[Eoas-seminar] MET Seminar - Thursday September 21 - Prof. Hristo Chipilski (FSU Dept. of Scientific Computing)

[eoas-seminar at lists.fsu.edu]

Dear all,

Please join us next Thursday September 21 for a Meteorology seminar, given by Prof. Hristo Chipilski of FSU’s Department of Scientific Computing. Prof. Chipilski will speak about “Data Assimilation Meets AI: New Horizons for Earth System Modeling”.  (abstract below)

Prof. Chipilski will be joining us in person in 1044. If you are interested in meeting with the speaker, please contact Allison Wing (awing at fsu.edu<mailto:awing at fsu.edu>).

**Graduate students, please contact Allison Wing if you would like to join a pizza lunch with the speaker at 12:30 PM.**

DATE: Thursday September 21
SEMINAR TIME: Refreshments at 3 PM, Talk 3:15 PM - 4:15 PM.
SEMINAR LOCATION: EOA 1044 (Speaker in person)
SPEAKER: Prof. Hristo Chipilski

TITLE: Data Assimilation Meets AI: New Horizons for Earth System Modeling

ABSTRACT:
Our understanding of the Earth system hinges on the effective use of observations and numerical models. However, each comes with its own set of limitations: observations provide an incomplete description of the physical state, while numerical models can deviate from reality due to their inherent simplifications. The field of data assimilation (DA) reconciles these problems by merging observations and models in a statistically optimal way. Despite their numerous advantages, traditional DA methods are derived from crude statistical assumptions which do not allow them to make full use of the available observations.

In this seminar, I will elucidate these limitations through convective-scale simulations initialized with ground-based remote sensing retrievals. For a DA  system based on the popular ensemble Kalman filter (EnKF), I will demonstrate that systematic improvements in the forecast skill occur only when assimilating measurements that capture both the thermodynamic and kinematic characteristics of the lower atmosphere. Motivated by this finding, I will then present a new ensemble-based DA framework which combines traditional estimation methods and emerging AI tools. The new theory leverages the special mathematical properties of invertible neural networks to generalize the EnKF algorithm to arbitrarily complex distributions. Idealized experiments reveal that the most substantial gains from the new DA framework occur when observation errors are small and model variables are strongly correlated.

We look forward to seeing you there!

——————————————————
Allison Wing, Ph.D.
Werner A. and Shirley B. Baum Professor
Associate Professor, Department of Earth, Ocean and Atmospheric Science
Florida State University
awing at fsu.edu





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fsu.edu/pipermail/eoas-seminar/attachments/20230915/fd1b31d5/attachment.html>