GRAPES的新初始化方案

The new initialization scheme of the GRAPES

中文关键词:  初始化  数字滤波  四维变分同化  GRAPES

英文关键词:Initialization  Digital filter  Four-dimensional variation assimilation  GRAPES

中文摘要:

      四维变分同化由于引入预报模式作为一项约束，理论上它的分析场已经具有较好的平衡性，但实施时还会有诸多因重力波导致的高频振荡过程，因此，四维变分同化（4DVar）分析仍需要初始化。文中描述了GRAPES全球四维变分同化系统（GRAPES-4DVar）的新初始化方案的科学设计、公式演绎以及试验结果。GRAPES-4DVar的新初始化方案采用数字滤波方案作为代价函数的一项约束控制重力波引发的不平衡结构，约束强加在分析增量上与极小化迭代过程同步进行。新的初始化方案是变分同化系统的一部分，数字滤波的积分时间与4DVar的同化时间窗一致，不会对4DVar产生额外的计算资源消耗；并能适应长时间窗的同化，不会因为时间窗的延长而削弱慢波过程。新初始化方案中，模式轨迹的光滑程度可在变分同化中通过重力波控制项的权重系数方便控制。GRAPES全球四维变分同化的理想和循环同化批量试验都表明，在四维变分同化中，重力波的控制依然非常重要，具有初始化的GRAPES试验，无论分析还是预报技巧都较无初始化的有明显优势。与以前分析和滤波独立实施的旧初始化方案相比，新方案的分析和预报效果略优，同时有效地节省循环同化系统的运行时间，这对四维变分同化来说非常重要。

英文摘要:

      Theoretically, the analysis of a four-dimensional variational (4DVar) data assimilation system is in better balance as the forecast model is its constraint. In actual implementation of 4DVar, however, there are many processes that can lead to high-frequency gravity waves. Thus, the initialization process is still necessary for 4DVar. In the present paper, the scientific design and formula derivation of the new initialization scheme of GRAPES (Global and Regional Assimilation Prediction System) numerical weather prediction system and experiments using the new scheme are presented. The four-dimensional variational data assimilation system of GRAPES (hereafter GRAPES-4DVar) adopts the digital filter scheme for initialization. The digital filter is added to the penalty function as a weak constraint to efficiently diminish the unbalanced structure associated with gravity-inertia waves. The constraint is only imposed on the incremental analysis and run synchronously with the minimization. The weak constraining term does not add extra computation cost since the integration period of the digital filter is the same as the 4DVar's time window. Furthermore, the new initialization scheme is suitable for long time window assimilation and will not weaken slow wave processes as the time window extends. It is easy to control the smoothness of model trajectory by tuning digital filter coefficient in the variational assimilation procedure. Results of GRAPES-4DVar assimilation cycling experiments show that the initialization is necessary in the 4DVar analysis, and the analysis and forecast skills have been improved significantly in the experiments with initialization compared to those without initialization. Compared to the old initialization scheme, in which the analysis and filter are two independent procedures, the new scheme efficiently saves the running time of assimilation cycling, although its impact on analysis and forecast is similar to that of the old scheme.