Improving Complex Models through Stochastic Parameterization and Information Theory
Thursday, November 10, 11AM – 12:30PM
Andrew J. Majda, Department of Mathematics and Climate Atmosphere, Ocean Science (CAOS), Courant Ins
In many situations in contemporary science and engineering involving complex systems, the analysis and prediction of phenomena often occur through complex dynamical equations that have significant model errors compared with the true signal in nature. Clearly, it is important to improve the imperfect model’s capabilities to recover crucial features of the natural system and also to accurately model the sensitivities in the natural system to change in external of internal parameters. These efforts are hampered by the fact that the actual dynamics of the natural system are unknown. Important examples with major societal impact involve the Earth’s climate and climate change where climate sensitivities are studied through a suite of imperfect comprehensive (AOS) computer models; other examples occur in neural science, material science, and environmental engineering. This lecture surveys three different uses of stochastic parameterization and/or information theory developed by the speaker and collaborators to improve model fidelity and predictive skill for complex systems with model error: 1) Improving tropical convective parameterization through the stochastic multi-cloud model; 2) Improving model fidelity and long-range forecasting skill through empirical information theory and stochastic parameterization; 3) judicious model errors in filtering/data assimilation of turbulent dynamical systems through suitable stochastic parameterization extended Kalman filters (SPEKF) in order to avoid curse dimension and curse of ensemble size. All papers available at Majda NYU faculty website: http://www.math.nyu.edu/faculty/majda/publicationrevised.html.
Hosted by Clint Dawson