代刊,朱跃建,毕宝贵. 2018. 集合模式定量降水预报的统计后处理技术研究综述[J]. 气象学报, 76(4):493-510, doi:10.11676/qxxb2018.015
集合模式定量降水预报的统计后处理技术研究综述
The review of statistical post-process technologies for quantitative precipitation forecast of ensemble prediction system
投稿时间:2017-11-10  修订日期:2018-02-07
DOI:10.11676/qxxb2018.015
中文关键词:  集合预报  定量降水预报  统计后处理  误差订正
英文关键词:Ensemble prediction system  Quantitative precipitation forecast  Statistical post-process technologies  Bias calibration
基金项目:公益性行业(气象)专项(GYHY201306002)、国家科技支撑(2015BAC03B00)、关键集成(CMAGJ2015Z06)。
作者单位
代刊 中国国家气象中心, 北京, 100081 
朱跃建 美国国家环境预报中心/环境模式中心, 马里兰, 20740 
毕宝贵 中国国家气象中心, 北京, 100081 
摘要点击次数: 333
全文下载次数: 441
中文摘要:
      集合数值模式预报已在定量降水预报业务中广泛应用,以获得预报不确定性、最可能预报结果以及极端天气预警。由于集合系统的数值模式不完善,且不能提供所有的不确定性信息,常表现出系统性偏差以及欠离散或过离散(如对于多模式集合)。为此,需要发展统计后处理技术,在尽量保持集合预报解析度的条件下,提高预报的技巧和可靠性。近年来,各种集合预报统计后处理技术得到快速发展。针对定量降水预报,依据技术方法的途径和成熟度将后处理研究归纳为3方面进行总结,包括:(1)不基于统计模型的非参数化后处理,包括集合定量降水预报偏差订正、多成员或模式信息集成以及基于空间分析的对流尺度模式后处理;(2)基于概率分布统计模型的参数化后处理,包括集合模式输出统计和贝叶斯模型平均两种方法框架;(3)考虑预报量的时间、空间和多变量间依赖关系或结构的处理方法,包括参数化和经验连接概率法。最后,讨论发展统计后处理技术需要关注的问题,包括考虑不同来源、不同尺度的多模式信息集成;提供高质量、高分辨率的降水分析资料;发展再预报技术扩充训练样本;基于不同的订正目的和应用场景来使用不同的后处理技术;发展面向海量预报数据、捕捉极端降水以及考虑预报量结构的新技术。
英文摘要:
      Ensemble forecast has been widely used for daily quantitative precipitation forecast (QPF) for the purposes to analyze forecast uncertainty, to reveal the most possible forecast scenario and to issue warning for extreme weather. Ensemble forecast could be biased; under-dispersion for single model ensemble and/or over-dispersion for multi-model ensemble due to possible problems in the modeling system and physical parameterizations, etc. Therefore, it is necessary to develop a statistical post-process method that can improve the forecasting skills and reliability while retaining the forecast resolution. In the past decade, many methodologies have been developed to generate numerical guidance. For ensemble QPF, the post process methodologies are summarized from three aspects based on their properties, i.e., (1) the nonparametric techniques not based on statistical models, including ensemble QPF bias correction, ensemble or multiple-model QPF integration, and generating probabilistic forecasts from convection-allowing ensembles; (2) the parametric techniques for building statistical models, including the framework of ensemble model output statistics (EMOS) and Bayesian model averaging (BMA); (3) the calibration techniques that account for inter-variable, and spatial and temporal structures, including parametric and experienced copula methods. Finally, we discuss the highlights of precipitation calibration and application, such as the integration of ensembles from different models and different spatial scales, the quality of high resolution precipitation analysis, and a set of reforecasts to provide more reliable training samples. Meanwhile, we need to consider "situation dependent" cases when applying these methodologies, and develop new technologies to handle huge forecast data, capture the extreme events and preserve the dependence structure of the forecast variables.
HTML   查看全文   查看/发表评论  下载PDF阅读器
分享按钮