基于超级集合预报的定量降水概率预报方法研究

Aiming at the problems of quantitative precipitation prediction without uncertainties specified are hard to be incorporated into operations and decision making of other downstream applications such as early warning of floods, and probabilistic precipitation prediction simply fitted from the output of ensemble prediction without bias correction of model errors, and also fitted probabilistic precipitation prediction are not making full use of all the information available in an ensemble forecast, a solution is proposed in the project. The goal of this research is to develop a new statistical and methodologies of probabilistic quantitative precipitation based on super-ensemble forecasting to generate skillful probabilistic quantitative precipitation forecasts. The globe ensemble prediction system output data set of the world meteorological center and rain gauge data with high spatial resolution and long history records are employed. The research analyze the distribution of bias ensemble prediction system, constructs nonlinear and multi-goal theory model, develops a new statistical methodologies and combines the uncertainty of multi-system and multi-spatial scale into probabilistic precipitation forecasting and also considering the uncertainty caused by model system. To reach the aim, the characteristics of distribution of rain gauge data will analyses using Gamma Distribution, carry out bias correction with post processor. The relationship between ensemble member and observations will address using nonparametric regression. The contribution size of different ensemble member is estimated using bayesian Informativeness Score method. The uncertainties of multi-system and multi-spatial scale will combine into generalized additive model and bayesian model averaging. The novel contribution of project is promoting the probabilistic precipitation forecast skill by combining the useful information of multi-model and multi-spatial scale sources, and promoting the application of multi-model ensemble prediction into, such as early warning of flood risk, and uncertainties of Hydrological Forecasting and others fields. This project provides strong scientific support for application of probabilistic quantitative precipitation forecasting for public and the field user.

针对我国目前提供给用户决策和其他领域应用的定量降水预报大多缺乏不确定信息、集合预报直接拟合概率预报不能包含系统误差且未能充分利用集合预报全部信息的问题。本项目利用历史降水观测资料和全球集合预报资料，研究集合预报系统降水预报误差规律，降水预报不确定特征；研究基于多模式多空间尺度集合预报定量降水概率预报不确定信息集成理论和方法，构建模式资料、观测资料不确定性集成的定量降水概率预报理论模型。为此，通过伽马分布函数分析降水预报的不确定性和模式误差的分布规律。通过统计模型处理器校正预报偏差。通过非参数回归方法分析模式降水的空间尺度与集合预报成员降水预报的交互作用关系。采用贝叶斯方法确定各成员贡献大小和集合多模式多空间尺度的不确定信息。本研究所发展的定量降水概率预报方法为洪水风险预警方法、水文预报不确定性研究、电力等各个领域提供科学支撑，具有重要的科学意义和应用价值。

本项目针对我国目前提供给专业用户使用的定量降水预报大多缺乏不确定信息、集合预报直接拟合概率预报不能包含系统误差且未能充分利用集合预报全部信息的问题开展研究。主要包括降水观测和集合预报成员的降水概率分布及集合预报成员不确定性之间的关系分析；集合预报系统的误差规律分析；集合预报各个成员中有效信息提取方法和有效信息估计的研究；多模式、多成员集合预报的定量降水概率预报方法的建立等几个方面。主要研究成果如下：建立了我国小时降水的概率分布模型，揭示了我国小时降水的分布特征。从与集合概率预报降水检验对象匹配的角度，提出了观测概率法和观测百分位法两种降水观测概率的表达方法，达到了在评估环节中考虑了降水观测概率计算方法不确定性的目的。建立了三种模式气候累积概率分布方案。证明了在区域气候特征比较均匀的区域，运用空间扩展的方式来增加模式气候样本数，生成的模式气候累积概率分布能提高极端强降水预报指数（EPFI）对极端强降水的识别技巧，并提前8d发出长江中下游地区极端强降水预报信号。分析了不同观测概率下的集合预报定量降水的概率密度分布及对集合预报分辨性影响，确定了不同量级概率降水预报的误差来源；建立了基于BMA模型-统计降尺度的综合定量降水概率预报模型；从降水观测先验信息和集合预报各模式先验信息的角度，建立了不同先验信息下BMA模型的集合预报降水概率预报方法。. 本项目发展的具有应用价值的定量降水概率预报方法，为集合预报向实际应用的转化、为用统计方法提升集合预报在定量概率降水预报的预报价值、为挖掘集合预报在暴雨、洪水、泥石流等灾害早期预警决策中的应用提供科学支撑。本项目共发表论文17篇，其中SCI论文3篇，EI论文1篇。