Climate sensitivity is IMHO, the most important unresolved issue in climate science. A number of recent papers, including the IPCC AR5 leak, plus the recent Economist leak of a later AR5 draft are pointing to lower climate sensitivities than what have been sold in the past. Now, we have an older model that seems to do a better job of hindcasting than even the models run on the new Met Office supercomputer.
Steve McIntyre writes at his blog:
Results from a Low-Sensitivity Model
Anti-lukewarmers/anti-skeptics have a longstanding challenge to lukewarmers and skeptics to demonstrate that low-sensitivity models can account for 20th century temperature history as well as high-sensitivity models. (Though it seems to me that, examined closely, the supposed hindcast excellence of high-sensitivity models is salesmanship, rather than performance.)
Unfortunately, it’s an enormous undertaking to build a low-sensitivity model from scratch and the challenge has essentially remained unanswered.
Recently a CA reader, who has chosen not to identify himself at CA, drew my attention to an older generation low-sensitivity (1.65 deg C/doubling) model. I thought that it would be interesting to run this model using observed GHG levels to compare its success in replicating 20th century temperature history.
The author of this low-sensitivity model (denoted GCM-Q in the graphic below) is known to other members of the “climate community”, but, for personal reasons, has not participated in recent controversy over climate sensitivity. For the same personal reasons, I do not, at present, have permission to identify him, though I do not anticipate him objecting to my presenting today’s results on an anonymous basis.
In addition to the interest of a low-sensitivity model, there’s also an intrinsic interest in running an older model to see how it does, given observed GHGs. Indeed, it is a common complaint on skeptic blogs that we never get to see the performance of older models on actual GHGs, since the reported models are being constantly rewritten and re-tuned. That complaint cannot be made against today’s post.
The lower sensitivity of GCM-Q arises primarily because it has negligible net feedback from the water cycle (clouds plus water vapour). It also has no allowance for aerosols.
This is a must read. See the results here:
Results from a Low-Sensitivity Model
richard verney says:
July 22, 2013 at 4:09 am
“Unless back radiation has no effect, it is impossible to resolve climate sensitivity without first fully resolving natural variation.”
Ian W says:
July 22, 2013 at 4:15 am
“The relationship does not apply if a phase change is encountered, because the heat added or removed during a phase change does not change the temperature.”
Part of what I commented at Steve M’s site (and embellished here) concerns the fact that the earth’s crust was thinner and was hotter in the Archean ~3B ybp evidenced by extrusion of komatiite lavas at 1700C compared to modern ~1100C basalts. The crust thickened and the surface cooled over the next 1B yrs or so, reaching an “equilibrium” around which natural variation of 7 to 9C has oscillated ever since. These limits on temp appear to have had nothing or very little to do with CO2 and are supported by the unbroken chain of life. Supermodels should be constructed that give a horizontal trend with the differences in extremes of 7-9C – probably a combination of orbitals and phase changes of water (as long as a substantial part of the ocean doesn’t freeze the lower temp is essentially capped and as apparent from the maximum sea surface temp achievable of 31C a la Willis’s Thermostat Hypothesis expanded, so is the upper temperature limit). On the very long term, I can’t see any reasonable argument against negative feedbacks being the main players in constraining temperature change. If this is the case, then negative feedback on the micro scales of days to decades is logically acting, too (max 31C SST). Even CO2 oscillations are feedbacks. If not, then there would be nothing to stop us from boiling the seas or freezing solid. These could happen but it would be due to an astronomic cataclysm, not a puff of CO2.
If you take the model predictions at each point and subtract from them the actual data, you are left with the “residuals”. What a modeler would be looking for are residuals that are normally distributed about “zero” with small variance.
After digitizing the data in the figure (using Engauge) I performed this calculation for both models. Both gave residuals that are centered at about 0.11 with a standard deviation of 0.16. This tells me that both models are biased on the warm side by 0.11(i.e. overall point for point the models overpredict the actual temperature by 0.11 throughout the range of measured values).
Why is that?
If the model can’t at least predict the known range of measured values without a bias how can you trust forecasting with it?
Bill Illis, first of all I appreciate all your hard work. I use many of your charts to inform alarmists of the truth. Thank you.
However, I think you are missing one natural change because it is almost unmeasurable. What I’m referring to is a general equilibrium factor. Given we have seen many times as warm or warmer than the present over the last 10K years without any additional CO2, it’s not too difficult to believe there’s a baseline temperature determined in general by solar energy.
The LIA might very well have taken us below this equilibrium condition and part of the change over the last 250 years is simply a slow return, most likely due to a return to normal solar input. Since there’s no clear way to understand exactly why the LIA led to cooling (although an inactive sun during the Maunder or volcanoes could both be the culprit) it is difficult to determine what the equilibrium condition should be. If it is represented by the MWP, RWP, etc. then returning to this value needs to be factored into the determination of CS.
The point is, it doesn’t take a more active sun to cause warming from a point below the equilibrium. It only needs to be consistent. Time takes care of the rest.
And I betcha a model ignoring CO2 would outperform them both.