Defending the Uncertainty Monster paper

by Judith Curry

I’ve completed a revised draft of my response the to Reply to our Uncertainty Monster paper.

When I first received the Reply, my inclination was not to respond.  The Reply spouted the IPCC “party line,” about which I felt that our paper raised significant concerns.  Further, this came at the worst possible time in terms of my crazy travel schedule a few weeks ago.  However, the editors of the journal convinced me to reply.  We dashed something off rather quickly.  The reviewer of the Reply and response was quite critical of our response, and further reiterated the IPCC party line.  So we’ve taken the opportunity to rework our response, and I now think it is much better.

We need to resubmit our revised version by Sunday.  It would be inappropriate for me at this point to mention the authors of the Reply or post their Reply, but you can get a sense of it from our response.  I would appreciate your comments on this, you will definitely see the impact of our earlier traceability discussion.  Thanks in advance for your comments.

Reply to X et al.’s Comment on “Climate Science and the Uncertainty Monster”

Judith A. Curry and P.J. Webster

Abstract.  X et al.’s comment provides us with a further opportunity to emphasize and clarify our arguments as to why the treatment of uncertainty in the IPCC AR4 assessment regarding attribution is incomplete and arguably misleading.

We would like to thank the authors of the Comment, all of who played leadership roles in the IPCC AR4, for their interest in our paper. The authors are correct that since the Third Assessment Report, the IPCC has placed a high priority on communicating uncertainty, and perhaps their methods have been effective in communicating with the public.  However, communicating uncertainty is a very different endeavor from actually characterizing and understanding uncertainty, an effort in which we feel the IPCC falls short.

Curry and Webster (2011) raise the issue of how the IPCC has actually undertaken to investigate and judge uncertainty.  X et al.’s comments focus on section 4  Uncertainty in attribution of climate change of our paper (Curry and Webster, 2011), which addresses the IPCC AR4 conclusion regarding attribution: “Most of the observed increase in global average temperatures since the mid-20th century is very likely due to the observed increase in anthropogenic greenhouse gas concentrations.”

The text in the IPCC AR4 (chapter 9) referenced by X et al. describes general issues and methodology for investigating uncertainty on the topic of attribution, including listing uncertainty locations. In preparing our original manuscript, we read these passages and the cited references numerous times.  Listing a large number of uncertainty locations, and then coming up with a ‘very likely’ likelihood statement using expert judgment in the context of a consensus building approach, is at the heart of our concern regarding the IPCC’s treatment of uncertainty.

X et al. object to our statement in the original manuscript: “Figure 9.4 of the IPCC AR4 shows that all models underestimate the amplitude of variability of periods of 40-70 years” on the basis that we do not consider the uncertainties presented in the chapter.  Figure 9.4 is presented on a log-log scale, and the magnitudes of the uncertainties for both the model simulations and the observations are approximately a decade (factor of 10).  Accounting for uncertainty, a more accurate statement would have been:  The large uncertainties in both the observations and model simulations of the spectral amplitude of natural variability precludes a confident detection of anthropogenically forced climate change against the background of natural internal climate variability.

X et al. state: “The remaining uncertainty in our estimates of internal climate variability is discussed as one of the reasons the overall assessment has larger uncertainty than individual studies.” Translating this uncertainty in internal climate variability (among the many other sources of uncertainty) into a “very likely” likelihood assessment is exactly what was not transparent in their assessment.  We most definitely “do not appreciate the level of rigor with which physically plausible non-greenhouse gas explanations of the recent climate change are explored,” for reasons that were presented in the original manuscript. It is our assessment that the types of analyses referred to and the design of the CMIP3 climate model experiments do not support a high level of confidence in the attribution; we do not repeat the argument that we made in our original manuscript.  Circular reasoning regarding 20th century attribution is associated with any model whose parameters were tuned to the 20th century climate simulations and whose 20th century forcing was not chosen in an a priori manner prior to actual referenced attribution simulations.

X et al. take issue with our statement “The high likelihood of the imprecise ‘most’ seems rather meaningless.”   Whereas X et al. disagree with our statement, the InterAcademy Council Review[1] of the IPCC seems to share our concern: “In the Committee’s view, assigning probabilities to imprecise statements is not an appropriate way to characterize uncertainty.”  While this passage from the IAC Review was cited in our original manuscript, we repeat it here for emphasis.  Assigning a ‘very likely’ likelihood to the imprecise ‘most’ is not an appropriate way to characterize uncertainty.

With regards to the issue of traceability, the IAC made the following recommendation:

“The IPCC uncertainty guidance urges authors to provide a traceable account of how authors determined what ratings to use to describe the level of scientific understanding and the likelihood that a particular outcome will occur. However, it is unclear whose judgments are reflected in the ratings that appear in the Fourth Assessment Report or how the judgments were determined. How exactly a consensus was reached regarding subjective probability distributions needs to be documented.”

Xet al.’s assertion that they provided a traceable account of their attribution statement raises the issue of exactly what is meant by IPCC’s traceability guidelines, and what kind of traceability is actually suitable for IPCC’s assessments. A general description of the method and multiple lines of evidence ‘traceable’ to published papers is not adequate for traceability of the assignment of the confidence/uncertainty statement. Traceability allows an independent person or group to trace back to understand how the result came to be and to walk through the decision process and achieve the same result. Commonly used practices in system engineering include the traceability matrix and document control. Some fields (e.g. medical science, computer science, engineering) have stringent traceability requirements, particularly for issues that are mission critical or having life and death impacts.  However one size doesn’t fit all, and the level and type of traceability required are related to the complexity of the subject matter and the criticality of the final product.  With regards to the IPCC assessment reports, an independent team should be able produce the same result (assessment and assignment of uncertainty/confidence level) from the same material. Traceability is fundamentally about accountability and openness.

We clearly disagree with X et al., and we believe that our position is justified by our arguments.  We are not the only people that are unconvinced by the IPCC’s attribution assessment.  The existence of this disagreement is not surprising given the complexity of the issue.  The existence of this disagreement implies, at the very least, one of two things:  1) the IPCC AR4 has done an inadequate job in articulating the evidence for their position; or 2) the uncertainty is much greater than acknowledged by the IPCC AR4.

Acknowledgements.  Comments from the Denizens of Climate Etc. judithcurry.com are greatly appreciated.  Particular thanks to Steve Mosher, John Carpenter and Pekka Perila for their input on traceability.


122 responses to “Defending the Uncertainty Monster paper

  1. Since the UN’s IPCC reports are used in making policy decisions, uncertainties need to be stated so clearly that politicians cannot conveniently ignore the uncertainty.

    The failure of UN’s IPCC reports to meet that criterion is obvious.

  2. Judy – I believe you have provided compelling reasons why uncergtainty and traceability need to be addressed rigorously, and it is hard to disagree with your perspective on that. That said, however, I continue to believe that you have chosen the wrong example in challenging the AR4 attribution statement that it is very likely that most warming after about 1950 was due to anthropogenic GHG emissions. I won’t reiterate what I see as strong evidence in support of that attribution, having discussed it in several threads, but I will only summarize one aspect of that evidence by suggesting that if a different interval had been chosen, a similar attribution would have been unjustified based on both the forcing data and the magnitude of net unforced variability between the beginning and end of the interval.

    In my view, an emphasis on uncertainty would be enhanced by choosing a different attribution as a focus for concern.

    • I agree with you Fred.

      I think the attribution statement is actually relatively weak. ‘Very likely’ and ‘most’ seems to imply the corollary that ‘less than half but up to half’ of the warming can be attributed to ‘other’, with the distinct possibility (‘significant’) that the ‘other’ contributed more than half. And the evidence for that is where, precisely?

      If the majority of climate scientists think this overplays the certainty (or rather underplays the uncertainty) of AGW then I fear I have been misreading the literature.

      • I don’t think that’s the way most people read “very likely” and “most”. And remember who the target audience for the AR is.

      • I agree about the target audience.
        Aren’t the ‘very likely’ and ‘most’ characterisations of ‘with greater than 90% probability’ and ‘more than half’? – I thought AR4 was quite specific about those?
        Or have I misconstrued that this also means ‘up to 10% probability’ and ‘up to half’ for the attribution of warming to ‘other/natural variability’. ?

      • Those are risible equivalences in scientific literature. The point of having numeric confidence levels is to control the potent effects of data selection, confirmation bias, etc., in shaping and assessing data and speculations/hypotheses/theories. Even “99%” is inadequate to control these dangers, as amply demonstrated in the last century or two of scientific reporting. A 90% statistical confidence level is “very likely” to be a fudged and meaningless conclusion, not a correct one.

    • @Fred Moolten…

      I can’t agree with you, Fred. If we look at the figure referenced, which I’m assuming is somehow Figure 9.7 of AR4 from the IPCC Website (I don’t understand why everybody is calling it Figure 9.4, but since everybody else seems to understand I’m assuming it’s my ignorance), we see enormous error bars (as the text above notes), along with this quote from the caption:

      All models simulate variability on decadal time scales and longer that is consistent with observations at the 10% significance level.

      Given that the line itself for observed spectral density is higher than any simulation in the range of ~40-70 years, a key time frame involved in determining attribution for the last 50 years, the translation of this 10% significance level to a >90% probability is intuitively unreasonable, unexplained, and results from an untraceable, undocumented methodology.

      Looking for a reason for this (probably false) confidence, I find a reference to Figure 9.5 which compares observed and modeled global average temperatures with and without all presumed anthropogenic forcings. No error bars are present. This presents a powerful visual message in favor of anthropogenic influences, despite being (arguably) highly deceptive.

      The deceptiveness isn’t quite what is at issue here, rather it’s that WG1 has failed to document any sort of audit trail between the 10% significance in Figure 9.4/9.7 and the 90% confidence of their attribution. No explanation why the attribution confidence isn’t smaller than 10% (no matter how confident they were otherwise), nothing to stop me (or any other observer) from thinking that somebody looked at Figure 9.5, nodded their head(s), and forgot all about the error levels in Figure 9.7. (And don’t tell me 10% significance and 90% confidence is comparing apples and oranges, show it to me in the text of AR4. We’re talking about communications here, as well as traceability.)

      • P.S. I’d sure like to see Figure 9.5 with the BEST results overlaid, including error margin(s).

      • the translation of this 10% significance level to a >90% probability is intuitively unreasonable, unexplained, and results from an untraceable, undocumented methodology.

        I am surprised more people do not find that to be a significant shortcoming. Going from 10% significance level to 90% confidence in the hypothesis is a common simple error.

        No error bars are present.

        Again, I am surprised that more people do not find this objectionable.

    • What I believe that some miss is that the method the IPCC largely used for attribution was their climate models. They could only see the human contribution of increased CO2 to account for the CO2 increase and it was the additional CO2 that seemed fit to make their model work best to account for the additional heat.

      I believe in a nutshell that is how it was done, but I appreciate it if someone points out where I am wrong.

      • Rob, you did not include that to reach the conclusion and to justify that the circular assumptions entailed in using the models were acceptable based on the premise that paleoclimate reconstructions were considered and indicated agreement with the models..

        I don’t know how much Dr. Curry wishes to accent this, but without the methodology of this comparison, then Fred’s comment above is moot. In other words, Fred’s methodology is not the methodology stated in AR4. Not only do we have the variability and uncertainty of natural variations, and model performance, but uncertainty must be examined with respect to uncertainty of the reconstructions and how this effects the conclusions per attribution. Dr. Curry above states “”Circular reasoning regarding 20th century attribution is associated with any model whose parameters were tuned to the 20th century climate simulations and whose 20th century forcing was not chosen in an a priori manner prior to actual referenced attribution simulations.””

        There is still the circular argument that Dr. Curry is correct on

      • John, AK, Rob Starkey and others – Without the model uncertainties, I expect it would have been possible to attribute substantially more than half of post-1950 warming to anthropogenic GHGs – the fairly weak reference simply to “most” of the warming reflects uncertainties. Some of the uncertainties are largely irrelevant, however, because they end up canceling out between the beginning an end of the interval – the exact magnitude, for example of aerosol forcing makes little difference to the final attribution even though it contributes a high proportion of the model uncertainty. In my previous comments in several threads, I also referred to the internal modes – AMO, PDO – where again, a different interval might have lent them greater proportional weight, but where their post-1950 net contribution was far less than the maximum amplitude of either. For the interval post-1950, it is hard to make a plausible case for anthropogenc GHGs to have played less than a majority role when all the forcings and unforced variations, along with their uncertainties, are added up. That would not be completely impossible, of course, but that is why it was appropriate to use the term “very likely” rather than “certain”.

      • Fred

        Singer wrote that the use of GCM’s was the primary basis for the IPCC attribution assessment and not other means.
        You write: “I expect it would have been possible to attribute substantially more than half of post-1950 warming to anthropogenic GHGs”

        I might agree that your perspective seems “reasonable”, but there is no means to prove it so scientifically as far as I can tell. If you listen to people like Eli you will hear claims of isotropic data being used, but that is not possible within the margin of error required. How would you PROVE that the additional heat came from CO2 and then PROVE that the additional CO2 came from humans?

        It is possible that non human released CO2 increased and that natural absorption decreased isn’t it?

      • It is possible that non human released CO2 increased and that natural absorption decreased isn’t it?

        I think some of this was discussed previously on the Salby thread. Very short term fluctuations of the sort you mention are possible, be we know that over the 55 years from 1950-2005, anthropogenic emissions continued to add well quantified annual amounts of CO2 to the atmosphere (in excess of what persists there) and that the airborne fraction has remained fairly constant (with some very recent evidence suggestive of a slight increase). This tells us that all or almost all increases in atmospheric CO2 continue to be of human origin. There are many more details in the previous thread.

      • Fred

        I question your statement and conclusions.

        You wrote: “we know that over the 55 years from 1950-2005, anthropogenic emissions continued to add well quantified annual amounts of CO2 to the atmosphere (in excess of what persists there) and that the airborne fraction has remained fairly constant (with some very recent evidence suggestive of a slight increase). This tells us that all or almost all increases in atmospheric CO2 continue to be of human origin. There are many more details in the previous thread.

        My response: I agree that we know = or – maybe 25% how much CO2 were emitting during the period you reference. (I hardly think that is “well quantified” however)

        When you wrote- “that the airborne fraction has remained fairly constant” This tells us that all or almost all increases in atmospheric CO2 continue to be of human origin.

        How does the airborne fraction remaining fairly constant (your words) have anything to do with attribution to humans?

      • The problem is that models in AR4 are not actually physical, as in physics. Whether they use hyperviscosity, convective adjustments, or other parameterizations such as aerosol adjustments, they have one standard to test against, the historical temperature. In this sense they are “tuned.” The problem is one of circularity. The IPCC in AR4 chapter 9 developed a methodology that recognised this inherrent problem. Your claim of “”For the interval post-1950, it is hard to make a plausible case for anthropogenc GHGs to have played less than a majority role when all the forcings and unforced variations, along with their uncertainties, are added up. That would not be completely impossible, of course, but that is why it was appropriate to use the term “very likely” rather than “certain”.”” with just models is what the IPCC determined in Chapter 9, due to their correct admission that without the paleo support ,it was a circular argument wrt the assumptions made that are inherrent in an attempt to model with only one realizable event, the historical record.

        The very likely and much less certain wrt models was not agreed nor claimed in AR4, except as reasonable within the framework they posited which included paleo reconstructions and conclusions. They state “”No climate model that has used natural forcing only has reproduced the
        observed global mean warming trend or the continental mean warming trends in all individual continents (except Antarctica) over the second half of the 20th century.”” p665. “”While many aspects of these past climates are still uncertain, key features have been reproduced by climate models using boundary conditions and radiative forcing for those periods. A substantial fraction of the reconstructed Northern Hemisphere inter-decadal temperature variability of the seven centuries prior to 1950 is very likely attributable to natural external forcing, and it is likely that anthropogenic forcing contributed to the early 20th-century warming evident in these records.”” and “”Better understanding of instrumental and proxy climate records, and climate model improvements, have increased
        confidence in climate model-simulated internal variability.However, uncertainties remain.””p666.

        “”The objective of this chapter is to assess scientific understanding about the extent to which the observed climate changes that are reported in Chapters 3 to 6 are expressions of natural internal climate variability and/or externally forced climate change.”” and “”In this chapter, the methods used to identify change in observations are based on the expected responses to external forcing (Section 9.1.1), either from physical understanding or as simulated by climate models. An identified change is ‘detected’ in observations if its likelihood of occurrence by chance due tointernal variability alone is determined to be small.”” Introduction p667.

        The point occurs here “”Both detection and attribution require knowledge of the internal climate variability on the time scales considered, usually decades or longer. The residual variability that remains in instrumental observations after the estimated effects of
        external forcing have been removed is sometimes used to estimate internal variability. However, these estimates are uncertain because the instrumental record is too short to give a well-constrained estimate of internal variability, and because of uncertainties in the forcings and the estimated responses.Thus, internal climate variability is usually estimated from long control simulations from coupled climate models. Subsequently,an assessment is usually made of the consistency between the residual variability referred to above and the model-based
        estimates of internal variability; analyses that yield implausibly large residuals are not considered credible (for example, this might happen if an important forcing is missing, or if the internal variability from the model is too small). Confidence is further increased by systematic intercomparison of the ability of models to simulate the various modes of observed variability (Chapter 8), by comparisons between variability in observations and climate model data (Section 9.4) and by comparisons between proxy reconstructions and climate simulations of the last millennium (Chapter 6 and Section 9.3).””

        So as to Craig’s point about tuning, yes it is specified here. Not all aspects of the tuning just one where it is implied that all their models wrt attribution are tuned. This represents a circular argument. Other points about the necessity of apleo and tunign can be found in the rest of ch 9.

      • @Fred Moolten…

        Fred, I’m not sure you read my comment.

        While I’m perfectly willing to debate the level of certainty involved, the issue at hand is the audit trail showing how the AR4 calculations of uncertainty managed to ignore that 10% significance level. You can make your argument (WRT uncertainty) and I can make mine, but neither of us can compare it to the AR4 argument, because there isn’t any. They lay out a bunch of data, document their conclusion, but don’t say anything about how they got from one to the other.

        You mention “…the internal modes – AMO, PDO – where again, a different interval might have lent them greater proportional weight, but where their post-1950 net contribution was far less than the maximum amplitude of either,” but AFAIK these are 0-2 decade “internal modes“, while the subject under scrutiny is 4-7 decade modes. Eyeballing the graphs from BEST, I see a strong possibility that some number of oscillations in the 4-7 decade range may be combining to push the latest warming trend to a level higher than anything since the MCO. (i.e. Without any anthropomorphic contribution.)

        Now, I don’t want to debate (here, in this thread) the relative probabilities of my hypothesis vs. yours. What I want to do is determine, from the documentation of AR4, whether this explanation was considered and factored into their determination that there was a >90% probability that at least half of the warming was due to anthropomorphic “causes”. If they didn’t consider it, how would their conclusions change if it were factored in? If they did, which reference should I look at to falsify my hypothesis (or not)?

        When I do look, I find all the models as well as the observed showing a high spectral density in the 4-7 decade range, I find the statement that “All models simulate variability on decadal time scales and longer that is consistent with observations at the 10% significance level,” and I find no explanation why they came up with a >90% confidence in their attribution. They state that there’s no other plausible explanation, but don’t explain what’s implausible about the hypothesis that the combination of 4-7 decade oscillations in the real world just happened to reinforce to produce the real-world observed record.

        Presumably they used some sort of Bayesian analysis on the “58 simulations produced by 14 models with both anthropogenic and natural forcings.” (Didn’t they? If not, why not?) What were their priors? What probability did they assign to the possibility that all 14 models shared incorrect assumptions causing their handling of 4-7 decade oscillations to be wrong all in the same way? If I wanted to assign a higher probability to that source of error, is there enough information to re-run their calculations and get a new value for confidence in their >90%/>50% conclusion?

        There appears to be a tacit assumption built into their whole process, that any source of error not studied and dismissed in the reports they chose to include has no chance of explaining the rise. Is this true? I don’t mean do you think this is true, I mean can you find an answer in the AR4 documentation? How easily?

        Is there a list of possible sources of error? How easy to find? (I looked, and couldn’t find it quickly.) Is there an audit trail where I can find how each possible source of error is evaluated, with references? And so on.

        The >90%/>50% conclusion is the key conclusion, the take-home from the whole exercise. It may or may not be the weakest, but any weakness it has is far more important in terms of selling the science. This is the point of the crowbar trying to pry major political institutions out of their do-nothing attitude, and this is therefore the one that everybody opposed to any specific action on AGW will be looking closely at. And it’s shaky. Very shaky. Not the science so much, that’s debatable, but the audit trail.

      • AK – The audit trail is worth pursuing as you note. However, the original challenge was to the attribution statement itself that it is very likely most post-1950 warming was attributable to ghgs. I believe there is enough evidence in AR4 to justify that statement, and certainly enough in the literature itself, but you are right that we should determine whether that evidence is adequately summarized in one place. I will have to revisit AR4 to see if I can get a handle on that. I doubt that the figures cited above are helpful, because they don’t distinguish between anthropogenic ghgs and anthropogenic black carbon. I’m not sure, however, that the models are even directly necessary to justify the attribution. I will try to hunt down more attributable evidence regarding the audit trail question.

      • @Fred Moolten…

        However, the original challenge was to the attribution statement itself that it is very likely most post-1950 warming was attributable to ghgs.

        My understanding is that xxx et al. were actually challenging the traceability, as well as the uncertainty estimate. At least, that was the thrust of their stooge reviewer(s). WRT the traceability, it’s not so much whether the evidence was cited, as whether an audit trail was in place so we can see how it was used.

        My guess is that they couldn’t lay out an audit trail, because it would have shown a bunch of unjustifiable assumptions. Either they didn’t include it at all, or they buried it well enough that nobody looking for it could find it, and the reviewer(s) dragged a red herring across the trail rather than cite it. Why?

        WRT evidence, can you explain why they translated a 10% significance level for 4-7 decade internal variation (with error bars on the order of 1000%!) into a >90% confidence level that this variation didn’t explain more than half the 1950-2000 increase?

      • AK,

        I might be misunderstanding your point here but ‘confidence level’ and ‘significance level’ have an inverse relationship so it’s simply a tautology that 10% significance = 90% confidence. See here

      • @Paul…

        I might be misunderstanding your point here but ‘confidence level’ and ‘significance level’ have an inverse relationship so it’s simply a tautology that 10% significance = 90% confidence.

        Thanks Paul. I’m more used to seeing things like “p<0.001” and so on. But does this mean that because they were 90% confident that each model was replicating the observed spectral density, they were 90% confident that the models had correctly replicated reality. What about all the other sources of error?

        Does this mean they were 100% confident that the models had included every possible factor? Should I assume this is their audit trail, they simply subtracted this particular significance from 1? Ignoring any other sources of error?

      • AK,

        I very much doubt it. All that figure says is GCMs are able to produce similar levels of internal variability to what has been seen in observations, and therefore models can reasonably be used to determine whether, and how far, a trend is appearing above natural variability. I would guess this piece of evidence is used more for the ‘most’ part than the ‘very likely’.

        There are many other pieces of evidence described in the chapter – spatial fingerprints of GHG warming, comparisons with natural forcings etc. – and I would suggest it’s all used to build to a conclusion. That’s given me an idea for traceability of thought processes leading to an expert conclusion: would it be possible to attach some kind of bayesian weighting to each morsel of evidence showing to what extent it strengthened or weakened belief in a particular conclusion?

      • @Paul S…

        Now we’re getting somewhere.

        I would guess this piece of evidence is used more for the ‘most’ part than the ‘very likely’.

        There are many other pieces of evidence described in the chapter – spatial fingerprints of GHG warming, comparisons with natural forcings etc. – and I would suggest it’s all used to build to a conclusion.

        which was the thrust of my argument: where’s the audit trail.

        That’s given me an idea for traceability of thought processes leading to an expert conclusion: would it be possible to attach some kind of bayesian weighting to each morsel of evidence showing to what extent it strengthened or weakened belief in a particular conclusion?

        An excellent idea. Something the IPCC evidently didn’t do with AR4. If they’d done that, I could pinpoint which assumptions I disagreed with their weighting (and calculation).

        Even more important, what about their priors? Do you suppose in each report (FAR,SAR, TAR, AR4) they’ve used their previous conclusions as priors? That would be one way to build a Kuhnian paradigm. It would also, IMO, be a fraud. At least unless it was explicitly documented, with opportunities for people with alternate ideas to question the developing paradigm And even more especially since their skullduggery tended to censor alternative explanations.

        If you started with a prior assumption that there was a 90% confidence it was all internal variation, what would you end up with for the probability that it wasn’t? Just because it would take an improbable combination of 3-7 decade oscillations (perhaps on top of a 1-3 century oscillation) to create the current situation doesn’t make it unlikely that it happened. After all, what were the a priori chances that you would be born?

      • Pekka Pirilä has posted a long answer to my latest comment here

      • At this point, I’ve visited WG1 sufficiently to have the impression that Chapter 9 does a good job in justifying the main competition in attribution – between solar and anthropogenic ghg forcings. However, black carbon isn’t mentioned – it appears in Chapter 2 but is weak compared with the ghgs. Some of the internal modes are mentioned in Chapter 9, but those data are scattered elsewhere as well. The only relevant interval for the apportionment of unforced vs forced variability to 1950-2005 warming is 1950-2005 – this point is important because although unforced variations occur at all times and may sometimes have large amplitudes, their net effect depends on the selected interval, and is minor for 1950-2005 from the published data.

        In summary, I think AK and others can make a good case that the attribution statement is not explained well enough in any one place, although I think Chapter 9 does a not unreasonable job. On the other hand, it is such a conservative statement in limiting the warming role of anthropogenic ghgs to the word “most” when it’s probably a good deal more than 50% that it may be unfair to be excessively critical of the audit trail. Even so, it would be a worthwhile exercise to assign quantities to the most relevant warming phenomena – ghgs, black carbon, solar total and spectral irradiance, the net AMO and net PDO (assuming the latter two are not partly anthropogenically forced) – to get a more precise idea of how the apportionment should go. It would also be useful (but complicated) to separate out the observational measurements on the strength of each influence from the model-based estimates of its relative contribution. I think this will show that model uncertainties don’t translate into commensurate uncertainties about the relative role of different warming phenomena even if they are off on the total effect. It will also confirm, I believe, the conclusion that much model uncertainty relates to cooling aerosols, whose net effect was either close to zero or a slight cooling over the entire 1950-2005 interval, and was therefore irrelevant to the relative strength of the different warming phenomena.

    • Norm Kalmanovitch

      It is very difficult to have an attribution statement for anthropogenic GHG emissions causing global warming when there hasn’t been any global warming for over a decade as CO2 emissions continue to increase at unprecidented rates.
      It is also difficult to attribute the 500% increase in CO2 emissions from 1942 to 1975 to global warming because there was no global warming during this 33 year period either.

  3. Judith,
    I like that you emphasized the IAC criticism. But I am a little surprised you did not spend more time developing the false confidence the IPCC expresses in understanding the natural climate system. I would advise you to emphasize the fact IPCC scientists are more willing to discuss their lack of knowledge in private emails than in public documents and to cite Trenberth’s email about the missing heat.

    Based on the abstract alone I cannot tell if you have cited the published criticism by Pat Frank and William Briggs which i provided in the previous thread. I hope you do.

    • To state this another way, no IPCC assessment report will be honest until it discusses Trenberth’s missing heat and the current lack of understanding of the natural climate system.

      You might or might not feel it is appropriate to discuss possible specific solutions. For example, the infrared iris hypothesis has been offered by Lindzen, observed by Spencer and potentially answers, at least in part, the question of where Trenberth’s missing heat went. Yet, infrared iris has not gotten any attention by researchers or is modeled in any AOGCM model. At least Svensmark’s cosmic ray theory is being studied by serious people. I don’t understand why the infrared iris is being ignored.

    • Ron,

      Trenberth actually published his ‘missing heat’ ideas in August 2009, before anyone saw the emails.
      http://www.cgd.ucar.edu/cas/Trenberth/trenberth.papers/EnergyDiagnostics09final2.pdf

  4. Dr. Curry

    I can see multiple lines where the polish of the argument reflect significant re-consideration and edits that improve text the authors feel very strongly about.

    Which isn’t going to place, I am certain, the response above criticism from those it is directed to.

    Personally, I think the traceability passage is one of your stronger points, and ought have gone first in the order of arguments, as it is one where the IPCC report is most technically lacking, and can lead into the other points once the issue of technique or formalism is introduced.

    As someone whose own writing is often awkward, I find the passage, “However one size doesn’t fit all, and the level and type of traceability required are related to the complexity of the subject matter and the criticality of the final product,” to be a little odd-sounding.

    “However, it is not our intention to recommend specific traceability standards for IPCC reports, which may be more appropriately done by an independent expert, but to note that in this case the traceability of key elements of the report is not fit to purpose as stated in the founding charter of the IPCC,” might be closer, if that’s what you’re trying to say, I think.

  5. Judith –

    emphasis added:

    The large uncertainties in both the observations and model simulations of the spectral amplitude of natural variability precludes a confident detection

    of anthropogenically forced climate change against the background of natural internal climate variability.

    What standard do you apply to determine a confident detection?

    We are not the only people that are unconvinced

    by the IPCC’s attribution assessment. The existence of this disagreement is not surprising given the complexity of the issue. The existence of this disagreement implies, at the very least, one of two things: 1) the IPCC AR4 has done an inadequate job in articulating the evidence for their position; or 2) the uncertainty is much greater than acknowledged by the IPCC AR4.

    What about a 3rd implication – that some number of people being unconvinced might be an inevitability considering the political and ideological context of the IPCC report.

    Or 4th – that some number of people being unconvinced is simply a function of the complexity of the issue rather than a reflection of the inadequacies of the articulation or acknowledgment of uncertainty by the IPCC? Wouldn’t some number of people remain unconvinced no matter how the uncertainties were articulated or acknowledged by the IPCC? As an instructive example, despite your careful attention to articulation and acknowledgement of uncertainties, there are some number of people (many of whom have highly technical scientific background) who disagree with your viewpoint on whether CO2 emissions can, theoretically, warm the climate.

    • “The existence of this disagreement is not surprising given the complexity of the issue. The existence of this disagreement implies, at the very least, one of two things: 1) the IPCC AR4 has done an inadequate job in articulating the evidence for their position; or 2) the uncertainty is much greater than acknowledged by the IPCC AR4.”

      This is a bit odd – Judith, it sounds as if you are arguing for 100% concensus.

      The existence of disagreement in complex matters is directly related to the complexity.

      As it stands, the stated reasons for disagreement at 1) and 2) sound more like assertions than arguments.

      Traceability is probably the most productive argument, but I’d suggest giving the “systems engineering” reference away as it’s a very poor fit with what the IPCC is trying to do, and opens the author to rebuke on the grounds that they making recommendations on matters of which they have no experience, and only superficial knowledge.

      • The existence of disagreement in complex matters is directly related to the complexity.

        Thanks for more succinctly re-framing what I was trying to say.

      • I thought about this a bit more.

        All you need is two people in a room for a disagreement.

        I’ve been involved in SR’s and disagreement is unavoidable.

        Colleagues report similiar experiences in cases involving RCTs despite the use of quality assessment tools and standardised protocols.

        Disagreement is the norm.You don’t need some crazy ‘systems engineering’ solution, just a formal mechanism for dealing with disagreement.

      • but I’d suggest giving the “systems engineering” reference away as it’s a very poor fit with what the IPCC is trying to do,

        FWIW, I disagree. She simply uses that as one example of a field that has traceability standards.

        OTOH, I think that Bart R’s rewrite of the “one size fits all” sentence clarifies it.

      • There are just much better fits, and this one is a world away. It would be unneccesarily cumbersome and overkill for the purpose.

        But, yes, one size fits does not fit all…..and some items are just inappropriate whatever the size.

      • Michael

        Is this something to do with the “All scientists should never wear spandex” rule?

      • Spandex is definitely out.

        Leggings too.

    • I agree, Josh. Some people are just too dumb to get it. Leave the IPCC alone. They are doing a good job. If they stay on the path they are on, we will be saved.

  6.  “Accounting for uncertainty, a more accurate statement would have been: The large uncertainties in both the observations and model simulations of the spectral amplitude of natural variability precludes a confident detection of anthropogenically forced climate change against the background of natural internal climate variability.”

    When run through the disambiguation machine the above reads as follows:

    Our observation of natural variability prejudices our view of simulated knowledge and that is why narrow-minded skeptics discriminate against academia’s log of the log of measures of uncertainty based on adjusted data with adjustments to the adjustments to accord with academia’s natural internal preconception that all global warming is the result of human-forcing.

  7. doskonaleszare

    > It is our assessment that the types of analyses referred to and the design of the CMIP3 climate model experiments do not support a high level of confidence in the attribution; we do not repeat the argument that we made in our original manuscript.

    At least part of the argument from your original manuscript was wrong, as I have demonstrated here:
    http://judithcurry.com/2011/09/10/uncertainty-monster-paper-in-press/#comment-124677
    http://judithcurry.com/2011/10/17/self-organizing-model-of-the-atmosphere/#comment-125240

    > Circular reasoning regarding 20th century attribution is associated with any model whose parameters were tuned to the 20th century climate simulations and whose 20th century forcing was not chosen in an a priori manner prior to actual referenced attribution simulations.

    Can you name the models that were tuned to the 20th century climate simulations? Or maybe you’re just stating a tautology?

    By the way, I am m glad to notice that you at least retracted your original claims about the CCSM3 tuning.

  8. Thanks Judith, that mention was a nice birthday present.

  9. I think I can identify some disconnects in POV between yourself and the reviewers.
    1) they deny that GCMs are “tuned” because they don’t know what that term means and do not recognize that “unrealistic” runs being dropped or submodels being “calibrated” is tuning.
    2) they do not see that the actual decision process is a black box to outsiders because they think their decisions are so logical and inevitable that any computing brain would come to the same conclusion given the [selective] “data” they consider.
    3) they have never heard of decision theory or logic traceability
    4) they never heard of and deny groupthink/bias exist for them
    An example of what they need to document would be how and why they fail to weight the factors pielke sr points to that are left out like LULCC, as well as GCR effects (and ultraviolet etc), and possible existence of long term cycles (longer than ENSO) which would bias their result. This is a concrete example you could point to where a decision was made (which is a priori for IPCC authors and largely subconscious but bugs the heck out of critics).

    • doskonaleszare

      > 1) they deny that GCMs are “tuned” because they don’t know what that term means and do not recognize that “unrealistic” runs being dropped or submodels being “calibrated” is tuning.

      I don’t deny it, I just want some evidence. So far, dr. Curry hasn’t provided any, which is pretty ironic considering how much she had written about “traceability”.

  10. Some fields (e.g. medical science, computer science, engineering) have stringent traceability requirements, particularly for issues that are mission critical or having life and death impacts.

    On the question of traceability, and the field of computer science (and engineering) as a relevant example, I thought of the situation with the “Pentium FDIV bug” where computer science engineers and personnel in other departments at Intel spend valuable time pointing finders back and forth about who was responsible. Seems that if traceability were so well-defined in that industry (and among engineering processes), Intel might not have lost hundreds of millions of dollars.

    Of course, picking one example may just be citing an exception that proves the rule, but was that situation really unique WRT accountability in the world of computer science?

    • heh heh my own experience in engineering suggests that the traceability while well defined is not necessarily well adhered to. lots of things slip when people are careless. the IPCC would do well to follow documentation procedures of the like that engineering companies are supposed to follow. hell let’s use the NRC regs for documenting the design of a nuke plant. not that there have never been problems with a nuke plant, or that the regs have always followed. one could argue that in terms of the number of lives potentially at stake the IPCC reports are every bit as critically needing transparency than a nuke plant design. i don’t want to make this a long thread, but I have no problem holding IPCC to higher standards than industry DOES follow and probably than most industry SHOULD be required to follow IMO.

      • in respect to your intent to not wanting to make this a long thread (painful for me, as you know) – I agree, as is not infrequently the case with your posts. More traceability = good. I still think that the practicality of traceability, in multiple respects, is an open question – and that simplifying those questions does a disservice to the realization of traceability.

      • My experience was that industry does a better job than academics at trying to ensure traceability during all stages of projects. It was much easier in the companies I worked for to go to physical volumes and disk records to recover a data audit trail (which matters to me as a statistician), and a slight majority of the people I talked to about this in academia were oblivious to the necessity.

        That errors occur still is not a reason to endorse positive laxity.

      • Norm Kalmanovitch

        Lives are at stake here! The world is taking 6.5% of the global grain supply and using it as feedstock for the 85billion litre global annual ethanol production. When 6.5% of the worlds grain is taken away the wealthy pay double for food and the poor simply starve.
        How many of the now 7 billion people on the planet are now victims of this biofuel for carbon credit fraud spawned by t6he Kyoto Accord?
        The real stupidity is that when the CO2 released in the fermentation process is added to the fact that ethanol only produces 64% of the energy of the gasoline that it replaces and add again the vast amount of energy mostly from fossil fuels used in the distillation process; ethanol produces significantly more CO2 than the gasoline it replaces!

    • Latimer Alder

      @Joshua

      All the human constructs like checks and balance, traceability, auditability, acccountability, conflict of interest etc are really there to guard against human fallibility.

      But that they are not always perfect, nor followed to the letter does not seem to me to be a reason to junk them altogether. And I am deeply deeply suspicious of any human organisation (climatology for one) who wishes to continue without them and

      The old Russian proverb – beloved of both Lenin and Reagan has a lot to tell us.

      ‘doveryai, no proveryai’. Trust, but verify.

      • But that they are not always perfect, nor followed to the letter does not seem to me to be a reason to junk them altogether.

        If you see anything that I’ve written that seems to be in disagreement with that point, then I didn’t express my opinions clearly. And if so, please point it out.

    • BlueIce2HotSea

      Joshua –

      Seems that if traceability were so well-defined in that industry (and among engineering processes), Intel might not have lost hundreds of millions of dollars.

      No. Traceability was not well defined 17 yrs. ago. And Traceability is not Testing and Verification.

      The relevance to Dr. Curry’s paper is that stronger adherence to Traceability could help insure replicable estimates of uncertainty. And in so doing, it would help resolve another type of uncertainty, one related to the trustworthiness of IPCC conclusions.

      • BlueIce2HotSea

        Me:

        “Traceability was not well defined 17 yrs ago.”

        Well, that’s a pretty stupid statement. I was thinking of the traceability systems I worked on in the 90’s. They all had excrutiatingly detailed requirements, however the traceability of the requirements was all but nonexistant (not well-defined).

        Sorry about that.

    • Steven Mosher

      The flaw in the chip was discovered by Intel and reported by someone outside the company. In product marketing chips for example, you do this analysis. there is a 1 in 9 billion chance of the error effecting a computation. You calculate the cost of returns and you make a go/no decision. The error was very traceable, the lots and steppings of the chips which had the problem were traceable, thats how intel or any chip guy can do a recall. Having a traceable process allows you to make the decision to ship with a flaw. You look at your fab line and say.. holy crap we have 10,000 wafers in process at a cost of 2K per wafer,
      5000 of those have half of the metal layers deposited, we can scrap them or build them out and suffer the returns.. oh crap we have 100,000 in packaging and test, do we stop the packaging.. Then wennies crunch numbers and you look at the cost of junking the stuff versus the cost of shipping and risking a return. Its all traceable. It’s how you know who to fire!
      What is standard. On my first product I made a bad call. Everyone knew the problem. we knew how it happened, but the decision to ship was mine. I wasnt fired, but I did have to clean up my mess. So the decisions were traceable, there was accountability, I’m a nice guy so they didn’t kill me.

      because you have traceability you can do this

      http://www.cs.earlham.edu/~dusko/cs63/fdiv.html

      The other way to look at it is this. Whatever standard Intel had they clearly, on your argument, where not good enough to prevent a mistake.
      If your goal is a mistake free process for the IPCC, we could build something based on this:

      http://www.everyspec.com/DoD/DoD-STD/DOD-STD-2167A_8470/

      If you want examples of how things are done in computer science, well you can look at any number of change files on the R Forge project and see who reported what bugs, where the fixes happened, what the fix was.

  11. The authors are correct that since the Third Assessment Report, the IPCC has placed a high priority on communicating uncertainty, and perhaps their methods have been effective in communicating with the public. However, communicating uncertainty is a very different endeavor from actually characterizing and understanding uncertainty, an effort in which we feel the IPCC falls short.

    I think the presentation of your point, if I understand your point correctly, is woolly here. Emphasis on communicating uncertainty, which is itself poorly characterised, is not actually effective communication of uncertainty at all – if being effective implies accurately representing and communicating.

    AR4 did not, on balance, effectively communicate uncertainty. Instead, it only seemed to introduce an unfortunate additional blanket layer of uncertainty over the very subject of uncertainty itself. This obfuscation did not positively impact the communication of uncertainty in the science at all in AR4.

  12. I think it would help to disambiguate the terms “communication of uncertainty”. What IPCC did was try to communicate their CONCLUSIONS about uncertainty, but not the process by which they arrived at these conclusions. These are not the same thing. Believing their conclusions with no traceability requires trust, and this is not a case where one can replicate anything experimentally (and is a case where critics come to very different conclusions).

  13. Norm Kalmanovitch

    Science starts with certainty and moves into the realm of uncertainty through the development of well founded hypotheses which are tested by experiment to remove uncertainty and advance the hypothesis to proven theory if it is in agreement with all observations or discard the hypothesis if it fails even in the slightest way to meet this criterium.
    Climate change started off as an unproven computer model projection based on an uncertain correlation between CO2 emissions and global temperature through an uncertain assumption of radiative flux balance being affected by an uncertain process of thermal radiation capture by increasing CO2 concentration based on the uncertain assumption that this increase in CO2 concentration is primarily human sourced.
    It is a certainty that the increased CO2 concentration is not primarily human sourced, it is a certanty that the 14.77micron band is so close to saturation that increased CO2 cannot possibly have the forcing statede by the climate models, it is a certainty that there is no detectable decrease in OLR from observed increased CO2 concentration It is an absolute certainty that there is no possible correlation between CO2 emissions andf global warming, and it is an absolute certainty that the 1988 scenario “A” of Hansen’s climate model projection underestimated the CO2 emissions over estimated the resultant CO2 concentration and predicted warming for the past nine years as the Earth continues to cool.
    It boggles the mind to think that with all this uncertainty people are still defending this IPCC fraudulent conjecture and even worse are demanding government action to arrest global warming nine years after all global temperature data clearly demonstrates that global warming ended!
    None of the intelligentsia that Chicken Little advised that the sky was falling questioned the uncertainty of his warning, which puts may here in that same category!

    • “Climate change started off as an unproven computer model projection based on an uncertain correlation between CO2 emissions and global temperature ”

      Wrong. its started from first principles over a hundred years ago and is confirmed ( not proven) by observations of warming. if you want to show that C02 or Ghgs will not warm the planet, you need to overturn working physics.

      • Aren’t you overstating the case? Rather than overturning the phyics of absorbtion and re-emission, wouldn’t it be sufficient to show that there are negative feedbacks, either in the CO2 concentration itself, or because of competition in other greenhouse gases that share part of CO2’s absorbtion spectrum (e.g. water), or in albedo (again, e.g. water).

        In fact, isn’t it true that over geological time scales, CO2 changes lead temperature changes, suggesting that the cause and effect relationship really is the opposite of the “accepted” one? Yes, I understand that there’s a convoluted explanation for this that has become part of the concensus. But it wouldn’t really be “overturning the working physics” if someone showed this hypothesis to be mistaken, would it?

      • Steven Mosher

        nope. not overstating the case

      • its started from first principles over a hundred years ago

        correct

        and is confirmed ( not proven) by observations of warming.

        Only if there is no other important process on going, such as, perhaps, increases in cloud cover somewhere. Or as they say sometimes, cetaris paribas. You could make a case that the physics involved in the daily evaporation of water and its recondensation (resulting in energy transfer from the lower troposphere to the upper troposphere) is both really important and incompletely known. It’s not hard to imagine a totally realistic possibility that increased CO2 results in increase transfer of heat and water from lower to upper troposphere, with a resultant net increase in cloud cover and net reduction in insolation. It wouldn’t take much to negate or overturn the radiative budget hypothesized in the principles over a hundred years ago.

      • Matt, good post, but Paribas is – or was, until recently – a French bank. I think you mean “ceteris paribus?”

      • Norm Kalmanovitch

        Wrong! Svante Arrhenius in his 1896 paprer, stated trhat he made no measurements of wavelengths greater than 9.5microns and CO2 only has an effect from 13.5 to 17.5 microns. Arrhenius measured the effect of watret vapour believing it to be the effect of CO2 which he called carbonic acid in his paper
        It was refured by Angstrom who showed the absorptive spectra for CO2 as well as by several others.
        http://www.applet-magic.com/arrhenius.htm
        We are a lot smarter now at least some of us are.
        If you want to show how CO2 warms the planet you have to violate the first and second laws of thermodynamics as was done with the energy balance that forms the basis for the AGW fabrication
        If you want to use a term GHG you first haver to define it and the only gases that have a definitive effect on the Earth’s radiative spectrum are water ozone and CO2 and only CO2 is labelled as a greenhouse gas in the Kyoto Accord.

  14. Judith,

    Notwithstanding your remarkable courage and continued progress at searching through the weaknesses of CC science and the IPCC process, you have shied away from discussing the topic of scientists playing politics with science and other scientists.

    Humans are a political animal. Office politics and research politics are as much a part of science as they are of any other social enterprise. Your criticism of the IPCC is that they are playing advocacy politics. In many respects you are mounting a political rebuttal in response to their political assertions.

    In Tamiso’s post of Fake Skeptic Criticism of “Decadal Variations in the Global Atmospheric Land Temperatures” and it’s respondents, we see the following type of skeptic bashing:

    Keenan only offers “spin” on Muller et al. (2011). What’s probably most offensive is the hubris which permeates his post, the insultingly condescending tone with which he implies the Berkeley team should go “back to school” and even suggests introductory textbooks on time series analysis.

    Hubris and condescension are what we’ve come to expect from posts at WUWT. The only real surprise is just how fervently WUWT is attempting to smear the Berkeley team, their results, and Richard Muller now that the Berkeley team has had the audacity to contradict what Watts & Co. have been asserting for years. The Berekeley group have become WUWT’s current favorite target of ridicule. What a strange reaction from a blogger who originally declared he would “accept whatever result they produce, even if it proves my premise wrong.”

    When experts mock non-experts they are tacitly saying that non-experts are ignorant and incapable fools who deserve to be dismissed and disparaged for their foolishness at questioning expertise. What is ignored is that the criticism of lousy science is supposed to come from with science. If non-scientists can manage to raise a challenge to a scientific product, that deserves respect and encouragement, It provides a contribution which scientists have neglected to care for in their open right.

    There is ZERO excuse for credible researchers to mock non scientists for poor ability.

    It is imprudent for scientists to mock no scientists for being motivated by vested interests. The politics of science and research leave scientists open to comparable preexisting vested interests.

    Let me be both ‘personal’ and brutally honest here …

    Speaking personally, I find the topic of CC to be very difficult and confusing to understand. Reading this blog does not help much to clarify that confusion. Whether it is confusion brought on by my own laziness or lack of familiarity … or whether it is confusion arising from the many sources of inherent uncertainty is ‘probably’ unimportant. Education ability and scholarship cannot resolve, nor vastly improve adeptness regarding the understanding of ‘climate’. … ‘Climate’ and climate change is intensely and inherently confusing and uncertain. That is the nature of the beast! Nothing is gained by experts insisting that the experts understand. The experts don’t understand. The experts struggle to make their way through the confusion and uncertainty, gaining whatever appreciations might be gleaned in the course of doing so.

    In the main the experts do not understand climate science. It is ludicrous to embrace a model of consensus and declare that the science is settled.

    The failing of the IPCC and the scientists who contribute to it, is that they desire to construct an ‘objective’ estimation for a complex process which at the present state of knowledge resists objective description.

    The IPCC is absolutely adamant about describing AGW in cold hard certain objective terms. They demand reducing all considerations down to a single grade of increasing temperature brought about by rising CO2.

    The IPCC demands that climate change be described a simple ‘grade’ of increasing temperature caused by anthropogenic emissions of CO2 that directly harms life and planet.

    The IPCC unilaterally and unequivocally shun any description of a complex, multi-direced, uncertain evolving process. They only want to manufacture a deliberately misleading illusion of ‘objective certainty’ so as to beat humanity with undeniable subjective propaganda.

    • Tamino writing
      Keenan only offers “spin” on Muller et al. (2011). What’s probably most offensive is the hubris which permeates his post, the insultingly condescending tone with which he implies the Berkeley team should go “back to school” and even suggests introductory textbooks on time series analysis.

      Hubris and condescension are what we’ve come to expect from posts at WUWT.

      is truly the pot calling the kettle back.

    • you have shied away from discussing the topic of scientists playing politics with science and other scientists.

      Thank goodness.

  15. Regarding the paper overall, I think you did well, but I personally would have included more on a few points.
    You wrote: X et al. object to our statement in the original manuscript: “Figure 9.4 of the IPCC AR4 shows that all models underestimate the amplitude of variability of periods of 40-70 years” on the basis that we do not consider the uncertainties presented in the chapter.

    I am guessing an engineer helped in the part about the traceability matrix and document control. That is true and a great seemingly obvious point-that was not done. It was good to point that our clearly.

    My thoughts: what you wrote is accurate but it would have been even stronger imo if you added something about the variability in the outputs of the models depending upon the number of times the model was run. My understanding is that the models can exhibit orders of magnitude differences between runs, which is why it was suggested that they should be run multiple times each. I also understand this was not done for all the models due to cost etc.

    I would also have liked to see something in the response citing that we have no data regarding the relative accuracy of the models used to estimate specific conditions. Have any of the models used been tested against observations and fully characterized? What was their relative accuracy for each characteristic over what time period?

  16. The IPCC has been confirming how little it knows. Look at SAR, TAR, and AR4 for many repeated statements that much has been learned since the preceding report. If they’ve learned so much… how little did they know when they issued their preceding certain reports? How much do they claim uncertainty has decreased in each report?

  17. Dr. Curry can you support the claim I added? And should it be included as a lead up to the tuning etc.?

    The large uncertainties in both the observations and model simulations of the spectral amplitude of natural variability precludes a confident detection of anthropogenically forced climate change against the background of natural internal climate variability. {add} This would tend to be true even if one had comparisons of many independent runs rather than the one actual record of historical temperature due to the variance. The existance of one independent expression of climate largely invalidates X et claim that uncertainty can be determined, but rather re-enforces our point of an inherrent circularity detailed below.

  18. John Carpenter

    Judith,

    Thanks for the acknowledgment. I’m glad you found my comments worthwhile. I think you captured the essence of how traceability is viewed in other technical fields very well in your response.

  19. Judith, in the fields I inhabit — ethics and personal responsibility->>behavior — uncertainty plays a critical role. This may seem a diversion, but it is worth the 2 minutes to think through. Our mis-assignment of cause [we can read here both failure to consider the uncertainties of our lives and the less-than-rational attribution of causality that results] leads to many of the more tragic results in our inter-personal relationships. Thus I have led many a directed session of behavioral therapy to rooting out the specific uncertainties of causality in a situation and the resultant aberrant actions a person has taken as a result.
    My subjects have often tried to explain how a chain of multiple uncertainties ‘cancel one another out leading to a near certainty’ of causality. Sometimes, in voicing it, they realize how irrational that idea is. Sometimes they get it when I direct them gently at the idea. Sometimes they never get it.

  20. Dr Curry, you wrote:

    Traceability allows an independent person or group to trace back to understand how the result came to be and to walk through the decision process and achieve the same result. Commonly used practices in system engineering include the traceability matrix and document control. Some fields (e.g. medical science, computer science, engineering) have stringent traceability requirements, particularly for issues that are mission critical or having life and death impacts.

    I’d suggest adding a sentence here “They do so because, as commercial activities, litigation is always a risk when things go wrong and being able to substantiate processes and conclusions is a critical risk mitigation factor”.

    I make this suggestion because my own view, as a lawyer, is that the reason behind much of the uncertainty and imprecision in climate science is that, unlike those other fields you mention, climate scientists have no ultimate liability to anyone because they don’t deliver any service (no one pays Trenberth or Hansen, for example, for their output for the purpose of relying on it and being able to sue them if it’s wrong). The L’Aquila, Italy, litigation about earthquake predictions is the first I’m aware of that is turning a cold legal eye to a climate science activity suggesting liability on the part of the scientists. I think that if climate science had had lawyers crawling all over it for decades, as in the commercial sector, the uncertainty monster would have been slayed a long time ago.

    Just a suggestion. Thanks for being invited to comment. This blog rocks.

  21. JC

    The question in life is not whether you get knocked down. You will. The question is, are you ready to get back up… And fight for what you believe in?

    JC, yes, keep on fighting for what you believe in.

    We must all “contain” the scare mongering of AGW.

  22. JC, to demonstrate the “uncertainty is much greater than acknowledged by the IPCC AR4” why not include the following evidence?

    IPCC Fourth Assessment Report projection:

    Since IPCC’s first report in 1990, assessed projections have suggested global average temperature increases between about 0.15°C and 0.3°C per decade for 1990 to 2005. This can now be compared with observed values of about 0.2°C per decade, strengthening confidence in near-term projections.

    Actual observation:
    http://bit.ly/rYp5OM

    This observation demonstrates IPCC AR4 “confidence in near-term projections” is unwarranted.

  23. Judith, Your reviewers obviously had their own reasons for rejecting your uncertainty monster ‘paper’.

    Call me old fashioned, if you like, but I’d expect scientific papers to have at least some of the following: data, graphs, mathematics, computer models, descriptions of work done, conclusions from results, suggestions for further work.

    Wouldn’t ‘op-ed’ be a better description for your submission?

    • tempterrain

      Call me rationally skeptical, if you like, but (as you wrote) the “reviewers obviously had their own reasons for rejecting [[the Curry + Webster] uncertainty monster ‘paper’.

      The attempted rebuttal was a feeble attempt to defend the IPCC conclusion on the anthropogenic attribution of late 20th-century warming. It first tried to marginalize C+W by stating that this was just the opinion of two scientists. But it failed to rebut the two specific key points brought up by C+W: lack of traceability regarding the IPCC treatment of uncertainty, where “uncertainty of climate models is interpreted in the context of model inadequacy, uncertainty in model parameter values, and initial condition uncertainty” and use of expert judgment rather than formal attribution studies.

      IMO the C+W argument is compelling and well presented and the rebuttal was weak

      Do you agree?

      If not, why not?

      Max

    • “compelling” ? My first reaction would be that it wasn’t for me to say. I’m not qualified to referee a climate science paper. But, as I said above, I’d just question if it can really be termed that

      I’d suggest Judith might have a better chance of publication if co-authors included other climate scientists rather than just her husband. It makes it seem too much of a family concern IMO.

      If Judith can’t find anyone else to publish there is always Energy and Environment. Or maybe the Heartland annual conference. They’d be quite likely to find it compelling too.

  24. Judith Curry

    Looks like it is basically a simple “cut and paste” exercise for IPCC.

    Delete highlighted paragraph in AR4 WG1 SPM (p.10):

    Most of the observed increase in global average temperature since the mid-20th century is very likely due to the observed increase in anthropogenic greenhouse gas contributions. This is an advance since the TAR’s conclusion that “most of the observed warming over the last 50 years is likely to have been due to increase in greenhouse gas concentrations. Discernable influences now extend to other aspects of climate, including ocean warming, continental-average temperatures, temperature extremes and wind patterns (see Figure SPM 4 and Table SPM 2).

    And replace with the one you propose above (plus a short lead-in):

    Observations have shown that both anthropogenic greenhouse gas concentrations as well as global average temperature have risen since the mid-20th century. However, the large uncertainties in both the observations and model simulations of the spectral amplitude of natural variability preclude a confident detection of anthropogenically forced climate change against the background of natural internal climate variability.

    Looks like an easy fix to make the next SPM report a bit more “scientific” and a bit less “political”.

    Let’s see if IPCC is up to the task…

    Max

  25. Judith Curry

    As other have indicated here, I believe that the strong point of your argument is the lack of traceability to substantiate the IPCC claim that “most of the observed increase in global average temperatures since the mid-20th century is very likely due to the observed increase in anthropogenic greenhouse gas concentrations”.

    You summarized this well in Curry + Webster:

    Listing a large number of uncertainty locations, and then coming up with a ‘very likely’ likelihood statement using expert judgment in the context of a consensus building approach, is at the heart of our concern regarding the IPCC’s treatment of uncertainty

    where “uncertainty of climate models is interpreted in the context of model inadequacy, uncertainty in model parameter values, and initial condition uncertainty”

    and

    Since no traceable account is given in the AR4 of how the likelihood assessment in the attribution statement was reached, it is not possible to determine what the qualitative judgments of the lead authors were on the methodological reliability of their claim.

    You have very well identified the weak spots: lack of traceability to substantiate the claim by reducing uncertainty including the use of qualitative expert judgments by the lead authors.

    And it appears to me from the first rebuttal attempt that these points have not been effectively refuted.

    Hang in there – you are on the right track, and what you are doing is extremely important.

    Max

    • manacker

      http://xkcd.com/970/

      While it’s valuable to get traceability right, it’s also good to get it right in proportion to the thing being traced, too.

      In science before this decade, the fashion often was to strip out traceability of the who (of the expert opinion), and maintain only traceability thread of conclusion back to raw data by steps of fact from fact from fact, on the premise that even suggestion of authority ought never interpose in the line of reasoning, and the sole determination of correct scientific traceability in the physical sciences is merely that an independent researcher reproduce the same conclusions from the same procedure.

      Given the special nature of traceability in science, an expert might consider the following in determining standards for traceability in IPCC reports:

      Clearly, where there is any issue that is discussed in the report pertaining questions outside the physical sciences, that merely reproducible standard of traceability is inappropriate.

      Also, where reproducibility is difficult due any reason, the mere reproduction standard is insufficient.

      Where one has elected to draw together multiple lines of physical inquiry, mere reproducibility is directly contraindicated by the most elementary reasoning.

      So, for separate and distinct physical arguments that are merely reproducible, it is proper to relax traceability where it provides protection from the influence of authority on physical research; conclusions from such distinct physical units can be traced adequately to the study itself with only appropriate further details where mere reproducibility applies.

      Everything else of significance ought be non-invasively connected to the hands it passes through and the dates on which it passed, with links to adjunct materials where they might be plausibly relevant to conclusions and key decisions.

      This documentary authentication standard of integrity is easily supported by software in project or publishing systems.

      This standard will not extend to external references that are not the sole source used in significant or key decisions or conclusions. Where there is a sole source of key information and extension of the traceability standard is feasible, it ought be considered.

      You can see this would be a standard higher than IPCC has met, but still not costly, not onerous, nowhere close to the level of an FDA, medical, engineering or even most IT system tracking guidance, at the same time adding substantially to the value of reports.

      Imagine being able to track down typos in seconds instead of weeks or months? Identifying what is grey literature and what is physical science and what is peer reviewed and whether a statement has been audited?

      • Bart R

        I would agree with you that there are many more problems with statements made in IPCC AR4 WG1 SPM (and other reports) than simply the statement on attribution of 1950-2005 global warming (exaggerations, distortions, cherry-picking of reports, etc.) – but this is the one key statement, which Curry and Webster critiqued in the Uncertainty Monster paper, which is the topic of this thread.

        Max.

  26. Judith,

    Politicians cannot afford to look at anything to do with uncertainty.
    They know their decisions have cost massive amounts of funding to be put into technology based on the fear of AGW.
    Many politicians have that hero complex of being the first or leader in the fight against the fear imposed by the AGW scare.
    You should investigate the countries that put into new policies and vast amounts of funding into the curbing of greenhouse gases.
    When “hell has frozen over” is when they politicians will have no choice into admitting mistakes and the world of science crashes as a bunch of guys with guesses that were wrong.

  27. http://fivethirtyeight.blogs.nytimes.com/2011/10/27/herman-cain-and-the-hubris-of-experts/
    On uncertainty – from the best political analyst I’ve ever seen. And I usually disagree with him.

    • The problem with Nate Silver is that although he knows his statistics, he doesn’t understand science, apart from whatever soft sciences he has studied, including economics, baseball, whatever. Everything that he writes about has the potential to be gamed, and that is where the vast majority of the uncertainty comes into play with the soft sciences.

      On the other hand, the physical data from the hard sciences cannot be gamed since nature does not display any intelligence; at most it has hidden variables that we have yet to root out. The uncertainty only rests on these variables and the aleatory and epistemic considerations that we need to apply to the data.

      Like those Freakonomics guys, Nate Silver would be completely out of his element should he undertake studies of something like climate change. The Freakonomics guys go outside their comfort zone occasionally and they fail miserably.

      • “Everything that he writes about has the potential to be gamed, and that is where the vast majority of the uncertainty comes into play with the soft sciences.” I would like to see evidence for that. I assume that the uncertainty in soft sciences comes from the fact that they are really, really complicated, and have lots of Black Swans hiding out there that we don’t understand at all. That’s what Silver is saying.

        Hard sciences are more mature, and we expect many fewer massive surprises per decade. But climate science may be “hard”, but it is far from mature, and as a result I think plenty of Black Swan surprises are to be expected.

      • Gaming refers to game theory, in that most of the data from the soft sciences comes from human behavior. The uncertainty is that no one knows what the actual behavior is because it changes due to psychological manipulation. This happens during the answering of surveys and establishing models for behavior. For specific cases, look up the description for Lucas criteria and Goodhart’s law. Nothing even close to this comes up in the hard sciences.

        Sure, the soft sciences may actually be harder than the hard sciences to model (and may in fact be intractable, as realistic models of game theory are proving out to be), but thank goodness climate science falls under the hard science category.

  28. Please, please, please, Dr. Curry, continue working on the Uncertainty Monster. My time on this blog convinces me that this is the true crux of the issue.

  29. AK wrote higheer up in this thread:

    The deceptiveness isn’t quite what is at issue here, rather it’s that WG1 has failed to document any sort of audit trail between the 10% significance in Figure 9.4/9.7 and the 90% confidence of their attribution. No explanation why the attribution confidence isn’t smaller than 10% (no matter how confident they were otherwise), nothing to stop me (or any other observer) from thinking that somebody looked at Figure 9.5, nodded their head(s), and forgot all about the error levels in Figure 9.7. (And don’t tell me 10% significance and 90% confidence is comparing apples and oranges, show it to me in the text of AR4. We’re talking about communications here, as well as traceability.)

    As far as I can say there’s no direct connection between the 10% and 90% brought up by AK. The issues are related, but not the numbers.

    The basic argument considered is, what’s seen in Figure 9.5. That tells, how models agree with data fairly well, when AGW is included, but disagree strongly without. That leaves the question of the representativity of the set of models for all possible models that are not excluded by other knowledge. AK noted that there are no error bars. Instead we can see only, how much the numerous model runs differ from each other. If the models are really representative of all possible outcomes, the distribution of the outcomes in the lower figure tell, how much of the warming can be explained by natural variability and with what level of confidence.

    One of the questions related to the representativity is the ability of the chosen models to produce similar variability as seen in history data. This single factor is considered in Figure 9.7. Here we see that the models may have somewhat less variability than data in the range 10-50 years. The 10% significance tells that there’s an estimated 10% possibility that the disagreement is due to a random effect in data, while it’s more likely that the models have too little variability. More essential is, however, that the models do have nearly as much variability as the data (although the logarithmic scale may be misleading for that). The random effect of data might be in opposite direction, i.e., correct models might have even more variability, but even so Figure 9.7 presents evidence for the conclusion that the models have variability of the right order of magnitude. It gives evidence for that conclusion, but does certainly not prove that the models have correct variability.

    Now we come back to Figure 9.4. The discrepancy without AGW is very large, much larger than the differences between models. If we now accept that models have almost the right level of variability, then it’s easy to conclude that it’s 90% certain that the observations cannot explained without AGW. This conclusion is based of several assumptions related to the representativity of the models. Therefore the confidence is not based on fully traceable

    This is a clear example of problems of traceability in the IPCC handling of uncertainties. The scientists know that there are weaknesses in their models, but they maintain that the models are anyway good enough to provide evidence, because the margins between the variability of model results and the deviation from empirical data are large without AGW. All their attempts without AGW have failed to explain much, if any, of the warming. The conclusion is strong for the set of models that exist, the question is, do the models miss something essential. For this crucial question providing traceability is not possible. There basis for strong subjective thrust may be good, but it’s not traceable. Additional objective arguments may perhaps be presented, but combining evidence objectively to form a quantitative PDF or a traceable overall level of confidence is and is likely to remain impossible.

    The dilemma is fundamental. Even very strong reasons for confidence may remain partly non-traceable forever. It’s not correct to insist that the strong confidence is not justified without a fully traceable logical argument, but it’s also important to admit that strong confidence requires strong justification even, if it’s too diverse to be fully traceable.

    Uncertainties and risks will remain difficult issues that cannot be handled fully objectively, but many disagreements can be reduced, when people do their best to understand, what others are saying and, why they think as they do. Making all that objective that can be made helps also, but the subjective factor remains.

    Most decision making is decision making under uncertainty. Often the uncertainty is of familiar nature, those uncertainties the decision makers have learned to handle. When the uncertainties are unfamiliar as are very low probability very large risks, or they are related to very long term potentially very serious trends, as the climate issue, then the experience doesn’t help far. The only way to improve decision making is then to improve analysis. That requires data from specialists, but even more that requires deep thinking of the basic setup and a good overall view of the situation. Much more discussion and analysis is needed on handling the final steps of reasoning.

    • @Pekka Pirilä…

      If the models are really representative of all possible outcomes, the distribution of the outcomes in the lower figure tell, how much of the warming can be explained by natural variability and with what level of confidence.

      You’ve put your finger on one of the key sources of uncertainty. Quite a few assumptions besides “simple physics” are built into the models, and some of those assumptions may well be driving errors. For instance, quoting from the article referenced in the Candid comments post, regarding the “missing heat”:

      Hansen now believes he has an answer: All the climate models, compared to the Argo data and a tracer study soon to be released by several NASA peers, exaggerate how efficiently the ocean mixes heat into its recesses. Their unanimity in this efficient mixing could be due to some shared ancestry in their code. Whatever the case, it means that climate models have been overestimating the amount of energy in the climate, seeking to match the surface warming that would occur with efficient oceans. They were solving a problem, Hansen says, that didn’t exist. [my bold]

      Now, as it happens Hansen may well be wrong, as “the observational analyses[5] inherently include these effects,” (Meehl et al. 2011) however, their own endnote [5] (Trenberth and Fasullo 2010, note the overlapping authors) points out:

      The difference between the incoming and outgoing energy—the planetary energy imbalance—at the top of the atmosphere is too small to be measured directly from satellites.

      All these differences introduce possible points of uncertainty. There seems to be a tacit assumption in the AR4 uncertainty calculations of >99% confidence that there are no such distortions common to all models due to “shared ancestry in their code“. There’s also a tacit assumption of >99% confidence that no other assumptions common to all the models might be introducing error.

      For instance, based on geological factors, there’s an assumption that increased CO2 (and other GHG’s) produce a significant warming on sub-century timescales. What is the confidence that we can extrapolate interpolate from geological to decadal timescales? I changed from extrapolate to interpolate because based on various observations and models we can be somewhat confident (66%?) that any correlation between CO2 and “global average temperature” on a roughly annual time-frame (e.g. ENSO) is driven in the other direction.

      Another unjustified assumption is that the feedback due to clouds is “constant”, that is cloud characteristics can be determined by a few climate-related variables (e.g. air and SSTemperature, humidity, etc.). Aerosols are known to have indirect effects on feedback via clouds (e.g. Rosenfeld et al. 2008), but how much, and in what direction is unknown. Aerosols aren’t just anthropogenic, biogenic aerosols play a major role in cloud formation (Poschl et al. 2010), and we can only guess at the magnitude of their effect. We have no idea what long-term variations in natural and anthropogenic production of aerosols might to to the “feedback” between clouds and temperature.

      Evidently, the tacit assumption is that there’s no effect, with >99% confidence each. With an audit trail of where these assumptions enter the calculation leading to the overall >90% confidence, we might identify where possibly invalid assumptions have been made, and what the effect would be of correcting them (or applying potential corrections for debatable factors).

      Ref’s:

      Meehl, G.A., Arblaster, J.M., Fasullo, J.T., Aixue Hu, A., Trenberth, K.E. (2011) Model-based evidence of deep-ocean heat uptake
      during surface-temperature hiatus periods
      Nature Climate Change 1, 360–364 (2011) doi:10.1038/nclimate1229

      Poschl, U., Martin, S.T., Sinha, B., Chen, Q., Gunthe, S.S., Huffman, J.A., Borrmann, S., D. K. Farmer, D.K., Garland, R.M., Helas, G. Jimenez,, J.L., King, S.M., Manzi, A., Mikhailov, E., Pauliquevis, T., Petters, M.D., Prenni, A.J., Roldin, P., Rose, D., Schneider, J., Su, H., Zorn, S.R., Artaxo, P., Andreae, M. O. (2010) Rainforest Aerosols as Biogenic Nuclei of Clouds and Precipitation in the Amazon Science, Vol. 329, No. 5998. (17 September 2010), pp. 1513-1516. doi:10.1126/science.1191056

      Rosenfeld, D., Lohmann, U., Raga, G.B.,O’Dowd, C.D., Kulmala, M., Fuzzi, S., Reissell, A., Andreae, M.O. (2008) Flood or Drought: How Do Aerosols Affect Precipitation? Science 5 September 2008: Vol. 321 no. 5894 pp. 1309-1313 DOI: 10.1126/science.1160606

      Trenberth, K.E., Fasullo, J.T. (2010) Tracking Earth’s Energy Science 16 April

  30. In reviewing recent discussions about the AR4 statement that it is “very likely” most warming from about 1950 to 2005 is attributable to anthropogenic greenhouse gases (hereafter simply “ghgs”), I’ve found myself in agreement with the claims that tracing the basis for that assertion in WG1 is difficult, but I also believe that the requisite information is buried within the report in chapter 9, along with other relevant data in chapter 2 (on black carbon) and scattered references elsewhere on unforced variability. It’s fair to say that traceability has not been made convenient.

    On the more general question as to whether the AR4 attribution is justified, I think there is more than enough evidence to say it is despite the uncertainties that have been mentioned here. One problem about recent statements on uncertainty is that they are general enough to be valid in many circumstances, but are probably overstated for the post-1950 interval.

    Two salient examples involve cooling aerosols and unforced natural variability. The attribution statement relates to how we divide up the observed post-1950 warming, and is therefore limited to climate phenomena that exerted a warming effect during that interval. This almost certainly does not include cooling aerosols (mainly from anthropogenic activity but also including some recent increases in volcanic forcing). Observational data show a substantial cooling aerosol effect from 1950 through much of the 1970s, and some reversal of that effect later, but none of the data indicate that the result of the reversal was a net warming – rather, a net residual cooling, perhaps small but still a cooling, is indicated. While this cooling will have reduced the effect of the warming phenomena, there is no obvious mechanism whereby it would greatly alter their apportionment. It is possible to contrive scenarios for a small alteration (e.g., in the contribution of the black carbon warming component of aerosols), but not a major reapportioning through any plausible mechanism I’m aware of. Aerosol uncertainty in the models, for example, is therefore largely irrelevant to the attribution statement. At the same time, it is a major source of model uncertainty, and so when that component is omitted, the models become less a source of uncertainty than has been claimed.

    Similarly, unforced natural variability in the form of the AMO and PDO may be a significant contributor to some intervals, but this generalization can’t automatically be applied to 1950-2005 (I’m neglecting here the possibility that there is an anthropogenically forced component to these oscillations, and treating them as independent climate mechanisms). Significant climate effects may occur with these mechanisms at the height of their warming or cooling amplitudes, but over the relevant interval, inspection of their behavior implies little net effect. It is hard to quantify this precisely, because each of these oscillations is regional and detrended, but over the entire interval, the magnitude and duration of behavior when they exceeded their zero baseline values and were exerting a warming influence were similar to the duration and magnitude of cooling – something that would not necessarily be true for other intervals during the past century. The same conclusion can be reached via data unrelated to the reported values of the AMO or PDO indexes – ocean heat content, which increased significantly over the 1950-2005 interval. This finding is very difficult to reconcile with a major contribution to surface temperature from heat redistribution mechanisms whereby the ocean transfers heat to the surface, because the latter would require the ocean to lose heat. Isaac Held has addressed this issue quantitatively at Heat Uptake and Internal Variability.

    Beyond the AR4 evidence cited in chapter 9 and elsewhere, additional data (some since AR4) provide a different form of confirmation for the correctness of the attribution statement that is largely independent of the GCM data cited in chapter 9. Probabilistic estimates of “transient climate sensitivity” (TCS) or “transient climate response” (TCR) permit us to estimate temperature changes concurrent with the climate response to a forcing without a need to estimate what might happen at equilibrium many centuries later. The TCR data reported by Gregory and Forster 2008 yield a 95% confidence interval of 1.3 – 2.3 C warming for a CO2 doubling. Based on the fractional CO2 increase from 1950 – 2005, and the logarithmic relationship with forcing, the estimated effect of this increase is about 1/3 of the doubled CO2 warming. For the 1.3 -2.3 C range cited, this yields a lower bound estimate of 0.44 C, a mid-range estimate of 0.6 C, and a high end estimate of 0.77 C. Adding the effects of methane to the CO2 effect would magnify the increase further by a small amount.

    If we now look at the observed global temperature trend for the interval, we find a warming of about 0.5 – 0.6 C. For the ghgs to have accounted for less than half (0.3 C) of the 0.6 C figure would require a very large cooling offset to all the ghg warming mechanisms operating during the interval. It seems unlikely that even a strong negative aerosol forcing would have been sufficient to do this, and so there is no obvious way to reduce the 0.44 – 0.77 C warming from CO2 below 0.3 C. If we assume, however, the existence of some unappreciated cooling sufficient to accomplish that reduction, we remain with the problem that it would have offset all the other warming influences competing with the ghgs for dominance. There is nothing in WG1 chapter 9 or elsewhere to support a strength from solar and black carbon forcing between 1950 and 2005 that would allow them to raise temperatures by 0.44 to 0.77 C, even with a small effect from unforced variability added in. Rather, the most likely estimates would place them well below the CO2/methane effects.

    For these reasons, I see the attribution statement as highly conservative, albeit with a proper recognition for remaining uncertainties. A more likely apportionment would probably put the ghg contribution closer to 70 or 80 percent. Even so, overlooked possibilities can never be excluded, including serious miscalculation of all the estimates cited above. This should preclude any attempt to assign responsibility to the ghgs with absolute certainty. Calling that attribution “very likely”, however, would appear to be justified.

    • @Fred Moolten…

      Fred I suspect you’re evading the question, WRT traceability.

      …I also believe that the requisite information is buried within the report in chapter 9, along with other relevant data in chapter 2 (on black carbon) and scattered references elsewhere on unforced variability. It’s fair to say that traceability has not been made convenient.

      When you say information, I assume you mean data. Not the algorithm by which the data is translated into a final confidence level. AFAIK that’s not there, therefore there’s no effective traceability.

      The attribution statement relates to how we divide up the observed post-1950 warming, and is therefore limited to climate phenomena that exerted a warming effect during that interval.

      Since you didn’t add any caveats, I can only say that your use of the term “divide up” WRT such a non-linear phenomenon as climate invalidates your whole statement.

      Observational data show a substantial cooling aerosol effect from 1950 through much of the 1970s, and some reversal of that effect later,

      Subject to a whole bunch of unstated assumption about the effects of aerosols. AFAIK we don’t even have any idea what their indirect effects are. Do we?

      It is possible to contrive scenarios for a small alteration (e.g., in the contribution of the black carbon warming component of aerosols), but not a major reapportioning through any plausible mechanism I’m aware of.

      The indirect effect of aerosols was left out of AR4. Aerosols have a substantial effect on clouds, and therefore probably the effective feedback from them.

      Aerosol uncertainty in the models, for example, is therefore largely irrelevant to the attribution statement. At the same time, it is a major source of model uncertainty, and so when that component is omitted, the models become less a source of uncertainty than has been claimed.

      No, the models become more of a source of uncertainty, since they have less a priori probability of actually representing the real world. To reduce uncertainty, the models would have to include both direct and indirect aerosol effects, without kluging, yet still replicate the 20th century observations.

      Similarly, unforced natural variability in the form of the AMO and PDO may be a significant contributor to some intervals, but this generalization can’t automatically be applied to 1950-2005 (…).

      Why are you assuming there’s only those 2? It’s an extremely complex non-linear system, seems to me we should start by assuming hundreds of (at least slightly) independent feedback structures with a 4-7 decade natural cycle. The question is whether more than those two provide enough amplitude in their independent variation to matter. If you start with a default hypothesis of “no”, then you should multiply any confidence level by 50%.

      This finding is very difficult to reconcile with a major contribution to surface temperature from heat redistribution mechanisms whereby the ocean transfers heat to the surface, because the latter would require the ocean to lose heat. Isaac Held has addressed this issue quantitatively at Heat Uptake and Internal Variability.

      I’m not completely sure what Held is saying, but I want to point out one early assumption:

      The assumption is that this is a linear superposition of two independent pieces, so I’ll write the heat uptake as H = HF + HI

      AFAIK this is a very questionable assumption in a non-linear system.

      In addition, he seems to be saying that with internal variation there would be less heat loss from the oceans than with forcing, based on the models. But if the models have some sort of shared error “due to some shared ancestry in their code“, or even the basic assumptions built into it, this assumption may well be incorrect.

      I would also question the assumption that long-term variability has to be “due to ENSO, AMO, volcanoes” which “do not all share the same horizontal structure as the forced response to CO2.” There could well be other sources of variability on a 3-10 decade time-frame that have not been observed because their “phases” canceled out (to the extent that oscillations in a complex non-linear system have such things as “phases”). The climate system alone is very complex and interconnected in so many ways that the assumption that we know what they all are is, IMO, untenable. Add more uncertainty, please.

      And the climate system is interconnected with the biosphere at a number of points, many of which have just begun to be studied. This expands the size of the interactive network substantially. It also supports my contention that the default hypothesis should be that there are undiscovered mechanisms of feedback between the biosphere and the climate, with unknown effects.

      The TCR data reported by Gregory and Forster 2008 yield a 95% confidence interval of 1.3 – 2.3 C warming for a CO2 doubling.

      But you’re assuming a >99% confidence that their models actually represent the real world.

      For the ghgs to have accounted for less than half (0.3 C) of the 0.6 C figure would require a very large cooling offset to all the ghg warming mechanisms operating during the interval.

      You seem to be arguing in circles. First you assume “all the ghg warming mechanisms“, then you say there’s no known cooling mechanism to counter it. But what if the sensitivity is much lower than that? Then, of course, the models are wrong, perhaps “due to some shared ancestry in their code“, but we already know they’re wrong because they didn’t predict the last 10 years. If your assumed “ghg warming mechanisms” are much smaller, than that extra cooling isn’t needed.

      A more likely apportionment would probably put the ghg contribution closer to 70 or 80 percent.

      You’re probably right, if the models are really representative.

      Calling that attribution “very likely”, however, would appear to be justified.

      Doubtful. The confidence that the models actually represent reality should probably be kept under 50%.

      • AK – I think that anyone who reads what I wrote and visits the linked sources will judge that you have either misunderstood or chosen to misrepresent what I said. I’ll let interested readers review the material with that in mind. I’m afraid you completely misunderstood my point about aerosol uncertainty, as well as climate sensitivity ranges outside of GCM simulations. Even beyond that, though, your emphasis on the GCM models is misplaced for reasons I stated – it’s not merely the irrelevance of aerosol uncertainty, but even more that I provided evidence to show that conclusions drawn from those models could also be derived without them – and all the other conclusions I drew involve multiple well-quantified different lines of evidence, which is why the conclusions, while not 100 percent certain, are very unlikely to be wrong. The Held analysis is one of many pieces of evidence along these lines. I don’t think you truly understood it, and I don’t think you accurately described the overall sense of it by quoting one small part. Incidentally, I think his analysis is very well worth reviewing even outside of the context of the 1950-2005 interval we’re discussing, and is relevant to all unforced variability rather than just the AMO or PDO. Held is a brilliant, careful analyst, and very much not an ideologue. His overall conclusions should be reread (but again, it’s only one part of multiple lines of evidence arriving at the same conclusions).

        I truly believe you are struggling too hard to resist an overall conclusion that considerably more than 50 percent of 1950-2005 warming was due to ghgs in the face of compelling multiple lines of independent evidence for that conclusion – I suspect 70 to 80 percent is about right but wouldn’t attempt to prove it to that degree of accuracy, except to say that to get more than a 20 to 30 percent contribution from the alternatives would be very problematic. Rather than go through every one of your statements, I’ll let other readers make their own judgments, but I think you should review what I wrote with an eye toward understanding why your statements don’t invalidate it.

    • Fred Moolten

      You write (pardon the condensation):

      In reviewing recent discussions about the AR4 statement that it is “very likely” most warming from about 1950 to 2005 is attributable to anthropogenic greenhouse gases (hereafter simply “ghgs”)…It’s fair to say that traceability has not been made convenient.

      Curry and Webster say essentially the same thing, i.e. there is “no traceability” in AR4 WG1 for the attribution uncertainty related to this claim. In fact, this appeared to be a principal criticism of the AR4 statement.

      Did I misunderstand you or are you basically in agreement with Curry + Webster?

      Max

      • I wouldn’t say there is “no traceability”, but I agree you have to work to find all the evidence. In fairness to the IPCC, however, their attribution statement was only one of many conclusions they drew in Chapter 9 based on the multiple sets of data they described, and perhaps they didn’t anticipate that they would be asked to provide a separate listing of the justifications for each conclusion. I think the subsequent discussion has been useful in alerting IPCC authors to the need to anticipate these challenges in the future and to provide a clearer route to the evidence specific to each conclusion.

      • As I understand “traceability” it means that a chain of logic is presented, where each step can be quantified and the formal process that leads to the conclusion can be reconstructed objectively. That kind of traceability cannot be presented for the attribution or for any of the projections for the future climate. It cannot be found from the IPCC report, because it doesn’t exist.

        It’s possible to list a number of scientific results and related arguments that support the conclusions presented, but only some minor pieces in that are really objectively quantifiable and those pieces alone cannot tell much, if anything without support from the less objective arguments.

        I do believe that “the balance of evidence” gives clear results, but that’s a belief, not something that I can prove objectively or in fully traceable way. I believe strongly that the whole climate science community cannot prove that objectively in a fully traceable way.

        The IPCC report might have been more clear and systematic in listing supporting evidence and pinpointing remaining gaps in argumentation, but that would not change the basic logic of the presentation. It’s also quite common that, when such issues become clear in one’s own mind, rereading the IPCC report reveals, why some particular wording had been chosen. The writer was obviously aware of the issue, but failed to explain it so that readers could easily understand it. That may be unintentional, but sometimes the impression is that the complication was left implicit by purpose. (The limits on allowed number of pages influences also the choices.)

        It’s a big mistake to insist that nothing should be taken seriously without a traceable proof. A really large part of most certain scientific knowledge has similar problems. There’s a lot of evidence to support the most certain knowledge, but there are holes in the evidence. Therefore science is always subject to doubt, even when the possibility of error in earlier conclusions is judged to be small.

        The recent CERN neutrino experiments that appear to violate Einsteins theory of relativity are a good example. The result is not totally dismissed although very few really expect that it will be found to be correct. People do still believe in Special Relativity, but they agree that more experiments are warranted and that there’s the remote possibility that the theory turns out to be wrong.

  31. Pekka Pirilä

    I would agree with your first statement (and hence disagree with Fred Moolten’s assessment):

    As I understand “traceability” it means that a chain of logic is presented, where each step can be quantified and the formal process that leads to the conclusion can be reconstructed objectively. That kind of traceability cannot be presented for the attribution or for any of the projections for the future climate. It cannot be found from the IPCC report, because it doesn’t exist.

    So you basically agree with the assessment of Curry + Webster on the question of whether there is “traceability”.

    Your second point is more difficult to follow, although I do appreciate that this is your considered opinion. You compare the IPCC premise that most of the warming since 1950 has very likely been caused by human GHG emissions to the theory of Special Relativity and, using this analogy, state your opinion that “traceability” is not really required:

    It’s a big mistake to insist that nothing should be taken seriously without a traceable proof. A really large part of most certain scientific knowledge has similar problems. There’s a lot of evidence to support the most certain knowledge, but there are holes in the evidence. Therefore science is always subject to doubt, even when the possibility of error in earlier conclusions is judged to be small.

    IMO your logic that hypotheses, which lack traceable proof, should still be taken seriously should come with a BIG “depends on” clause. In the case of the IPCC hypothesis that human GHG emissions have very likely been the cause of most of the warming since 1950, I would say that this depends on whether or not there is any major uncertainty that other factors may have had more impact than assumed by the models, this invalidating the “most” (as well as the “very likely”)..

    This is the crux of the C+W “uncertainty monster” paper, as I understand it.

    Rather than side-stepping to the validity of “Special Relativity”, I think we should stick to the topic at hand, and there I would have to support the logic of C+W.

    Max

    • Max,
      I agree on lack of full traceability, but I emphasize that full traceability is rare and almost totally absent with all complex issues. Therefore that’s again a point, where neither “yes” nor “no” is a relevant answer. The right issue to consider is the balance of evidence, which cannot, unfortunately, be handled fully objectively.

      This is an area, where I seem to disagree with Judith (the difference in our views has come a couple of times also earlier). She has expressed her views on uncertainties several times in a way that I consider too narrow and absolute. That applies also to the concept of “Italian flag”, because that adds any one third alternative to yes and no, while I consider it essential that many more grades of uncertainty are considered. A very important part of knowledge is not known with certainty, but is still not unknown. The issues of attribution and climate sensitivity are examples of such knowledge.

      It’s right to require that all arguments in favor and against any conclusions are presented explicitly and openly and as precisely as the nature of each argument allows, but it’s not right to require that a fully traceable construct can be presented before drawing conclusions is allowed. Part of the justification is and will remain subjective. Everybody for whom the conclusions matter must judge, how to value the conclusions with their subjective component. I don’t think that IPCC has done a good job in this area, but it’s has not been as bad as many seem to think.

      The impossibility of presenting full traceability means that quantitative estimates of the level of uncertainty are often impossible to reach. Evidently people, who formulated the IPCC guidelines for presenting uncertainties understood this well, as the guidelines seem to reflect such understanding. The guidelines have, however, never been clear enough or enforced strongly enough to produce good results.

      Risks and uncertainties are a so difficult concept that they are seldom if ever been handled in a way that most knowledgeable people could agree as the right one.

      • @Pekka Pirilä…

        I seem to have a problem with you, Fred Moolten, and, to some extent, Paul S regarding this traceability issue. It seem pretty straightforward to me. We start by assuming there are “multiple independent lines of evidence”, so:

        1.  Idendify all the original assumptions going into the “proof”.

        2.  Develop independent subjective confidence levels for each assumption.

        3.  Idendify all the “demonstrated” points, with their confidence levels.

        4.  Identify all the intermediate “logical conclusions” leading to the “proof”.

        5.  Map all their dependencies.

        6.  Calculate the confidence level associated with each independent, noncircular line of “proof”.

        7.  Combine the levels of certainty for all the parallel lines of “proof” to produce the final level of certainty.

        I’d expect a picture of the result to look something like a cross between a logic diagram and a programming flowchart. (Not that they’re all that different.)

        Some comments:

        Paul S mentioned having “an idea for traceability of thought processes leading to an expert conclusion” that may duplicate or overlap with what I described above.

        I said straightforward, but not simple, easy, or quick. In my experience doing this kind of analysis in support of business Information Systems, I’ve found that multiple iterations are necessary, along with making sure anybody with something to contribute gets heard. Skipping steps in the interest of saving time usually costs more time than you thought you were going to save.

        Circularity is going to be a real problem. For example, take the subject of 0-dimensional models. AFAIK the original models were little more than educated guesses. More complex models (e.g. GCM’s) were built with these 0-dimensional models in mind, and appeared to support them. Thus 0-dimensional models are now considered worthy of much more confidence. You’ll need to have multiple levels of object for the 0-dimensional model (e.g. 0-DM1.0, 0-DM2.0, 0-DM3.0, etc.) with different confidence levels due to different levels of support. This will eliminate the circularity, without sacrificing the increased confidence in the latest generation.

        The end result (and all but the earliest intermediate iterations) will be very large, and complex. The result will be one more point of controversy. Large numbers of knee-jerk ideologues on either side will insist that “it’s got to be right, cause it’s so big and complex”, or “it’s got to be wrong because I don’t understand it.” But at least it will provide an audit trail.

        My guess is that when you’re finished, including all the assumptions now “taken for granted”, you’ll discover the calculations give a much lower confidence level in the result. But whether or not, it will no longer be a guess.

      • I seem to have messed up my blockquotes.

      • AK,

        My view is that the process that you described cannot be done quantitatively. Many inputs to that are subjective and not quantifiable, That’s a real problem for estimating the uncertainties, but I don’t think that the problem can be solved. Much of what you describe is justified and should be done as far as possible, but one should not draw too strong conclusions from the failure of the process.

        There are good reasons to believe that quite a lot is known, but it’s not possible to tell precisely how much.

        The best way forward is in my view to drop the linear approach, where quantitative estimates of uncertainties are needed from one step before the next step is entered, Instead there is a need for people who can look at all factors influencing the decision making simultaneously. The uncertainties of attribution are likely to be one of the smaller problems, when they try to do their task.

      • Well, using percentages for a subjective judgement is probably not a good idea. But we should still nail down in detail each separate subjective judgement. I’m guessing that if people saw how many assumptions they were treating as >99% confidence, they might lose some in the result.

        Thing is, if you pin down the entire network of logic, assign confidence levels to all the assumptions (and research conclusions), and program the whole thing into an automatic calculation, then you can try dropping (or raising) your confidence level on any one assumption, and see how it impacts the results.

        And if we’re going to use any sort of “consensus” judgement of confidence, shouldn’t we use it on the detailed assumptions and conclusions, and let the judges (and their peers and the public) see how the detailed calculation matches up with their “consensus” of the whole in aggregate.

        Personally, I suspect the field(s) will benefit tremendously just from going through the process. It always helps to have your assumptions detailed in black and white. It would probably also be very helpful in selling the result to politicians and business people.

      • We have both problems. Some people have too much trust in their best estimates, but we have also people, who do not trust in what they should trust.

        The inherent impossibility of producing fully traceable and objective uncertainty estimates leads to an uncertainty on second level: We have very much uncertainty in the uncertainty (or accuracy) of the existing knowledge.

        How we should behave, when we have strong second level uncertainty is at the core of difficulties of decision making. That problem can be approached best in context of specific decisions, because the nature of the alternatives and the nature of uncertainties mix in a way that makes it impossible to handle them properly separately.

        Active skeptics use the conceptual difficulties of all that to conclude that no policy measures are warranted, while some activists of the type of Hansen go to the other extreme. Neither of these groups is willing to take the issue seriously and try to analyze, what’s the wisest solution. Judith has taken the uncertainties seriously, but to me her approach is still too narrow and too close to the skeptics end of spectrum.

      • Pekka

        I can understand your “several shades of the truth” premise, yet it leaves the door open for declaring something as “very likely” that might, in actual fact, be less than “more likely than not” or declaring that “most” of an observed phenomenon is attributable to root cause X, when in actual fact it was only “a small portion of” this phenomenon.

        This then becomes problematic when these exaggerated assumptions are applied to predictions for the future, especially in a highly politicized field such as climate science by a highly political organization, such as IPCC.

        Thus we see a 20th century observed warming of 0.6C, with no net change over the first decade of the 21st century (despite increase of CO2 to record levels) blown up to a projected warming over the next 9 decades of – believe it or not – up to 6.4C!.

        Obviously, something does not pass the “reality test” here.

        I believe our host has simply stated that if assumptions on anthropogenic attribution are not clearly traceable they are too uncertain to be used as the basis for long-term projections.

        This lack of traceability plus the worrisome use of “expert judgment” make the IPCC projections for the future no more than simple “crystal ball” predictions.

        If you believe that they are in any way reasonable, so be it. I don’t.

        Max

      • Max,

        You are discussing real problems, but are the risks presently exaggerated or belittled?

        Noting that the issue is difficult doesn’t tell, which way we or somebody else is erring.

  32. Pekka

    Thanks for your comment.

    IPCC has the brief to warn us of human-induced climate changes and their negative impacts on society or our environment.

    The tendency will be, by definition, to find such human induced climate changes and their negative impacts.

    This will mean that naturally induced climate changes will be de-emphasized, while human-induced ones will be accentuated.

    Therefore, without robust traceability the certainty of claims of anthropogenic versus natural attribution will be suspect to the rational skeptic.

    You are not a rational skeptic as I am, so you give IPCC a bit more slack on this question than I would or than Judith – who is not a skeptic, but a climate scientist – would, as a matter of fact.

    So we agree to disagree on this one, Pekka.

    Max

    PS Our disagreement is a different one than the disagreement I have with Fred Moolten.

    Fred believes that the traceability is there, but is difficult to find, while you believe the traceability is not there, but is also not required. My standpoint is that the traceability should be there for the attribution uncertainty to be resolved, but that it is not there.