Composing a surrogate observation operator for sequential data assimilation

Abstract

In data assimilation, state estimation is not straightforward when the observation operator is unknown. This study proposes a method for composing a surrogate operator for a true operator. The surrogate model is improved iteratively to decrease the difference between the observations and the results of the surrogate model, and a neural network is adopted in the process. A twin experiment suggests that the proposed method outperforms approaches that use a specific operator that is given tentatively throughout the data assimilation process.