Journal cover Journal topic
Geoscientific Model Development An interactive open-access journal of the European Geosciences Union
Journal topic

Journal metrics

IF value: 5.240
IF5.240
IF 5-year value: 5.768
IF 5-year
5.768
CiteScore value: 8.9
CiteScore
8.9
SNIP value: 1.713
SNIP1.713
IPP value: 5.53
IPP5.53
SJR value: 3.18
SJR3.18
Scimago H <br class='widget-line-break'>index value: 71
Scimago H
index
71
h5-index value: 51
h5-index51
Preprints
https://doi.org/10.5194/gmd-2020-350
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/gmd-2020-350
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Submitted as: methods for assessment of models 26 Oct 2020

Submitted as: methods for assessment of models | 26 Oct 2020

Review status
This preprint is currently under review for the journal GMD.

Efficient Bayesian inference for large chaotic dynamical systems

Sebastian Springer1, Heikki Haario1,2, Jouni Susiluoto1,3, Aleksandr Bibov1,5, Andrew Davis3,4, and Youssef Marzouk3 Sebastian Springer et al.
  • 1Department of Computational and Process Engineering, Lappeenranta University of Technology, Lappeenranta, Finland
  • 2Finnish Meteorological Institute, Helsinki, Finland
  • 3Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA, USA
  • 4Courant Institute of Mathematical Sciences, New York University, New York, NY, USA
  • 5Varjo Technologies Oy, Helsinki, Finland

Abstract. Estimating parameters of chaotic geophysical models is challenging due to these models' inherent unpredictability. With temporally sparse long-range observations, these models cannot be calibrated using standard least squares or filtering methods. Obvious remedies, such as averaging over temporal and spatial data to characterize the mean behavior, do not capture the subtleties of the underlying dynamics. We perform Bayesian inference of parameters in high-dimensional and computationally demanding chaotic dynamical systems by combining two approaches: (i) measuring model-data mismatch by comparing chaotic attractors, and (ii) mitigating the computational cost of inference by using surrogate models. Specifically, we construct a likelihood function suited to chaotic models by evaluating a distribution over distances between points in the phase space; this distribution defines a summary statistic that depends on the attractor's geometry, rather than on pointwise matching of trajectories. This statistic is computationally expensive to simulate, compounding the usual challenges of Bayesian computation with physical models. Thus we develop an inexpensive surrogate for the log-likelihood via local approximation Markov chain Monte Carlo, which in our simulations reduces the time required for accurate inference by orders of magnitude. We investigate the behavior of the resulting algorithm on model problems, and then use a quasi-geostrophic model to demonstrate its large-scale application.

Sebastian Springer et al.

Interactive discussion

Status: open (until 25 Dec 2020)
Status: open (until 25 Dec 2020)
AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment
[Subscribe to comment alert] Printer-friendly Version - Printer-friendly version Supplement - Supplement

Sebastian Springer et al.

Sebastian Springer et al.

Viewed

Total article views: 211 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
151 58 2 211 5 3
  • HTML: 151
  • PDF: 58
  • XML: 2
  • Total: 211
  • BibTeX: 5
  • EndNote: 3
Views and downloads (calculated since 26 Oct 2020)
Cumulative views and downloads (calculated since 26 Oct 2020)

Viewed (geographical distribution)

Total article views: 160 (including HTML, PDF, and XML) Thereof 155 with geography defined and 5 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 

Cited

Saved

No saved metrics found.

Discussed

No discussed metrics found.
Latest update: 01 Dec 2020
Publications Copernicus
Download
Short summary
The model predictions always contain uncertainty. But in some cases, such as weather forecasting or climate modelling, chaotic unpredictability increases the difficulty to say exactly how much uncertainty there is. We combine two recently proposed mathematical methods to show how the uncertainty can be analyzed in models that are simplifications of true weather models. The results can be extended in the future to show how forecasts from the large-scale models can be improved.
The model predictions always contain uncertainty. But in some cases, such as weather forecasting...
Citation