"Data assimilation in coupled chaotic dynamics and its combination to machine learning to infer unresolved scale error"

The International Symposium on Data Assimilation Online (ISDA-Online) May 7, 2021
"Coupled Data Assimilation"
International Symposium on Data Assimilation - Online

By using the covariant Lyapunov vectors we show that they are related to coupling mechanisms and are natural manifestation of the atmosphere-ocean interactions. As opposed to the uncoupled case, they must be included in a coupled data assimilation with EnKF to achieve good performance. Finally, we show how coupled DA and machine learning can be combined to infer parametrization of the unresolved scales. It generates a data driven parametrization that is then added to the physical model to form a hybrid physical-data-driven model. Results show the goodness of the hybrid model and the importance of using CDA in the training so to embody coupled mechanisms otherwise absent in the raw data.