the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Nov 2025
A warming adjustment method for CORDEX RCM simulations
Abstract. Regional climate model (RCM) projections provide high-resolution information essential for climate impact assessments and adaptation planning. Currently available Coordinated Regional Climate Downscaling Experiment (CORDEX) simulations, driven by CMIP5 global models, exhibit considerably weaker European warming throughout the 21st century compared to the latest Coupled Model Intercomparison Phase 6 (CMIP6) models. This discrepancy arises from multiple factors: RCMs tend to underestimate warming compared to their driving models in part due to the use of aerosol climatologies, while CMIP6 models themselves exhibit higher climate sensitivity than their CMIP5 predecessors. Here, we present a method to adjust existing CORDEX simulations toward the large-scale warming simulated by CMIP6 models while preserving fine-scale spatial structure and physical coherence. The method operates by reassembling complete annual fields from the original simulations according to smoothed temperature trajectories, maintaining temporal monotonicity without requiring interpolation. We demonstrate its application to both temperature and precipitation, including mean conditions and extremes over Europe, though the approach is applicable to any regional domain and warming-sensitive climate variable. As an extension, the method can be applied sequentially to establish consistency from prescribed global warming levels through CMIP6 regional patterns to high-resolution RCM projections. This bridges the temporal gap between regional and global model development cycles, making existing high-resolution climate information compatible with both updated model generations and specific warming targets.
Competing interests: Sonia I. Seneviratne is a member of the editorial board of Earth System Dynamics. The authors declare that they have no other competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(11813 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-4968', Anonymous Referee #1, 18 Dec 2025
-
RC2: 'Comment on egusphere-2025-4968', Anonymous Referee #2, 24 Dec 2025
Comments on Börsig et al., eguspere-2025-4968
General comments
This paper presents a method to adjust CMIP5-based RCM simulations to better align with CMIP6 GCM simulations. I perfectly see the value in combining or merging scenario generations, I also see the need of speeding up the process of downscaling CMIP ensembles with CORDEX RCMs. The number of people struggling with these issues is quite large.
As I see it this study build on a premise that is, if not false, at least no proven. You assume that CMIP5-based RCM simulations can be related to CMIP6 GCMs in a consistent way. There are a few problems with this, however. 1) The CMIP5 and the CMIP6 models are not the same, and they have different climate sensitivity. You state this yourself in the paper. 2) RCP8.5 and SSP5-8.5 do not prescribe the same GHG concentrations (Wyser et al., 2020). 3) In the reference period you use, 1991-2020, the scenario period starts 2006 in RCP8.5 and 2016 in SSP5-8.5, thus the proportion of observed and assumed emissions is different.
Taken together, this raises a few questions. If the forcing is different and the models are different, why should we expect the results to be the same, and why is it motivated to adjust CMIP5-based RCMs to better fit CMIP6 GCMs? It can be suggested that the CMIP6 scenarios are better because they are newer, but when you decouple GHG forcing and climate response, as I see that you do with the RCM simulations, the physical consistency breaks down.
I might have missed something, but for this paper to be published I think these questions need to be properly explained.
Specific comments
L10 “physical coherence” I don’t see how this is achieved. See general comments above.
L31-31 It would be good to specify that this statement applies to Europe (specifically west central Europe). The aerosol forcing problem is not a problem in northern Europe. I don’t know how it is in the rest of the world.
L41 and other occasions. Please specify that you are talking about CMIP5-based RCM (and not GCM) simulations here, because it explains some of the differences to CMIP6 GCMs.
L44-46 Another reason for this could be that RCMs better resolve some processes, and that the results from RCMs actually are better. Are all differences between CMIP5-based RCMs and CMIP6 GCMs a proof that the GCMs are better?
L63 Again, CMIP5-based EURO-CORDEX simulations.
L64-66 I find this sentence a bit awkward, unecessarily long. Consider rephrasing to something like: “CMIP5-based EURO-CORDEX simulations as 0.11 degree horizontal resolution, the best available”.
L71-77 See general comments. I would argue that there are reasons for at least some of these differences, and that this not necessarily mean that one ensemble is wrong. It’s just that they are based on different data and different models.
Fig 1a “CMIP6-CORDEX” could mean CMIP6-based RCM simulations. Consider using “CMIP6 GCMs – CMIP5 RCMs” or something like that.
Fig 1b “C6” -> “CMIP6”
Fig 1c.3 It would be helpful if the increments are the same on the x-axis and the y-axis.
Fig 1c.3 The text mentions “grey dots”. The grey dots I see are the vertical lines, which I don’t think are referred to. I think the text refers to the grey dots that look like a line. Consider using different colours.
Fig 1c.4 What is the blue line, and the thin orange line? Add to legend.
L151 Does this approach assume a constant relationship between global and local warming? It’s not clear to me.
Sec. 2.4 Considering the lengthy description of Fig 3 here, should the ordering of the figures be changed so that Fig 3 comes before Fig 2? That seem logical to me. Fig 2 is only briefly mentioned before this.
L162 Please explain the “IPCC GMT trajectories” more. Where do they come from, and what are they based on? How is it different from the CMIP6 GCM ensemble?
L164 “chatateristics” -> “characteristics”?
L165 “trajectoeies” -> “trajectories”?
L192 “intermediate behaviour” feels like an awkward formulations. Consider rephrasing.
L194 I don’t agree with the use of “modifications” here. Modifications are something you do, not a result of your method, nor a part of the climate system.
L201-211 These paragraphs seem a bit redundant. Was this not explained already in sec 2.4? I don’t see any real results here. Consider to move to, and merge with, 2.4.
L223 I don’t fully understand how you did the adjustment of TXx. Is the adjustment made on TXx? I thought that only annual Tm was adjusted, and don’t see how TXx could be derived from that.
Fig 3b It would be more logical to put IPCC range and IPCC mean together in the legend.
Fig 3c Consider using colours that are more different and to use dashes and dots.
Fig 4a and 4b Could be more clear that “IPCC” in the title and legend actually means RCM simulations.
Fig 4 caption “responsedatasets” -> “response data sets”?
L240 What are modifications if not adjustments? I don’t think that you need to demonstrate this, I think you should demonstrate how your modifications effect the climate change signal.
L243 Again, it sounds a bit strange to say that adjustments led to modifications. They are the same to me. I see what you want to say. Isn’t it enough to just say: “The adjustments produce shifts ...”?
L244 “These outcomes reflect an additional information layer ...” I don’t understand this sentence.
L246 “These findings underscore...” Do they? In what way? I thought they showed the importance of adjustments.
L250-251 How useful for adaptation planning and impact modelling are the new annual ensemble means over large areas that you produce here? They usually require higher spatial and temporal resolutions.
L255 “replace CMIP6-CORDEX” Isn’t the question rather whether they should replace CMIP5-CORDEX RCMs?
Fig A1 caption “like Fig, 1a. (b) Like (a), but for” is a bit difficult to understand. Please rephrase.
References
Wyser, K. et al., 2020 :Warmer climate projections in EC-Earth3-Veg: the role of changes in the greenhouse gas concentrations from CMIP5 to CMIP6, ERL, 05402 5, 15, 1748-9326, https://doi.org/10.1088/1748-9326/ab81c2
Citation: https://doi.org/10.5194/egusphere-2025-4968-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 280 | 106 | 29 | 415 | 21 | 17 |
- HTML: 280
- PDF: 106
- XML: 29
- Total: 415
- BibTeX: 21
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of “A warming adjustment method for CORDEX RCM simulations” by Stefanie Börsig, Dominik L. Schumacher, and Sonia I. Seneviratne
In this paper, the authors propose a novel approach to combine the results of EURO-CORDEX regional climate models (RCMs), which are forced by CMIP5 global climate models (GCMs), with the results of CMIP6 global climate models, as constrained by observations in the IPCC AR6, to provide more accurate and up-to-date regional climate information. The RCMs indeed tend to underestimate regional warming compared to global climate models (both CMIP5 and CMIP6), as documented in the literature. However, they still provide useful high-resolution climate information over Europe.
The authors aim to produce regional climate transient projections that are consistent with global climate models and, ultimately, with the constrained CMIP6 projections proposed by the IPCC. This involves a two-step process where global projections are first matched with the constrained IPCC projections, and then the regional projections are matched with the global projections.
The method consists of stitching together complete annual fields from the RCM simulations trying to temporally match the warming simulated by the GCMs. This approach ensures spatial and inter-variable consistency.
The authors describe the method and then illustrate its interest, for changes in temperature, precipitation, and their extremes.
This paper presents an innovative and practical methodology with clear potential applications, for climate services for example. While the paper is interesting and generally well-written, several aspects of the methodology lack sufficient clarity, making them difficult to follow. Additionally, the discussion of the method’s limitations remains underdeveloped.
Major revisions are required to address these issues before the paper could be published.
Major comments
The method is not always clearly explained, and the details are sometimes difficult to follow. The authors should therefore improve the description of the method (see the minor comments for the points where improvements are particularly necessary).
The limitations of the method concerning snow cover, and the estimation of cold, dry, or wet spells in winter, which could be disrupted when years are reassembled, should be better discussed and illustrated. For example, snow cover in January will have nothing to do with the snow cover from the previous December, which could be a major issue for some applications. Also, since some RCM years can be repeated, what is the impact on temporal variance at different time scales: daily, seasonal, and decadal? The authors should analyze at least some of these limitations, and discuss all of them thoroughly.
As I understand, the authors use standard calendar years (January to December), but it would make more sense to at least start the year in December so that the winter averages in the regional projection retain climatological meaning (i.e., to avoid stitching December to January and February from another winter). At the very least, the snow season would not be artificially split. More generally, I think that for many applications, using "hydrological years" (e.g., years starting in October for example) would help avoid issues related to snow and reduce problems related to soil moisture. What is the rationale for using standard years?
The method also makes the implicit assumption that the regional warming of GCMs over Europe is accurate (for a given level of global warming), but an increasing number of studies suggest this is not the case, i.e. that GCMs themselves tend to underestimate warming over Europe. The method proposed in this paper corrects for the fact that CMIP6 models, as an ensemble, may overestimate global warming, but it does not account for the possibility that they underestimate warming over Europe. As a result, the global "cold" correction may amplify the regional underestimation of warming in the final adjusted projections. This should be noted and discussed.
Minor comments
l48-52. This paragraph is not very clear. It could be useful to better explain the method of Tebaldi et al. (2022), as it is used here and at other places as a reference for the explanation of the authors’ method. For example, "harmonizing existing high-resolution regional climate projections with updated global and regional warming constraints" is not clear or precise.
Introduction: I think it would be useful to cite Ribes et al. (2022), who show another line of evidence for the underestimation of regional warming over Europe (France in that case) by RCMs (and the GCMs). Also, it would be interesting to discuss how climate services across Europe try to deal with the issue of reconciling global and regional projections (e.g., Corre et al., 2025, for France). I also think that the approach of Corre et al. has the advantage of taking into account the underestimation of regional warming over Western Europe by all models (global and regional), compared to the approach proposed by the authors. This could also be discussed in the conclusion, for example.
Corre et al. (2025) Using regional warming levels to describe future climate change for services and adaptation: Application to the French reference trajectory for adaptation. Climate Services
Ribes et al. (2022) An updated assessment of past and future warming over France based on a regional observational constraint. Earth Syst. Dyn
line 72. SSP5-8.5 and RCP8.5 are not exactly identical, which can lead to some differences. The authors should discuss this and cite relevant literature.
l82. The authors should provide the exact properties of the locally weighted regression: which order is used? (e.g., linear or quadratic?).
l87. Which markers? They are hard to see in the figure. Also, I don’t see repeated years at the beginning of the mapping. The RCM years seem to increase from the beginning to 2075, so why are they filtered (in grey) at the beginning? The authors should clarify this point.
l89-92. Not clear. Also, I don’t understand exactly how the linearization works or why a long window of 20 years is needed. Also, the evolution of years in Fig 3c3 does not seem totally linear over 20 year periods. The sentence on line 91 that justifies this is not clear. And is there not a risk of inaccuracy by forcing all the 20-year lines to connect. ? Additionally, the regression does not provide entire years I suppose, but years with decimals (e.g., 1998.54). Please explain how this is dealt with.
Figure 1c (panels 3 and 4). It is better to avoid placing the legend over the curves.
l102. Is it always possible to enforce a 0.75 slope in the GCM-RCM mapping? If the RCM warms much slower than its driving GCM, how is it possible to maintain a 0.75 slope? What happens if it is not the case?
l104-107. Not clear. And what is the scientific justification for this? Why do the authors avoid consecutive duplicate years if it is the best choice to match the GCM?
Near l125. The limitations of the method could be better explained. If I understand correctly, 31/12/2012 could be followed by 01/01/2011 or 01/01/2013. In any case, there is a discontinuity, with possible large and unphysical shifts in snow cover (as we are in winter). This should be more clearly mentioned (rather than referring vaguely to “memory issue”). Would it not have been preferable to start years in fall, for example, to limit this issue with snow cover (and soil moisture) i.e. to use "hydrological years"? Also, if someone is interested in cold, wet, or dry spells in winter, this will be problematic. This should be noted.
I’m not sure to understand the point of section 2.3, as no result of the “validation” described is shown.
l130. Is it not an issue to have to truncate the adjusted time series? Information from the same ensemble of RCMs cannot be used consistently through time periods because each regional climate simulation has to be truncated at a different time. Is this not an issue, for example, when characterizing model uncertainties, as the ensemble of models considered changes with time? Please also provide a table (or a figure) with the last (and first) year of each adjusted RCM simulation.
For example, how can the adjusted time series in Figure 2 be extended to 2100? In Figure 4, when the text refers to 2060, what is the exact time period represented? Is it for example a 20-year average (e.g., 2050–2070). Are all the adjusted regional projections available until 2070, or is there a cutoff earlier for some simulations? Is it not an issue?
l136. What is the size of the window to compute the slope here? Is it 20 years also? Are they running windows?
Figure 1. The numbering is inconsistent, with panels labeled as a, b, and then c1, c2, c3, etc.
L150–170. It is not really clear how the adjustment for global warming is done. The explanation should be detailed, explained step by step.
Also, when the adjustment is applied to the 95th percentile from the IPCC estimate, is the ensemble mean of CMIP6 models adjusted to the 95th percentile? If so, the resulting multi-model distribution extends well above the IPCC 95th percentile, which does not really make sense, I think. Am I misunderstanding something?
Note: Figure 3 is not particularly clear (especially 3a). Also, please add the ensemble spread in Figure 3c with shading.
Conclusions. Please discuss the limitations of the method more thoroughly in the conclusion.
I also found quite a few typos. A careful proofreading prior to the submission of the new version of the manuscript could be beneficial.
“chatateristics” (l164)
“trajectoeies” l165
“warming-adjustement” (Fig. 3 caption)
“tragectory” (Fig. 3 caption)
“warming responsedatasets"” (Fig. 4 Caption)
“temperatues” (Fig. 4 caption)
“unter” (Fig. A1 caption)
Figure A1, title: 20250 instead of 2050
“componentn” (Fig. A2 caption)