Exploring precipitation pattern scaling methodologies and robustness among CMIP5 models

Kravitz, Ben; Lynch, Cary; Hartin, Corinne; Bond-Lamberty, Ben

doi:https://doi.org/10.5194/gmd-10-1889-2017

Articles | Volume 10, issue 5

https://doi.org/10.5194/gmd-10-1889-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/gmd-10-1889-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 10, issue 5

Methods for assessment of models

|

12 May 2017

Methods for assessment of models |

| 12 May 2017

Exploring precipitation pattern scaling methodologies and robustness among CMIP5 models

Ben Kravitz, Cary Lynch, Corinne Hartin, and Ben Bond-Lamberty

Download

Final revised paper (published on 12 May 2017)
Supplement to the final revised paper
Preprint (discussion started on 23 Nov 2016)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Pattern scaling', Anonymous Referee #1, 16 Dec 2016
- AC1: 'Response to reviewer #1', Ben Kravitz, 16 Feb 2017
RC2: 'Review for Kravitz et al', Anonymous Referee #2, 14 Jan 2017
- AC2: 'Response to reviewer #2', Ben Kravitz, 16 Feb 2017

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Ben Kravitz on behalf of the Authors (16 Feb 2017) Author's response Manuscript

ED: Referee Nomination & Report Request started (21 Feb 2017) by Olivier Marti

RR by Anonymous Referee #2 (04 Mar 2017)

Suggestions for revision or reasons for rejection

The paper is improved since the previous iteration. The exclusion of the 'physically based' metric takes away the most serious issue from the previous draft.

However, I'm still left with some concerns. Firstly - some of the paper still seems to be devoted into demonstrating the trivial - that the latter part of a 1pctCO2 simulation or RCP8.5 simulation can be used to reconstruct the same years from the same simulation. Why it would be expected for any scaling approach to produce anything other than a near-perfect reconstruction? As such, are these plots useful?

Given that the paper is written as a best-practice guidance for pattern scaling precipitation, it would be useful if the authors provided the reader more concise guidance in the abstract and in the main text on best practice. The authors show that predicting RCP2.6 from RCP8.5 is obviously stretching the approaches because the relative mix of forcings is so different at the end of the 21st century. How different do forcing mixes have to be before the methods are no longer useful (i.e. would it be acceptable to try and predict RCP4.5 from RCP8.5?). Which of the two methods is better in what circumstance? How many models do you need to make your pattern?

The assessment of inter-model prediction skill could potentially be useful, but it's under-explored at present. The 'group 1' and 'group 2' distinctions seem arbitrary - and in many cases contain models which are very closely related, meaning that they are questionable as independent samples. It's not clear at present why most of the paper is conducted with only one of these groups.

This is a potentially rich topic, and in future the authors could ask some relevant questions - like, for example, is there an optimal subset of models used to generate the pattern which minimizes errors? i.e. if there is one really terrible model in there, are you better to leave it in or take it out of your mean pattern?

It's interesting that the authors don't see the same relationship between number of models used and skill in reconstruction for the two methods. Some analysis of why this is the case would be useful.

The non-CO2/CO2 separation makes more sense than in the previous iteration, but the supplemental material requires some more clarification. In all RCPs, aerosols decline during the 21st century - if this is associated with a patterned precipitation response, then one would never expect this aerosol-driven pattern to be a simple function of global mean temperature. Rather, you would expect the pattern to scale with delta_T(non-GHG).

By deriving the quantities by regressing delta_T(non-GHG) against delta_T(CO2), the authors are hard-coding a fixed relationship between forcings. But, as the authors saw - there is no need for these quantities to be linearly related. It should be noted that the relationship isn't 'quadratic' or 'exponential', it's just how the relative forcings happen to change in the RCP8.5 scenario over the 21st century, and because these relative forcings are very different in RCP2.6 to RCP8.5, the method produces large errors.

It's also only fortuitous that delta_T(non-CO2) and delta_T(CO2) lie on a straight line in RCP8.5 (I suspect that this would not be true if the patterns were derived in RCP2.6). As such, it should be noted that the equation for P_nonCO2 in the additional material is not general - it is tied to a particular scenario.

Hide

RR by Anonymous Referee #1 (08 Mar 2017)

ED: Publish subject to minor revisions (Editor review) (29 Mar 2017) by Olivier Marti

AR by Ben Kravitz on behalf of the Authors (29 Mar 2017) Author's response Manuscript

ED: Publish as is (19 Apr 2017) by Olivier Marti

AR by Ben Kravitz on behalf of the Authors (19 Apr 2017)

Short summary

Pattern scaling is a way of approximating regional changes without needing to run a full, complex global climate model. We compare two methods of pattern scaling for precipitation and evaluate which methods is better in particular circumstances. We also decompose precipitation into a CO₂ portion and a non-CO₂ portion. The methodologies discussed in this paper can help provide precipitation fields for other models for a wide variety of scenarios of future climate change.