We previously highlighted Roy Spencer’s poor practices in comparing models with observations, but we’ve now dug down a little deeper, and it’s not pretty.

A few weeks ago, the email exchanges of the CWG authors were published after a court ordered them to be made public. In them, there is an interesting exchange between Steve Koonin and Roy Spencer. Koonin wanted Spencer to address the (obvious) complaint that Roy’s comparison of ‘Corn Belt temperature trends’ figure was a cherry pick.

Roy agrees to look into it, but whether he ever got back to Koonin is unclear. In any case, no public statements or responses have been made. The conversation did however reveal where the data came from and Roy’s method for making the comparison, inspiring me to try and replicate the analysis more appropriately. So let’s see what we can find out.

NOAA Climate Divisions

NOAA has a great website with its Climate Division data (ClimDiv) which is an aggregated product from the individual station data, but averaged at the division, state, and regional levels. It has averages for 9 regions of CONUS, the big river basins, NWS areas, and multiple agricultural regions. The Corn Belt map for averages is below:

Spencer most likely used this precomputed index (across 11 states, though his figure says 12 – he might have tried making his own 12-state index but I don’t think it matters much, other than making it trickier to replicate). Anyway, the Corn Belt data (code 261 for the area weighted observational data) can be downloaded directly and the JJA trends computed for the 50 year period 1973-2022 (following Spencer’s graph). I’m sure it will come as a great shock to our readers that of all the regions computed in the NOAA ClimDiv dataset over this period, the Corn Belt index has almost the lowest summer trend (0.1ºC/dec though with a wide (95%) confidence interval [-0.07,0.27]). Quel surprise!

Note that the annual trends are much less noisy and more similar across the regions (but that wouldn’t be as useful, now, would it?).

This can be compared with the CMIP6 models data from a roughly analogous area (100-81ºW, 40-46ºN). This is relatively easy to extract from the ClimateExplorer website for 144 individual simulations (using the historical + SSP245 scenario). Spencer (I think) chose the option to download the ensemble mean for each model (or perhaps just a single run from each model). In either case, he discards very relevant information from the ensemble for each particular model.

So let’s plot the JJA trends for the Corn Belt region, including two elements that Spencer ignored. First, the uncertainty in the OLS trend in the observations (which is large) and, second, the spread across the individual model ensembles. As we did previously, we can plot the trends against the climate sensitivity (here I’ll use the Transient Climate Response) to see the difference that the ‘hot models’ make (IPCC AR6 assessed that the likely range of TCR was 1.4-2.2ºC, very likely 1.2-2.4ºC). Note that noisier the statistic (short periods, small regions, etc.) the less clear any difference related to climate sensitivity will be.

As in Spencer’s original plot, the bulk of the simulations have a larger trend than observed (all but 2 in fact). However, we are not done. Recall that the simulations are an ensemble of opportunity. Some of the models have sufficient ensemble members to reasonably estimate the standard deviation and the 95% spread for the trend in that model, but others have only one member and no spread can be calculated. If we assume that the average standard deviation (in this diagnostic it’s 0.11ºC/dec), is a good estimate in those cases, we can plot the ensemble means, along with an estimated 95% spread, for each model.

Now we have a slightly more regular statistical comparison, and we can see that this observation is within the 95% spread of about half a dozen models. Thus it’s a plausible (if not likely) match. Given the large number of comparison one could theoretically do, if one area (specifically chosen) is an outlier is not so surprising.

However, the point of the criticism of Spencer’s figure to begin with was not that these Corn Belt temperature trends have been well predicted, but rather that this index was cherry picked to provide the worst possible comparison. So what would a different index have given? I have not checked them all [The other crop regions are not so easily translated into lat/lon rectangles!], but I did look at the NorthWest region (124-111ºW, 42-49ºN) – this region has warmed more than the CONUS average by about the same degree that the Corn Belt region was below.

And… there is a much stronger coherence with the ensemble. But this would not have provided as useful a talking point of course. Does this mean the models are perfect? Of course not. A worthwhile analysis would have looked at a spread of such comparisons and made a statement about the utility of the models based on that collective analysis. Just looking at the one or two small areas or single seasons is not going to be informative.

Useful vs. Useless model-observation comparisons

My point here is not to discuss the utility of the CMIP6 projections for the corn, wheat, soy regions of the US etc. That is best left to people who have better domain knowledge about those applications. However, I do want to (again) stress a few points:

• Observations have uncertainty too. Whether it is in the linear trend and/or structural, it needs to be accounted for in any comparison.
• Observational data comes from a single realization. You need to think about what the irreducible uncertainty associated with internal variability means.
• The model ensembles are complex. You cannot substitute what is easily downloadable from ClimateExplorer for an analysis in and of itself.
• Whether the observed trend is or is not close to the model ensemble mean(s) is not a good test of the skill of the model(s) or multi-model ensemble.
• The appropriate test is whether the real world observation is exchangeable with an ensemble member from the model. The visual comparisons shown above address this, but it can also be done quantitatively in ways that take into account both the observational uncertainty and spread.
• Comparisons such as presented by Spencer in the DOE or Heritage Foundation reports are fundamentally flawed since they never deal seriously with any of this, and will continually be called out as cherry picking.

One last thought. Steve, take a moment and think about why Roy didn’t ever address the critiques. Just spitballing here, but the intersection between people generally thought of as ’eminent’ scientists and folks that engage in this kind of hackery is empty.

The post Spencer’s Shenanigans: Part II first appeared on RealClimate.

The mystery of why the last million or so years of glacial variability are so different to what came before just got more mysterious…

It’s easy to understand why the ice ages have such a hold on our imaginations. Putting aside the cavemen, woolly mammoths, and sabre-toothed tigers of popular culture, the scientific questions around the pacing of the glacial cycles, their magnitude, variability, and impacts are truly profound.

Despite huge strides in understanding the ice ages – from the ground-breaking work of Hayes, Imbrie and Shackleton (1974) that demonstrated the skill of the Milankovitch model in the 1970s, the paradigm-busting results from the Greenland Ice Cores in the 1990s, the discovery of the Heinrich events, etc., there remain plenty of real and abiding mysteries including:

• Why are the 100kyr cycles so strong?
• What are the details of the carbon feedbacks on glacial-interglacial cycles?
• What triggered the ice ages in the first place? (i.e. why did the impact of Milankovitch cycles get much larger over the last 2.5 million years?)
• Why didn’t humans develop agriculture in the last interglacial?
• What triggers the Dansgaard-Oeschgar oscillations?
• and… what caused the change from lower magnitude 40kyr cycles to 100kyr cycles across the Mid-Pleistocene Transition (MPT)?

We have good evidence from the deep Antarctic ice cores of the coupling between CO₂ and temperature over the last 800kyrs and from ocean sediment proxies, we have reasonable estimates of the coupling between CO₂ and temperature over the long cooling during the Cenozoic (the last 65 million years). But, until now, we haven’t been able to really examine that intervening period – the early Pleistocene.

Theories, of course, abound. The obvious one is that the long term declines in CO₂ crossed a threshold that allowed for larger ice volumes that had more resonance with the 100kyr cycles. Another is that the early ice advances (which were more spread out but less voluminous) scraped all the soils off the rocks and that subsequent ice sheets were less mobile. I think most folks expected the data (when it arrived) to basically confirm what people expected.

But sometimes the observations don’t confirm your preconceived notions. The nice thing about science is that scientists (ideally) tend to get excited at this point (instead of, say, trying to deny the new information). So what has just happened?

Two new papers, Marks-Peterson et al. (2025) (direct link) and Shackleton et al. (2025) (direct link) in Nature this week report on analyses of very old Antarctic ice. These samples come from the “blue ice” in the Allan Hills in Antarctica where multi-million year old ice surfaces after having been deposited and transported over large distances. This is quite distinct from deep drilling in places where you hope the ice has not moved much, and while it doesn’t have the nice stratigraphy of the cores, you can sample snapshots of the atmosphere over a much longer time – in this case, almost 3 million years – albeit with coarser dating.

There are two main measurements presented. The first are the GHG concentrations in the air bubbles trapped in the ice (Fig. 1), and the second is a record of mean ocean temperature inferred from the ratio of noble gases in the air bubbles (Fig. 2).

The first and most dramatic (or rather, non-dramatic) result, is that CO₂ levels appear to have barely changed (on average) over this key period – dropping only 20-30ppm over the onset period. That isn’t nothing, but it’s only about 0.45-0.7 W/m² in forcing, and would lead to around 1ºC in global surface cooling. The CH₄ levels might have been expected to fall too, but they seem to be static. [Note that this method is not sampling the glacial/interglacial variations which are apparent in the more recent records]. The second, and somewhat confounding, result is that the global ocean seems to have cooled by about 2ºC over the same time period (with the global surface temperature change would have been larger).

So we have a conundrum. The onset of NH glaciation did happen as the planet cooled (as might be expected), but the first guess for what caused that cooling (long term trends in CO₂ and/or CH₄) does not appear to work.

How might this be resolved?

There are always multiple potential ways out of a conundrum: subsequent analyses might find an issue with the observations, there might be a hyper-sensitivity to the small CO₂ changes at this time (but why?), there might be something else driving the change (volcanism? dust aerosols?), or… what? None of these possibilities are obvious winners, and of course, they are not mutually exclusive. Eric Wolff (direct link) in his commentary seems to think that the ocean is doing the driving, but I think that might be backwards.

The funny thing is that paleo-climatologists have been wanting these old ice analyses for a long time – with the anticipation that they would help answer these questions. But they seem to be posing many more questions than they have answered.

Broader issues

One thing this shows is that scientists can’t be complacent. As we’ve seen with surprising climate events even over the last few years (2023, Antarctic sea ice, the increases in the Earth’s Energy Imbalance), the more you look at the planet (or even the universe) the more surprising things you find. Science is an active search for deeper understanding – and we are not done yet.

Final thought

At face value, these results seem to suggest that CO₂ declines were not the dominant/only cause of the cooling at the onset of the ice ages, despite expectations. Some of the usual suspects are certainly going to claim (fallaciously) that this means that CO₂ can’t be the cause of anything. This is obviously a stupid argument so feel free to judge anyone that makes it.

Nonetheless,…

There are more things in heaven and earth, Horatio, than are dreamt of in your philosophy.

References

1. J.D. Hays, J. Imbrie, and N.J. Shackleton, "Variations in the Earth's Orbit: Pacemaker of the Ice Ages", Science, vol. 194, pp. 1121-1132, 1976. http://dx.doi.org/10.1126/science.194.4270.1121
2. H. Heinrich, "Origin and Consequences of Cyclic Ice Rafting in the Northeast Atlantic Ocean During the Past 130,000 Years", Quaternary Research, vol. 29, pp. 142-152, 1988. http://dx.doi.org/10.1016/0033-5894(88)90057-9
3. J. Marks-Peterson, S. Shackleton, J. Higgins, J. Severinghaus, Y. Yan, C. Buizert, M. Kalk, R. Beaudette, V. Hishamunda, D. Eves, A. Carter, A. Kurbatov, J. Epifanio, J. Morgan, I. Nesbitt, M. Bender, and E. Brook, "Broadly stable atmospheric CO2 and CH4 levels over the past 3 million years", Nature, vol. 651, pp. 647-652, 2026. http://dx.doi.org/10.1038/s41586-025-10032-y
4. S. Shackleton, V. Hishamunda, Y. Yan, A. Carter, J. Morgan, J. Severinghaus, S. Aarons, J. Marks-Peterson, J. Epifanio, C. Buizert, E. Brook, A.V. Kurbatov, M.L. Bender, and J. Higgins, "Global ocean heat content over the past 3 million years", Nature, vol. 651, pp. 653-657, 2026. http://dx.doi.org/10.1038/s41586-026-10116-3
5. E.W. Wolff, "Climate snapshots trapped in ancient ice tell a surprising story", Nature, vol. 651, pp. 592-593, 2026. http://dx.doi.org/10.1038/d41586-026-00636-3

The post The Puzzling Pleistocene first appeared on RealClimate.

Guest commentary from Nathan Lenssen (Colorado School of Mines)

A new analysis of historical temperatures suggests that things are getting warmer faster, but what does it mean for the future?

A study (Foster & Rahmstorf 2026) was published on Friday claiming evidence that “Global Warming Has Accelerated Significantly”. This study is an update by the authors of a similar study they published in 2011 where they found no statistical evidence for an acceleration in global warming. Both studies sought to determine if there is a detectable acceleration in warming, after statistically removing the effects of ENSO, volcanoes and changes in solar forcing from the observed global mean temperature (GMT) series (through to 2024).

As I’ll discuss further below, there was no detectable acceleration in the raw GMT series – this doesn’t mean there isn’t any, but that the noise (internal variability etc.) doesn’t allow us to see if there is clearly. Thus, the study has detected an acceleration in the rate of warming of inferred long-term trends – which we can pretty confidently attribute to anthropogenic effects. This study has understandably gotten substantial attention in the media. Here, I will outline what I think we have learned from this study, what this means for our understanding of the current state of the climate system, and what it means for projections of climate change (Hint: not much).

FR26 make three contributions in this recent work: (1) the production of an “adjusted” GMT series that removes statistically estimated impacts of a few short term changes in GMT, to hopefully leave just the warming associated with changes in anthropogenic forcings, (2) the detection of an acceleration in the rate of warming on this series using three different statistical methods, and (3) a forecast that 1.5ºC warming will be reached by ~2030. The methods used here are generally sound, particularly by engaging with the state of the art in changepoint detection methods as one of the methods for acceleration detection (Beaulieu et al. 2024). The figure below shows the three statistical methods for detecting changes in trend, all of which provide statistically significant evidence that the recent trend is faster than previous trends. 

 Given the assumptions made by the authors, this provides statistically robust evidence that acceleration has been detected. On first glance, this may be surprising or alarming as, to the zeroth order from our understanding of the Earth’s system’s response to CO₂, we expect a roughly generally linear warming in GMT given the exponential rise in CO₂ due to the log-scaling of GMT with CO2. Acceleration could be the result of the decrease of cooling anthropogenic forcings (as is hypothesized for some regional accelerations detected in Beaulieu et al. 2024) or substantial feedbacks/tipping points that are causing the Earth to warm faster than the simple CO₂ forcing physics dictates. Note though that the climate models that are used to inform our future projections also expect an acceleration around now (of course, given the assumptions that went into them).

However, as the authors point out, their method of ‘removing’ ENSO could be improved (for instance, Compo and Sardeshmukh (2010)), and there is still some imprint of natural climate variability in their adjusted time series. Note that an estimate of the “true” natural variability of the climate system, and correspondingly the “true” forced response, is one of the white whale problems in climate science! FR26 does an credible, but necessarily imperfect, job of isolating the forced response, but don’t account for this uncertainty in their statistical tests.

While we can’t know the true internal variability perfectly, we have climate models which provide an estimate of this variability. The figure below shows that the CMIP6 models (screened for a likely Transient Climate Response (TCE)) have a spread that fully contains the observed climate signal. Notably, the ensemble mean of these models demonstrates a slightly greater than linear warming (minus the effects 1991 eruption of Mount Pinatubo).

We can look at this more closely. If we look at the trends in individual model simulations for the last 13 years (2013-2025) and the 13 years before that (2000-2012), on average, the models show a slight acceleration over the same period highlighted by FR26 (0.18ºC to 0.30ºC). However, while there is a difference in the mean of these distributions, they are not clearly separate. This shows that, at least in model land, the acceleration in trend (given the internal variability and model uncertainty) is going to be difficult to detect. Note that comparisons between the models and the real world are complicated by any divergences in the forcings in the scenarios (designed more than 15 years ago) and what actually happened (Hunga Tonga, the IMO regulations, Chinese aerosol decreases etc.).

So where does this leave us? There is no detectable acceleration in the raw observed GMT, but there is an acceleration in GMT when removing the linear effects of ENSO, volcanoes, and solar variability, and there is slight acceleration in GMT when estimated using a multi-model ensemble of climate models. John Kennedy recently discussed some of these results in the context of FR26, expanding to a wider discussion of estimates of warming rate. He hits the nail on the head by pointing out two key open questions: “If there is an acceleration, what is physically driving it?” and “What will happen to the warming rate in the future?” The question about mechanism is key to trustworthy predictions of the future rate, and this is not addressed in the new paper.

The prediction of 1.5ºC warming by ~2030 made in FR26 is made in this context by estimating the rate of warming in this adjusted GMT. While made in the imperfect context discussed here, this estimate is reasonable when compared to a more comprehensive attempt to estimate this date . However, as John states, we already know the planet was warming, we have some evidence for acceleration, but we need a better path forward to predict how GMT and subsequent regional climate will change under continued CO₂ emission

References

1. G. Foster, and S. Rahmstorf, "Global Warming Has Accelerated Significantly", Geophysical Research Letters, vol. 53, 2026. http://dx.doi.org/10.1029/2025GL118804
2. G. Foster, and S. Rahmstorf, "Global temperature evolution 1979–2010", Environmental Research Letters, vol. 6, pp. 044022, 2011. http://dx.doi.org/10.1088/1748-9326/6/4/044022
3. C. Beaulieu, C. Gallagher, R. Killick, R. Lund, and X. Shi, "A recent surge in global warming is not detectable yet", Communications Earth & Environment, vol. 5, 2024. http://dx.doi.org/10.1038/s43247-024-01711-1

The post How robust is our accelerometer? first appeared on RealClimate.

This month’s open thread for climate related topics.

The post Unforced Variations: Mar 2026 first appeared on RealClimate.

The EPA has announced its final^* ruling on the CO₂ Endangerment Finding.

^*not even close to final.

Notably, they have completely abandoned any reliance on the DOE’s CWG report.

The EPA is not relying on new findings by the Administrator with respect to global climate change concerns under CAA section 202(a)(1) as a basis for the rescission or repeals and is not finalizing the alternative basis set out in section IV.B of the preamble to the proposed rule.

…

With respect to commenters’ precautionary arguments, the EPA is not finalizing the proposed alternative basis for rescission and repeal based on a new climate science finding by the Administrator.

…

Although the Administrator continues to harbor concerns regarding the scientific determinations underlying the Endangerment Finding, the EPA has decided not to finalize this scientific alternative rationale at this time.

…

For similar reasons, and in light of concerns raised by some commenters about the draft report authored by the U.S. Department of Energy’s Climate Working Group (CWG), the EPA is not relying on the May 27, 2025 CWG draft report entitled “Impact of Carbon Dioxide Emissions on the U.S. Climate” or the July 23, 2025 CWG report entitled “A Critical Review of Impacts of Greenhouse Gas Emissions on the U.S. Climate” for any aspect of this final action.

This is good news, since it leaves them with only a legal argument that for some reason the law is different now than it was when Mass. v. EPA was decided in 2007. That ruling forced the EPA to conduct the Endangerment finding in the first place, and still stands as legal precedent. The EPA argument now is that recent rulings (Loper-Bright and W.V.) from the Supreme Court mean that the original ruling can no longer be relied on. However, only the Supreme Court really has the power to overturn previous SC rulings and so the next set of lawsuits (in D.C. District Court) will likely find for the plaintiffs and possibly enjoin this ‘final’ rule. If this then goes to appeal to the SC, they would have to agree to hear it, and then folks would basically have to re-litigate the whole thing. Only lawyers are likely to gain from this.

Science will still play a role in this since a) the fact that CO₂ and the other five greenhouse gases (the Kyoto gases – CH4, N2O, CFCs, etc.) do endanger public health and welfare is now unchallenged, and b) EPA is now relying on a de minimus argument (which will likely also be challenged) about the impact of regulations on motor vehicles (which isn’t really a valid part of the endangerment finding) that comes from climate modeling. The CWG folks will likely not play any further role in any of this.

The ruling is long and legalistic, and so interested parties will need to take some time to digest what it says and what to do. As they say on Broadway, this one will run and run…

The post EPA’s final* ruling on CO2 first appeared on RealClimate.

For the first time, the Federal Judicial Center (FJC) commissioned a chapter on climate science for the manual they put out (with the NASEM) for judges, the Reference on Scientific Evidence (4th Edition). This week, a month after it was published, they pulled the chapter out after being pressured by 27 Republican Attorneys General. You can nonetheless read it here.

Some background. The FJC is “the research and education agency of the judicial branch of the United States Government”. As one of its roles, it is tasked to provide educational materials to judges and other court workers about issues that might come up in court, and in particular, on scientific matters that one might not expect judges or lawyers to be expert in. They have codified this information in the Reference Manual on Scientific Evidence, which is now in it’s Fourth Edition. (Previous editions were issued in 1994, 2000, and 2011).

The 4th Edition had its genesis in a workshop in 2021, and was finally published (after extensive peer review) on Dec 31st 2025. It covers legal scholarship on the use of expert testimony in court cases (noting the Supreme Court’s Daubert standard), as well as primers in the current state of the science across multiple fields (forensics, DNA evidence, mental health, neurology, epidemiology, exposure, statistics, regression, eye witnesses, engineering, computer science, AI, etc.). Notably, it included a chapter on climate science, covering topics such as the greenhouse effect, atmospheric circulation, detection and attribution, and the issues being raised in an increasing number of climate-related cases in the courts. The authors, Jessica Wentz and Radley Horton are a respected and mainstream lawyer/scientist team and the resulting chapter is a clear and concise summary of the topic. So far so good.

Of course, there are groups that would rather not have climate change discussed knowledgeably in the courts, and after the publication of the 4th Edition of the manual, the Republican-led House Judiciary Committee started sending threatening letters to all involved (FN – sorry!) (Jan 16th). Additionally, a group of 27 Republican Attorneys General (led by West Virginia) sent a letter (Jan 29) to the FJC claiming that Wentz and Horton were biased because they have (correctly) stated that the “political sphere in the United States continues to be clouded with false debates over the validity of climate change”. Additionally, they were upset that there are no references to the recent DOE CWG report (Lol).

The real target of the AGs ire is the discussion of attribution, and the notion that there is an emerging consensus that partial attribution of climate damages can be assessed on emitters. This line of thinking is exemplified by recent papers (such as Callahan and Mankin (2025), but is based on more than a decade of work on this topic, and of course is a direct threat to the fossil fuel companies that the WV AG is trying to protect.

The Republican AGs demanded that the FJC remove the chapter, arguing that any official acknowledgement of the science in the Manual would prejudice their cases that are based on, let’s say, “contrary” interpretations of the scientific evidence (or no evidence at all). And without much ado, or even consultation, the FJC did exactly that, putting out an amended Manual on Feb 6th. The only note to mark the deletion is:

No explanation or excuse was noted.

As stated above, this chapter is actually well-written, appropriately peer-reviewed, and deserves a far better fate than to be cowardly disappeared into a memory hole for being inconvenient, so you can download it here. The nice thing about science is that it doesn’t change based on whether a report is published here or there, so feel free to share.

Update (March 4): An excellent response to the criticisms from the Republican AGs from Wentz and Horton has been posted, and a letter from the other authors decrying political interference in the manual has also appeared. The Republican AGs have also written to the National Academies to demand that they take down the chapter too. (Note that no-one has emailed us!).

Further update (March 8): The authors have a blog post at the Sabin Center website.

Another update (March 17): Apparently the NASEM told the AGs to pound sand.

References

1. C.W. Callahan, and J.S. Mankin, "Carbon majors and the scientific case for climate liability", Nature, vol. 640, pp. 893-901, 2025. http://dx.doi.org/10.1038/s41586-025-08751-3

The post The Climate Science reference they don’t want Judges to read first appeared on RealClimate.

At a public event debating the DOE CWG report, Steve Koonin embarrasses himself further.

This week there was a bit of a peculiar event at the Civitas Institute at UT Austin, with three of the CWG authors (John Christy, Steve Koonin and Ross McKitrick) being rebutted by Andy Dessler (working solo).

The event itself was a rehash of the CWG’s reports ‘findings’ (or rather, a repeat of their cherry picks, uncontextualized statements, and ignoring of the literature), and Dessler somewhat successfully pointing this out. The event seemed a bit rushed (too much content being crammed into too short a time) and is a great example of the applicability of the Brandolini’s Law.

There would be a lot to criticise in the presentations if one wanted (most of this was gone over in the Scientists response to the DOE report that Andy helped organise), but the presentation by Koonin went even further into nonsense territory than the CWG report itself. Apparently, “internal variability” (something noticeably ignored in many claims by the CWG) is the “last refuge of fools and scoundrels” (at least according to Koonin)!

What this stems from is Koonin’s reliance on Nicola Scafetta’s work on evaluating climate models – readers here will know that is a very bad idea, and we went through a lot of this in respect to a GRL paper that Scafetta published in 2022. That led to a whole saga, which took so long that while we were trying to get the 2022 paper retracted on the grounds of being totally wrong, Scafetta basically published the same analysis again (with almost all the same errors and some new ones) in another journal. Our enthusiasm to go another round pointing out his mistakes was limited, and so the second paper still stands nominally unrebutted in the literature despite having been pre-rebbutted by our comment on the first paper (Schmidt et al., 2023). This came up in the ‘internal review’ of the CWG report, where one of the reviewers said that the CWG should deal with our criticism of Scafetta’s work (pointing to the published comment), and were blown off by the CWG who claimed that because they cited the second paper (not the first), our comment was moot. Classic dissembling.

Anyway, Koonin’s presentation at the Civitas event (starts around 20:20 in the video) repeats the errors, but goes even further. First, he notes that some CMIP6 models have climate sensitivities that are too high. That’s fine – I have made the same point here, and in Nature Hausfather et al., 2022. But then he elides from ‘some models’ to ‘the models’ without even taking a breath (Hmm…). He doubles down on Spencer’s cherrypicking (itself not peer-reviewed of course), and claims that people pointing out that something has been cherry-picked are trying to “change the topic”. Yes, that metric that no-one had ever mentioned before Spencer did this analysis is *the* topic that the assessment was designed to address /sarc.

Koonin additionally claims that the mainstream scientists are blaming model-observation discrepancies on internal variability for the last twenty years, while ignoring it for the previous twenty years. Of course, he provides no citation nor evidence that anyone has ever done such a thing. Worse, in response to a suggestion that they utilise the uncertainty in the modeling (esp. the internal variability), he makes an incredible statement (starting at 30:47):

Well, if you do that, it effectively broadens the uncertainty so much as to be almost essentially useless.

Let’s parse this out. He isn’t claiming that the internal variability isn’t real (it is of course). He is claiming that his model-observation comparison doesn’t show any discrepancy if you include the uncertainties and that therefore it’s useless! To repeat, Koonin is stating that he isn’t including the uncertainties because it would undermine the conclusion he is trying to draw.

This is as clear an admission of scientific misconduct as I’ve heard.

He then illustrates this with reference to Tokarska et al. (2020) (Fig 3, Panel A) which is not really trying to do the same thing, but fine. [I think there must be a second half to that slide showing individual runs – but I’m not sure where that would have been from]. However, we addressed this exact issue with the comment on the first Scafetta paper:

The question being asked is whether there is a discrepancy between any specific model and the observations. An initial condition (IC) ensemble starts the model with a different weather pattern, but each run has the same forcing. The standard deviation of the IC ensemble is a reasonable measure of the internal variability (i.e. the spread that could occur only as a function of the (unpredictable) weather. The real world can be considered a single realization of the real world climate, so the standard way to assess whether the a model is consistent with the real world is to estimate the probability that the real world result could be part of the model distribution. In practice, one can calculate the 95% confidence interval for the model (based on it’s ensemble) and ask whether the real world data falls within that range. Wherever it is, you can calculate the probability of getting that result, assuming that model distribution. The further away the observation from the model spread, the less likely that it could have generated by that distribution.

So if the real world falls inside the 95% CI, it is clearly consistent with the distribution, even if the ensemble mean is different from the observations. As the signal grows, the spread due to the internal variability will shrink, and discrepancies might emerge more clearly. But no-one is arguing that internal variability should be ignored for one period, and used in another. Rather, it should be used consistently at all times. If that prevents Steve Koonin from trashing the models, so be it.

To go back to the claim though, there are multiple models with sensitivities up to about 5ºC that have surface temperature trends that are compatible with the observations. A few models don’t have sufficient simulations to say, and a few are clearly incompatible. This is what Koonin says:

They say that the fact I can find one starting point that agrees with the data is enough to validated that model. In fact that doesn’t sound right at all. I don’t think that would pass peer review – at least among my peers.

This is not quite an accurate reflection of the mainstream position, nor do his feelings on the issue make sense. The mainstream position is first more nuanced (as explained above); it is not that seeing that observations fall within the spread validates the model, rather if this happens you should not reject that model (a much less onerous claim). But why does this sound strange to Koonin? Is he in the habit of rejecting models that are consistent with observations? And of course, this position has passed peer-review many times, though I will accept that his peers might not agree (which is a statement about his peers, not the claims).

To wrap this up, I updated the figure above to look at a slightly longer period (the change to 2015-2025) using the latest observations from ERA5.

It is still clear that some models are not consistent with ERA5 (notably the five models with the highest sensitivity), but it is also clear that many of them are – and that Koonin’s claims (like Scafetta’s before him) are hogwash. His implicit claim that you should ignore uncertainty if that gets in the way of your preferred conclusion is simply embarrassing for someone who likes to think of himself as an “eminent” scientist.

References

1. G.A. Schmidt, G.S. Jones, and J.J. Kennedy, "Comment on “Advanced Testing of Low, Medium, and High ECS CMIP6 GCM Simulations Versus ERA5‐T2m” by N. Scafetta (2022)", Geophysical Research Letters, vol. 50, 2023. http://dx.doi.org/10.1029/2022GL102530
2. Z. Hausfather, K. Marvel, G.A. Schmidt, J.W. Nielsen-Gammon, and M. Zelinka, "Climate simulations: recognize the ‘hot model’ problem", Nature, vol. 605, pp. 26-29, 2022. http://dx.doi.org/10.1038/d41586-022-01192-2
3. K.B. Tokarska, M.B. Stolpe, S. Sippel, E.M. Fischer, C.J. Smith, F. Lehner, and R. Knutti, "Past warming trend constrains future warming in CMIP6 models", Science Advances, vol. 6, 2020. http://dx.doi.org/10.1126/sciadv.aaz9549

The post Koonin’s Continuing Calumnies first appeared on RealClimate.

This month’s open thread about climate topics. Try to be substantive and respectful to other commenters.

Note: Comments that are abusive and/or repetitive to the point of tedium will be moderated. If you’ve made the same point ten times already in the thread, and people are still arguing with you, repeating it another ten times isn’t going to help.

The post Unforced variations: Feb 2026 first appeared on RealClimate.

Our various graphics and model-observation comparisons have been updated with 2025 data. There are a few version updates that make some difference (particularly in sea ice extent), but the basic story is similar to last year. Feel free to comment and/or suggest changes for the future, including potential new comparisons.

The post 2025 Updates first appeared on RealClimate.

New email releases from the EDF/UCS lawsuit against the DOE provide a rarely-seen behind the curtain look at how the climate contrarians work.

The new releases of non-governmental emails were ordered released by the judge in the case and illustrate clearly the desire for the five scientists of the Climate Working Group (CWG) (and their Cato Institute handler working temporarily for the DOE) to avoid using their government emails (they were all Special Government Employees) and attempt to do an end run around FACA regulations that prohibit unbalanced groups meeting in secret to advise the federal government.

The emails are (at present) in three tranches (some overlap, but mostly distinct) and may be added to as the lawsuit proceeds.

• Climate Working Group Emails Part I
• Climate Working Group Emails Part II
• Climate Working Group Emails Part III

Among other highlights, we have Steve Koonin (an ex-undersecretary of the DOE, who really should know better), telling his friends to “keep it to themselves”:

We should be mindful that our email communications that go to DOE addresses are subject to FOIA. While I don’t think we’ve been saying anything untoward in our recent group exchanges, one never knows how they might be twisted by those of nefarious intent.
I’d therefore urge that we keep our future email communications restricted to the authors (except, of course, for matters that directly involve the DOE — like the recent Al query from the New Yorker).

[Oddly, that was sent on August 4th 2025, months after they started work – and communicating using their gmail and hotmail emails.].

There is some slightly spicy discussion about Roy Spencer’s disdain for Will Happer’s arguments, Judith Curry’s dislike of Marcia McNutt (then president of the National Academies), Roy’s conspiracy theories about the original Endangerment Finding, their collective delusion about how this report would be reviewed, and the sweet irony that the President’s EO on ‘Gold Standard Science’ meant they couldn’t pretend to have their report ‘peer reviewed’.

There is one moment when Ross McKitrick almost gets to the point of realising what they are doing (this is his paraphrase of what a NASEM review would say):

“While the report makes some valid points, including that climate science must continue to improve the models used to study climate dynamics, it is unfortunately biased and incomplete, and fails to provide a comprehensive summary of the current evidence regarding the seriousness of the climate crisis.”

Yup.

Almost as an aside, even Roger Pielke Jr. makes an appearance – getting a secret briefing from the politicos at DOE on the work of the CWG on June 24th, months before this was made public (or even before it was known these folks had been hired). We’re sure that Roger’s well-known concerns about FACA, the proper procedure for climate assessments, and conflicts of interest mean that he wrote about this at the time, though for some reason we can’t find it. How odd.

The third tranche includes their responses to the mostly hostile comments from some (still unknown AFAIK) internal DOE folks. The responses are almost entirely non-substantive, and only led to trivial edits.

Anyway, dive in, and brace yourself for the whiplash from people who relished every detail in the hacked Climategate emails but who will now insist that this court-mandated release is grossly improper and conspiring to avoid FOIA is perfectly fine actually. Lol.

Update (Jan 30): The judge has ruled against the DOE (NYT). (Text of judgement).

The post A peek behind the curtain… first appeared on RealClimate.