2pm at NERSC (Copernicus).
Talk by Francois Counillon (NERSC).
Abstract:
Supermodelling is a novel approach in which different models are run interactively and are connected to achieve a model with superior performance. The method exploits model diversity and reduces model error by training the connection terms using observations. Structural differences between the models and the amount of data exchange among models are technical challenges and have limited the applicability of supermodelling to Earth system models. Here, we show that data assimilation can handle these issues. A demonstration is done with three Earth system models having different resolutions and structural differences and using a limited amount of data exchange (monthly ocean surface temperature). Synchronisation is achieved in several key regions, and model error is reduced. We discuss the limitations of the current framework and the prospect of improvements.