2011-08-28 04:47:05 | Lessons from Past Climate Predictions: IPCC SAR | |

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
Next one in the series is ready for review. As other investigations have found, the SAR actually underestimated the warming over the past two decades. I think the most likely explanation is that their "best estimate" sensitivity (2.5°C) was too low. 3.5°C sensitivity would account for the difference. But there may be another factor involved in the discrepancy too. | |

2011-08-28 13:14:48 | ||

Alex C coultera@umich... 67.149.101.148 |
>>>For example, perhaps the IPCC overestimates the cooling effect of aerosols. Yes, the 2007 figures from SAR are higher (in magnitude) than AR4 puts them at, by about 0.2-0.3 W/m^2. However, the CO2 figures are also overestimated - the 1992 SAR levels were practically on par with the 2007 AR4 levels, only a tad less - comparing 2007 to 2007, you can see the CO2 prediction from SAR was a whopping 2 W/m^2, compared to 1.66 from AR4. I think an underestimation of climate sensitivity is a more likely candidate, or the TCS v. ECS estimate. In any case, the net forcing I think is what matters more, rather than component mis-estimates (unless you want to get into efficacy - not sure how that would play out, maybe the average effect of each efficacy dampens the forcing overestimation so that the best ECS is lower than 3.5, though still higher than 2.5). As you had noted, the net forcings are almost identical. A couple suggestions, don't say "significant" when talking about SAR v. realized temperatures, as it often implies a statistical meaning (try, notable). Also, can you remodel the SAR predictions incorporating the ECS change as you said, and include a graphic to show us like you did last time? | |

2011-08-28 16:08:48 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
Actually I was just reading the TAR and working on that post, and found out that in the FAR and SAR, they thought the radiative forcing for 2xCO2 was 4.37 W/m2. In the TAR they corrected the value to 3.7 W/m2. I'm trying to figure out if that correction would alter the projections. Because they still have the forcings in 2010 correct. But they said the sensitivity to 2xCO2 was 2.5C, but I guess that's the sensitivity for 4.37 instead of 3.7 W/m2. In that case, their model sensitivity was actually lower than they thought, which would account for some of the discrepancy. So maybe correcting for that and the low temperature projection would bring the adequate sensitivity to 3 instead of 3.5C. I'll have to look at it a bit more tomorrow. But yeah, I'll plot it with the adjustments too, and take out the significant comment. | |

2011-08-28 16:24:35 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
Oh and Alex, you win a good "told you so", because I know you made a comment that I should check the forcing calcultion on the FAR :-) | |

2011-08-29 04:14:40 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
Yep, the actual sensitivity of the SAR "best estimate" was thus 2.12°C. Adjusting it to match the observed trend brings the sensitivity up to 2.98°C for 2xCO2. Very cool. | |

2011-08-29 15:12:50 | ||

Alex C coultera@umich... 67.149.101.148 |
>>>the actual sensitivity of the SAR "best estimate" was thus 2.12°C What? If you have this equation: ∆T = L*A*ln(C/C_0) and remember that A is derived from RTMs, and C values and ∆T values from observations or proxies, the true dependent variable here is L. If you make a mistake in your radiative transfer models so that A is actually supposed to be smaller, you correct this mistake by dropping A and recomparing against your observations to redetermine what L is - in other words, you don't concomitantly decrease ∆T and A, you increase L as you decrease A to what it ought to be. Once you determine L as per historical context, you apply L to your equation with A and hypothetical C's to determine the temperature response. You need to remember that L is the sensitivity factor, and that its value in an equation only makes any physical sense if it is in the context of the A value from which it is derived. If A is wrong, L cannot possibly stay the same, you cannot divide ∆T by the ratio of your A values. When the issue comes down to a discrepancy in A, the model must be applied as if ∆T is an independent variable, C and C_0 are independent, A is independent, and L is dependent. If it comes down to ∆T being the discrepancy, that does not affect your models or C values and thus L varies with ∆T (or, ∆T varies with L, it does not matter in this case). A better application of the model would be to correct the equation to have the currently accepted A value, and thus tweaking L to match the 2.5˚C ECS they derived. Now, since you have determined the dependent variable, you can properly flip the model around so that L is the independent variable and you can alter ∆T by switching L a bit. In other words, tweak the sensitivity factor until the model better fits temperature observations. I guess you would have to take into account TCS/ECS discrepancy, what ratio do they use in SAR? Assuming ECS is 50% larger than TCS I get a final ECS of 3.6˚C.
Edit: My figure is almost certainly wrong as I used a 1990 temperature baseline (as opposed to a baseline of a system in better equilibrium, like pre-industrial temps) and didn't take into account previous forcing that was still pushing temperatures to a higher equilibrium (i.e. all the previous CO2 increase). It will be lower. | |

2011-08-30 01:05:09 | Recommendation | |

John Hartz John Hartz john.hartz@hotmail... 98.122.98.161 |
Add "Related SkS Articles" and "Suggested Reading" tabs. | |

2011-08-30 03:53:31 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 64.129.227.4 |
Alex, my thought is that the sensitivity is inherent to the climate model. The sensitivity being "a" in ∆T = a*∆F. Thus if they got ∆F wrong, it means they got ∆T for 2xCO2 wrong by the same factor. The model sensitivity "a" (in °C per W/m2) is the same, it's just that they used the wrong W/m2 for 2xCO2. So they said it was 2.5°C/4.37 W/m2, which is equivalent to 2.12°C/3.7 W/m2. Do you think that's not right? | |

2011-08-30 06:16:49 | ||

Alex C coultera@umich... 67.149.101.148 |
The equation can act as either a deductive model from which you can determine sensitivity, or a predictive model from which you can determine (well, predict) a temperature response. You must first have the scaling component though between the forcing and temperature response, which is either the L in my equation or the "a" in yours, to use it as a predictive model. If there is a problem with ∆F, that is a problem with the radiative transfer models, and thus there will be an inherent problem with the deduced climate sensitivity. This translates into a problem with L, not a problem with ∆T. If they said that climate sensitivity was 2.5˚C per doubling of CO2, that is derived from past observations of temperature v. CO2 levels, and also a physical estimate (the radiative transfer models) of how concentration relates to forcing. So, if you based your deduction off of a proxy record that shows, say, a degree of temperature change as a result of an increase in CO2 from 200ppmV to 250ppmV, you apply those figures with your RTM coefficient to figure out the scaling factor L (the sensitivity factor) between temperature and forcing. "a," in your equation again. Once you know L/a, you figure out a standard comparison, which is typically a doubling of CO2, and get what the temperature response would be (using a 2xCO2 figure, your coefficient, and your L/a). This is the 2.5˚C figure they get. If there is a problem with A, then the step affected by that is not the figuring of your standard comparison (2.5˚C/2xCO2), but the figuring of L from your proxy observation. Thus, L must change. That is when you can use the model as a predictor. >>>So they said it was 2.5°C/4.37 W/m2, which is equivalent to 2.12°C/3.7 W/m2. ECS is temperature responding to a forcing, usually expressed as an easy standard like 2xCO2. Work backwards from what you just did, if you increase the forcing from 3.7W/m^2 to 4.37W/m^2 you will get 2.5˚C of warming, the exact same as the other scenario. These climates can't possibly have different sensitivities! The scaling factors are the exact same, and the scaling factors are the sensitivity determiners. | |

2011-08-30 06:55:11 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 64.129.227.4 |
Sensitivity is an output of, not an input into a model. And 2.5°C was just their "best estimate" - they also ran models with 1.5°C and 4.5°C sensitivities.
Well I'm not saying it has a different sensitivity to 4.37 W/m2, I'm saying it has a different sensitivity to 2xCO2. That was my point - that the model has a built-in sensitivity to a given forcing, but when they cited the sensitivity to 2xCO2, they used the wrong forcing. So I maintained the same model sensitivity, but re-scaled the 2xCO2 response to 3.7 W/m2. I hear your point about the RTMs. They thought the forcing from doubled CO2 was 4.37 W/m2, but that's incorrect, and that's a fundamental basis of the model. But I'm still not convinced that my re-scaling approach doesn't adequately account for that. | |

2011-08-30 08:49:07 | ||

Alex C coultera@umich... 67.149.101.148 |
>>>But I'm still not convinced that my re-scaling approach doesn't adequately account for that. Hm, maybe I should try more equations, where L is the sensitivity factor lambda and A is the coefficient, as if the 5.35 in Myhre's ∆F = 5.35ln(C/Co): ∆T = L*A*ln(C/Co) ∆T / [A*ln(C/Co)] = L and so the sensitivity factor is determined, by looking at proxies for ∆T and C values, and the RTMs for A. You can modify the model by changing the values of certain variables and seeing how others should respond so that the equation remains true, but you can only adjust some as makes physical sense. For example, if you changed (C/Co) so that it is higher, you increase the denominator in that fraction and can compensate by either increasing ∆T, or by decreasing A, or decreasing L (in a strictly mathematical sense). It's ridiculous to modify L though, afterall future projections of temperature rely upon a constant climate sensitivity. For similar reasons you cannot adjust A, as A's value was determined independently of what the C values were when you were running the RTMs; it is always 5.35 in Myhre's equation, for instance. No, you can only adjust temperature, which again makes sense: add more CO2, get more heat. So, that's a precedence that only certain variables can be adjusted. It's actually a very pertinent precedent, as ∆T and (C/Co) are related through a ratio: ∆T/ln(C/Co) = L*A This ratio of T to C is constricted to a range by observational evidence, the proxy records (or, mini natural experiments like volcanoes, so on), and a best estimate within that range is given. You musn't change ∆T or C when you change L or A as the ratio between the former two must still agree with the best estimate from observations. L must change in response to A, and vice versa (though vice versa won't happen in real life, as again from the first derivation above, L must be determined experimentally based on A in the context of the observations).
I don't know a better way to explain this, I hope someone can give a third voice in this matter if this is still a point of contention. | |

2011-08-30 12:46:23 | ||

dana1981 Dana Nuccitelli dana1981@yahoo... 69.230.106.125 |
I think we've got a different perspective on how models work. And I'm pretty darn sure my adjustment is correct. Let me try to explain why I think so more clearly. What I want to know is the climate sensitivity parameter of the model. My "a", your "L", climate scientists' "alpha". And that's built into the model. Paleoclimate data shouldn't come into the discussion, except that model sensitivity should fall within the range of paleoclimate sensitivity estimates, which fortunately is the case. But that's not really relevant here. So we know a = dT/dF. Forget 2xCO2 - the SAR "best estimate" model sensitivity is 2.5°C/4.37 W/m2 = 0.57°C per W/m2. So if we then want to know what the model sensitivity is to doubled CO2, it's 0.57*3.7 = 2.12°C. |