[p2] Model-related issues


Information contents of a 4DVar analysis based on a reduced order model approach

Tsuyoshi Wakamatsu (JAMSTEC)


Ocean/Atmosphere 4D-Var systems estimate a large size control vector such as an initial model state and external forcing fields. The amount of meaningful information in the estimated control vector beyond observational noise, called observable modes, has been studied in geophysical inverse problems. Previous studies showed that the observable modes are useful to assess performance of a 4D-Var system. However, the separation of observable modes depends on a condition that measurement error covariance and background error covariance are both explicitly known, which is usually not the case in ocean/atmosphere data assimilation problem. In this study, we propose a method of designing measurement and background error covariance by truncating information contents in model dynamics to the amount extractable from a given observation array. The model truncation is conducted based on POD model reduction method and the observable modes of a 4D-Var system.