Journal cover Journal topic
Nonlinear Processes in Geophysics An interactive open-access journal of the European Geosciences Union
Journal topic

Journal metrics

IF value: 1.558
IF1.558
IF 5-year value: 1.475
IF 5-year
1.475
CiteScore value: 2.8
CiteScore
2.8
SNIP value: 0.921
SNIP0.921
IPP value: 1.56
IPP1.56
SJR value: 0.571
SJR0.571
Scimago H <br class='widget-line-break'>index value: 55
Scimago H
index
55
h5-index value: 22
h5-index22
Preprints
https://doi.org/10.5194/npg-2020-38
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/npg-2020-38
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

  25 Sep 2020

25 Sep 2020

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

Training a convolutional neural network to conserve mass in data assimilation

Yvonne Ruckstuhl1, Tijana Janjić1, and Stephan Rasp2 Yvonne Ruckstuhl et al.
  • 1Meteorological Institute Munich, Ludwig-Maximilians-Universität München, Germany
  • 2ClimateAi, San Francisco, USA

Abstract. In previous work, it was shown that preservation of physical properties in the data assimilation framework can significantly reduce forecasting errors. Proposed data assimilation methods, such as the quadratic programming ensemble (QPEns) that can impose such constraints on the calculation of the analysis, are computationally more expensive, severely limiting their application to high dimensional prediction systems as found in earth sciences. In order to produce from a less computationally expensive, unconstrained analysis, a solution that is closer to the constrained analysis, we propose to use a convolutional neural network (CNN) trained on analyses produced by the QPEns. In this paper, we focus on conservation of mass and show in an idealized setup that the hybrid of a CNN and the ensemble Kalman filter is capable of reducing analysis and background errors to the same level as the QPEns. To obtain these positive results, it was in one case necessary to add a penalty term to the loss function of the CNN training process.

Yvonne Ruckstuhl et al.

Interactive discussion

Status: open (until 20 Nov 2020)
Status: open (until 20 Nov 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

Yvonne Ruckstuhl et al.

Yvonne Ruckstuhl et al.

Viewed

Total article views: 294 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
247 45 2 294 4 3
  • HTML: 247
  • PDF: 45
  • XML: 2
  • Total: 294
  • BibTeX: 4
  • EndNote: 3
Views and downloads (calculated since 25 Sep 2020)
Cumulative views and downloads (calculated since 25 Sep 2020)

Viewed (geographical distribution)

Total article views: 223 (including HTML, PDF, and XML) Thereof 221 with geography defined and 2 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 

Cited

Saved

No saved metrics found.

Discussed

No discussed metrics found.
Latest update: 29 Oct 2020
Publications Copernicus
Download
Short summary
The assimilation of observations using standard algorithms can lead to violation of physical laws such as mass conservation, which has been shown to have detrimental impact on the system's forecast. We use a neural network (NN) to correct for this mass violation, using training data generated from expensive algorithms that can constrain such physical properties. We found in idealized setup that the NN can match the performance of these expensive algorithms, at neglectable computational costs.
The assimilation of observations using standard algorithms can lead to violation of physical...
Citation