对于中尺度数值天气预报来说,初始条件的准确与否已成为影响预报技巧的主要因素之一。现有的大气观测资料在时空分布上的不均匀,以及存在的观测误差,使得我们必须引进资料同化方法,为中尺度数值模式提供最优的初始场。由于传统的三维变分同化(3DVar)方法缺乏模式约束以及背景误差协方差矩阵(B矩阵)不具有流依赖性,因此本文提出一种基于历史样本投影的3DVar(HSP-3DVar)方法,它不仅具有流依赖的B矩阵,而且比传统的3DVar简单易行。为了评价HSP-3DVar的同化性能,我们基于区域暴雨预报模式AREM(Advanced Regional Eta Model)对其进行了观测系统模拟试验(OSSE),结果表明:HSP-3DVar能够有效融合观测信息,模式初值在各层的均方根误差都显著地降低。
As an important role in mesoscale numerical weather forecasts, initial conditions (ICs) of atmosphere affect the prediction skills directly. Due to the existing noises in observation, and the irregularly spatial and temporal distributions of observation, data assimilation approaches are required to provide mesoscale numerical weather forecasts with optimal ICs. Because the standard three-dimensional variational data assimilation (3DVar) method lacks model constraints and uses a background error covariance matrix (simply B-matrix, hereinafter) without flow-dependence, the authors proposed a new 3DVar scheme based on historical sample projection (HSP-3DVar) with a flow-dependent B-matrix, which is easier to be implemented. In order to test the performance of HSP-3DVar, an observing system simulation experiment (OSSE)we carried out using the Advanced Regional Eta Model (AREM). The results show that the HSP-3DVar could effectively fit observing information and significantly reduce the root mean square errors (RMSE) of ICs on every level.