Seminar Title: Improving Numerical Weather Prediction through Bayesian Inference with External Information
Seminar Abstract: Numerical weather prediction (NWP) is essential for a wide range of military and socioeconomic activities (e.g., green energy production and disaster warnings). Improvements in NWP will thus benefit both defense and society. Since NWP is an initial value problem involving chaotic dynamics, the utility of NWP depends on the accuracy and precision of the initial conditions. These initial conditions are constructed by synthesizing a prior estimate of the atmospheric state with atmospheric measurements. The methods used for such synthesis are called Data Assimilation (DA) and are a form of high-dimensional Bayesian inference. DA is a crucial part of operational NWP systems. As such, research that improves the outcomes of DA has the potential to tangibly benefit society and defense.
In this seminar, Chan will begin with motivating the problem of DA by introducing the chaotic nature of the atmosphere. DA will then be introduced from a Bayesian perspective, and its (absurd) effectiveness will be illustrated via DA experiments involving infrared measurements from satellites and astrophysical cosmic radiation measurements. Finally, Chan will discuss the curse of dimensionality intrinsic to DA, the essential role of heuristic information to overcome that curse, and new approaches to leverage that information.
No prior knowledge of DA or atmospheric science is needed to understand this seminar.
