The Effect of Model Uncertainty and Error on the Forecast Uncertainty

模型不确定性和误差对预测不确定性的影响

• 批准号: 1237613
• 负责人: Istvan Szunyogh
• 金额: $ 36.37万
• 依托单位: Texas A&M University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2012
• 资助国家: 美国
• 起止时间: 2012-08-01 至 2016-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1237613&HistoricalAwards=false
• 关键词: Effect; Model; Uncertainty; Error; Forecast

This project examines the impact of uncertainty and error in model formulation on errors in weather forecasts produced by ensemble prediction systems (EPSs). In an EPS, forecasts are produced by running an ensemble of forecast models in which each model starts out with a slightly different initial condition (models can also differ in their formulation), and the resulting ensemble of forecasts is analyzed statistically to produce an optimal forecast, and estimate of the error in the forecast, and (in combination with real-world observations), a set of initial conditions for the next forecast cycle. The work is based on the hypothesis that errors in model formulation (principally errors in parameterization and truncation) introduce errors into the model integrations at the scale of the parameterized processes, presumably at or near the truncation limit of the model, and these errors are propagated upscale by resolved model dynamics until they produce forecast uncertainty at synoptic scales. Because upscale propagation determines the forecast impacts at substantial lead times (day three, for instance), forecast errors due to model errors do not have any particular characteristics that would distinguish them from forecast errors due to initialization errors (which would likely undergo the same upscale propagation before affecting the forecast). Based on the above results, the PIs conjecture that the effect of model errors could be accounted for, at least approximately, by modulating the magnitude of the different error patterns in the low-dimensional vector space which contains most of the forecast uncertainty from all sources. The research has a three part agenda, in which the first part will test the hypothesis in forecasts archived in the THORPEX Interactive Ground Global Ensemble (TIGGE) data set. The TIGGE archive contains forecasts produced by a variety of ensemble prediction systems using a variety of techniques to account for errors in model formulation and intial conditions, thus allowing numerous tests of the hypothesis. The second part consists of a suite of "perfect model" experiments, in which the "true" state of the atmosphere will be taken from the same model used in the ensemble forecast system. The perfect model configuration enables experiments in which there is no model error, as the "true" system can have exactly the same physics and truncation as the forecast model. Such experiments are useful for considering other sources of forecast errors. The third part consists of forecast experiments using a state-of-the-art data assimilation system to assimilate real-world observations, and the PIs will attempt to specific challenging forecast cases, such as prediction of cyclogenesis produced from a warm-core tropical cyclone.In addition to its scientiifc merit, the work will have societal benefit by developing a strategy to improve the quality of weather forecasts issued to the general public. The work also seeks to improve understanding of the uncertainty inherent in weather forecasts, so that information regarding the likely accuracy of forecasts can be included in forecast guidance. The work may also have applicability to climate and earth system models used to produce climate projections and long-range forecasts, and to understanding and predicting the behavior of other complex systems. In addition, the project provides support and training to a graduate student, thereby developing the workforce in this research area.

本项目研究模式制定中的不确定性和误差对集合预报系统(EPSS)产生的天气预报误差的影响。在EPS中，预报是通过运行预报模型集合来产生的，其中每个模型从略有不同的初始条件开始(模型的表述也可能不同)，并对产生的预报集合进行统计分析，以产生最优预报，并估计预报中的误差，以及(与真实世界观测相结合)下一个预报周期的一组初始条件。这项工作是基于这样一种假设，即模式中的错误(主要是参数化和截断错误)在参数化过程的尺度上引入了模式积分中的误差，假定是在模式的截断极限或附近，并且这些误差通过分解的模式动力学向上传播，直到它们在天气尺度上产生预报不确定性。由于高级传播在相当大的提前期(例如，第三天)确定预测影响，因此，由模型错误引起的预测误差不具有任何特定特征，以将它们与由于初始化错误(在影响预测之前可能经历相同的高级传播)的预测错误区分开来。基于上述结果，PI推测，模型误差的影响至少可以通过调制低维向量空间中不同误差模式的大小来解释，低维向量空间包含来自所有来源的大部分预测不确定性。这项研究有三个部分的议程，其中第一部分将检验THORPEX互动地面全球集合(TIGGE)数据集中存档的预报中的假设。TIGGE档案包含由各种集合预报系统产生的预报，这些预报系统使用各种技术来解释模式形成和初始条件中的错误，从而允许对假设进行多次检验。第二部分由一套“完美模式”实验组成，在这些实验中，大气的“真实”状态将从集合预报系统中使用的同一模式中获取。完美的模型配置使没有模型误差的实验成为可能，因为“真正的”系统可以具有与预测模型完全相同的物理和截断。这样的实验对于考虑其他预测误差来源是有用的。第三部分是使用最先进的数据同化系统进行的预报试验，以同化真实世界的观测数据，个人信息系统将尝试具体的具有挑战性的预报案例，例如预报暖核热带气旋产生的气旋生成。这项工作除了具有科学价值外，还将通过制定一种策略来提高向公众发布的天气预报的质量，从而产生社会效益。这项工作还寻求提高对天气预报固有的不确定性的理解，以便在预报指导中纳入有关预报可能准确的信息。这项工作也可能适用于气候和地球系统模型，用于产生气候预测和长期预报，以及理解和预测其他复杂系统的行为。此外，该项目还为一名研究生提供支持和培训，从而发展这一研究领域的劳动力。