This talk is presented as a Zoom Webinar and requires a passcode. For anyone outside the department; If you would like to attend, email [email protected]

How to handle nonlinearity in multiscale problems: pushing the frontier of data assimilation methodology 

Yue Ying, Researcher, Nansen Environmental and Remote Sensing Center, Norway 

Data assimilation (DA) methodology combines information from observations and model forecasts to obtain the best estimate of the model state, which is an important step in making numerical predictions of the Earth system. Many Earth system components display multiscale properties, in that the key features of the system (eddies, lines of convection, plumes, etc.) are equally prominent at small scales and large scales. At small scales, position errors of the features give rise to high nonlinearity, which cannot be handled well by traditional DA methods based on linearization. To resolve both the large and small scales, the model grid dimension is also large, making nonlinear DA methods struggle because of the curse of dimensionality. While we have abundant satellite observations and ever-improving model physics, the lack of effective DA method is the current bottleneck. Motivated by these grand challenges, I introduced a method called “multiscale alignment” in 2019 and have been developing a multiscale DA framework to push the frontier forward. In this talk, I will share my experiences in method development, how I borrowed ideas from other disciplines, how I am now testing the new methods in real problems, and my vision for the future of the Earth system science from a DA research perspective.