Odd Contradiction in the IPCC Report

Earlier today, I noticed an odd contradiction in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5). The Working Group II (WGII) Summary for Policy Makers says:

Global economic impacts from climate change are difficult to estimate. Economic impact estimates completed over the past 20 years vary in their coverage of subsets of economic sectors and depend on a large number of assumptions, many of which are disputable, and many estimates do not account for catastrophic changes, tipping points, and many other factors. With these recognized limitations, the incomplete estimates of global annual economic losses for additional temperature increases of ~2°C are between 0.2 and 2.0% of income (±1 standard deviation around the mean)

While citing Chapter 10. The Final Draft of Chapter 10 says:

Globally aggregated economic impacts of global warming are a small fraction of income up until 3°C [10.9.2, medium evidence, high agreement]. A global mean average temperature rise of 2.5C may lead to global aggregated economic losses between 0.2 and 2.0% of income (medium evidence, medium agreement) and losses increase with greater warming.

The SPM says “additional temperature increases of ~2°C” will cause a loss of 0.2 and 2.0% GDP. Chapter 10 says a “global mean average temperature rise of 2.5C may lead to global aggregated economic losses between 0.2 and 2.0% of income.” Why does the SPM say those damages will happen half a degree sooner?

I don’t know. What I do know is the previous version of the SPM said damages for a:

global mean temperature increase of ~2.5°C above recent levels are between 0.2 and 2.0% of income

Meaning the SPM was changed at the last minute to say those damages will happen sooner, contradicting the section it cites.

Now then, the revision-display for that version of the SPM has the ~2.5°C highlighted, which I believe indicates it will be rewritten. That would mean the decision to edit it was disclosed in the revision process. It just doesn’t seem to have been explained. How could they? You can’t justify citing a chapter which lists one value while randomly listing a different value? The only way they could justify it is to change the value given in the “Final Draft” of Chapter 10 to match the SPM. But isn’t that backwards?

I don’t know what’s going on. To try to figure things out, I decided to look into the origin of this ~2.5°C/~2°C value. The previous quote from Chapter 10 is in its Executive Summary section, and it cites Section 10-9. None of those numbers show up in the text of that section. In fact, they don’t show up anywhere else in the chapter.

They also don’t appear to show up anywhere in the figures for Section 10-9:

AR5-10-1

FD-10-2

Which are the only figures with any numerical values in the chapter. As far as I can see, those values have been pulled out of thin air. It seems impossible to tell whether we should expect to lose 0.2 – 2.0% of our global GDP at 2°C or 2.5°C because neither value has any documentation to support it.

For thoroughness, we should try checking earlier versions of Chapter 10 to see if perhaps a reference got lost. Unfortunately, that won’t work. The previous draft, the Second Order Draft, doesn’t have any text about this value in its Executive Summary. That means the text I quoted at the start of this post was only added after the last round of external reviews.


To sum up, an apparently calculated value was seemingly pulled out of thin air and added to IPCC WGII Chapter 10 after the last round of external reviews. It was then cited in a draft of the Summary for Policy Makers (WGII), the most read document for the IPCC Report (WGII). It was then inexplicably changed to be 20% more alarming for the final version of that document.

Advertisements

20 comments

  1. Joe Public, I think I actually have a decent grasp of where the .2-2% part comes from. The problem I have is understanding how 2.5C shrank to 2C and how the underlying work can go completely uncited.

    Well, I also take issue with the underlying work being complete and utter rubbish. But that’s a story for another day.

  2. The “between 0.2 and 2.0% of income” seems to describe the points on Figure 10-1 at the abscissa value 2.5. While the SPM revision draft has “~2.5°C above recent levels”, it also shows by its markups that the prior version had “2.5°C above preindustrial levels.”
    My speculation — and it’s only that — is that the writers of the SPM were under the impression that the the chapter 10 reference to a temperature change of 2.5°C was with respect to pre-industrial. After having changed the reference point in the SPM sentence to “above recent levels”, they realized that they should also have changed the 2.5°C figure to ~2°C.

  3. HaroldW, that would be incredibly strange, but I suspect the explanation for this will be incredibly strange.

    To add to the oddness, Richard Tol has cited the value from the SPM several times without making any comment about it being off. He’s seemingly okay with the change.

  4. As I said in the other thread, IPCC AR4 synthesis report has this: “a 1 to 2°C increase in global mean temperature above 1990 levels (about 1.5 to 2.5°C above pre-industrial)”. That’s the same 0.5 differential we see in this case. I doubt it’s the right correction to the estimates in question, though.

  5. Dale Stephenson, I don’t see how that could explain the discrepancy I’ve highlighted. Under your interpretation, why would a chapter of the IPCC report use one value and the SPM another? Are you saying the chapter made a mistake/the SPM made a faulty correction, or is there some other explanation?

  6. I’m saying that the 0.5 differential from “1990 levels” is the same differential being used for this adjustment. Found a similar instance of the same in Tol 2002b:

    1. The optimal temperatures in Table I are relative to 1990. The model, however, uses temperatures relative to pre-industrial times. 1990 was about 0.5 ◦C warmer than the pre-industrial climate.

    1990 is not the same as “1990s level”, but again we have a straight 0.5 difference from “pre-industrial”. That’s how a temperature increase of 2.5C could become an additional temperature increase of ~2C. I’m not saying it’s the correct adjustment.

    I’m still not sure what the source of -0.2% to -2.0% is, though. That’s not the range plotted in Figure 10-1 for +2.5C. It also isn’t close to matching the ridiculous confidence intervals in Tol ’09s Figure 1, original or revised.

    But if you look at the range between the mean lines on the deleted figure 19-7 that Tol removed in his rewrite, it looks to me like +2.5C there is pretty close to that range. If you pushed out to about +2.8C and dropped the SPEC line, I think the other four would be a dead match. Maybe those models are the source of the range?

  7. Figure 19-7 also says in the caption that the differentials are C above pre-industrial, so assuming that figure was correctly labeled and the models correctly plotted, that would be consistent.

  8. Dale Stephenson, I think there was a bit of a communication problem, but I can see how your interpretation would work now. I reread HaroldW’s comment upthread and saw:

    it also shows by its markups that the prior version had “2.5°C above preindustrial levels.”
    My speculation — and it’s only that — is that the writers of the SPM were under the impression that the the chapter 10 reference to a temperature change of 2.5°C was with respect to pre-industrial.

    Because you quoted AR4, I didn’t make the connection. To be honest, I didn’t look too closely because you had previously quoted AR4 for the multi-model mean of GCM responses to the A1B scenario (which as you’ve since realized, were irrelevant). My mind saw “AR4” and didn’t look too closely.

    I’m sorry for that. I’m normally better about how I read people’s comments. I was responding from my phone in the middle of the night. That normally wouldn’t be a problem, but if you look at my Twitter feed, you’ll see I’ve been distracted by having my head buried in books. I’m on my fifth one in 38 hours now. That’s no excuse. Neither is the vodka I’ve been drinking, Still, it’s at least a reason.

    But if you look at the range between the mean lines on the deleted figure 19-7 that Tol removed in his rewrite, it looks to me like +2.5C there is pretty close to that range. If you pushed out to about +2.8C and dropped the SPEC line, I think the other four would be a dead match. Maybe those models are the source of the range?

    I’m pretty sure that’s not the case. I haven’t tracked down an exact match, but I can make a pretty compelling case for the source. As you might expect (given the values were added to Chapter 10 only after Richard Tol added tons of stuff from his own work), I believe it’s from on work done by Richard Tol. There are a couple of his papers which show very similar values in graphs.

    I was going to write a post about the work, but I’ve come to the conclusion practically nobody really cares about any of this. I’m still willing to write about it (because writing about whatever interests me is the purpose of this blog), but I don’t know if I have the interest to do an actual writeup. Sometimes just figuring out things for myself is enough.

    For a short version, I suggest starting with this comment Tol left at another blog. He references the same values as a 1.1% loss, the central value for the range, while referring to “~2K of warming.” (That’s one example of I was referring to above when I said he references the IPCC SPM numbers without any sign of objection.) In the same comment, he says:

    The big difference with the old and the new results is not what happens at modest warming. The change reflect the limits of parametric statistics. See http://ideas.repec.org/p/sus/susewp/6914.html

    The “change” he mentions isn’t the same one we’re talking about here, but the comment suggests the numbers in AR5 are related to the paper. If you check out the paper, you’ll find him comparing (basically) the same estimates we’ve been talking about all along, just with a different approach. Along with the different approach to combining the estimates, he’s begun using a different approach to calculating uncertainty values. The results he gets (on both accounts) are very close to what we see in AR5. They don’t seem to match perfectly, but there are tons of quirks and tweaks to his approaches which could explain the minor discrepancies.

    In addition, I’d suggest checking out the precursor to that paper (found here). I’d also like to note while I’ve discussed Tol 2009 a lot, that’s not because I believe it is the most relevant citation. I probably should have clarified this before, but a more direct link to what’s found in AR5 is Tol 2013. Tol 2013 is basically a follow-up to Tol 2009. The reason I discussed Tol 2009 is the “correction” published for Tol 2009 was more informative than the one published for Tol 2013.

  9. Okay, I can’t leave my reference to those two papers by Richard Tol unadorned. I do think interested people should look at them for themselves. I also think anyone who does will see Richard Tol has gone far beyond Tol 2009 in comparing the estimates, to a point which is indefensible.

    But most importantly, I think I need to say those papers are worse than dreck. Reading them makes it clear Tol does not understand the math he’s using. A couple of his results, including at least one he describes as expected, are so wrong as to be insane.

    In effect, all Tol has done in those papers is apply a overly-complicated smoothing algorithm to ~20 values which aren’t remotely comparable. And of course, he finds the same “pattern” of moderate warming being beneficial because his calculations are skewed because of that one outlier from his Tol 2002 paper. If you remove that one outlier he created, the quintessential “pattern” of his vanishes.

    In other words, once you strip away the unnecessary complexity, all you have is the same basic problem as before with completely bogus uncertainty calculations.

    It’s like seeing Michael Mann follow up on his original hockey stick with Mann 2008. Exact same problems dressed up with newer methodology which adds nothing. I swear, Richard Tol is a mini-Michael Mann.

  10. Found another interesting comment on what “pre-industrial” means at Judith Curry’s site, in this thread:

    http://judithcurry.com/2013/10/01/negotiating-the-ipcc-spm/

    She quotes the IPCC AR5 WG1 negotiations as follows:

    On temperature change between 1850-1900 and 1986-2005, Canada, supported by Belgium and the US, proposed providing context for the two time periods, referring to the former as the early instrumental period, and the latter as the AR5 reference period used for projections. Delegates debated at length whether to place this text in the observations or in the projections section. Delegates also discussed whether to use the term “pre-industrial” for the period 1850-1900, with some countries suggesting this would lead to confusion as in other places, “pre-industrial” refers to 1750.

  11. Chapter 10 refers to temperature since pre-industrial times. The Summary for Policy Makers to temperature since recent times. The difference is 0.61K.

    I think this is terribly confusing, and I wish the authors of the SPM had not introduced this.

  12. Brandon, thanks for the links, I don’t have time to read them in full yet, but the discussion of the AR languages in the paper linked from Tol’s comment does shed a lot of light on the -0.2C to -2.0C range. And yet another thing relevant to the pre-industrial differential:

    The SPM of AR5 refers to temperature increases relative to “recent times” which is 0.6°C above pre-industrial.

    If true, it means the 0.5C adjustment we see between warming and additional warming is off by 0.1C.

    For technical summaries:

    AR5 “Globally aggregated economic impacts of global warming are a small fraction of income up until 3°C [10.9.2, medium evidence, high agreement]. A global mean average temperature rise of 2.5C may lead to global aggregated economic losses between 0.2 and 2.0% of income (medium evidence, medium agreement) and losses increase with greater warming. Little is known about aggregate economics impacts above 3°C.” (Arent et al. 2014)

    Compared to the state of knowledge at the time, AR2 was overconfident. AR3 more accurately reflected the uncertainties, but peculiarly suggested that the best guess at the time was zero impact with a symmetric confidence interval. AR4 stresses the differences with AR3, which are in fact insignificant. If anything, AR4 should have been more optimistic about the impact of larger warming. AR5 omits the initial benefits, shows the 67% confidence interval (rather than the 95% one as in Figures 1 and 2), and admits to ignorance about the impacts of larger climate change.

    And for the SPM:

    AR5 “With these recognized limitations, the incomplete estimates of global annual economic losses for additional temperature increases of ~2°C are between 0.2 and 2.0% of income (±1 standard deviation around the mean) (medium evidence, medium agreement). Losses are more likely than not to be greater, rather than smaller, than this range (limited evidence, high agreement). Losses accelerate with greater warming (limited evidence, high agreement), but few quantitative estimates have been completed for additional warming around 3°C or above.” (Field and Canziani 2014)

    This is also an interesting complaint:

    Although there has been little advance in knowledge over the 19 year period between AR2 and AR5, the IPCC suggests otherwise.

    Figures 1 and 2 are the “Restricted Nadaraya-Watson kernel estimates of the impact”. I think comparing those to the source damage paths (finally illustrated!) would be an enjoyable post for you. My first impression is that it makes no more sense than the regresssion line did.

  13. Richard S.J. Tol, you say:

    Chapter 10 refers to temperature since pre-industrial times. The Summary for Policy Makers to temperature since recent times. The difference is 0.61K.

    I think this is terribly confusing, and I wish the authors of the SPM had not introduced this.

    Would you care to explain where the numbers came from in the first place? I’m glad to know the statement in Chapter 10’s Executive Summary is in relation to pre-industrial times, even though it does nothing to say so. I’m just not sure of the value of that knowledge when it’s seemingly impossible to know where the numbers came from in the first place.

  14. Richard Tol, thank you for the comment. When the damage paths are plotted for the models, does 0C represent the -0.61K “pre-industrial” level and do the various models actually estimate beginning at the 0C level?

  15. Dale Stephenson:

    If true, it means the 0.5C adjustment we see between warming and additional warming is off by 0.1C.

    They used the ~ symbol to show the value is an approximation so I can’t complain about a difference that small. If what Richard Tol says is true (and I have no particular reason to think it isn’t), the discrepancy seems understandable. The problem is until we can actually see where those numbers came from, there’s no way to know what Tol says is actually true.

    Plus, even if what he says is true, I don’t see how any reader of the IPCC report could verify it. I can’t find anything in Chapter 10 to justify what Tol says. He may be correct, but I don’t think uncited and unverifiable claims are okay.

    Figures 1 and 2 are the “Restricted Nadaraya-Watson kernel estimates of the impact”. I think comparing those to the source damage paths (finally illustrated!) would be an enjoyable post for you. My first impression is that it makes no more sense than the regresssion line did.

    I’m not sure where you’re seeing “source damage paths.” Richard Tol didn’t show any. I think you may have misunderstood what he showed in his papers. All he showed is (improper) calculations for various subsets of the same estimates we’ve been looking at.

    Those two papers aren’t showing any new information. All they’re doing is applying different algorithms to the same data we’ve been looking at all along.

  16. I’m not sure where you’re seeing “source damage paths.” Richard Tol didn’t show any. I think you may have misunderstood what he showed in his papers. All he showed is (improper) calculations for various subsets of the same estimates we’ve been looking at.

    Paper 64-2013. Figure 1 has “six alternative fitted models”. One of those is the kernel estimate, but the other five look like honest to goodness damage paths to me. Following figures illustrate them individually, figure 5 shows them together and calculates an average. For an actual semi-apples-to-apples comparison, I would think the average of these five paths would be a reasonable estimate, though it would be a reasonable estimate of this *kind* of estimate, not of the entire field.

  17. Dale Stephenson, I think you are way, way misinterpreting those figures. Those “six alternative fitted models” are just six different regressions on the same set of estimates we’ve been looking at all along. They’re nothing but different ways of drawing lines through the same ~20 data points. If you think me qualitatively contrasting the various data points in AR5 WGII’s Figure 10-1 was wrong, those graphs are far, far worse.

    Those lines you are looking at are just five of the infinite number of lines I could draw to “fit” the ~20 data points. Section 3 of the paper even gives equations showing the functional form of (some of?) them.

  18. Richard Tol,
    Strictly speaking, the 0.61 K is the difference between “recent” — as used in AR5 WG1 at least, this is 1986-2005 — and the latter half of the 19th century: “Based on the longest global surface temperature dataset available, the observed change between the average of the period 1850–1900 and of the AR5 reference period is 0.61 [0.55 to 0.67] °C.”

    I’ll stipulate, however, that the difference between that period and “pre-industrial” is minimal, perhaps 0.1 K. The RCP scenarios, which are referenced to zero forcing in 1765, show an average of 0.2 Wm-2 in 1850-1900. If we take TCS to be, say, 1.5K/doubling, and doubling=3.44 Wm-2 (following Forster et al.), the forced change is slightly under 0.1 K.

    I agree with Brandon that expressing the change as “~2 °C” covers (to me, at least) such minor discrepancies, especially given all the uncertainties. You wrote that it was unfortunate that the SPM chose to refer temperatures to recent levels. But isn’t the usage in chapter 10 the odd man out? If I remember correctly, AR1 chose to refer temperatures to pre-industrial, but I think the general practice in AR5 (and AR4 as well) was to use a recent baseline.

    On a different topic, are you asserting that all the studies listed in WG2 chapter 10 are referenced to pre-industrial temperatures? The one study which I looked at — Nordhaus 1994 expert elicitation — seemed to peg its 3 K warming from (then-)current temperatures, see comment in previous thread. It’s listed in Table 10.B.1 with a 3K warming, and you appear to claim that the reference period for those entries is pre-industrial.

    Perhaps I’m seeking more precision than is reasonable from what is arguably a very imprecise exercise…

  19. Brandon, now that I’ve actually had a chance to read the paper, I see you’re right. I saw Figure 5 describe it as “Expected impacts for individual models” and thought I was finally seeing an estimated damage path. But it’s based on the same incomparable data points that are unsuitable for the purpose. Not an improvement on IPCC figure 10-1 at all.

    It looks like the closest thing to what I want to see is the deleted Figure 19-7. I’ll have to look up the papers it references when I get a chance to see if those really are estimated damage paths plotted correctly.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s