In IPCC land “very likely” means 90% – 100% probability, while extremely likely means 95% – 100% probability. In light of that, I was wondering if anyone had any insights as to why, in Chapter 10 of the IPCC’s WGI report, it says
More than half of the observed increase in global mean surface temperature (GMST) from 1951 to 2010 is very likely due to the observed anthropogenic increase in greenhouse gas (GHG) concentrations.
with respect to anthropogenic GHGs, but says
It is extremely likely that human activities caused more than half of the observed increase in GMST from 1951 to 2010.
when it comes to human activities overall.Judging from the figure on the right, it seems that if you combine all anthropogenic influences then there is only a small chance (< 5%) that non-anthropogenic influences could have produced more than 50% of the observed warming since 1950. If, however, you consider anthropogenic GHGs alone, then the chance that other factors (OA, NAT, Internal Variability) could produce more than 50% of the observed warming goes up slightly, and hence one can't reject the null at the 95% level.
The odd thing is that it seems likely that anthropogenic GHGs alone would have produced much more warming than that observed and that other anthropogenic influences would have had a cooling influence. Therefore it seems slightly strange that it's extremely likely (> 95%) that anthropogenic influences overall caused more than 50% of the observed warming, but only very likely (> 90%) that anthropogenic GHGs caused more than 50% of the observed warming. Maybe I’m missing something but this seems to be a consequence of sticking rigidly to a frequentist statistical analysis, rather than following a Bayesian approach, as many seem to think would be better.
So, if anyone else has any insights as to why there is this apparent discrepancy between anthropogenic influences overall and anthropogenic GHGs only, feel free to point it out in the comments. FWIW, this post was motivated by Fabius Maximus’s most recent post in which he appear to be arguing that because the confidence about the anthropogenic GHGs is only very likely (> 90%) that it is insufficient for policy purposes.
Paul S, in the comments, points to Figure 10.4 which I’ve included below. The left-hand panels shows the trends due to anthropogenic GHGs, other anthropogenic influences, and natural influences, while the right-hand panel shows the trends due to all anthropogenic influences combined and natural influences. It seems clear that when you combine all the anthropogenic influences, you get a much clearer attribution then when you consider anthropogenic GHGs and other anthropogenic influences in isolation.
On Twitter, Doskonale Szare points to this comment from Gavin Schmidt, which may be consistent with what I was getting at here
For a visual description of what’s happening there is Figure 10.4, showing the individual model results which are combined to produce Figure 10.5.
Thanks. I’ve added that to the end of the post.
All I can say is that Fabius Maximus must miss out on an awful lot of opportunities. I expect he scorns insurance, and never packs an umbrella or a raincoat either, unless he sees it’s already pelting cats and dogs.
Fabius Maximus don’t understand the risk management. You policy is based on the product of the probability of occurrence and the consequence. Normally, most of the time probabilities are low (<.5), but the consequences are severe. If the product of the two is significant, you act.
For example, there is a policy that building have fire exit. Probability that a fire occur is very low (P <.0,01), but the perspective of having hundred of people dying in a fire is enough to justify the policy of having fire exit.
Yes, I agree that is also a factor. I was simply trying to get across that rigidly applying a 95% criteria risks a Type II error, especially if you have additional understanding of the system. That argument didn’t appear to work.
I find this confusing. I’m not sure I have any insights, but Gavin Schmidt probably does, and covered this at realclimate
It came up again there as a result of one of Judith Curry’s posts:
I made an attempt to tease out the understanding. My key takehome from this was that the attribution is NOT strictly quantitative but applies judgement; this judgement makes the attribution significantly *less* than the straightforward numbers would give. It also I think answers your question on why the uncertainty is different.
Gavin’s quote given by ATTP in the first comment makes sense to me, but perhaps an analogy helps some others. Mr. Greenhouse and Ms. Aerosol have bought mixed packs of tasty M&Ms for a meeting – Mr Greenhouse bought a special edition where you get many more red M&Ms and Ms Aerosol bought a special edition where you get many more blue ones. Their colleagues Enso, Pido, Amo, Eruptiony and Sunny bought jelly beans and mix packs.
Everyone pours their contribution into the meeting bowl, but people are a little selfish and also keep some for themselves. The boss enjoys her treats wants to know whom to thank and promote so she checks out the bowl. It’s overflowing with M&Ms with the odd jelly bean speckled through. She stirs it a couple of times and M&Ms just keep coming up, so she’s sure that Mr Greenhouse and/or Ms Aerosol are the biggest contributors. There are also more red ones than blue ones so it looks like Mr Greenhouse is the biggest contributor, but to be sure Ms Boss sits down with her abacus and does the maths. There’s some uncertainty in the fraction of colours in each packet of M&Ms and she also doesn’t know what fraction of their sweets each person contributed and kept for themselves. It turns out that there are various combinations where it’s possible that Mr Greenhouse no longer contributed the majority of all the sweets, even though it’s obvious that most of them are M&Ms.
Basically, if the anthropogenic stuff looks really different to natural stuff (which it generally does, like M&Ms & Jelly beans) then you can used observed patterns to pick them apart with confidence. But if two anthropogenic factors look similar (a mixture of M&Ms from two packs where each one has more of one colour), then it’s harder to pick them apart for sure. (of course, I ignored negative M&Ms, perhaps Ms Aerosol is greedy and keeps stealing sweets from the bowl when no-one is looking?)
If you build up from the physics of radiative forcing alone then you’d actually expect the opposite ease of attribution, but that’s a slightly different approach from fingerprinting which kind of works backward from observed patterns. From the fingerprinting approach Gavin’s comment makes sense to me.
Of course, at the end of the meeting, a blogger who wasn’t there and didn’t do any of the calculations says that these M&Ms don’t really exist and anyway, this “claim” that Mr Greenhouse and Ms Aerosol contributed M&Ms is probably not true, despite receipts, witnesses and security camera footage confirming that they did. His blog becomes the most-viewed site on the topic of sweets in meetings.
A very good analogy overall, but this puts it in a class of its own
Michael Tobis has a good post on this general topic of humans probably causing more than all: http://initforthegold.blogspot.com/2015/01/more-than-all.html
Yes, that is a good post, although it was partly motivated by my Twitter discussion with Judith Curry, so maybe I’m biased 🙂
“1)Models produce a good simulation of natural variability. AR5 section 9.5.3 concludes “ Nevertheless, the lines of evidence above suggest with high confidence that models reproduce global and NH temperature variability on a wide range of time scales.”
2)Model spread of natural variability in the 1950-2010 timeframe is ca zero +/- 0.1 degC
3)Therefore the rest must be anthro”
WRT #Tell that to 1900-1940.
I think that would be Judith’s argument.
What would your argument be Steven?
I’m not convinced by that argument either: 1) models simulate natural variability > 2) it’s not big enough > 3) therefore rest is anthro.
There are too many difficulties in getting natural variability from models IMO, especially over multiple decades. It’s easier to quantify the anthropogenic component using methods like fingerprinting, plus evidence from palaeoclimate, recent temperatures changes, emergent constraints etc. It’s extremely hard to find a sensible analysis that comes up with realistic values of temperature response to anthropogenic forcing that gives much chance of <50% anthropogenic attribution.
It's possible that some kind of natural variability is to blame, but the probability must be very small given the amazing coincidences that would be required to fool all of the methods based on fingerprinting AND observed energy budget AND palaeoclimate AND emergent constraints. For me, the combination of these methods and the similarity of their results means we can confidently say that global warming is mostly human-caused.
Curry's arguments leave me really confused sometimes. The best estimate is that human factors have caused warming equivalent 110% of that observed, so the best estimate for everything else all together is -10% because the total must be 100%. Curry says things that I interpret as meaning she believes that observed warming can be more than 100% of observed warming. I find it really hard to wrap my head around this.
No matter how many hypothetical natural warming factors Curry invents, the best estimate of the cumulative effect must be -10%. There might be a +220% contribution from the PDO, but then other natural factors must sum to -230% (ignoring cross correlation). You can invent as many hypothetical natural factors as you want and the attribution statement doesn't change. As far as I can see, the only way to change it is to show systematic errors across all the independent methods of attribution to human-caused factors.
Tell that to 1900-1940.
1) 1900-1940 is a shorter period than 1950-2010 so the estimated range of internal variability influence will be larger.
2) Pretty much all CMIP5 model runs simulate significant warming over 1900-1940. Many produce greater warming than observed. It’s possible that the level of warming over 1900-1940 could have been produced by known basic forcing factors alone (anthropogenic and natural). It’s almost certain that known basic forcing factors were a significant influence on the warming trend over that period.
Therefore, trying to promote 1900-1940 warming as some of kind of unexplainable wildcard of “natural variability” doesn’t work. Once you account for known factors which we would expect to cause warming over that period the amount remaining is comfortably consistent with IPCC internal variability estimates.
One question I have about the tightness of the all-anthropogenic scaling factor produced by fingerprinting exercises is whether or not that’s simply being achieved by pairing with a much smaller natural forced response.
In the tests with separate GHG and OA regressions the responses are at least roughly at the same order of magnitude, which means the pattern of warming could be caused by many very different combinations of the two factors, leading to a lack of precision and sometimes entirely unconstrained results.
When there is just one Ant and one Natural you have one factor with a substantial warming signal and one factor with basically no signal at all. As long as the simulated pattern of anthropogenic warming is consistent with observations surely it’s always bound to return a result that anthropogenic forcing explains all the observed warming (because there’s nothing else available in the test to explain it).
From what I’ve seen simulated patterns of surface warming/cooling due to varying forcing factors are just not that different. One test I would propose is to produce a simulation in which solar forcing is scaled to the magnitude of anthropogenic forcing and then use that in an Ant/Nat fingerprinting exercise. Would the method identify the solar forcing as erroneous and again return a near zero natural contribution? If not (and I suspect it wouldn’t), what do these results actually tell us that we didn’t already know? I feel like I’m missing the point somehow.
I’m not sure I’m quite getting what you’re suggesting. If you were to amplify the solar forcing wouldn’t that then not produce stratospheric cooling, for example, and so one could eliminate that as a major contributor?
Of course. I’m talking about the studies which contribute towards Figure 10.4, which only take into account surface warming patterns.
Okay, I see you said that explicitly 🙂 So, what you’re suggesting is that these studies are really just confirming what should be obvious given that our current understanding is that solar and internal variability can’t have contributed substantially since 1950. Hence they don’t explicitly rule that out since we’ve essentially assumed this anyway.
That’s basically the gist of my thinking, yeah.
I guess this is why I think some people regard attribution studies as falling a bit of a trap. Given our current understanding, it’s really hard to see how it can’t be mostly anthropogenic. An attribution study, however, tries to quantify this. However, if the result ends up being less than 95%, people then regard this as somewhat unsure, even though all the evidence points towards it being mostly anthropogenic.