Another paper shows that climate models and climate reality vary – greatly

A new paper has been published in Geophysical Research Letters that shows once again, that climate models and reality significantly vary. It confirms what Dr. John Christy has been saying (see figure below). The paper also references Dr. Judith Curry and her work.

Pronounced differences between observed and CMIP5-simulated multidecadal climate variability in the twentieth century

Plain Language Summary

Global and regional warming trends over the course of the twentieth century have been nonuniform, with decadal and longer periods of faster or slower warming, or even cooling. Here we show that state-of-the-art global models used to predict climate fail to adequately reproduce such multidecadal climate variations. In particular, the models underestimate the magnitude of the observed variability and misrepresent its spatial pattern. Therefore, our ability to interpret the observed climate change using these models is limited.


Identification and dynamical attribution of multidecadal climate undulations to either variations in external forcings or to internal sources is one of the most important topics of modern climate science, especially in conjunction with the issue of human-induced global warming. Here we utilize ensembles of twentieth century climate simulations to isolate the forced signal and residual internal variability in a network of observed and modeled climate indices. The observed internal variability so estimated exhibits a pronounced multidecadal mode with a distinctive spatiotemporal signature, which is altogether absent in model simulations. This single mode explains a major fraction of model-data differences over the entire climate index network considered; it may reflect either biases in the models’ forced response or models’ lack
of requisite internal dynamics, or a combination of both.

Some key quotes:

Here we show that state-of-the-art global models used to predict climate fail to adequately reproduce such multidecadal climate variations. In particular, the models underestimate the magnitude of variability in the twentieth century.

Our study documents pronounced differences between the observed and CMIP5-simulated climate variability in the twentieth century. These differences are dominated by a coherent multidecadal hemispheric-scale signal present in the observed SST and SLP fields but completely missing in any of the CMIP5 simulations.

Our results are also broadly consistent with recent analyses of Cheung et al. [2017], who documented substantial mismatches between their estimated internal components of the observed and CMIP5-simulated AMO, PMO, and NMO variability. However, these authors used subtraction of the scaled CMIP5 MMEM signal to deduce the internal variability in historical simulations of individual CMIP5 models. Kravtsov et al. [2015] and Kravtsov and Callicutt [2017] showed that the residual variability so defined misrepresents the true internal variability in CMIP5 simulations and is, in fact, dominated by model error, that is, the differences between the true forced response of individual models and the MMEM response. The magnitude of the CMIP5 “internal” variability estimated by this method is, hence, much larger than that of the true simulated internal variability, and the spectral characteristics of the true and estimated internal variability are entirely different.

Despite our explicit decomposition of the climate variability into the forced and internally generated components, dynamical attribution of the multidecadal model-data differences still remains uncertain. On one hand, if our derived CMIP5-based forced signals are realistic, these differences must arise from internal climate system dynamics presumably misrepresented in CMIP5 models, such as sea ice dynamics [Wyatt and Curry, 2014], oceanic mesoscale eddies [Siqueira and Kirtman, 2016], positive cloud and dust feedbacks [Evan et al., 2013; Martin et al., 2014; Brown et al., 2016; Yuan et al., 2016], or SST-forced NAO response [Kushnir et al., 2002; Eade et al., 2014; Stockdale et al., 2015; Siegert et al., 2016]. On the other hand, however, it is possible that CMIP5 models underestimate multidecadal variations in the true response of the climate system to external forcing or misrepresent the forcing itself [Booth et al., 2012; Murphy et al., 2017]; if this is true, the model-data differences reflect the mismatch between the actual and CMIP5-simulated forced signals, whereas the real world’s internal climate variability may be consistent with that simulated by the models. In either case, we strongly believe that model development activities should strive to alleviate the present large discrepancies between the observed and simulated multidecadal climate variability, as these discrepancies hinder our fundamental understanding of the observed climate change.

The paper is here:

The SI is here:

One of the figures from the SI shows the differences between the models forcing response and observed natural variability seen in AMO, NAO, etc cycles:

Figure S4: Raw observed indices (thin lines) and their estimated forced components — ensemble mean (thick lines) and uncertainty (error bars) — with the forced-signal estimates based on the Community Earth System Model (CESM) Large Ensemble Project (LENS) simulations (Kay et al., 2015). Forced signals were estimated using Kravtsov and Callicutt (2017) methodology, as (a) the rescaled (unfiltered) ensemble mean over the 40 historical LENS simulations (left panels), or (b) — as the rescaled 5-yr low-pass filtered ensemble means for 20 synthetic sub-ensembles of 5 simulations, each randomly drawn from the parent 40-member LENS ensemble. The index abbreviations are given in panel captions. Comment: The forced signals based on the entire LENS ensemble and its 5-member sub-ensembles are consistent.

But here is the real smoking gun:

Figure 1. Standard deviations (STDs) of the estimated observed (blue) and CMIP5-simulated historical (red) and control-run (black) internal variability for the five indices considered; top-to-bottom rows correspond to the results for the AMO, PMO, NMO, NAO, and ALPI indices, respectively. Also included are the estimates of the observed internal variability based on the one-, two-, and three-factor scaling methods of Frankcombe et al. [2015]; see legend. The STDs were computed for raw and boxcar running mean low-pass filtered time series using different window sizes of 2 ×K + 1 yr, K = 0 , 1, … , 30 (shown on the horizontal axis); K = 0 corresponds to raw annual data, K = 1—to 3 year low-pass filtered data, and so on. Error bars show the 70% spread of the STDs, between 15th and 85th percentiles of the available estimates of internal variability (see text for details). Shading indicates the range in which the observed internal variability is statistically larger than its historical (light shading only) or control-run counterparts (dark shading and light shading regions combined), at the 5% level; here KC2017 methodology was used to estimate the observed and simulated internal variability over the historical period. The NAO plot also includes the results (heavy black curve) based on an alternative, station-based observed NAO index ( /climate-data/ hurrell-north- atlantic-oscillation-nao-index-station-based). (left column) The results based on the full annual data; (right column) the results based on the anomalies with respect to the leading M-SSA pair of the corresponding observed or simulated realization of internal variability (see text for details); the M-SSA embedding dimension M = 20. Comments: (i) The simulated multidecadal variability is much weaker than observed (Figure 1, left column). (ii) Much of this model-data difference is rationalized by the leading M-SSA pair (Figure 1, right column).

h/t to Dr. Leif Svalgaard


175 thoughts on “Another paper shows that climate models and climate reality vary – greatly

    • You mean the dystopian era Trump inherited? He’s clearly working on reconnecting science to reality so that Obama’s fear of a dystopian planet arising from CO2 emissions no longer has its influence on any legislation or regulation.

    • That’s one of the most ignorant and bass-ackward comment I’ve ever read. Science (correctly practiced) is an investigative method for humans to understand reality. Science does not create reality (though many on the warmunist side seem to think it does).

      • Every time I see the acronym CMIP-5 for the “ensemble” of computer climate models, I can’t help but read it as “CHIMP-5”

        Which leads to fond memories of stand-up comedian Bob Newhart’s routine (1960s), in which he imagines a group of scientific technicians actually testing the hypothesis: “If an infinite number of monkeys type on an infinite number of typewriters, they will eventually recreate all the great books of history”

        One of the techs looks at the output and says, “Hey, Charley, I think we’ve got something over here!”

        “This chimp just typed “To be or not to be, … that is the … GAZORNENPLATT”

        A billion bucks worth of computer climate models is worth exactly GAZORNENPLATT

      • Jim, it’s no accident that “The New York Times” is an anagram of “The Monkeys Write.” Gazornenplatt in, gazornenplatt out.

    • I propose a new acronym to provide a shorthand for the atrocious, future destroying, career padding, Science damaging, Fascist, independent thought killing, Socialist, politician posturing, Gorian, Mannian, rational thought attacking and finally FAILING climate paradigm of AGW.
      Climate Rationality And Projections or CRAP

      It’s just quicker and pretty sufficiently accurate. More accurate than their predictions, anyway!

  1. No problem, first rule of climate ‘scince’ when the models and realty differ in value. It is always realty which is error.

    • Which means that ‘climate scientists’ continually need to adjust reality to match the models. This includes the recently discovered Roman and Medieval Not-so-Warm Periods.

  2. I think it would be helpful to include on the chart the date at which the models were produced. This would show where the models were “tuned” to fit known climate data and at what point it moved into “predictive” mode.

    • The CMIP5 experimental design is published was finalized Nov 2011 IIRC. The mandatory hindcast was from YE2005 back three decades. Safe to assume that is also the tuning period. The only ‘experimental design’ option was to initialize with Dec 31 or an average of December.

      • So, why the models run hot even during the hindcast/parametrization period? They start diverging around 1985, Nobody noticed that they were badly parametrized?

        Or were the models hindcasting properly and the 1985-2005 divergence is due to temperature data set adjustment? Is the red line showing what the temperature was before the big adjustments were made?. Chiefly, HadCRUT 3 to HadCRUT 4 in 2012, the big non airport weather station drop in the nineties and the bucket thing. Not sure about the last one.

      • I see, thanks, I did not notice the legend.

        Then, the question is, why the models are tuned to surface temps but compared to radiosonde or satellite temperatures? It would be more logical to compare the output of the models to the temperature dataset used in the tuning.

      • “… the question is, why the models are tuned to surface temps but compared to radiosonde or satellite temperatures?”
        The question is why the troposphere is warming slower than the surface instead of the opposite as predicted by the models?

    • the models were “tuned” to fit known climate data …

      Has anyone tried to compare the models to raw unadjusted temp data…that is providing it even exists

      Models seem to be projecting/predicting the same slope the past was adjusted down to make.

      • Today, after so many failures, I think that raw temperature data must now be tuned to match model output.
        Mainly because they need that final unchanging model that accurately forecasts thermageddon more than they need more real data that clearly doesn’t..

        When I see the letters CMIP5 the following Superbowl advertisement comes to mind… temperatures are DOWN…enjoy!

      • The models where “tuned” to fit the political established UNFCCC claim of future CAGW and that the only salvation is international socialism.

    • Per CAGW theory, the surface warming is a result of the troposphere warming overall 20 percent more then the surface. As it is barely waing at all then the surface warming CANNOT BE CAUSED BY CO2!

      UHI, adjustments, land use change etc… but NOT CO2…..

  3. Would have liked to send this info by direct email, but could not find the way.
    “Scientific American” sent an email to me titled:
    “Learn to Blog About Science”
    Which seems to be encouraging scientists to become activists.
    How did I get an email from SA–don’t know.

    • When a scientist becomes an activist, they stop being a scientist. The two activities are mutually exclusive. Though I doubt SA cares…..

      • Once again Griff is confused. Let me explain. A scientist explains what the science tells us (and what it can’t tell us), and then lets us come to our own conclusions. An advocate will not stop there. They will proceed to tell us what actions we should take in response. If we don’t agree, they will try to force us through the power of government; for our own good of course.

      • The clock starts!
        Every time Griff shows up here he babbles on, getting braver and digging himself deeper; until someone smart takes the time to respond with a thorough, logical and irrefutable argument that utterly destroys him. Then he slinks away like Gollum to report his failures to his masters and hide in the closet for a while.

      • Griff is not wrong in the slightest. Folks on WUWT don’t like being called out on their hypocrisy. Note how not one of the commenters addressed his main point that Curry is both a scientist and an activist. Sad.

      • Chris: Assuming that was Griff’s point, Joel Snider’s response is pretty direct, she’s not an activist. Markw, not quite as direct but still direct enough, she’s not an activist. That’s two, did you overlook them in your haste to qvetch about us folks?

      • No, Griff, Judith Curry is NOT an activist. An activist makes stupid, lying movies like one Al Gore. An activist doesn’t know what a chinook wind is like Leonardo. An activist says stupid things like “The oceans will boil” as did James Hansen. He at least had the courtesy to quite being a scientist and become an activist full-time, perhaps recognizing the two activities are mutually exclusive. Activists are emotional, hysterical people who care nothing about the truth. Judith Curry disagrees with the conclusions of climate “science”, but until she starts a group that marches like the idiots from the other side do in Washington, or starts wearing animal outfits, or some other emotional stupidity, no, she is not an activist. She also must exaggerate the danger and scream like Chicken Little). Of course, you can try redefining the word activist, kind of like the Dems do with words like “bully” (now means you disagree with the Dem and said so and will not back down and cower), but no one who actually understands science will fall for it. Your emotional groupies might, but they are beyond hope.

      • Griff is having a “relativity” problem:

        Griff views himself as a scientific authority; Dr Curry’s opinions are in opposition. So, relative to Griff, Dr Curry is an activist.

    • Nothing says ‘we’re losing badly’ more eloquently than a plea like that. Next phase will be drafting in the Climate Youth and a scorched server directive.

  4. This study will not be popular with the climate modelling community. Clearly more adjustments are required to the 20th century temperature records.

  5. If there was ever a point of real evidence for calling global warming a scam it is these climate models. It is an inherent part of developing computer math models that you correlate to real measured data when it becomes available. The fact that these models are not correlated to real data makes them a joke in the minds anybody with common sense, but the fact that predictions of these models are quoted as part of justification for generating climate study related income through grants and book publishing on the subject, etc, makes many of us think of global warming as a scam. A scam in that people are deliberately trying to fool others into thinking they need to spend more money on global warming related studies and policies than is necessary. Like a pest control company deliberately showing you data that exaggerates the amount of pests in your neighborhood and the consequences of not buying their pest control services. It is completely unethical and a scam. The proof is right in our face, but the media never talks about it. In all the hundreds of news media stories on global warming I have NEVER seen a graph showing measured global temperatures plotted versus time, much less against climate model predictions, but I have heard the vacated 97% consensus study mentioned in almost all of those articles.

    • With some of the adjustments to the historical record temperature, it you might say that the pest control company was not only exaggerating the number of pests, but also bringing a number of them onto your property and letting them loose as evidence of what they are telling you.

    • Agreed – never seen such a chart in the popular press.

    • When the first Earth Resources satellites were launched, the users had to correlate the multi-spectral data with what was actually on the ground. This is called the “Ground Truth”, and was essential for the ERT satellite data to be of any use. Now, it seems that we need some “Ground Truth” to correlate the so-called temperature record against what the temperatures actually were (or are).

  6. As more and more credible studies surface proving that the models greatly overestimate the actual warming, sadly only President Trump is taking the correct action of getting out of Paris. All the other senior politicians in developed countries continue towards the economic cliff as if nothing has happened whilst the scientific basis for Paris is galloping into oblivion.

    • Most world leaders only know what they read in the ‘funny papers’ (aka Summary for Policy Makers) and the same goes for their science advisers. The IPCC and allies in the press are very good gate keepers, keeping any contradictory studies out of sight as much as possible. Most Chief Executives don’t have time to peruse the latest journals on their own, instead they rely on others to process and summarize it for them in digestible bites. Pres. Trump’s only difference is that he has advisors skeptical of the CAGW meme.

    • John

      Wait until developed country senior politicians have to pony up serious cash to meet their commitments, let alone attempt to make up for the loss of USA cash.

      Free-riding on USA over-contributions (UN, NATO, underfunded EU defense budgets…) since WWII has become a habit; some folks just haven’t fully made the connection that when the USA left “Paris”, so did its money.

      We are approaching Josh’s “Where’s my Money!? moment.

  7. The models are clearly running hot. The fact that they can hindcast but not forecast suggests their inner workings are not just trimmed wrongly but are bunkum.

    • ssat

      I think it is worth pointing out that the key portion of the paper above is discussing the internal mechanisms (AMO etc) that are poorly represented. It is one thing to ‘fix’ the temperature match by adding some aerosol cooling or reducing the heating effect of CO2. It is quite a separate thing to correctly model the PDO, which in theory might not even affect temperature. We know of course that it does, but it is separate.

      My inclination, not being a modeler, is to adjust the CO2 effect to reduce the temperature forecast to match the actual temperatures. That would be ‘normal’. But it is also fair to highlight in this article the other significant failures of the models that could be contributing to their skill-free outputs.

      Your assessment as ‘bunkum’ of the inner workings is (apparently) correct. Fixing these workings may or may not bring the temperature predictions into line with reality while maintaining a high CO2 GHG forcing value. Who knows? Until they are fixed the best thing do to is take the model outputs and divide by three. That is apparently more accurate than whatever else they are doing.

      • I could be wrong but I only see two options to the obvious, major failings of model projections:
        Identify and add negative feedbacks to try out on the models, or
        Reduce the effect of CO2 in the models
        Must be issues with that approach I guess ;)

      • John

        A better understanding & incorporation into models of the hundreds of climate variables (not just “parameters) might be advisable, as well.

        Someone might also want to do some work on proving “climate” is a type of system that actually can be modeled.

    • SSAT, there is a simple technical explanation. Correctly modeling important things like convection cells (thunderstorms) is computationally intractable by about 7 orders of magnitude. (See my previous guest post on models here for details.) So such processes are paramterized. The parameters are tuned to best hindcast, for CMIP5 expressly from 1975-2005. This drags in the attribution problem. AR4 WG1 figure SPM 4 expressly says the warming from ~1920-1945 was mostly not AGW; not enough change in CO2. It was mostly natural variation. Yet the warming from ~1975-2000 is essentially indistinguishable to that from 1920-1945. CMIP5 attributes all the 1975-2000 warming to AGW. The attribution problem is that natural variation did not stop in 1975. So since 2000 the models have run hot.

  8. As for Figure S4: Aren’t these oscillation indices known for at least 10 years past where the graphs shown end? The graphs don’t include most of the pause period. If they did, then we can get a better view of the contribution of these oscillations to global temperature change, and a better idea how much of the warming of the past several decades was caused by something other than them.

  9. The question is, “Why don’t the models work?”

    Apology #1: the ocean ate it.

    Apology#2: that doggone chaos makes modeling impossible

    Apology#3: unknown modes of internal variability confound our otherwise perfect models

    Or…you explore what IS known–that CO2 has unusual spectral properties.

    1. The strongest bands were saturated at 280ppm.
    2. The “wings” that remain unsaturated are orders of magnitude weaker.
    3. “Pressure” broadening does not further reduce radiation to space in saturated bands, and it expands the range of saturation.
    4. “P” rotational bands are destructive, they reduce the energy of the molecule.
    5. The Schwarzschild equation improperly assumes an emissivity of 1.

    CO2_just_don’t_work like they think.

    A schedule of IR CO2 transitions:

    • “The question is, “Why don’t the models work?””

      The simple answer is they don’t understand the natural world like they claim to. Can’t accurately model what you don’t understand.

      • and if its chaotic can’t accurately model what you do understand.

        That is probably the most important thing in science that people don’t understand and desperately need to.

        we got ‘correlation is not causation#, but we have yet to understand that ‘deterministic does not imply predictable’..

      • The models don’t work because they are building in all the assumptions about the global warming theory.

        CO2 causes warming at = 0.81C * 5.35 ln (CO2/280)

        Water vapor = increases at 7.0% per 1.0C from CO2 causes warming.

        Water vapor = feedbacks on itself to cause even more warming.

        Clouds = decrease at 1.0 W/m2 per 1.0C from CO2 warming.

        They are not magic black boxes (like the scientists like to pretend). They have rules incorporated into them that were built by Hansen in 1979. .

    • “Why don’t the models work?”

      Because they’re far too complicated, yet far from being complex enough to accurately determine what they’re attempting to predict. Therefore, there are far too many knobs and dials whose values are based on faith, hope and assumptions passed off as an ‘educated’ guess supported with self righteous indignation, for example, an absurdly high climate sensitivity. Others are tweaked to hindcast a short window of the past in a vain attempt to predict the future, except that if you want to predict something 20 years out, hindcasting better be for at a good match going back least 60 years in order for the forecast have any probability of being close and there’s just not enough accurate data going back far enough to do this.

      • I think Bill nails it on the most fundemental level. The have the net affect of water vapor and clouds wrong.

        Clouds are net cooling. SW radiation penetratung the ocean is net warming. LWIR striking the surface is absorbed in evaporation in the first few microns of the surface, causing increased convection, and increaded radiation to space at elevation.

        June in San Diego is a good example. Every day of high T we had there was no cloud cover. Every day 10 to 15 F cooler was overcast. Night time lows were almost the same. High pressure, no clouds, equals warming.

        This does not even consider the SW radiatin on clear days which penetrates the ocean surface. Some of that energy is in our earths system, but lost to the atmosphere for days to years, decades, to centuries!

        Water vapor, even in clear skys, prevents a significant amount of insolation from reaching the oceans.

      • David,
        Yes, the IPCC and its cohorts don’t have the effects of clouds anywhere near correct. But it really doesn’t matter, relative to trying to understand what the sensitivity is and trying to account for clouds and couplings at a microscopic level only adds layers of complexity, more things to get wrong and more wiggle room to support that which can not be supported with the laws of physics.

        All you need to do is examine the planet by its macroscopic yearly averages of input (solar energy), output (planet emissions) and state (surface temperature). How these 3 things interact is easily predictable and testable since their ultimate relationships MUST conform to known physics.

  10. The paradigm of postmodern Science.Inc considers model outputs to be superior to physical facts.

    • If only it were like that Star Trek episode where the populace had to report to disintegration sites because a computer model told them they had been killed in a bombing.
      Perhaps we could persuade the ardent adherents of AGW that they must now report to death camps because sea level rise has inundated their homes.

    • Thanks—excellent. (Brown should be a member of Pruitt’s red team.)

      Since comments on it are closed, I’m posting this one here:

      TYPO in the article: change “pen” to “pan” in:
      “that balances this teetering pen of a system on a metaphorical point”

  11. “If the theory doesn’t match observations then it is WRONG. Simple as that.” Feynman

      • BallBounces,
        What is wrong, the theory or the observations?
        if you are a scientist the theory IS wrong.
        If you are a warmista then the observations ARE wrong.

  12. In 2011 Santer published a revision to the 2008 AMS position that a 15 year discrepancy signaled model problems. Santer said 17 years. It has now been 17 years. And Santer just published his pause paper. Wheels coming off the bandwagon.

    • “In 2011 Santer published a revision to the 2008 AMS position that a 15 year discrepancy signaled model problems. Santer said 17 years. It has now been 17 years. “
      Completely garbled. The AMS paper was about surface temperatures after adjustment for ENSO. Santer’s 2011 paper said (no ENSO adjustment)
      “Our results show that temperature records of at least 17 years in length are required for identifying human effects on global-mean tropospheric temperature.”

      “It has now been 17 years”
      of what?

      • Nick,

        If CO2 be the control knob on global climate, then why don’t three decades of cooling and two decades of flat temperature under steadily rising CO2 falsify the CACA hypothesis? If so-called “climate science” were a real science, the failed CACA predictions would indeed have shown the conjecture false.

        For about 20 years after the PDO flip of 1977, rising CO2 accidentally coincided with slightly rising temperatures. But since CO2 took off after the end of WWII, its steady rise correlated with pronounced cooling for 32 years. The subsequent slight warming for about 20 years was followed by flat temperatures for another approximately 20 years.

        Every other line of evidence also shows the CACA conjecture false.

      • Surely you’re aware that there was no statistically significant global warming for 17 years.

  13. I have an issue with averaging the models’ output as shown in the first chart. I know it is frequently done, but it squashes together all the different initial conditions and parameter selections of each model. That comparing apples, oranges, cherries, grapefruit, lemons, etc. The predictive value of the averages is therefore nil, if it wasn’t already for many other reasons.

    • It’s not comparing apples to oranges, each are attempting to model the same thing. The predictive value of the average is theoretically better than most of the models themselves, and when the models are used to persuade policy, it’s very pertinent to test the models vs reality.

      • RW

        Do you feel that the predictive power of the models, individually, should be tested?
        Do you fell the predictive power of the average of the models should be tested?

        Does the average represent anything real?

        It is interesting to consider that the average has no meaning. It is like holding a popular song writing contest and concluding that the average of all the popular songs will make the most popular song. I suppose there are things in the world that is stupider than that, but it is not easy to find examples.

      • Crispin, how about the worlds average telephone number? Or even better the worlds average user password!
        OMG the average password should be able to unlock all the systems! [sarc]

      • Those are false analogies that have nothing to do with this. A more proper analogy would be to compare the modeled average lap time of a car around a track to the actual times, and by doing so you see if your models are generally too fast or too slow, and then you move on to more advanced analyses, but at least you know in general which direction you should adjust your variables.

        Does the average represent anything real? Yes, it is what it is, an average of the GCMs. By comparing reality to the average of the models you are testing whether the models tend to run too hot or too cold on average, and that is the purpose of doing so, you are trying to read too much into this simple exercise.

        It quite clearly shows that the average is too high, and that’s the only point being made. This in no way keeps anyone from comparing each individual model to reality.

      • @rocketscientist
        Average user name = user
        Average password = password

        You know that’s true!! :-D

      • D.J. Hawkins,
        Those would not be the average user names and PWs, but perhaps the modes (most frequently occurring) .
        Precision please :)

    • It would be nice to also see the “best fit” and “worst fit” model outputs graphed and compared.

    • From the figure captions:

      Figure S4: Raw observed indices (thin lines) and their estimated forced components — ensemble mean (thick lines)

      The timing of NOA or ALPI are not aligned in climate models. So comparing observations vs the model ensemble mean (average of many models) averages out all the short term variability. To compare variability, you need to compare a single model run.

      • If the mean ensemble shows no variability, because the model’s averages have cancelled each other out, then that’s a clear indication that the models are not accurately modeling those indices at all. Notice how all the other indices’ model average show variability and some agreement with reality, those are being modeled much better than the ones I mentioned above…

  14. A bit of a conundrum!

    Adjusting hind cast model data and assumptions so the model output matches observations means that the model is increasingly detached from current scientific and physical reality.

    Building a model based on the underpinning physics and science means that hind cast model output varies with reality.

    We should not jump to conclusions based on either flawed approach but understand what is wrong – there is clearly a disconnect between reality/observations and the science included in the model.

    It is all to easy to adjust parameters within a complex model to generate a wide range of answers. Current models certainly conform and are probably tuned to output the preconceived view of the climate change community. We should try to keep an open mind on the issue until a coherent and consistent story emerges, not bang the table in favour of any one range of outcomes.

  15. Arrgh…they do not ‘vary greatly’ they differ greatly.


  16. As an engineer with some significant experience in using computer models in structural design, I find the application of GCMs in climate science (e.g. CMIP5) to be ludicrous. If I had two models for a truss design that gave significantly different answers for identical inputs I wouldn’t trust either. But if I built, tested and measured the reaction of the design and it agreed with one of the models, l would know which was better. Indeed engineering models are extensively tested and validated in this way before they can be used in practice.

    The GCMs should be evaluated in the same way. If there are, say 39 models that produce widely differing results, it is impossible to determine if any of them have predictive skill. Certainly the average of garbage is still garbage. Now it might be reasonable to compare each individual model’s projections to actual observations and conclude which, if any, are validated – i.e. predicted outcome matches observed outcome within a prespecified limit at a prespecified confidence level. In engineering, for example, we might be satisfied with a deviation of < 2% at 95% confidence.

    When looking at the CMIP5 spaghetti graph, it appears to me that the 'Russian model' comes closest to matching the observations. This would argue for simply throwing out all the other models. Still doesn't mean that the Russian model should be considered validated. I have yet to find any objective standard that defines appropriate criteria for validation of GCMs.

    • I agree. If the model does not “work”, there is something in the model that is wrong.

    • In aerospace, we extensively model aircraft using CFD (Computational Fluid Dynamics) analyses, and then we go test the vehicles in wind tunnels to verify the accuracy of the CFD models. If all looks copacetic we can then place some level of trust in the CFD analyses, to a point. We fully understand the limitations of the equations and realize that extrapolation beyond their validated ranges, especially when flow goes critical and chaotic. Beyond the models capability we do actual tests and make actual observations.

    • But it is a political system, so you can’t throw out anyone’s work. All models are equally “important”. Everyone gets a participation trophy.

    • Rick C PE and others,

      Logically, there can only be one best model, assuming the results are unique and can be repeated. Averaging its results with other models that have poorer predictive abilities provides results that are less than optimum. What should be done is to determine how the model(s) with good predictive abilities differ from those with the worst ability, to understand why the poor models are underperforming.

      • Clyde, logically there can be only one best model if the model actually includes all of the correct variables(which we don’t know). The paper on Stochastic Resonance, the link by RWTurner, points out a lesser know mechanism as to why the climate is a candidate example of an effect that can’t be completely explained through modelling and why the climate appears to have multiple stable states in a long term variable cycle. The Milankovitch cycle dominates the glacier core sample records, but even so, resonance doesn’t always occur at the right time to trip a shift.
        It would be Really Nice To Know if we currently are in a long slide into glaciation or are in one of the interludes that got cold but didn’t really make it to down to the -10degC 100,000 year drop.

    • Agree. Wells said, Rick C PE!
      My experience as a metallurgical engineer in research and development aerospace projects included working with design engineers to sort out ‘short life’ structural test failures. Pattran models of the parts stress/strain performance were revisited, and often run at a finer mesh, to see if anything was inadequate in the part design. Heattran models were employed to assess any thermally induced stress/strain effects on part performance. Without confidence (due to demonstrated predictive accuracy) in these structural design models, much more physical testing would be required before a design could be validated and implemented.

      In contrast, the various climate models provide little to no confidence they represent measured climate variability or performance. They demonstrate low predictive accuracy when compared to measured climate data. We should not commit to significant reductions in CO2 emissions and risk global economic disaster, given the poor predictive performance of global climate models.

    • One thing you would never ever ever do in engineering is take the average of your 50 say altimiter models which all said the plane was flying at 10000 feet, when in fact it was flying at 5000 feet, and use your CONSISTENTLY WRONG answer as the product for your pilots.

      The climate models do this. All attribution studies ( the majority of climate scientists who know diddly about atmospheric physics) use the wrong model mean to make wild ass projections of sea level rise, animals dying, heat waves, etc… It is a giant scam.

    • “A denier will declare “aha, the models are wrong, therefore we don’t need any climate policies!” A skeptic will ask what’s causing the difference between the observational estimates and model simulations.”
      A scientist will go back to the data and theory and find his mistake after admitting the mistake is there. Apparently, Nuccitelli has no idea how science works—but we knew that.

      It’s interesting that the argument always goes like this: Okay, our estimation of what a horse looks like based on all the evidence we have shows two front legs shorter than the back legs, a very long neck, long hair, a short nose and it walks on its hind legs.
      When shown an actual horse, the response is: It has four legs, a head, hair and it can stand on it’s back legs. So WE WERE RIGHT.

    • Dang, I quoted the article and am in moderation. Seems “quoting” requires modification, too, presumably noting I modified it. Sigh……

    • “…about two-tenths of a Watt per square meter per decade.”

      So, Chris, they have made the case for an utter lack of climate sensitivity, though I can’t see how a one thousandth degree Kelvin per decade is even measurable.

  17. Could you unpack the NMO and ALPI acronyms? I know that AMO = Atlantic Multidecadal Oscillation, PMO is the same for the Pacific. NAO=North Atlantic. NMO and ALPI escapes my googling…

  18. Cam_S:

    OK. I read the entire Guardian article you linked and I fail to see it having “debunked” any “argument”.
    What do you think it has “debunked”?

    Please note that the article you have linked does not mention the most pertinent fact concerning so-called ‘climate science’ which is that
    There is no evidence for discernible man-made global climate change; no evidence, none, zilch, nada,


    • In the Skeptical Science world of Cook and Nuccitelli, models are always correct.
      Just an observation… I think it is interesting timing, that the Guardian article was posted about models being accurate. The Mann and Santer paper was published a few days ago, June 19. The paper reported by this WUWT post, published June 15.

    • Richard, you’re out on a limb there, even most sceptics acknowledge there is a CO2 greenhouse effect. As for evidence, you can look at the work of Tyndall or Plass or Arrhenius or any physicist who worked on quantum theories of emission/absorption.

      What aspects of their observations do you find unconvincing?

    • “There is no evidence for discernible man-made global climate change; no evidence, none, zilch, nada,”
      Of course there is.

  19. Jo Nova points out that climate sensitivity estimates have decreased. link

    Given decreasing sensitivity estimates (by non-skeptic authors), the models are looking less and less viable.

  20. I don’t the think the discrepancies ‘hinder’ our understanding. They do illustrate our lack of understanding though!
    Should be “In either case, we strongly believe that model development activities should strive to alleviate the present large discrepancies between the observed and simulated multidecadal climate variability, as these discrepancies hinder illustrate our fundamental lack of understanding of the observed climate change.

    • Here is the “Plain Language Summary” as printed after the Abstract and before the Introduction of the article published by Kravstov and discussed here. Nothing more needed:

      “Plain Language Summary: Global and regional warming trends over the course of the twentieth
      century have been nonuniform, with decadal and longer periods of faster or slower warming, or even
      cooling. Here we show that state-of-the-art global models used to predict climate fail to adequately
      reproduce such multidecadal climate variations. In particular, the models underestimate the magnitude of
      the observed variability and misrepresent its spatial pattern. Therefore, our ability to interpret the observed
      climate change using these models is limited.”

    • rd50,
      I disagree. The poor predictive skills of the global climate models graphically illustrate our fundamental lack of understanding of observed climate change. Until we can identify the true root cause variables and their interactions that drive observed climate change, our attempts to build Global Climate Models will continue to yield rubbish. Setting CO2 emission standards and crippling abundant low cost energy based on such rubbish is irrational to the point scientific schizophrenia.

  21. The whole premise that a ‘model’ is proof of anything is bunkum. The only fact is they are getting away with alarming people, redistributing wealth, ruining efficient and economical energy production, and smearing/shaming non believers and getting away with it. Meanwhile others are sitting around twiddling their thumbs waiting for LIA II to be the savior of science. Sorry folks, but it’s time to fight back.

    • Poor Dana, Nutcase of the SkS propaganda site.

      If you believe a word he says.. then you are stupidly GULLIBLE.

      • ‘But if we can reduce human carbon pollution, we’ll shift to a scenario with a long-term global warming slowdown.’

        Theoretically, natural variables have overwhelmed the AGW signal temporarily, but in the Southern Hemisphere there is a huge global warming signal which is being ignored.

        I speak of the intensification of the subtropical ridge.

  22. It would be nice to see more attention given to the disagreement between IPCC climate models and observation, but this paper is a poor place to start that conversation since it makes a very fundamental mistake right out the gate by “averaging” the output of 102 different climate models, then comparing that value with observation.

    The idea behind an average value is that, due to the “law of large numbers”, error in the observation of a measure will cancel out as more observations are made and, assuming the error is normally distributed, will eventually converge on the true value with greater precision.

    This concept depends on making multiple measurements/observations of the same thing. If you measure the same table 100 times, your measures will be made more precise by averaging. If you measure 100 different tables 1 time, the average of those measures is completely meaningless (literally).

    Sergey Kravtsov makes this error. As a result, his critique is essentially baseless from the very start. A much better approach to demonstrating this problem might be to calculate the correlation coefficient (r-squared) for each individual model with respect to observation. R-squared is a concise measure of the percentage of variability in the dependent variable described by the independent variable and most stats packages produce that metric as a byproduct of a regression. For example a value of r-squared equal to .93 would tell us that 93% of the variability observed in temperature is described by the model. Producing the range of r-squared values over the set of 102 CMPI5 models would give a useful indication of how well the models agree with observation and ranking the models by r-squared provides an objective method for selecting the model that best fits observation. Any model that failed to describe at least 60% of the variability in observed temperature should, in my opinion, be rejected outright.

      • I am responding to your comments on the Kravtsov article.
        Here is what you wrote:

        “It would be nice to see more attention given to the disagreement between IPCC climate models and observation, but this paper is a poor place to start that conversation since it makes a very fundamental mistake right out the gate by “averaging” the output of 102 different climate models, then comparing that value with observation.”

        Try to find the “averaging” the output of 102 different climate……” in the Kravtsov article.
        The models he used are in Table 1 of the Kravtsov article. See how many were used and how they were used. You are simply wrong assuming he used the average output of 102 …..

        You simply did not read the article.

      • RD, I read this article. I did read the abstract of the other. If the assertion and graphic in this article is the one I was commenting on (and it was) the comment stands. I read the article.

      • RD, if you feel the Kravtsov paper has been misrepresented in this article I strongly encourage you to take that issue up with its publisher.

      • RD50 writes: “You are simply wrong assuming he used the average output of 102 …..”

        I made no assumption. The first graphic presented in this article (the one I am commenting on) very specifically labels the red chart line as the average of 102 CMIP 5 model runs. There’s no assumption; that’s the legend.

        My criticism is in response to this article. If this article has deliberately, or in error, misrepresented the research being reported, that’s a problem you have with its publisher, not me.

      • I asked you:
        Did you read the article?
        You responded; YES
        Now you pretend that the article you responded to was the graph published introducing the published article!
        A pretty graph with a red line going up being the average of 102 models.
        This graph was NEVER EVER introduced by Kravtsov in his article. But you blamed Kravtsov for using it.
        Sure, now you have to try to defend yourself one way or the other that you were not responding to Kravtsov.
        Sorry, the Kravtsov article this what you were responding to. You selected the wrong graph, a graph he never used.
        Read again your response.
        You specifically wrote: “Sergey Kravtsov makes this error.”
        What was the error Sergey Kravtsov made? According to you: using the average of 102 models.
        So you were responding to the article of Sergey Kravtsov erroneously, pretending that you read the article!
        You never read the article by Kravtsov.
        Just give me a response of how many models he listed in Table 1 of his article and I will upload Table 1 of the article here.
        The list in his Table 1 is quite different from your 102 list of models you used to unfairly criticize him.

      • Waited long enough for your response.
        Here is a copy of Table 1 listing the models selected:

        Table 1. CMIP5 Twentieth Century Simulations Used in This Studya
        Model # Model Acronym Historical Historical GHG Historical Nat PI Control
        1 CanESM2 5 5 5 996
        2 CCSM4 6 3 4 501
        3 CNRM-CM5 10 6 6 850
        4 CSIRO-MK3–6-0 10 5 5 500
        5 GFDL-CM2.1 10 – – –
        6 GFDL-CM3 5 3 3 500
        7 GISS-E2-Hp1 6 5 5 540
        8 GISS-E2-Hp2 6 – – 531b
        9 GISS-E2-Hp3 6 – 5 431b
        10 GISS-E2-Rp1 6 5 5 550
        11 GISS-E2-Rp2 6 _ – 531b
        12 GISS-E2-Rp3 6 – 5 531b
        13 HadCM3 10 – – –
        14 HadGEM2-ES 5 3 4 575
        15 IPSL-CM5A-LR 6 3 3 1000
        16 MIROC5 5 – _ 770
        17 MRI-CGCM3 3 1 1 500
        Total 17 models 111 simulations 39 simulations 51 simulations 9306 (7282) years
        aWe selected the models with four or more historical realizations (fourth run for the MRI model was not available) and
        analyzed the runs for which sea surface temperature (SST), surface air temperature (SAT), and sea level pressure (SLP)
        outputs were all available. Listed are the following: The number of realizations in historical runs with all forcings included
        and in the runs with greenhouse gas (GHG) and natural (Nat) forcings only, as well as the length of the preindustrial control
        runs (in years).
        bThe (low-variance) PI control runs of GISS models were not included in the final control-run ensemble to compensate
        for the absence of the (high-variance) GFDL-CM2.1 and HadCM3 control runs (see KC2017).

        Difficult to read. I agree but a list of the models used is there. Much better presentation if you download the article with Table 1.
        Nevertheless it proves beyond a reasonable doubt that the average of 102 models was never used by Kravtsov.

      • RD50 writes: “This graph was NEVER EVER introduced by Kravtsov in his article. “

        This appears to be the root of our misunderstanding. My critique was of this article, the one we’re discussing. I expressed trust that the leading graphic, which is very clearly identified as representing the”Average of…” was a true and correct representation of the methods used by Kravtsov. I stand by the criticism I made in that context”.

        As I mentioned earlier, if you believe this publication has in some way misrepresented Kravtsov’s work, the issue isn’t with me, it’s with the editor of this publication. My criticism remains valid.

      • I’d like to add (@RD50), that had there been some number of models less than 102, for example 17, which where run several times each, then aggregated into an “average”, the criticism I’ve made still stands. This is a fundamental truth of valid statistical methods.

        It’s very important you understand the purpose of an “average” before you can comprehend the criticism I’ve made above. If that remains beyond your ken, this really isn’t the place to rectify that deficiency.

      • Barb,

        Yes. Even though the models and runs with ECS above 2.0 degrees C per doubling are clearly wrong, IPCC must leave them in so that the future looks scary.

      • Jones, while I’d like to use your reference I’m afraid I can’t really accept it as a curated source. I won’t be spending $1139 to read the original, most especially since it was completely funded by US taxpayers, of which I am one.

        If it shows up on ResearchGate, a curated source I trust, I’ll be happy to review it in full.

    • Nobody should be subject to Peter Sinclair’s condescending factually incorrect videos.

      (unless your like your condescending incorrect information which lots of people do).

      • Pete is a second generation enviro-N@zi trough-feeder with a Bachelor of Fine Arts degree.

        Hilarious that the worse than worthless, waste of oxygen dweeb imagines he can breathe the same air scientifically with Anth@ony.

  23. From key quote # 3 :

    The magnitude of the CMIP5 “internal” variability estimated by this method is, hence, much larger …

    I am confused by the use of the phrase this method – ” which ” method is ” this method ” above, referring to ?? Anybody can help me ?

  24. It would have been polite to tell us that the article was by Sergey Kravtsov.

    I love the plain language summary.

  25. If the hiatus continues for another ten years then we can safely say the Lukewarmers win.

    • In that case, the non-warmers win.

      Despite a physical “greenhouse effect”, that would tell us that in the actual climate system, it doesn’t happen, probably because of net negative feedbacks, which is just what should be expected on a self-regulating, watery planet.

      • Lose they must because Mother Nature says so.

        The sad truth is that the cr!minals will suffer not penalties for their cr!mes.

      • If the Lukewarmers win then the Klimatariat is let off the hook, its a sensitivity issue.

        CO2 may yet have a case to answer if a Gleissberg Minimum fails to show.

        ‘Solar cycle 24 has turned out to be historically weak with the lowest number of sunspots since cycle 14 peaked more than a century ago in 1906. In fact, by one measure, the current solar cycle is the third weakest since record keeping began in 1755 and it continues a weakening trend since solar cycle 21 peaked in 1980.’

        Paul Dorian

  26. The jig is up.

    The CACA charlatans lose. Reality wins.

    The question now is, are Hansen, Schmidt, Jones, Mann, Trenbeth and co-c@onspirators charged only with fraud, or with crimes against humanity, as is totally warranted.

    Using RICO, the Trump DoJ should sweat the underlings to get to the ring leaders, just as with the Mafia. Is Briffa an underling or a capo di tutti capi? If he sings like a canary, then he’s a mere soldier, not a capo. Overpeck, however, I have to go with capo.

  27. The habit of referring only to the cold season NAO as if the rest of the year doesn’t matter, only serves to confuse its relationship with the AMO. The 3 month running mean shows a positive NAO regime from around 1963, and shifting negative from around 1995, well in time with the AMO phase shifts. While the JFM cold season alone shows little such congruence with the AMO envelope.

  28. Gabro

    “……. he can breathe the same air scientifically with Anth@ony.”

    Nice to hear that Anth@ony knows science


    I have a scientific background and education but got dragged into finance out of school and thus haven’t worked in the field and greatly appreciate all that Anthony and the knowledgeable posters here have taught me. I have run across this link being thrown around to support the argument that the climate models work well and would appreciate if one or more could provide me with a succinct rebuttal.

  30. Remote Sensing Systems (RSS) just released their updated lower troposphere temperature data set from v3.3 to v4.0:

    The result is a hugely increased rate of warming over the course of the data compared to the previous data set. The decadal tend since 1979 is 0.184 C/dec, very close to, though a little faster than the surface data sets, including GISS (0.176 C/dec since 1979). Still slightly below the CMIP5 multi-model average over the past 20 years, but much warmer than UAH’s equivalent TLT.

    This change is primarily due to the changes in the adjustment for drifting local measurement time:

  31. The models work exactly as they are designed to work and provide the precise results they are created to show.

  32. I just want to highlight that Kravtsov is one of the main researchers on the Stadium Wave Theory. I believe much of this work actually falls out of the ongoing work related to the Stadium Wave Theory. Kravtsov was an advisor to Marcia Glaze Wyatt who developed the Stadium Wave Theory as her PHD dissertation. Curry was also an author to some of the relevant papers, but Kravtsov has been doing a lot of the heavy lifting both initially and now as the Theories testable claims are coming to the forefront.

    If the Stadium Wave Theory proves out we should see increasing sea ice across the top of Europe/Asia in the next 5-10 years.

    • Really? Bunches of drunk people standing up and sitting down while waving their hands in the air affectys the climate? Oy vay.

      • Wow, going by that graph the Gulf Stream is slackin’ off. Better dock its pay till it gets on the ball!

      • I told you. They ran out of beer in Mexico, so they quit peeing in the gulf.

        More seriously, 70-90 year cycles seem real. That the first place I’ve seen that has already topped out for this cycle. Next should be eurasian ice extent. (ie. the ice from scandinavia to Eastern Russia.) If the Stadium Theory is right, that part of the world should have topped out and be headed down soon. Look at that multi-colored sine wave chart I posted. The theory is the energy gets pushed from one cycle to the next. ngAMO is in the first set of waves. That’s negative AMO. In theory it will lead the rest of the world into a cooling phase over the next 20 years.

      • Oh, don’t even get me started on “cycles”, they inter/commingle. And do not even mention the SUN in any of this! That sets off the leftards like nobody’s business. Except for the whole “I can choose my gender” horsesh*t. First the LGBTQUACBRTMLBLAHBLAH said you are born gay and can not change it, now they say no matter what gender you are physically born you can “choose” what you are. Am I, honestly, the only one who sees how these two “philosophies” mesh together so conveniently?

Comments are closed.