Exact nonlinear filtering of geophysical systems

Hristo Chipilski

National Center for Atmospheric Research, Advanced Study Program

Tuesday, March 28, 2023, 2:00 pm

JOIN THE WEBINAR:

Abstract

This seminar will discuss the application and development of data assimilation (DA) methods for the estimation of nonlinear geophysical systems. I will first present systematic experiments which demonstrate the positive impact of assimilating ground-based remote sensors for improving the forecasts of organized night-time convection. However, comparison between single profiler experiments will reveal how standard (Gaussian-based) DA methods lead to inconsistent analysis results. This finding will then serve as a motivation for the second part of my talk where I will combine traditional DA methods and powerful AI/ML techniques to derive an exact nonlinear filtering theory. Specifically, I will describe and exemplify an ensemble-based DA algorithm which leverages the special mathematical properties of invertible neural networks to generalize the celebrated ensemble Kalman filter.