Stochastic Variation of Parameterized Physical Processes in Idealized Forecast Ensembles: Stochastic Physics

最近的研究表明，数值天气预报的一个主要误差来源是使用可解析变量的相对确定性函数来表示不可解析的子网格过程。改进集合预测的一项努力涉及扰动模型的物理参数化。为了解决这个问题，PI将研究各种验证措施中的子网格尺度特征，并使用它们的组合来改进集合预测。一个具有低维配置的理想模型将用于研究。研究重点将放在可解析变量和未解析过程的相互作用上，特别关注不同扰动幅度的影响，物理参数化随机元素的相关尺度，以及适当诊断中这些错误的可能特征。研究的总体目标是提供关于物理参数化中摄动结构的知识，并在集合预测中适当地表示它们。这项研究有可能为实际天气预报提供有用的资料。

Recently, it has been shown that one major source of errors in numerical weather predictions is generated by using comparatively deterministic functions of resolvable variables to represent the unresolved sub-grid processes. One effort in improving ensemble predictions involves perturbing the physical parameterizations of the model. To tackle the problem, the PI will study the sub-grid scale signatures in various verification measures and use a combination of them to improve the ensemble forecasts. An idealized model with low-dimension configurations will be used for the study. The research focus will be on the interactions of the resolvable variables and unresolved processes with particular attention on the effects of different perturbation amplitudes, correlation scales for the random element of physical parameterizations, and possible signatures of these errors in appropriate diagnostics. The overall goal of the research is to provide knowledge on the structure of perturbations in physical parameterizations and to represent them properly in ensemble forecasts. The research study has the potential of generating useful information for operational weather prediction.