the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
An evaluation of the LLC4320 global ocean simulation based on the submesoscale structure of modeled sea surface temperature fields
Katharina Martha Gallmeier
J. Xavier Prochaska
Dimitris Menemenlis
Madolyn Kelm
Abstract. We have assembled 2,851,702 nearly cloud-free cutout images (sized 144×144 km2) of Sea Surface Temperature (SST) data from the entire 2012–2020 Level-2 Visible Infrared Imaging Radiometer Suite (VIIRS) dataset to perform a quantitative comparison to the ocean model output from the MIT general circulation model (MITgcm). Specifically, we evaluate outputs from the LCC4320 global-ocean simulation for a one-year period starting on November 17, 2011 but otherwise matched in geography and day-of-year to the VIIRS observations. In lieu of simple (e.g., mean, standard deviation) or complex (e.g., power spectrum) statistics, we analyze the cutouts of SST anomalies with an unsupervised Probabilistic AutoEncoder (PAE) trained to learn the distribution of structures in SST anomaly (SSTa) on ~10-to-80-km scales (i.e., submesoscale-to-mesoscale). A principal finding is that the LLC4320 simulation reproduces well, over a large fraction of the ocean, the observed distribution of SST patterns, both globally and regionally. Globally, the medians of the structure distributions match to within 2σ for 65 % of the ocean, despite a modest, latitude-dependent offset. Regionally, the model outputs reproduce mesoscale variations in SSTa patterns revealed by the PAE in the VIIRS data, including subtle features imprinted by variations in bathymetry. We also identify significant differences in the distribution of SSTa patterns in several regions: (1) in the vicinity of the point at which western boundary currents separate from the continental margin, (2) in the Antarctic Circumpolar Current (ACC), especially in the eastern half of the Indian Ocean, and (3) in an equatorial band equatorward of 15°. It is clear that (1) is a result of premature separation in the simulated western boundary currents. The model output in (2), the Southern Indian Ocean, tends to predict more structure than observed, perhaps arising from a misrepresentation of the mixed layer or of energy dissipation and stirring in the simulation. The differences in (3), the equatorial band, are also likely due to model errors, perhaps arising from the shortness of the simulation or from the lack of high-frequency/wavenumber atmospheric forcing. Although we do not yet know the exact causes for these model-data SSTa differences, we expect that this type of comparison will help guide future developments of high-resolution global-ocean simulations.
- Preprint
(22446 KB) - Metadata XML
- BibTeX
- EndNote
Katharina Martha Gallmeier et al.
Status: closed
-
CEC1: 'Comment on gmd-2023-39', Astrid Kerkweg, 20 Apr 2023
Dear authors,
reading you code and data availability section I encountered the sentence
"Code and data availability. All of the data generated and analyzed in this manuscript is publicly available as parquet tables and hdf5 files
at Dryad (LINK TO APPEAR). [...]"So what is "Dryad (LINK TO APPEAR)" ?
The missing information should be provided asap, i.e., definitely before the end of the discussion phase.
Best regards, Astrid Kerkweg (GMD Executive Editor)
Citation: https://doi.org/10.5194/gmd-2023-39-CEC1 -
AC1: 'Reply on CEC1', Katharina Gallmeier, 25 Apr 2023
Dryad is a data archiving serviced provided by libraries in the U.S.
We don't believe we can edit what it served there. Therefore, we
want to wait until the paper is finalized before archiving these data.
In case the reviewer comments require us to make changesCitation: https://doi.org/10.5194/gmd-2023-39-AC1
-
AC1: 'Reply on CEC1', Katharina Gallmeier, 25 Apr 2023
-
RC1: 'Review on gmd-2023-39', Takaya Uchida, 24 Apr 2023
-
AC2: 'Reply on RC1', Katharina Gallmeier, 17 May 2023
Dear Dr. Uchida,
We thank you for your feedback and suggestions! We are still in the process of making changes to our manuscript but would like to get back to you on a few of your comments.
- Your comment for lines 46-47 has been noted. We revised the wording as suggested and added a reference to Leroux, et al., 2022, which seems to be the most appropriate.
- Your comment for lines 285-287 has been noted. We added text to the beginning of section 4.2.2 titled “Differences”, in order to clarify. However, we defer the details on where the thresholds arise to Appendix A.
- Your comment for line 324 and several figures has been noted. The dT is equivalent. Thank you for pointing that out. We have updated the figures and text, which will be seen in the final manuscript version.
- Your comment for lines 223-224 and 375-376 has been noted. We added a short paragraph at the end of the Southern Ocean discussion in Section 4.2.2 Differences and will be shown in the final manuscript version.
“One possible explanation for more energetic SST fields in LLC4320 relative to VIIRS is the fact that LLC4320 was inadvertently forced with tidal potential that is 10% more energetic than the real ocean (Yu et al., 2019; Arbic et al., 2022) resulting in, for example, more energetic internal tides than observed. A second, in our opinion more likely, explanation is that the 1/48-deg horizontal grid spacing of the simulation, although it is sufficient to enable the gravest modes of mixed layer instabilities to be captured, is insufficient to fully represent the smaller-scale instabilities that would damp the magnitude of the resolved instabilities.”
5. Your comment for lines 317-318 has been noted. We added a short paragraph at the end of the Equatorial band discussion in Section 4.2.2 Differences.
“Another possible explanation for lack of SST structure in the equatorial region is the limited spatiotemporal resolution of the prescribed atmospheric forcing (0.14deg and six-hourly). A recently-completed coupled ocean-atmosphere simulation (Torres et al., 2022; Light et al., 2022; Dushaw and Menemenlis, 2023) provides high-frequency/wavenumber interactive forcing (0.0625deg and 45 seconds) to the ocean. It would be interesting to repeat the present analysis on this simulation to determine whether the equatorial SST structures of this coupled simulation are more similar to VIIRS.”
Citation: https://doi.org/10.5194/gmd-2023-39-AC2 -
AC4: 'Reply on RC1', Katharina Gallmeier, 18 May 2023
To add on, we would like to get back to you on your question for lines 265-272.
In Section 4.2.1 titled “Similarities”, we do not think there is a reason for concern that correcting the LL_LLC will lead to an unfair analysis. The term “correcting” gives off the wrong impression. We hope to emphasize that correcting the LL_LLC values means we are just subtracting the spatial average of LL_LLC values and adding to it the spatial average of LL_VIIRS values. In other words, we're trying to remove these "large-scale" differences to see how similar the submesoscale structure is in between the datasets. This altered view of LL_LLC values is only used in the analysis for that section. The rest of the discussion (aka. Section 4.2.2) is not using the corrected LL_LLC. We also include a link here to the HEALPix software: https://healpix.sourceforge.io/index.php.
Citation: https://doi.org/10.5194/gmd-2023-39-AC4
-
AC2: 'Reply on RC1', Katharina Gallmeier, 17 May 2023
-
RC2: 'Comment on gmd-2023-39', Anonymous Referee #2, 26 Apr 2023
Review of “An evaluation of the LLC4320 global ocean simulation based on the submesoscale structure of modeled sea surface temperature fields” by Gallmeier et al.
This work uses an interesting and distinct method based on AI to evaluate the performance of the submesoscale-permitting global model LLC4320. Different from traditional evaluation methods, the method is novel and related to AI. Also, this paper involves a lot of work and lots of data need to be processed and analyzed. Overall, I believe this paper is suitable for publication in the journal GMD. Anyway, I have some minor comments addressed below.
1) Title: after reading the title, my first expression is the important role of submesoscale structure that resolved in the model in the evaluation. However, after reading the abstract, I cannot find any statements that highlight the role of submesoscale. Could the authors make it clearer about this point?
2) Line 10-20: here, the authors introduce the regional differences in the order of Gulf Stream, ACC, and Equatorial regions. But in the main text, the authors actually analyze them in Equatorial regions first, then ACC, and lastly Gulf Stream. To make it consistent, I would suggest the authors change the order here.
3) Line 220: is it possible to link the LL values to physical dynamic scales here? What does LL = -375 correspond to a spatial scale?
4) Line 260: the equation should be numbered. This is also true for the equations below.
5) Figure 10: is there any reason for the authors to choose the two regions to highlight the differences in the Equatorial region and the Southern ocean? It seems that these two are not the regions with the largest differences.
6) Line 315: could the authors give more specific explanations about the argument here? What is the possible reason for the model failing to reproduce submesoscale-to-mesoscale structure well?Citation: https://doi.org/10.5194/gmd-2023-39-RC2 -
AC3: 'Reply on RC2', Katharina Gallmeier, 17 May 2023
Dear Anonymous Referee,
We thank you for your feedback and suggestions as well. We again are in the process of making changes to the manuscript. For now, we would like to inform you of our current status for resolving your comments.
- Your comment on the title has been noted. We are of the belief we sufficiently addressed the role of submesoscale structure in our manuscript. Ulmo –our machine learning algorithm– was trained to identify submesocale SSTa patterns in cutouts. This was achieved by restricting the cutout size to ~100-100 km^2, thus forcing Ulmo to detect structure on the order of 10-80 km scales. We state this in Section 3.1 Creation of Comparable SSTa cutouts: “The size of these samples was chosen in part to focus on features at scales of ~30 km or smaller.” Line {108-109}. We also include in the introduction a reason for motivation to evaluate based on submesoscale structure, since it has not to our knowledge been done before for global, free-running OGCMs. Lines {52-53}
- Your comments for lines 10-20 have been noted. Changes have been made to the regional difference ordering in the introduction to match the manuscript's content. We thank you for your attention to detail.
- Your comment for line 260 has been noted. We numbered the equations, which will be seen in the final version of the manuscript. Thank you for calling us out on that.
- Your comment for Figure 10 has been noted. We added text to the introduction to clarify that this manuscript’s primary purpose is to present a relatively new statistical method in evaluating OGCMs. Thus, we do not provide an exhaustive evaluation of the LLC4320 SST outputs. In the case you mentioned, we investigated each of the groupings we mentioned for Figure 10, instead of investigating the largest differences. These largest LL differences were found to be around boundary currents. These boundary currents are of a dynamic nature that leads to lower LL values. Thus, slight differences in the current’s position or deviation from its dynamic nature makes those regions, in terms of LL differences, appear more drastic. We wanted to also investigate, in your language, the smaller LL differences (though still greater than -197/197) to showcase that it also picks up different SSTa patterns between LLC and VIIRS. We thank you for your observation and for the opportunity to clarify.
Citation: https://doi.org/10.5194/gmd-2023-39-AC3
-
AC3: 'Reply on RC2', Katharina Gallmeier, 17 May 2023
Status: closed
-
CEC1: 'Comment on gmd-2023-39', Astrid Kerkweg, 20 Apr 2023
Dear authors,
reading you code and data availability section I encountered the sentence
"Code and data availability. All of the data generated and analyzed in this manuscript is publicly available as parquet tables and hdf5 files
at Dryad (LINK TO APPEAR). [...]"So what is "Dryad (LINK TO APPEAR)" ?
The missing information should be provided asap, i.e., definitely before the end of the discussion phase.
Best regards, Astrid Kerkweg (GMD Executive Editor)
Citation: https://doi.org/10.5194/gmd-2023-39-CEC1 -
AC1: 'Reply on CEC1', Katharina Gallmeier, 25 Apr 2023
Dryad is a data archiving serviced provided by libraries in the U.S.
We don't believe we can edit what it served there. Therefore, we
want to wait until the paper is finalized before archiving these data.
In case the reviewer comments require us to make changesCitation: https://doi.org/10.5194/gmd-2023-39-AC1
-
AC1: 'Reply on CEC1', Katharina Gallmeier, 25 Apr 2023
-
RC1: 'Review on gmd-2023-39', Takaya Uchida, 24 Apr 2023
-
AC2: 'Reply on RC1', Katharina Gallmeier, 17 May 2023
Dear Dr. Uchida,
We thank you for your feedback and suggestions! We are still in the process of making changes to our manuscript but would like to get back to you on a few of your comments.
- Your comment for lines 46-47 has been noted. We revised the wording as suggested and added a reference to Leroux, et al., 2022, which seems to be the most appropriate.
- Your comment for lines 285-287 has been noted. We added text to the beginning of section 4.2.2 titled “Differences”, in order to clarify. However, we defer the details on where the thresholds arise to Appendix A.
- Your comment for line 324 and several figures has been noted. The dT is equivalent. Thank you for pointing that out. We have updated the figures and text, which will be seen in the final manuscript version.
- Your comment for lines 223-224 and 375-376 has been noted. We added a short paragraph at the end of the Southern Ocean discussion in Section 4.2.2 Differences and will be shown in the final manuscript version.
“One possible explanation for more energetic SST fields in LLC4320 relative to VIIRS is the fact that LLC4320 was inadvertently forced with tidal potential that is 10% more energetic than the real ocean (Yu et al., 2019; Arbic et al., 2022) resulting in, for example, more energetic internal tides than observed. A second, in our opinion more likely, explanation is that the 1/48-deg horizontal grid spacing of the simulation, although it is sufficient to enable the gravest modes of mixed layer instabilities to be captured, is insufficient to fully represent the smaller-scale instabilities that would damp the magnitude of the resolved instabilities.”
5. Your comment for lines 317-318 has been noted. We added a short paragraph at the end of the Equatorial band discussion in Section 4.2.2 Differences.
“Another possible explanation for lack of SST structure in the equatorial region is the limited spatiotemporal resolution of the prescribed atmospheric forcing (0.14deg and six-hourly). A recently-completed coupled ocean-atmosphere simulation (Torres et al., 2022; Light et al., 2022; Dushaw and Menemenlis, 2023) provides high-frequency/wavenumber interactive forcing (0.0625deg and 45 seconds) to the ocean. It would be interesting to repeat the present analysis on this simulation to determine whether the equatorial SST structures of this coupled simulation are more similar to VIIRS.”
Citation: https://doi.org/10.5194/gmd-2023-39-AC2 -
AC4: 'Reply on RC1', Katharina Gallmeier, 18 May 2023
To add on, we would like to get back to you on your question for lines 265-272.
In Section 4.2.1 titled “Similarities”, we do not think there is a reason for concern that correcting the LL_LLC will lead to an unfair analysis. The term “correcting” gives off the wrong impression. We hope to emphasize that correcting the LL_LLC values means we are just subtracting the spatial average of LL_LLC values and adding to it the spatial average of LL_VIIRS values. In other words, we're trying to remove these "large-scale" differences to see how similar the submesoscale structure is in between the datasets. This altered view of LL_LLC values is only used in the analysis for that section. The rest of the discussion (aka. Section 4.2.2) is not using the corrected LL_LLC. We also include a link here to the HEALPix software: https://healpix.sourceforge.io/index.php.
Citation: https://doi.org/10.5194/gmd-2023-39-AC4
-
AC2: 'Reply on RC1', Katharina Gallmeier, 17 May 2023
-
RC2: 'Comment on gmd-2023-39', Anonymous Referee #2, 26 Apr 2023
Review of “An evaluation of the LLC4320 global ocean simulation based on the submesoscale structure of modeled sea surface temperature fields” by Gallmeier et al.
This work uses an interesting and distinct method based on AI to evaluate the performance of the submesoscale-permitting global model LLC4320. Different from traditional evaluation methods, the method is novel and related to AI. Also, this paper involves a lot of work and lots of data need to be processed and analyzed. Overall, I believe this paper is suitable for publication in the journal GMD. Anyway, I have some minor comments addressed below.
1) Title: after reading the title, my first expression is the important role of submesoscale structure that resolved in the model in the evaluation. However, after reading the abstract, I cannot find any statements that highlight the role of submesoscale. Could the authors make it clearer about this point?
2) Line 10-20: here, the authors introduce the regional differences in the order of Gulf Stream, ACC, and Equatorial regions. But in the main text, the authors actually analyze them in Equatorial regions first, then ACC, and lastly Gulf Stream. To make it consistent, I would suggest the authors change the order here.
3) Line 220: is it possible to link the LL values to physical dynamic scales here? What does LL = -375 correspond to a spatial scale?
4) Line 260: the equation should be numbered. This is also true for the equations below.
5) Figure 10: is there any reason for the authors to choose the two regions to highlight the differences in the Equatorial region and the Southern ocean? It seems that these two are not the regions with the largest differences.
6) Line 315: could the authors give more specific explanations about the argument here? What is the possible reason for the model failing to reproduce submesoscale-to-mesoscale structure well?Citation: https://doi.org/10.5194/gmd-2023-39-RC2 -
AC3: 'Reply on RC2', Katharina Gallmeier, 17 May 2023
Dear Anonymous Referee,
We thank you for your feedback and suggestions as well. We again are in the process of making changes to the manuscript. For now, we would like to inform you of our current status for resolving your comments.
- Your comment on the title has been noted. We are of the belief we sufficiently addressed the role of submesoscale structure in our manuscript. Ulmo –our machine learning algorithm– was trained to identify submesocale SSTa patterns in cutouts. This was achieved by restricting the cutout size to ~100-100 km^2, thus forcing Ulmo to detect structure on the order of 10-80 km scales. We state this in Section 3.1 Creation of Comparable SSTa cutouts: “The size of these samples was chosen in part to focus on features at scales of ~30 km or smaller.” Line {108-109}. We also include in the introduction a reason for motivation to evaluate based on submesoscale structure, since it has not to our knowledge been done before for global, free-running OGCMs. Lines {52-53}
- Your comments for lines 10-20 have been noted. Changes have been made to the regional difference ordering in the introduction to match the manuscript's content. We thank you for your attention to detail.
- Your comment for line 260 has been noted. We numbered the equations, which will be seen in the final version of the manuscript. Thank you for calling us out on that.
- Your comment for Figure 10 has been noted. We added text to the introduction to clarify that this manuscript’s primary purpose is to present a relatively new statistical method in evaluating OGCMs. Thus, we do not provide an exhaustive evaluation of the LLC4320 SST outputs. In the case you mentioned, we investigated each of the groupings we mentioned for Figure 10, instead of investigating the largest differences. These largest LL differences were found to be around boundary currents. These boundary currents are of a dynamic nature that leads to lower LL values. Thus, slight differences in the current’s position or deviation from its dynamic nature makes those regions, in terms of LL differences, appear more drastic. We wanted to also investigate, in your language, the smaller LL differences (though still greater than -197/197) to showcase that it also picks up different SSTa patterns between LLC and VIIRS. We thank you for your observation and for the opportunity to clarify.
Citation: https://doi.org/10.5194/gmd-2023-39-AC3
-
AC3: 'Reply on RC2', Katharina Gallmeier, 17 May 2023
Katharina Martha Gallmeier et al.
Model code and software
Python Code J. Xavier Prochaska, Katharina Gallmeier, and Madolyn Kelm https://doi.org/10.5281/zenodo.7545904
Katharina Martha Gallmeier et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
580 | 255 | 22 | 857 | 7 | 4 |
- HTML: 580
- PDF: 255
- XML: 22
- Total: 857
- BibTeX: 7
- EndNote: 4
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1