Atmosphere Ocean Science Colloquium
Characterizing and predicting climate variability through the spectral theory of dynamical systems
Speaker: Dimitris Giannakis, CIMS
Location: Warren Weaver Hall 1302
Date: Wednesday, November 14, 2018, 3:30 p.m.
The spectral theory of dynamical systems characterizes the properties of dynamics through intrinsically linear evolution operators acting on spaces of observables (functions of the state). This perspective provides useful tools for statistical predictive modeling with little prior knowledge about the system, as well as identification of modes of variability exhibiting a coherent dynamical evolution. In this talk, we discuss the basic properties of this framework, implemented through machine learning approaches, and illustrate it with applications to climate dynamics. In particular, we discuss applications to non-parametric statistical prediction of the El Nino Southern Oscillation (ENSO), as well as a diagnostic study of Indo-Pacific and Antarctic variability on seasonal to decadal timescales. Results from this study identify a multiscale hierarchy of modulating relationships between the seasonal cycle, ENSO, and Pacific decadal variability.