GRAPES全球三维变分同化系统的检验与诊断

中国气象局数值预报中心新近升级的GRAPES全球三维变分同化系统的大气基本状态变量在物理属性与定义的网格和坐标上与预报模式保持一致，是一个完全针对GRAPES预报模式的同化系统。该系统不仅有利于减小分析误差，也是构建GRAPES四维变分同化系统的基本环节之一。该文通过与观测资料的对比、与国际其他业务中心分析场的对比，以及中期数值预报的检验，对新的GRAPES全球三维变分同化系统性能进行较全面讨论，并通过对这一系统的检验，探索资料同化系统性能的检验方法，尤其是观测资料同化效果的定量评价方法。诊断结果表明:在宏观特征上，GRAPES变分同化系统的分析场与欧洲中期数值预报中心和美国国家环境预测中心的分析场十分相似, 但细节上仍有差别。这些差别主要源自GRAPES同化系统中探空、地面报、掩星以及飞机报观测的贡献偏大，而卫星垂直探测仪观测资料的作用尚未充分发挥。从探测单要素来讲，风及湿度观测的作用发挥不够。此外，青藏高原周围地区、模式高层及赤道地区分析场偏差较大，它们与模式地形及高层的处理等有关系，这些问题有待进一步改进。

Verification and Diagnostics for Data Assimilation System of Global GRAPES

Numerical Weather Prediction Center of China Meteorological Administration has upgraded the global GRAPES (Global/Regional Assimilation and PrEdiction System) variation data assimilation system. The new data assimilation system employs the same coordinate, grids and atmospheric state variables as those of the GRAPES model. It can reduce analysis errors due to the interpolation and variable transformations, and also provide basics for developing GRAPES four dimension variation assimilation system. Some key characteristics of the new global GRAPES data assimilation system are discussed, and then the performance is evaluated in detail, by comparing with observations, analysis or reanalysis data from advanced operational numerical weather prediction centers, and the medium range forecast from background and analysis fields and different forecast models. Some guidelines for further optimizing the system is also given based on diagnosis and quantitatively estimating the impact of observations. Results show that the GRAPES data assimilation system assimilates conventional observations, satellite radiances and radio occultation observations effectively, making analyses closer to the real atmosphere and improving the forecast skill. The analysis of GRAPES are similar to those of European Centre for Medium Range Weather Forecasts and National Center for Environmental Prediction at the large scale circulation fields. However, some differences still remains, which actually expose issues of GRAPES. These differences are related to overlarge contributions from radiosonde, surface, ships, aircraft and radio occultation observations, and the relatively weaker influence of satellite radiance observations. There is broad consensus among the global numerical weather prediction centers that these types of observations tend to be the highest ranked contributors to forecast skill: Microwave temperature sounder, hyper spectral infrared sounder, radiosondes, aircraft observations, radio occultation and atmospheric motion vectors, although not necessarily uniformly in this order. However,contributions of the microwave temperature sounder and hyper spectral infrared sounder in GRAPES are not dominant, because GRAPES still uses less radiance data, and on the other hand, the bias correction effect is not so good.Contributions of wind and humidity observation are less in GRAPES. Additionally,biases in regions of the Tibet Plateau, upper levels of the model and the tropics are relatively larger compared to observations and the reanalysis, which are related to the treatment method of topography and upper boundary of model. To gain better analysis and forecast skill, there is a requirement to place more emphasis on the above issues.