the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Apr 2026
A flexible methodology to evaluate natural variability in ClimaMeter
Abstract. Anthropogenic climate change (ACC) is critically influencing numerous extreme events worldwide, leading to the development of rapid attribution frameworks that allow for the timely evaluation of the role of ACC in changes in the frequency and intensity of specific extreme events. ClimaMeter (Faranda et al., 2024) is one of the tools recently developed to contextualise extreme weather events relative to climate change. ClimaMeter analyses extreme events shortly after they occur and leverages the analogue methodology for conditional attribution to evaluate whether and how events similar to the one analysed have changed in the recent climate. In order to attribute such changes to ACC, natural variability and its contribution must be quantified. In ClimaMeter, three modes of sea surface temperature variability are considered: the El Niño–Southern Oscillation, the Atlantic Multidecadal Oscillation, and the Pacific Decadal Oscillation. These three modes are considered with equal weight regardless of the event’s location and type. Moreover, ACC is implicitly considered the primary factor influencing the occurrence of the event; therefore, changes not explained by natural variability modes are assumed to be attributable to ACC. Such an approach has potential limitations, which we address in this paper by proposing a refined and more flexible version, called ClimaMeter 2.0. First, we propose weighting the three modes of variability according to the strength of the teleconnection between the remote modes and the local hazards. Then, we test the hypothesis that ACC has critically influenced the observed changes by analysing long-term trends in specific quantiles of the local hazard variables. After extensive testing using pre-industrial climate simulations and observational data, we conclude that, while remaining within the same conceptual framework, ClimaMeter 2.0 provides greater flexibility and enables a more nuanced assessment of the influence of ACC on specific extreme events.
- Preprint
(6231 KB) - Metadata XML
-
Supplement
(326 KB) - BibTeX
- EndNote
Status: open (until 04 Jun 2026)
- RC1: 'Comment on egusphere-2026-1466', Vikki Thompson, 30 Apr 2026 reply
-
RC2: 'Comment on egusphere-2026-1466', Peter Pfleiderer, 11 May 2026
reply
The authors present an update ( ClimaMeter2.0) of an analogue-based extreme event attribution (EEA) framework ClimaMeter that relies on observations (reanalysis). It is an important method complementing other EEA methods that often rely on climate model simulations. The updates intend to fix flaws in the previous version of ClimaMeter by explicitly choosing a hypothesis based on a trend analysis and by weighting modes of natural variability based on the importance for the studied event. The methods are well explained and three examples help to understand the methodological changes. The paper is well written and structured. The synthesis of different parts of the analysis makes the interpretation of the results challenging. Before publication, some more reflection and evaluation on the importance of different parts of the analysis for the final attribution statement and the synthesis of different would be helpful.
I find that the tests (quantile regression, similarity in modes of internal climate variability, and difference signal in the analogues) are interesting for the interpretation of an observed extreme event. I also understand that the authors want to present a highly synthesized result for communication purposes. But I think that the synthesis of the three tests is challenging and needs some more testing and reasoning.
For instance, I find the interpretation of the results of the quantile regression problematic. The authors use a the quantile regression to set the confidence level for a test in similarity between modes of internal variability in factual (F) and counterfactual (CF) climate. Over shortish periods, ENSO, PDO and AMO can influence the results of the quantile regression. For instance, the AMO index increases over the reanalysis period. There could be cases, where you detect a significant trend in your extreme index due to this “trend” in AMO (if the event is influenced by AMO). Therefore I’m wondering, what level of evidence the statistical trend analysis brings and how much one should rely on this trend analysis to define the confidence level for later tests in the methodology.
Could there be cases of circular reasoning? AMO influences event and AMO is increasing in the studied period -> strong statistical trend due to AMO -> less strict test when comparing AMO values between analogues -> ACC intensified event
More generally, I would find it interesting to check how relevant the individual tests are for the final attribution statement. If it would turn out, that the final attribution statement (gauge value) heavily relies on the quantile regression I would suspect the results to be very similar to probabilistic EEA applied to observations only (GEV fit with the location varying with GMST). In my view that would be worth commenting as the analogues methods is seen as an important complementary approach to the probabilistic EEA. “Hiding” the result from the analogue comparison because of the trend analysis would be a shame in my opinion.
From the tests in long piControl runs, I do get the impression that the quantile regression is quite relevant in ClimaMeter2.0. I think that the piControl test was important to identify a flaw in ClimaMeter and it nicely shows that problems can be solved with ClimaMeter2.0. However, the approach only tests how often you get false positives. I think that an evaluation with in single model large ensembles would be very informative. Applying the method to single simulations of an ensemble and to evaluate well understood kinds of events (using the information of the ensemble mean) would be very useful.
I’m wondering whether only showing three simple statement, one for each of the tests, would be more interesting for the lay persons. For example for the European heatwave 2023 you could say something like:
“Heat extremes increase in this region” (quantile regression -> fig. 7d)
“Similar events were 0.86C cooler in the 1960s” (Analogues -> fig. 7c)
“Slow fluctuations in the climate system over the period 1950-now have influenced heat waves in this region” (fig. 7g)The last statement already gets a bit technical and I understand why you would want to avoid it. But I think that a few iterations with people outside of the field would lead to a wording that is understandable, interesting and correct. If you still want to show a synthesized main result I would strongly suggest to add such statements additionally.
Although I find the reasoning of the methodology quite convincing in the sense, that I don’t know how a synthesis targeting one single result value could be done better, I’m a bit skeptical whether such as synthesis is a reasonable representation of the results.
The examples are very useful to understand the methodological changes. The interpretation of the final result is however missing. A brief discussion on how the results can be interpreted in terms of physical mechanisms would greatly improve the confidence in ClimaMeter2.0. In the current state, the examples leave me with questions:
Heatwave 2023: Looking at fig. 7c I would have said, that only very few grid-points within the region of interest show a significant difference in temperatures. Is there a threshold for the significance of delta T? We know quite well, that heat waves in this region have intensified substantially and actually more than in other regions. This is due to circulation changes over the period of interest (forced or not), aerosol reductions and soil-moisture changes. How do you interpret, the result of ClimaMeter2.0 saying that the heat wave in 2023 was not influenced by ACC? Depending on how ACC is defined, the reduction in aerosols over the region could be seen as a reason for an intensified heatwave in this region where ACC did not play an important role. Since you do not treat aerosols explicitly, I’m however wondering how the method comes to this conclusion and how this should be interpreted. Does it mean that for heatwaves coming from this kind of circulation pattern the effect of ACC is negligible?
Storm Hans: The analogue comparison shows a northward shift of the precipitation with insignificant changes in the region of interest (except the parts at the border of the region). The region where extreme precipitation was observed are not affected by these differences. Why would you say that heavy precipitation was influenced by ACC? It seems that the result strongly depends on the extend of the region of interest.
Storm Ciaran: Figure 9 suggests that for similar large scale circulation patterns, local wind speeds are higher in F as compared to CF. In terms of conditional attribution one would say that ACC played a role for this kind of circulation type. There is no significant trend which could be to a decrease in frequency of weather patterns producing strong winds or winds being weaker for other types of weather patterns. I thought that with ClimaMeter you were targeting a conditional attribution for the relevant large scale circulation patterns. The synthesized result saying that ACC did not play a role is a surprise to me and to me the gauge does not represent the results shown in other panels of figure 9. I understand how the value of the gauge was obtained. I’m rather wondering whether this is what ClimaMeter should show us for cases where the trend is insignificant. To me it would make more sense to point towards the IPCC with a statement like: “ClimaMeter finds that wind extremes similar to Ciaran were intensified by ACC even though there is no overall intensification in wind extremes in this region”
Specific comments:
L35: New simulations are not required for each EEA study. Existing large ensembles or other simulations can be used.
L39: Reference for WWA is missing.
L71-75: I find this part a bit confusing and I think some clarifications would help. What about forced dynamic changes? I think that for this part of the argumentation, a separation between forced dynamical changes and forced thermodynamical changes would be helpful. Is the overall assumption in this article that dynamical changes are not forced? If that is the case please state it explicitly.
L173: You compare the modes of variability of the days from the selected analogues. I would reformulate this and not say “between the two periods”. I’m also wondering about the interpretation: there could be different cases leading to different modes in the selected analogue days: (i) the whole period has on average different values in these modes of variability (this you can actually check). (ii) analogues of the event occur in different phases of the modes in the factual as compared to the counterfactual climate. (iii) analogues in one period are by chance in different phases of the modes.
Would it make sense to discuss these cases explicitly?
L179: I’m a bit confused about the “Influenced by Natural Variability”: as I understand the test, you check whether the analogues in factual climate occurred in different phases of the modes of variability as compared to pre-industrial climate. With this you can make a statement about the 30 analogues in factual climate. Can you also make a statement about the observed event climaMeter is analyzing? Or how exactly has this “Influenced by Natural Variability” to be interpreted?
L219-221: Please also name one of the papers that discusses observed and simulated changes in ENSO.
L298: I don’t understand the reasoning behind option (b) in the ACC test. Since you already select analogues and control for modes of variability, what would be the fluctuation explaining the conditional difference? To me, this additional test is rather confusing because I thought the aim of climaMeter is specifically to do a conditional attribution. I understand the motivation to work on the null-hypothesis that ACC explains everything remaining after correcting for the modes. Maybe you could add an example of a case where this test helps to identify a mismatch between trend and conditional difference and how this can be explained “natural fluctuations of the climate system”.
L394: Beyond improvements to previous versions of the model, is the performance satisfactory?
L551: “It is important to stress that such statement is only valid for storm Ciaran, its dynamics and associated wind speed, and cannot be generalised to any other storm or event.” Is this the case because the trend significance depends on the region? Otherwise I find this statement misleading as this trend analysis does not depend on the circulation types.
Small details:
L37: one point to much
L44: “analyzed events” ?
L447: which pre-industrial heatwave?
L484: degree C (or K)
Citation: https://doi.org/10.5194/egusphere-2026-1466-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 201 | 109 | 20 | 330 | 33 | 14 | 16 |
- HTML: 201
- PDF: 109
- XML: 20
- Total: 330
- Supplement: 33
- BibTeX: 14
- EndNote: 16
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of Naldesi et al. A flexible methodology to evaluate natural variability in ClimaMeter
Following on from the current ClimaMeter methodology, this paper presents a new method for attributing changes in extreme weather events using analogues, remaining within the same conceptual framework.
The method presented, ‘ClimaMeter 2.0’, does provide improvements on the original method, and the paper also outlines the remaining limitations of such a method. The figures themselves are clear and nicely presented. The discussion of limitation of the original ClimaMeter method is comprehensive.
However, in my view the presentation of the method and particularly the comparison of results using the two methods should be enhanced before publication. Many of the most important findings are not adequately discussed within the text and suitable recommendations for implementing the method are not included.
Major Comments:
Minor comments: