中文摘要：

观测资料的质量控制直接影响数值预报资料同化的分析质量。本文针对GRAPES区域同化和预报的三维变分资料同化系统,发展了基于观测误差为“高斯分布＋均匀分布”模型的变分质量控制方案,讨论了该方案的初始启动和关键参数,并检验分析了其适用性与有效性。同时,以全球预报系统（GFS）资料作为背景场,利用探空、地面、船舶、飞机、云迹风等常规观测资料和COSMIC卫星反演资料进行同化和预报,分析了华南地区特大暴雨的个例试验和2013年8月共31 d的批量试验。试验结果表明：变分质量控制能够依据观测资料的不同质量对观测权重进行合理调整,对位势高度、气压、风、比湿的分析增量场和分析场改善显著,尤其在强降水区具有更加明显的效果;对降水落区、降水强度及中心位置的预报质量具有较好的提高,特别对暴雨、大暴雨等较大降水量级的预报能力反映出更好的改善效果,充分显示了变分质量控制在中小尺度剧烈天气过程中对同化分析和预报的重要作用。

英文摘要：

Quality control of observations directly affects the analysis quality of numerical prediction data assimilation. Based on the ＂Gaussian plus flat＂ distribution model of observation error and Bayes＇ probability theorem, this paper reports the development of a variational quality control scheme for the 3D-Var（three-dimensional variational） assimilation and forecast system in GRAPES （Global/Regional Assimilation and Prediction System）.It also discusses the initial startup and key parameters of this scheme, and furthermore analyzes and verifies its applicability and effectiveness.A heavy rainfall event in southern China is selected as a case for assimilating and forecasting the analysis using Global Forecast System（GFS） data as the background field and conventional observation data including TEMP, SYNOP, SHIPS, AIREP, SATOB and COSMIC satellite retrieval data. Also, we calculate the rain score（ETS and Bias） of batch tests with 31 days in August 2013.The results show that the ＂Gaussian plus flat＂ distribution model is a better match for the characteristics of real observation error than the Gaussian distribution.At the same time ,the variational quality control method is able to correct the observation weight in accordance with the size of the observation departure.This also proves the rationality of the non-Gaussian distribution assumption for real observation error and the correctness of variational quality control theory.The variational quality control method reasonably adjusts every observation weight according to different qualities of observation, and virtually classifies the observations.This is beneficial to identifying the quality of observations so that we can assimilate every observation with different weights, as available data, effective data and damaging data.The variational quality control method significantly adjusts the analysis increment field, which includes height, wind and specific humidity, especially in some areas where the damaging data are recognized.Due to the change