Bayesian online algorithms for learning data-driven models of chaotic dynamics
The advent of machine and more specifically deep learning techniques significantly boosts the capabilities of data assimilation and inverse problem techniques used in the geosciences. It also spurs new, more ambitious goals for data assimilation in high dimensions. One of the key, currently very pop...
Main Authors: | , , , , , , , |
---|---|
Format: | Conference Object |
Language: | English |
Published: |
2023
|
Subjects: | |
Online Access: | https://gfzpublic.gfz-potsdam.de/pubman/item/item_5017738 |
id |
ftgfzpotsdam:oai:gfzpublic.gfz-potsdam.de:item_5017738 |
---|---|
record_format |
openpolar |
spelling |
ftgfzpotsdam:oai:gfzpublic.gfz-potsdam.de:item_5017738 2023-08-27T04:11:55+02:00 Bayesian online algorithms for learning data-driven models of chaotic dynamics Bocquet, M. Farchi, A. Malartic, Q. Finn, T. Durand, C. Bonavita, M. Laloyaux, P. Chrust, M. 2023 https://gfzpublic.gfz-potsdam.de/pubman/item/item_5017738 eng eng info:eu-repo/semantics/altIdentifier/doi/10.57757/IUGG23-1847 https://gfzpublic.gfz-potsdam.de/pubman/item/item_5017738 XXVIII General Assembly of the International Union of Geodesy and Geophysics (IUGG) info:eu-repo/semantics/conferenceObject 2023 ftgfzpotsdam https://doi.org/10.57757/IUGG23-1847 2023-08-06T23:41:22Z The advent of machine and more specifically deep learning techniques significantly boosts the capabilities of data assimilation and inverse problem techniques used in the geosciences. It also spurs new, more ambitious goals for data assimilation in high dimensions. One of the key, currently very popular, area of research consists in learning data-driven models of dynamical systems. With the natural constraints of geoscience, i.e., sparse and noisy observations, this typically requires the joint use of data assimilation and neural networks. However, the vast majority of algorithms are offline; they rely on a set of observations from the physical system, which must be available before the start of the training. We propose new algorithms that update the knowledge of the surrogate (i.e., data-driven) model when new observations become available. We carry out this objective with both variational (weak-constraint 4D-Var like) and ensemble (EnKF and IEnKS like) techniques. We test these algorithms on low-order Lorenz models, on quasi-geostrophic models, the ERA5 dataset, and sea-ice dynamics. Remarkably, in several cases, the online algorithms significantly outperform the offline ones. This opens the way to adaptive surrogate models that progressively learn trends and conform to real-time constraints of operational weather forecasting. Conference Object Sea ice GFZpublic (German Research Centre for Geosciences, Helmholtz-Zentrum Potsdam) |
institution |
Open Polar |
collection |
GFZpublic (German Research Centre for Geosciences, Helmholtz-Zentrum Potsdam) |
op_collection_id |
ftgfzpotsdam |
language |
English |
description |
The advent of machine and more specifically deep learning techniques significantly boosts the capabilities of data assimilation and inverse problem techniques used in the geosciences. It also spurs new, more ambitious goals for data assimilation in high dimensions. One of the key, currently very popular, area of research consists in learning data-driven models of dynamical systems. With the natural constraints of geoscience, i.e., sparse and noisy observations, this typically requires the joint use of data assimilation and neural networks. However, the vast majority of algorithms are offline; they rely on a set of observations from the physical system, which must be available before the start of the training. We propose new algorithms that update the knowledge of the surrogate (i.e., data-driven) model when new observations become available. We carry out this objective with both variational (weak-constraint 4D-Var like) and ensemble (EnKF and IEnKS like) techniques. We test these algorithms on low-order Lorenz models, on quasi-geostrophic models, the ERA5 dataset, and sea-ice dynamics. Remarkably, in several cases, the online algorithms significantly outperform the offline ones. This opens the way to adaptive surrogate models that progressively learn trends and conform to real-time constraints of operational weather forecasting. |
format |
Conference Object |
author |
Bocquet, M. Farchi, A. Malartic, Q. Finn, T. Durand, C. Bonavita, M. Laloyaux, P. Chrust, M. |
spellingShingle |
Bocquet, M. Farchi, A. Malartic, Q. Finn, T. Durand, C. Bonavita, M. Laloyaux, P. Chrust, M. Bayesian online algorithms for learning data-driven models of chaotic dynamics |
author_facet |
Bocquet, M. Farchi, A. Malartic, Q. Finn, T. Durand, C. Bonavita, M. Laloyaux, P. Chrust, M. |
author_sort |
Bocquet, M. |
title |
Bayesian online algorithms for learning data-driven models of chaotic dynamics |
title_short |
Bayesian online algorithms for learning data-driven models of chaotic dynamics |
title_full |
Bayesian online algorithms for learning data-driven models of chaotic dynamics |
title_fullStr |
Bayesian online algorithms for learning data-driven models of chaotic dynamics |
title_full_unstemmed |
Bayesian online algorithms for learning data-driven models of chaotic dynamics |
title_sort |
bayesian online algorithms for learning data-driven models of chaotic dynamics |
publishDate |
2023 |
url |
https://gfzpublic.gfz-potsdam.de/pubman/item/item_5017738 |
genre |
Sea ice |
genre_facet |
Sea ice |
op_source |
XXVIII General Assembly of the International Union of Geodesy and Geophysics (IUGG) |
op_relation |
info:eu-repo/semantics/altIdentifier/doi/10.57757/IUGG23-1847 https://gfzpublic.gfz-potsdam.de/pubman/item/item_5017738 |
op_doi |
https://doi.org/10.57757/IUGG23-1847 |
_version_ |
1775355614272159744 |