Determination of the Adjoint Model of the NMC Global and NGMModels and Their Application to 4-D Data Assimilations

NMC全球和嵌套网格模式的伴随模式的确定及其在四维数据同化中的应用预测模型需要完整、准确的规范，或分析，在初始时间的大气状态。将时间上分布的观测值引入预报模式的过程称为四维数据同化。观测不能完美地描述给定时间的大气状态。理想情况下，同化应在满足模式中体现的控制方程的动力学特性的同时，产生最好的数据观测“匹配”。根据这一奖项，PI和他的同事将采用一种同化方法，该方法使用最优控制理论来最小化观测演变和模式的动力学约束之间的差异。这项技术需要确定所使用的模式的伴随项，在这种情况下，是国家气象中心(NMC)使用的模式。确定特定预报模型的伴随模型是一个复杂的数值问题。PI开发了一种分阶段的方法，致力于日益复杂的模型。在每个阶段，将与NMC的科学家合作测试四维同化的伴随方法。在研究过程中，私人投资促进机构必须发展新的方法，以处理因使用更复杂的模型而引致的问题。例如，确定伴随模型要求量是可微的，但某些物理过程，如对流，不是可微的。这项研究作为NSF-NMC合作努力的一部分得到支持，以支持数值天气预报的基础研究。

"Determination of the Adjoint Models for NMC's Global and Nested Grid Models and Their Application to Four Dimensional Data Assimilation" A numerical weather prediction model requires a complete and accurate specification, or analysis, of the state of the atmosphere at the initial time. The process of introducing observations which are distributed in time into a forecast model is known as four dimensional data assimilation. Observations cannot perfectly describe the state of the atmosphere at a given time. Ideally, an assimilation should produce the best "fit" of the data observations while satisfying the dynamical properties of the governing equations as embodied in the model. Under this award, the PI and his colleagues will employ an assimilation method which uses optimal control theory to minimize the differences between the evolution of the observations and the dynamical constraints of the model. This technique requires the determination of the adjoint of the models employed, in this case, models used at the National Meteorological Center (NMC). Determination of the adjoint model of a particular forecast model is a complex numerical problem. The PI has developed a staged approach, working on models of increasing complexity. At each stage, the adjoint method for four dimensional assimilation will be tested in cooperation with NMC scientists. In the course of the research, the PIs will have to develop new methods for dealing with problems introduced by the use of more complex models. For example, determination of the adjoint models requires that quantities be differentiable, but certain physical processes, such as convection, are not. This research is being supported as part of an NSF-NMC cooperative effort to support basic research in numerical weather prediction.