基于BGM的不同繁殖长度对集合预报的影响

引用本文

于永锋, 张立凤. 基于BGM的不同繁殖长度对集合预报的影响[J]. 应用气象学报, 2007, 18(1): 86-93. 复制到剪切板

Yu Yongfeng, Zhang Lifeng. The Effect of Different Breeding Length upon Ensemble Forecasting Based on BGM[J]. Journal of Applied Meteorological Science, 2007, 18(1): 86-93 复制到剪切板

基于BGM的不同繁殖长度对集合预报的影响

于永锋^1,2, 张立凤^2,3

1. 海军南海舰队海洋水文气象中心, 湛江 524001;
2. 解放军理工大学气象学院, 南京 211101;
3. 中国科学院大气物理研究所大气科学和地球流体力学数值模拟国家重点实验室, 北京 100029

2005-09-23 收到, 2006-08-15 收到修改稿.

资助项目: 国家重点基础研究发展规划项目“我国南方致洪暴雨监测与预测的理论和方法研究”(2004CB418300) 和中国科学院大气科学和地球流体力学数值模拟国家重点实验室开放课题共同资助

摘要: 利用T63L9全球谱模式和NCEP/NCAR再分析资料, 对BGM方法中增长模的繁殖长度对集合预报效果的影响进行研究。结果表明:与控制预报相比, 不同繁殖长度的集合预报都能使预报效果得到一定程度的改进, 特别是第4天预报以后, 改进程度随预报时效而稳步提高。三组不同繁殖长度的集合预报对控制预报的改进存在差别, 分析结果表明:繁殖长度为2 d的集合预报明显效果最差, 而繁殖3 d和4 d的集合预报差别并不明显。对集合Talagrand分布以及离散度的初步分析表明, 繁殖长度取为3 d似乎最为合理。

关键词: 增长模繁殖繁殖长度集合预报初始扰动

The Effect of Different Breeding Length upon Ensemble Forecasting Based on BGM

Yu Yongfeng^1,2, Zhang Lifeng^2,3

1. Hydrometeorological Center of South China Sea Navy, Zhanjiang 524001;
2. Institute of Meteorology, PLA University of Science and Technology, Nanjing 211101;
3. State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, CAS, Beijing 100029

Abstract: Ensemble of initial conditions is the most important method in ensemble prediction, and how to generate the initial perturbations is crucial. Among all the methods to create perturbations for ensemble forecast, the "breeding of growing modes" (BGM) method has gained more and more favor for its good performance and almost "free" cost of computation. The breeding method simulates how fast-growing errors are "bred" in the analysis cycle. When a breeding cycle starts, an random initial perturbation field is added upon the control analysis. After some time's breeding, the most non-growing or decaying perturbations are filtered and the remainders can be mainly the fast-growing modes (perturbations). Accordingly, the perturbation's growth rate reaches certain value and its shape is also getting to the phase of slow change. In some sense, the perturbation in the breeding cycle reaches "saturation". This saturated status of perturbation is regarded as the estimate of the fast-growing modes in the realistic analysis error, thus is applied as the initial perturbation. From the idea of BGM method, one of the crucial questions may be when the breeding cycle ends, which is called "the breeding length" problem. In fact, the breeding length is the total time of breeding of growing modes (initial perturbations) in the ensemble prediction system. The general steps to determine the breeding length are: defining the saturated character of growing modes; investigating the increasing and saturating process of the growing modes, and defining an approximate saturated time; according to the saturated time, doing a great lot of experiments with different breeding lengths, and determining the final length by the forecast skill. The model is the global media-range forecast spectral model T63L9 and the NCEP/NCAR 6-hour reanalysis data in 1998 are used. The initial modes in the beginning of a breeding cycle are random fields with unified distribution in [-1, 1]. The value of 6-hour is chosen as the length of the integration in each cycle. Besides, "rescaling" method to make the magnitude of growing modes comparable with the initial analysis error in RMSE sense is used. Some researches show that, if an appropriate breeding scheme is adopted, the breeding of the growing modes can present a clear saturation character in both the norm (magnitude) and the form (shape) after 3—4 days of breeding. Based on these results, numerical experiments are done with breeding length of 2, 3 and 4 days. Conclusions from these experiments show that the ensemble prediction with different breeding length can all reach a certain extent improvement upon the control forecast, especially the betterments increase steadily after 4 days of forecast lead time. Results also show the differences among the three breeding lengths in the improvement upon the control forecast. It shows a better improvement in ensemble mean with a breeding length of 3 days than 2 days, but the 4-day breeding seems pretty much the same as the 3-day breeding. A pilot study on the spread and the Talagrand distribution of the ensemble members with different breeding lengths is also made. It seems that the most appropriate way is to choose 3-day as the breeding length in the system.

Key words: breeding growing modes (BGM) breeding length ensemble forecasting initial perturbation

[1]	杜钧. 集合预报的现状和前景. 应用气象学报, 2002, 13, (1): 16–28.
[2]	Epstein E S. Stochastic dynamic prediction. Tellus, 1969, 21: 739–759.
[3]	Hoffman R N, Kalnay E. Lagged average forecasting, an alternative to Monte Carlo forecasting. Tellus, 1983, 35A: 100–118. DOI:10.1111/tela.1983.35A.issue-2
[4]	Toth Z, Kalnay E. Ensemble forecasting at NMC:The generation of perturbations. Bull Amer Meteor Soc, 1993, 74: 2317–2330. DOI:10.1175/1520-0477(1993)074<2317:EFANTG>2.0.CO;2
[5]	Molteni F, Buizza R, Palmer T N, et al. The ECMWF ensemble prediction system:methodology and validation. Q J R Meteor Soc, 1996, 122: 73–119. DOI:10.1002/(ISSN)1477-870X
[6]	Palmer T N, Molteni F, Mureau R, et al. Ensemble prediction, ECMWF research department. Tech Memo, 1992, 188: 1–45.
[7]	Houtekamer P L, Derome J. Methods for ensemble prediction. Mon Wea Rev, 1995, 123: 2181–2196. DOI:10.1175/1520-0493(1995)123<2181:MFEP>2.0.CO;2
[8]	于永锋, 张立凤. 基于增长模繁殖法的集合预报初始扰动饱和分析. 大气科学, 2005, 29, (6): 955–964.
[9]	付顺旗, 张立凤, 张铭. 谱模式T63L9正规模初值化方案及试验. 大气科学, 2001, 25, (5): 661–675.
[10]	于永锋. 用增长模繁殖法生成集合预报初始扰动的关键技术研究. 南京: 解放军理工大学气象学院, 2005.
[11]	Toth Z, Kalnay E. Ensemble forecasting at NCEP and the breeding method. Mon Wea Rev, 1997, 125, (6): 3297–3319.
[12]	柳崇健, 刘英, 康红文. 一种新的物理耗散技术及其在中尺度数值天气预报模式中的应用. 中国科学 (D辑), 2002, 32, (3): 255–264.
[13]	Talagrand O, Vautard R.Evaluation of probabilistic prediction systems.Proc of the ECMWF Workshop on Predictability, Reading, United Kingdom, 1997: 1-25.
[14]	Hersbach H, Mureau R, Opsteegh J D. A short-range to early-medium-range ensemble prediction system for the european area. Mon Wea Rev, 2000, 128: 3501–3519. DOI:10.1175/1520-0493(2000)128<3501:ASRTEM>2.0.CO;2


图 1. 集合预报对控制预报全球500 hPa高度场的改进量 (直方图) 和改进率 (曲线) Fig 1. Improvement quantity (column diagram) and quotiety (curve line) on the global 500 hPa geopotential height by the ensemble forecasts with respect to the control forecast


图 2. 北半球500 hPa高度场的改进 (其他同图 1) Fig 2. The improvement for the Northern hemisphere 500 hPa fields (others as in Fig. 1)


图 3. 亚洲范围500 hPa高度场的改进 (其他同图 1) Fig 3. The improvement for the Asian 500 hPa fields (others as in Fig. 1)


图 4. 500 hPa位势高度场的Talagrand分布图 (点划线表示期望概率f=0.125) Fig 4. The Talagrand distribution diagram of 500 hPa geopotential height (dash-dotted line represents the expected probability:f=0.125)


图 5. 各集合成员相对于集合平均预报的离散度 (给出的是500 hPa位势高度场32组试验个例的平均值, 计算范围取北半球) Fig 5. Spread of the ensemble forecasts relative to respective ensemle mean (shown are average values of 32 examples for geopotential height fields at 500 hPa and the Northern hemisphere is selected to be the analysis area)