I was at the NCAS conference today (since it was in Cambridge it would have been impolite not to go). Tim "Da Man" Palmer spoke about, ermm, sort of a merge of NWP and climate scales. But thats not the point... the point is that he showed a stratification of the Staniforth CP.net PDF in terms of modifications to the entrainment scheme. Increasing entraiment, or leaving it unchanged, produced the low values (up to 6). All the high sensitivities (6-12) came from experiments in which the convective entraiment was reduced (also, although he didn't mention this, all the high sensitivies had a distinctive level of difference from the std model climatology. Not expecially large, but remarkably constant).
"Reduced" wasn't defined, but TP then went on to show some results from the ECMWF model in which it was reduced by 4/5. And what that showed was that when you did that within the assimilation framework, the reduced entraiment was completely out of balance. Ie, within that framework, unrealistic. TP immeadiately said that he wasn't saying that the CP.net stuff was wrong; since it was a different model. But all the same...
[As usual, I've sprinkled a few tech terms around to see if you're awake. NWP=numerical weather prediction. CP.net=climate prediction.net. PDF=probability density function. Entrainment: relates to the parameterisation of convection: how much air is entrained into the upwelling plumes (at least I'm assuming thats what they men by it) -W]
Could you translate that into terms a layman can understand? ("entrainment"? "assimilation framework?" Not to mention the various acronyms.) Thanks.
Not sure if this is relevant but
Ben Sanderson has worked out where the realistic values of the entrainment parameter should be
[Very interesting. Would be nice to see as a paper, with accompayning words. The last figure would warm James heart... -W]
I was aware of a few parameters being highly correlated with hot runs going back to June 2004.
That Tim Palmer work was presented (by someone else) at the workshop I went to last summer.
The CPDN authors were well aware at the time of initial publication that all of the runs with sensitivity > 6.5C were the ones with the entrainment factor set to its lower extremum. They did, to be fair, almost mention this in the Nature paper if you look carefully enough...
[Ah, I didn't know it was "out". Certainly not very out... I don't have the Nature paper to hand: what is the "almost mention" text? -W]
the canonical text explaining climate sensitivity is here:
Climate sensitivity (click)
Well this is a nice coincidence. I've been going through my old AMC notes from 12 years ago, and they tell me that the UM (and thus the GM) uses/used a different method of assimilation to ECMWF. I realise that the GM has had at least one major re-write since then but how much do the different assimilation techniques affect something along these lines?
I've not re-read the bit on entrainment yet, but will do in the next couple of days.
[I don't think the assimilation method matters too much. TP was saying that, if you believe the assimilation is reasonable, it rules out these low entrainment values - W]
The relevant text is "If perturbations to another parameter (the entrainment coefficient) are omitted, the blue histogram in Fig 2a is obtained, with no model versions > 8K." (cut and paste not working, so any typos are my own) which would I think be naturally interpreted to imply that sensitivities of over 7.5K (rather than all being lower than 6.5) are obtained with the standard value, and also that both high and low values can give a high sensitivity (when it fact it's only the low values that do so). If you look carefully at the histogram, it is certainly consistent with an upper bound of 6.5K but of course the limited graphical resolution makes it impossible to be sure.
I think the "low" value they used was 20% of the standard value, which is presumably where the Palmer 4/5ths reduction comes from.
William, your writing leaves a lot to be desired. Could you define acronyms please? Could you write more clearly please?
James: yes the values used were 0.6 3 and 9 so 0.6 is 20% of 3.
>Entrainment: relates to the parameterisation of convection: how much air is entrained into the upwelling plumes (at least I'm assuming thats what they men by it)
Yes the video I linked above does give an explanation of entrainment and to try to simplify it is about how much air from outside a rising plume mixes itself into the plume. The video gives some explanation of the consequential effects.
> All the high sensitivities (6-12) came from experiments in which the convective entraiment was reduced
> which would I think be naturally interpreted to imply that sensitivities of over 7.5K (rather than all being lower than 6.5) are obtained with the standard value, and also that both high and low values can give a high sensitivity (when it fact it's only the low values that do so).
So James has basically repeated William's assertion. It does take long of looking through the thread I linked before James's post to find
And guess what? It has the high value for the entrainment co-efficient. I don't have the numbers to carry out the exponential extrapolation that CPDN have found produces little error but I estimate the sensitivity to be around 7C (20.25-13.25).
So do you now think it is appropriate to withdraw the accusation?
The other thing is the implication that the CPDN stuff is (or could be) wrong. I am inclined to take exception to this. Since you are using the same models isn't it only as wrong as your own work? Moreover, if there were no large ensembles like with CPDN would this sort of thing ever be found? If so, wouldn't the sensible reaction to be say that is why we need large ensembles and encourage people to join www.climateprediction.net (CPDN)?
It is hardly new for me to question the purpose of a post that casts aspersions for no good purpose (eg that article about wikipedia that just ranted about the politics).
The plot which illustrated that point was shown at a workshop at which at least one CPDN author was present: he agreed with the interpretation and confirmed that they had all been well aware of it at the time of publication - in fact he thought it was in the paper, which it just about was if you looked carefully. It is also in a paper that is currently in press (Rodwell and Palmer). There are clearly one or two experiments with sensitivities in the 6-6.5K range associated with each of high and standard entrainments, and no higher ones other than those with low entrainment. I guess it's not entirely out of the question that some plotted dots are hidden behind others but I would hope the authors have ensured that this is not the case. The histogram in the Nature paper should also have the resolution to show such a point if it really exists, and there is no hint of it there.
Maybe the plotted values were the average of the various perturbed IC runs (AIUI there were multiple runs done with the same parameters, weren't there?), and although you've found a single run with sensitivity marginally above that threshold, the others are lower and bring the average down. Or perhaps it was rejected for other reasons. These are just guesses at how to reconcile the situation. I agree that the one set of results you have pointed to look like a sensitivity of >6.5K (but still a long way short of 8K).
Maybe 3rd author Carl can explain further :-)
The Stainforth Nature paper was very careful to not claim that the graphs shown were PDFs. They are just histograms as they attached no probability to them.
PDFs using data from the same runs can be found in C. Piani, et al. 2006 http://www.climateprediction.net/science/pubs/Piani_GRL.pdf
[Interesting, thanks. Its not clear to me that 2.2-6.8 is consistent with the PDF on the X axis of fig 3 (they do says estimated CS) or even on the Y axis -W]
My understanding is that the high and low values for each parameter were chosen to be the extremes that experts considered to be physically posible. While the unchanged value is that found from previous Met Office runs of HadSM3 to give the best fit with past climate. There is no reason to limit the entraiment coefficient to just being unchanged or higher other than not liking the resulting climate sensitivity.
It would be interesting to see the result of more values of entrainment coefficient within the range previously used.
Finally I should point out that the Stainforth paper included variations of only a few of the parameters, combinations with more parameters being varied could be expected to produce even more extreme climate sensitivities.
>[Interesting, thanks. Its not clear to me that 2.2-6.8 is consistent with the PDF on the X axis of fig 3 (they do says estimated CS) or even on the Y axis -W]
My understanding (which could be completely wrong) is that the x and y axis shapes are histograms (not pdfs) of all the runs in the figure. There are quite a lot of runs with sensitivity less than 2.2 but figure 3 shows that these become rapidly poor models as the points move rapidly away from the 45% line. When you take the good runs which are close to the 45% line then you get a pdf with a 2.2 to 6.8 range.
Does that make sense?
[No. The figure caption clearly calls them PDFs, and the text is clearly using them to estimate CS: "The scatter plot, median and thin PDF are the same as Figure 2 but for climate sensitivity. The solid thick PDF is obtained from the thick PDF in Figure 2 by plotting it onto the inverse axis. The thick dashed line is obtained by using the median and the scatter about the median as a transfer function for the x-axis PDF. The shape of the median graphically explains why observationally constrained forecasts of CS are likely to be skewed to higher values." -W]
Err that is degree not % sorry.
(For James, I really meant pdf in the sense of what you can work out from this technique alone. There may be other techniques that might further refine the probabilities)