CIM Seminar with Gottfried Hastermann

Abstract: 

Data assimilation is the procedure of combining real world data and models based
on a-priori knowledge. It is widely adopted as data is abundantly available and
models in form of partial differential equations are able to accurately describe
complex phenomena across the natural and engineering sciences.

The especially popular class of sequential Bayesian data assimilation methods,
aim to solve the filtering problem by the means of Bayesian statistics and
frequently incorporates the model as a black box.  In the case of models in form
of non-linear partial differential equations expressing phenomena on multiple
scales, this approach not only requires unacceptably large sample sizes to
accurately predict the posterior probability density, due to the curse of
dimensionality, but also triggers fast modes of the model evolution
artificially.

We provide numerical methods to overcome these issues and develop
analysis towards robust combined data-model dynamics utilizing models
from classical mechanics, fluid dynamics and space weather applications.

Registration