全球业务数值模式存在偶发的中期预报时效误差极端大的问题,对其溯源可为模式和同化系统改进提供重要参考.分析2020年1—2月中国气象局高分辨率全球同化预报系统(CMA-GFS)和较低分辨率全球集合预报系统(CMA-GEPS)业务预报在中期时效(6 d)的预报误差,利用集合敏感性分析方法,对东亚地区具有极端中期预报误差的一个个例(2020年2月8日12 UTC起报)进行了预报误差溯源研究.由CMA-GFS预报误差的时空演变特征及基于CMA-GEPS系统的集合敏感性分析结果得到了一个关于预报误差关键源区的初步推断,即为位于东亚上游地区的大西洋及欧洲西部地区(20°～90°N、90°W～60°E).进而,将CMA-GEPS系统控制预报位于上述误差关键源区的初值替换为最优集合成员初值后,预报结果显示东亚地区500 hPa位势高度中期预报误差显著减小,不到原预报误差的50％,这进一步验证了识别出的关键误差源区的有效性.

To represent model uncertainties at the physical process level in the China Meteorological Administration global ensemble prediction system (CMA-GEPS), a stochastically perturbed parameterization (SPP) scheme is developed by perturbing 16 parameters or variables selected from three physical parameterization schemes for the planetary boundary layer, cumulus convection, and cloud microphysics. Each chosen quantity is perturbed independently with temporally and spatially correlated perturbations sampled from log-normal distributions. Impacts of the SPP scheme on CMA-GEPS are investigated comprehensively by using the stochastically perturbed parametrization tendencies (SPPT) scheme as a benchmark. In the absence of initial-condition perturbations, perturbation structures introduced by the two schemes are investigated by analyzing the ensemble spread of three forecast variables' physical tendencies and perturbation energy in ensembles generated by the separate use of SPP and SPPT. It is revealed that both schemes yield different perturbation structures and can simulate different sources of model uncertainty. When initial-condition perturbations are activated, the influences of the two schemes on the performance of CMA-GEPS are assessed by calculating verification scores for both upper-air and surface variables. The improvements in ensemble reliability and probabilistic skill introduced by SPP and SPPT are mainly located in the tropics. Besides, the vast majority of the reliability improvements (including increases in ensemble spread and reductions in outliers) are statistically significant, and a smaller proportion of the improvements in probabilistic skill (i.e., decreases in continuously ranked probability score) reach statistical significance. Compared with SPPT, SPP generally has more beneficial impacts on 200-hPa and 2-m temperature, along with 925-hPa and 2-m specific humidity, during the whole 15-day forecast range. For other examined variables, such as 850-hPa zonal wind, 850-hPa temperature, and 700-hPa humidity, SPP tends to yield more reliable ensembles at lead times beyond day 7, and to display comparable probabilistic skills with SPPT. Both SPP and SPPT have small impacts in the extratropics, primarily due to the dominant role of the singular vectors-based initial perturbations.

为了更好地理解不同随机物理扰动方案对全球中期集合预报的影响差异,本研究基于GRAPES全球集合预报系统(GRAPES-GEPS)对比分析了随机物理倾向扰动(Stochastically Perturbed Parameterization Tendencies,SPPT)、随机动能补偿(Stochastic Kinetic Energy Backscatter,SKEB)及联合使用SPPT与SKEB三种模式扰动方案所产生的扰动特征及其对集合预报的影响.为避免初值扰动影响,考察随机物理方案所产生的扰动特征时,不使用初值扰动.通过扰动与误差相关性分析(PECA)发现,不同随机物理扰动方案所产生的扰动对预报误差均具有一定的描述能力,而且联合使用SPPT与SKEB方案时,扰动对误差的描述能力最好.对所有扰动方案来说,扰动总能量最初主要集中在热带地区对流层中高层以及平流层低层.随着预报时效的延长,扰动总能量不断增大,其大值区不断向热带外地区转移.从扰动总能量的谱结构来看,扰动能量均呈现升尺度发展的特征.在基于奇异向量初值扰动的GRAPES-GEPS中,随机物理扰动方案的使用均能够显著增加不同地区等压面要素的集合离散度,并在一定程度上改善集合平均误差.由于集合离散度的增大,预报失误率显著减小.连续分级概率评分也有所减小,尤其是在热带地区,改进更为明显.此外,中国地区不同量级(小雨、中雨、大雨和暴雨)降水概率预报技巧在一定程度上得到改善.上述改进均在联合使用SPPT与SKEB方案时最好,这与扰动总能量、扰动与误差相关分析结果一致.

为描述GRAPES全球模式初始条件的不确定性,基于适合集合预报应用的GRAPES全球奇异向量技术,依据大气初始误差符合正态分布的特征,采用高斯取样奇异向量来构造全球集合预报初始扰动,在此基础上建立了GRAPES全球集合预报系统(GRAPES-GEPS).利用GRAPES全球同化分析场,对采用初始扰动的GRAPES-GEPS连续试验预报结果进行检验和分析.结果 表明:GRAPES-GEPS中高度场、风场及温度场预报的集合离散度能有效快速增加,集合平均均方根误差与集合离散度的关系合理;相对控制预报的均方根误差,集合平均的预报优势在预报中期非常显著.为进一步体现GRAPES-GEPS中模式物理过程的不确定性,发展了模式物理过程倾向随机扰动技术(SPPT).试验结果表明:SPPT方案的应用有效提高了GRAPES-GEPS在南、北半球和热带地区等压面要素预报的集合离散度,同时一定程度减小了集合平均误差,进而改进了集合平均误差与集合离散度的关系,其中SPPT方案在热带地区的改进最为显著.本文发展的基于奇异向量的初始扰动方法和模式扰动SPPT方案在中国气象局2018年12月业务化运行的GRAPES-GEPS中得到了应用.

For describing uncertainties in the subgrid-scale energy upscaling transfer, a Stochastic Kinetic Energy Backscatter (SKEB) scheme has been introduced into the Global/Regional Assimilation and Prediction System (GRAPES) global ensemble prediction system (GEPS) to represent model errors more reasonably and increase the ensemble spread. In this research, the SKEB scheme employs the stochastic patterns with temporally and spatially correlated characteristics along with the estimated local kinetic energy dissipation rates caused by numerical diffusion to construct the stochastic stream function forcing. According to the relationship between the streamfunction and the rotational component of horizontal wind, the streamfunction forcing in the SKEB scheme is then transformed into horizontal wind perturbations, which are suitable for the GRAPES global model. The results indicate that, on the one hand, the application of the SKEB scheme improves the simulations of the atmospheric kinetic-energy spectra in the GRAPES model; on the other hand, it leads to a better spread-error relationship, increases the spread of the ensemble and reduces the root mean square error of the ensemble mean to some extent and the improvement is the most pronounced in the tropics. This scheme also contributes to a significant improvement of the continuous rank probability score (CRPS) in the tropics. In terms of precipitation forecast, the results from the Brier score and Area under the Relative Operating Characteristics (AROC) show that the SKEB scheme helps to improve probabilistic forecast skills of rainfall in China for light rain [0.1 mm, 10 mm), moderate rain [10 mm, 25 mm) and heavy rain [25 mm, 50 mm); however, it has little impact on the forecast of rainstorm [50 mm, ∞) (24 h precipitation). On the whole, the introduction of the SKEB scheme ameliorates the probabilistic prediction skills of the GRAPES-GEPS.

以生成GRAPES全球集合预报业务系统的控制预报初值为目的,基于GRAPES全球模式,开展了控制预报初值生成方法研究,发展了高分辨率初值动力升尺度方法,并检验了不同方法的可行性.通过对比不同方法产生的初始场形态,证实了仅仅对高分辨率初始场进行二维水平插值存在不足,基于静力学方程对Exner气压变量进行三维插值至关重要.结果表明,动力升尺度方法利用静力平衡关系,对全场的温压场进行调整,使之协调平衡,可以改善二维水平插值方法导致的初始位势高度场和温度场的噪音问题,产生适用于GRAPES全球集合预报业务系统的控制预报初值.