Extending MESMER-X: a spatially resolved Earth system model emulator for fire weather and soil moisture

Quilcaille, Yann; Gudmundsson, Lukas; Seneviratne, Sonia I.

doi:https://doi.org/10.5194/esd-14-1333-2023

Articles | Volume 14, issue 6

https://doi.org/10.5194/esd-14-1333-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/esd-14-1333-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 14, issue 6

Research article

|

21 Dec 2023

Research article |

| 21 Dec 2023

Extending MESMER-X: a spatially resolved Earth system model emulator for fire weather and soil moisture

Yann Quilcaille, Lukas Gudmundsson, and Sonia I. Seneviratne

Download

Final revised paper (published on 21 Dec 2023)
Preprint (discussion started on 17 Apr 2023)

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2023-589', Claudia Tebaldi, 17 May 2023

Please see uploaded PDF.

Citation: https://doi.org/10.5194/egusphere-2023-589-RC1
- AC1:
  'Reply on RC1', Yann Quilcaille, 07 Jul 2023
  We would like to sincerely thank Claudia Tebaldi for the open, positive and constructive review. Her and the comments of the other Referee were completely integrated to the manuscript, which we believe has improve its quality.
  We detail in this document the modifications brought to the manuscript. Referees' comments are shown in black. The authors' response is shown in green text. The text quoted from the manuscript is shown between quotation marks in italics. Numbers of lines correspond to the version including tracked changes.
  
  Summary of modifications:
  Corrections in the main text following Referees’ recommendations, as detailed in the responses to the Referees.
  
  Updated figures 1, 4, 8 & 11 (selection of configuration): reversed colormap on CRPS, labels on color bars and bigger font sizes
  
  New appendices:
  1: Application of the Probability Integral Transform to discrete distributions
  
  2: Representation of the interannual variability
  
  3: Interpretability of the CRPS
  
  4: Maps of negative log likelihood for the retained configurations
  
  5 to 6.8: Equivalent of figures 2, 5, 9, 12 but for SSP1-2.6 and SSP2-4.5
  
  All details of the reply are provided in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2023-589-AC1
RC2:
'Comment on egusphere-2023-589', Anonymous Referee #2, 18 May 2023

Please see uploaded pdf.

Citation: https://doi.org/10.5194/egusphere-2023-589-RC2
- AC2:
  'Reply on RC2', Yann Quilcaille, 07 Jul 2023
  We would like to sincerely thank the Anonymous Referee 2 for the positive and constructive review. Its and Claudia Tebaldi’s comments were completely integrated to the manuscript, which we believe has improve its quality.
  We detail in this document the modifications brought to the manuscript. Referees' comments are shown in black. The authors' response is shown in green text. The text quoted from the manuscript is shown between quotation marks in italics. Numbers of lines correspond to the version including tracked changes.
  Summary of modifications:
  Corrections in the main text following Referees’ recommendations, as detailed in the responses to the Referees.
  
  Updated figures 1, 4, 8 & 11 (selection of configuration): reversed colormap on CRPS, labels on color bars and bigger font sizes
  
  New appendices:
  1: Application of the Probability Integral Transform to discrete distributions
  
  2: Representation of the interannual variability
  
  3: Interpretability of the CRPS
  
  4: Maps of negative log likelihood for the retained configurations
  
  5 to 6.8: Equivalent of figures 2, 5, 9, 12 but for SSP1-2.6 and SSP2-4.5
  
  All details on this reply are provided in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2023-589-AC2
- AC3:
  'Reply on RC2', Yann Quilcaille, 07 Jul 2023
  We would like to sincerely thank the Anonymous Referee 2 for the positive and constructive review. Its and Claudia Tebaldi’s comments were completely integrated to the manuscript, which we believe has improve its quality.
  We detail in this document the modifications brought to the manuscript. Referees' comments are shown in black. The authors' response is shown in green text. The text quoted from the manuscript is shown between quotation marks in italics. Numbers of lines correspond to the version including tracked changes.
  Summary of modifications:
  Corrections in the main text following Referees’ recommendations, as detailed in the responses to the Referees.
  
  Updated figures 1, 4, 8 & 11 (selection of configuration): reversed colormap on CRPS, labels on color bars and bigger font sizes
  
  New appendices:
  1: Application of the Probability Integral Transform to discrete distributions
  
  2: Representation of the interannual variability
  
  3: Interpretability of the CRPS
  
  4: Maps of negative log likelihood for the retained configurations
  
  5 to 6.8: Equivalent of figures 2, 5, 9, 12 but for SSP1-2.6 and SSP2-4.5
  
  All details on the reply are provided in the attached file.
  
  Citation: https://doi.org/10.5194/egusphere-2023-589-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Publish subject to minor revisions (review by editor) (01 Aug 2023) by Vivek Arora

Dear authors,

Thank for your thorough response to reviewers' comments. I am satisfied with your response. I have gone through the entire manuscript myself and I have few minor suggestions which I think will help improve the overall readability of the manuscript. Personally, I am extremely familiar with climate models yet I find equations 1 to 7 somewhat hard to follow because of the heavy statistics involved. I wasn't aware that emulators so heavily depend on statistical approaches. May I please suggest to add some additional information or simple language terminology for audience that is not familiar with emulators' terminology and helps follow equations 1 to 7.

-------------------------------------------------
Other minor comments

Line 103, what is timesteps t ? Is this the annual time step of the data from the comprehensive climate models?

Line 111 and 113. Vt is defined as "covariants" on line 112 and "changes in global climate" on line 113. I am able to follow line 113 but not sure what covariants mean in this context.

Line 123. It seems "innovations" is emulator terminology. I am unable to follow what "spatially correlated innovations" mean.

Line 132. "... we obtain the equivalent of normalized residuals ...". Residuals from what?

Line 160. "... require timeseries of anomalies in global climate". Do you mean anomalies from historical climate? If yes, what time period.

Equation (8). Unless, I missed it, it seems Y_s,t,n is not defined.

Similar to reviewers may I please suggest to describe in words what does CRPS actually quantifies.

Equation (10). Have the terms, mu, sigma, and xi (skewness?) been defined?

Please consider y-axis labels for Figure 2 (column c) and other similar figures.

It seems, the ESMs x regions are not defined/introduced.

Line 314. Please change Figure 4 to Figure 5.

Lines 316 and 317. This sentence about the HadGEM3-GC31-MM model refers to which figure.

Section 4. Please introduce units of soil moisture. My recollection is that soil moisture units are kg/m2. Yet in the figures the units of soil moisture are kg/m3. Please check. It also seems that you have included both liquid and frozen soil moisture in your analysis (hence the maximum on your colour scales goes to 5000). Can you please make this clear in the manuscript?

Section 4.2. In the context of lagged response of soil moisture to deltaT, I am wondering if precipitation could be factor here. Recall that higher warming also implies more precipitation so it is possible that higher precipitation in RCP 8.5 scenario is helping soil moisture not decrease as fast despite higher deltaT, compared to the other scenarios. If yes, please consider noting this.

In Figure 1, the darkest colours are associated with the highest values of CRPSS for the HadGEM3 model. Can you please clarify what does this implies? Does it mean that emulators fit this model the best?
-------------------------------------------------

I look forward to reading a revised version of your manuscript that takes into account reviewers' comments and my minor suggestions.

Many thanks,
Vivek

Hide

AR by Yann Quilcaille on behalf of the Authors (04 Aug 2023) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (review by editor) (01 Sep 2023) by Vivek Arora

AR by Yann Quilcaille on behalf of the Authors (19 Sep 2023) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (review by editor) (03 Oct 2023) by Vivek Arora

Dear authors,

Thank you for making the changes to your manuscript. Sorry, I have been going through your manuscript with a fine toothed comb and apologies for being so picky. Thank you for your patience with me.

I have few comments on your latest changes. I could have suggested these changes to be made at the proof stage and accepted the manuscript but when the changes are considered big they come back to the editor.

1.
Note the suggested changes in UPPERCASE letters below.

Here, 𝑟 REPRESENTS the ratio of geographical distance between points and a localization radius, and the next FEW paragraphs EXPLAIN how 𝛴 𝜈(𝑟) is obtained from the empirical covariance matrix.

2.
Note the suggested changes in UPPERCASE letters below. I changed "scape" to "scale".

For a normal distribution, the parameters 𝛼 introduced in equation (1) are the location and scale, written respectively 𝜇 and 𝜎 in Figure 1, corresponding to the mean and standard deviation of the distribution. For a GEV distribution, the parameters 𝛼 are the location, SCALE, and shape, written respectively 𝜇 , 𝜎 and 𝜉 in Figure 1.

3.
In the following sentence

Here, the parameters 𝛼 introduced in equation (1) are the rate 𝜆 and a shift 𝜇 . The training of the distribution gains in freedom using this shift of the distribution by 𝜇 , with its mean becoming 𝜇 +𝜆 , while the variance remains 𝜆

you have defined 𝜆 as both a rate and as variance. Can you please revise this sentence? In the usual Poisson distribution the minimum value is zero, and the mean and standard deviation are 𝜆. When you shift it by 𝜇 does the minimum value become 𝜇?

Here's one suggestion. Not sure if this satisfies what you intended to describe. Please check.

Here, the parameters 𝛼 introduced in equation (1) for the Poisson distribution are 𝜆 and a shift 𝜇 in mean . This modified form of the Poisson distribution has the same variance (𝜆) as the original Poisson distribution but the mean is 𝜆+𝜇 (as opposed to 𝜆), and the minimum value is 𝜇 (as opposed to zero).

Many thanks,
Vivek

Hide

AR by Yann Quilcaille on behalf of the Authors (06 Oct 2023) Author's response Author's tracked changes Manuscript

ED: Publish as is (17 Oct 2023) by Vivek Arora

AR by Yann Quilcaille on behalf of the Authors (24 Oct 2023)

Short summary

Climate models are powerful tools, but they have high computational costs, hindering their use in exploring future climate extremes. We demonstrate MESMER-X, the only existing emulator for spatial climate extremes (heatwaves, fires, droughts) that mimics all of their relevant properties. Thanks to its negligible computational cost, MESMER-X may greatly accelerate the exploration of future climate extremes or enable the integration of climate extremes in economic and financial models.