using NCEP reanalysis as control analysis. A series of experiments were performed to explore the possible path of establishing an operational MEPS by examining various IC perturbation strategies including random and breeding approaches as well as the relative impact between perturbation′s spatial structure and perturbation size. The heavy rainstorm case of May 24～25,2006 was chosen for this study because it was a bust of then operational forecast. The focus was the forecast of 24 h accumulated precipitation. In this case study,we can see the follows:(1) The spatial structure of IC perturbation played a critical and primary role in ensemble spread growth followed by the perturbation size. Dynamically well-constrained breeding perturbation performed much better than random one in all aspects of ensemble forecasting. (2)The ensemble forecasts provided more valuable information than the single deterministic control run. For example,a greater than 20% probability was predicted over an observed major heavy rainstorm center which was completely missed by the operational control forecast. Averaged over a major heavy rainstorm subregion,the observed rainfall amount fell within the envelope of ensemble forecasts but was severely underestimated by the control forecast during the most intensive precipitation period. The ensemble mean forecast is also a much improved forecast comparing to the control forecast in this case. (3)A reasonably good spread-skill relation was demonstrated in this experiment. For example,with breeding perturbation,the spatial structure of ensemble spread resembled quite well the absolute error of the ensemble mean forecast in large scale pattern although such relation was poor in quantity.