中文摘要：

合理估计背景场误差协方差矩阵（B）是做好变分同化的关键环节。利用控制变量随机扰动法（RandomCV）、增长模繁殖法（BGM）及NMC法等3种背景场样本模拟方法,基于WRFDA系统计算B矩阵,对B矩阵的特征及其对同化预报效果的影响进行了研究。B矩阵的特征分析和单点观测试验表明,NMC法与RandomCV法得到的B矩阵误差方差较大,在同化中观测的权重更大;RandomCV法得到的B矩阵,背景场误差中变量的长度尺度更大,说明同化中观测的水平影响范围更大。连续循环同化和预报试验表明：应用RandomCV法计算得到的B矩阵分析与预报的效果明显优于系统自带的以及BGM法得到的B矩阵,且效果与NMC法相当与NMC方法相比,采用RandomCV方法产生背景场样本具有时间和人力成本相对低的优点。

英文摘要：

A reasonable background error covariance（B） is the key to data assimilation.Three sample error simulation techniques for calculating B,such as,Randomized Control Variables（RandomCV）,Breeding of Growing Modes（BGM） and the NMC method,are used to simulate the background error,and different B are gotusing the WRFDA system,then the statistical characteristics of the background error covariance and their impacts on analysis and forecasting are discussed.The results show that there is more significant background error amplitude in NMC and RandomCV than in BGM,so the observation will carries a bigger weight in data assimilation when NMC and RandomCV are chosen,and the length-scale of RandomCV is the biggest,so the influence range of observation is wider when RandomCV is used.The background error covariance of RandomCV is generated from perturbing the initial condition by the CV3 in WRF model,which achieves almost the similar effect as NMC and better than CV3,but a lower computational and manual cost than NMC.