基于优化样本组合的集合-变分混合同化方案研究

STUDY ON ENSEMBLE-VARIATION HYBRID DATA ASSIMILATION BASED ON OPTIMIZED COMPOSITE SAMPLES

关键词:  数值天气预报  资料同化  混合同化  集合样本  流依赖

Key words:  numerical weather prediction  data assimilation  hybrid assimilation  ensemble sample  flow dependent

摘要:

为有效引入“流依赖”的背景场误差协方差，同时降低集合预报带来的计算量，尝试通过优选与同化时刻天气形势更相似的历史预报样本，并结合预报过程中的时间滞后样本，将两种样本引入集合-变分混合同化系统中，构建基于优选历史预报样本和时间滞后样本的集合-变分混合同化方案。单点观测理想试验表明，优选历史预报样本结合时间滞后样本，既能够缓解样本不足所导致的采样误差，又能够为同化系统提供“流依赖”的背景场误差协方差。连续一周的循环同化及预报试验结果显示，相较于ERA5资料和探空资料，三维变分方案整体表现稍差，样本组合混合同化方案分析场和预报场的均方根误差最小，且比仅用时间滞后样本的混合同化方案有所改进；降水评分整体也表现最优，尤其对中雨和暴雨的模拟改进较明显，较好地模拟出了强降水中心的强度和位置，且改善了降水过报的问题。

Abstract:

In order to introduce flow-dependent background error covariance effectively and reduce the computing cost of ensemble forecast, the present research attempts to select historical forecasting samples similar to the weather condition at the time of assimilation and combine them with time-lagged samples in hybrid assimilation system. The single observation tests indicate that using the combination of selected historical samples and time-lagged samples, the sample error caused by limited samples is mitigated and flow-dependent background error covariance could be introduced in hybrid assimilation system. Compared with ERA5 and sound dataset, cycling assimilation and forecasts for a week show that 3DVAR has slightly poor performance. Hybrid experiment based on composite samples has the lowest root mean square error (RMSE), and performs better than hybrid experiments which only use the time-lagged samples. Its overall precipitation score is the best, and performs especially well in the assessment of moderate and heavy precipitation. The intensity and location of heavy precipitation center could be predicted well and the problem of over-estimating is alleviated in the composite sample hybrid assimilation scheme.