地面气象要素多模式集成预报研究进展

目前,集合预报已成为天气预报业务的主要支撑。然而,由于数值模式本身的限制与不完善以及集合系统存在初值扰动、集合大小等方面的局限,常存在预报偏差。不同预报模式通常具有不同的物理过程参数化方案、初始条件等,导致其预报能力各有不同。为此,如何纠正预报偏差以及如何充分有效地利用不同模式的预报信息以获得更加准确的天气预报广受关注。近年来,利用统计理论与预报诊断,基于多个集合预报系统的多模式集成预报技术得到快速发展,已成为有效消除预报偏差从而提高天气预报技巧的一种统计后处理方法。针对气温、降水和风3个最基本的地面气象要素,首先依据预报形式将应用范围较广的简单集合平均、消除偏差集合平均、超级集合、贝叶斯模式平均、集合模式输出统计等加权或等权平均多模式集成技术,分成确定性预报和概率预报两大类,并做系统介绍。最后,讨论使用和发展多模式集成技术需要关注的问题,包括考虑参与集成的模式个数、发展降水及风速分级预报模型和发展基于机器学习的多模式集成新技术。

Research progresses of multimodel ensemble forecast of surface meteorological elements

Nowadays,ensemble forecasting has become the main support for operational weather forecasting.However,due to the limitations and imperfections of the numerical model itself,as well as the limitations of the ensemble forecast system in terms of initial perturbation schemes,ensemble size,etc.,the forecast results are generally biased.In addition,different forecasting models usually have different physical parameterization schemes,initial conditions,etc.,resulting in different forecasting capabilities.Therefore,how to correct the forecast deviation and how to make full and effective use of forecast information from different models to obtain more accurate weather forecasts has received extensive attention.In recent years,using statistical theory and forecasting diagnosis,multimodel ensemble forecasting technologies based on multiple ensemble prediction systems have been rapidly developed,and has become a statistical post-processing method to effectively eliminate forecast deviation and improve weather forecasting skills.For the three most basic surface meteorological variables (i.e.,temperature,precipitation and wind),the widely used multimodel ensemble technologies such as ensemble mean (EM),bias-removed ensemble mean (BREM),superensemble (SUP),Bayesian model averaging (BMA),and ensemble model output statistics (EMOS) are first introduced from the perspective of deterministic forecasting and probabilistic forecasting.Finally,this paper discusses the issues that need to be paid attention to when using and developing multimodel ensemble technologies,including considering the number of participating models,developing precipitation and wind speed classification forecast models,and developing new multimodel ensemble technologies based on machine learning.