Bishop Hill Lewis on Walport and the HadGem model
Sep 5, 2013 Climate: Models Climate: sensitivity In response to yesterday's post about Mark Walport's slide on climate projections, Nic Lewis sends some thoughts about the underlying model - HadGEM2-ES - and its projections of warming to 2100. Click on graphs for full size.
The Walport chart shows warming projections to 2100 from HadGEM2-ES, the Met Office's flagship climate model, used for policy advice (http://www.metoffice.gov.uk/climate-change/policy-relevant/advance).
;)
My graphic shows projected global warming at 2100 under Walport's highest emissions scenario, RCP8.5, and does show 5-95% uncertainty ranges.
;)
The red bar in the CMIP5 column shows HadGEM-ES. The flasks show, between the 5th and 95th percentiles of uncertainty distributions, the estimated probability density for each source: blue for observation-based studies, using data ending in two different periods for Otto et al; purple for the Harris et al 2013 Met Office study corresponding to the official UK climate projections (UKCP09); salmon for climate models - CMIP3 (last generation: IPCC AR4) and CMIP5 (current generation). Black bars are 50th percentiles (medians). The warming to 2100 is from pre-industrial times, and should be little different from warming from circa 1900.
The warming to 2100 for HadGEM2-ES on the RCP8.5 scenario is about 5.5°C both in my graphic and the Walport chart. Note that we have already had a global temperature rise of circa 0.8°C since the second half of the 19th century, so there is less to come than appears from the graphic.
My estimates are arrived at in the same way as in the Met Office July report "The recent pause in global warming (3): What are the implications for projections of future warming?", Figure 2, but after correcting errors in their calculations. ;)
The short white bars show the original Met Office central estimates, all miscalculated. Additionally, their Report states that these are all medians, which is incorrect. They are in fact a mixture of means and medians. In two of the three cases where they used the mean it is about 10% higher than the median, which provides a better central estimate.
In my figure, HadGEM2-ES is a clear outlier. And the Harris et al 2013 probabilistic estimates, which should correspond to the UK official climate projections, have a probability weighting very much concentrated towards the top end of what warming observational studies indicate is likely. The Harris et al. 2013 estimate was based on perturbing parameters controlling variable characteristics of the previous generation Met Office climate model, HadCM3, and comparing the resulting simulations with observations of the recent mean climate and historical climate change. But its results appear mainly to reflect structural characteristics of HadCM3.
Statistical uncertainty in projected warming between different runs of the HadGEM2-ES, arising from simulated internal variability (climate noise), will be minor compared to the differences between the warming to 2100 per HadGEM2-ES and the observationally-based estimates.
Reader Comments (26)
So, how many Angels can dance on a Meteorological Office Pin? Is it 7 or is it 9?
The Public must know........
Nic -
Thanks for this. It is somewhat larger than I had expected -- in rough terms, RCP8.5 results in two CO2-doublings, and at the Ottoe et al. median TCS I would have estimated around 2.5 K increase from baseline. I presume the difference is due to the tail of the impulse-response -- can you provide details of this? [Edit: just did the math, the "roughly two doublings" is really more like 2.2, which raises the naive estimate (TCS only) to around 2.8 K, fairly close to your figure. I'm still curious about the time-response curve you used, though.]
While it isn't fair to "assign homework", I'm wondering what the year 2100 response would be to the RCP6.0 scenario, if you have that result to hand.
Shouldn't the projections actually start from the end of the observations line?
HaroldW
Thanks for your comments:
"In rough terms, RCP8.5 results in two CO2-doublings, and at the Ottoe et al. median TCS I would have estimated around 2.5 K increase from baseline."
Per the RCP forcings database, RCP8.5 gives a forcing of 8.25 W/m2 at 2100, taking a 11 year running mean to even out the RCP forcings' solar cycle. Using the canonical estimate of 3.71 W/m2 forcing for a doubling of CO2, that equates to 2.22 CO2 doublings. Multiplying by the Otto et al 2000-09 TCR estimate of just over 1.3 K gives ~2.95 K, using the Met Office's method of calculating warming to 2100. That is consistent with the black line with my graphic.
" I'm wondering what the year 2100 response would be to the RCP6.0 scenario"
On the same basis, forcing at 2100 under RCP6.0 is 5.43 W/m2, implying a temperature rise by then of 1.9 K or so - of which ~0.8 K has already occurred.
David Schofield
"Shouldn't the projections actually start from the end of the observations line?"
I agree that would seem logical, but it is not what the Met Office did in their Report. For the observationally-based estimates, it doesn't make much difference, at least to the central estimates. That is because they give an estimated median warming to date (using the same method of calculation) of circa 0.8°C, the same rise as the global temperature record indicates has occurred, referred to in the post.
For the other three estimates the median projected warming to 2100 should really be reduced to reflect their overestimates of warming to date. Moreover, the uncertainty bands for all the projected warming estimates should be reduced in recognition that we know what warming resulted from the forcing increase to date arising from greenhouse gases etc - which amounts to 25-30% of the total forcing increase to 2100 even under the RCP8.5 scenario. But my graphic was simply intended to correct the obviously invalid calculations in the Met Office report Figure 2, not to change its whole basis to a more realistic one.
Nic -
Thanks. I didn't realize that this was only based on TCR, but thought there was some temperature impulse-response function involved.
So, I only needed to have have retained greater precision in the calculation. Plus looked up the Otto et al. TCR value rather than relying on memory.
I note that you used "the canonical estimate of 3.71 W/m2 forcing for a doubling of C" in your computation above. However, the Otto et al. main TCR estimate was based on using 3.44 W/m2/doubling (from Forster et al.). Table S2 line E indicates that 1.4 K is the TCR calculated using a 3.7 W/m2 per doubling. So perhaps the 2.95K should be more like 3.18K? [The latter figure being 2.4 doublings x 1.323 K/doubling.]
Nic:
Scary. As in not at all. Thanks as always for the clarity of your presentation.
I don't think you can have a better example of scientism than this, whoever - believer or sceptic - does the math ...
can any of them post me, or mail me, the equation, delta T = f ( co2 ) ?
thanks beforehand ...
"I didn't realize that this was only based on TCR"
Yes, the Met Office used this simplistic method, which I followed but with correct calculations.
"I note that you used "the canonical estimate of 3.71 W/m2 forcing for a doubling of C" in your computation above. However, the Otto et al. main TCR estimate was based on using 3.44 W/m2/doubling (from Forster et al.). Table S2 line E indicates that 1.4 K is the TCR calculated using a 3.7 W/m2 per doubling. So perhaps the 2.95K should be more like 3.18K? [The latter figure being 2.4 doublings x 1.323 K/doubling.]"
Good question. But actually, no.
TCR has been derived from estimates of the change in radiative forcing (ΔF), the corresponding change in global surface temperature (ΔT) and the forcing from a doubling of CO2 concentration (F_2x):
TCR = ΔT x F_2x / ΔF
The largest component by far of ΔF is forcing from increased CO2 concentrations. So in order for the calculation to be consistent it must use the same estimate for F_2x as is used for the numerator when deriving ΔF. Otto et al's main estimates did so: they used a CMIP5 multimodel mean estimate both for ΔF and for F_2x. I don't know why that CMIP5 mean estimate for F_2x was 3.44 W/m2 not 3.71 W/m2, but since the lower value was also reflected in ΔF it doesn't make much difference.
(The TCR and ECS figures in line E of Table S2 are on a basis that doesn't really make physical sense: if I recall correctly the main point being illustrated was the effect of a wider uncertainty band for F_2x, but probably someone asked for the effect also of a different F_2x to be illustrated.)
The Met Office actually went wrong on exactly the same F_2x -consistency point in their Report, but in another calculation. One would have expected them to get basic climate science calculations right in a published report. (To be fair, Met Office climate scientists I have dealt with seem to be generally highly competent, if perhaps sometimes under-critical towards climate models.)
The Met Office calculation deriving Warming to 2100 (ΔT_2100) for Figure 2 from TCR estimates was:
ΔT_2100 = TCR x ΔF_2100 / F_2x
where ΔF_2100 is forcing at 2100 (they used nominal forcing per the RCP8.5 emissions scenario).
For this to be acceptable, the value used for F_2x in the denominator must be the same as the value of F_2x used in the RCP forcings dataset to convert CO2 concentrations to forcing, being 3.71 W/m2. But the Met Office used a F_2x divisor of 3.44 W/m2. An embarrassing mistake to make – I doubt it was a deliberate attempt to show higher projected warming.
Sorry for a rather lengthy response!
Not only should their graph match observation, but their water vapour feedback assumption, whatever it is, should be compared to actuals. Any model which fails to match should be removed from consideration.
(Oh, and of course CS is an unsupported concept and this whole exercise is angel-dancing nonsense, but I don't want to dwell on it.)
Thanks Nic. I recalled being surprised at the Otto et al. F_2x, which is why it stuck in my mind. Your explanation is clear.
The graphic, although correct, is misleading because it states nowhere that the temperature increase is from (I think) the pre-industrial level rather than the present-day temperature. I think that information should be added into the graphic in an integrated fashion (not just a caption) to prevent further misinterpretation. Or perhaps better, redo the y-axis to show the increase from the present day. (i.e., subtract 0.8 C.)
Also, Nic Lewis states that his graphic shows the same 5.5-degree increase for the HADGEM-ES model as the Walport graph, but I read the median value as closer to about 4.4.
"Uncertainty of predictions", says the Government Office for Science chart. This is all based upon GCM methodology that has been falsified by nature, incorrect radiative physics (see the Trenberth discussion thread), ignorance about conductive, convective and phase change effects and with zero regard to the laws of non-equilibrium thermodynamics. How BH regulars can pontificate about "TCR" beggars belief...
My prediction for a doubling or trebling of atmospheric CO2? Deserts would get very slightly cooler by day and warmer by night. The effect on average surface temperature zero. The effect on the rest of the planet SFA. But it would move the planet closer to its biological optimum, and the plants would love it!
My prediction if the sunspot count continues to diminish - we're going to get bloody cold winters! The poor and the old will starve due to sky high fuel and food prices, and the fraudsters will be laughing all the way to the bank.
This is exactly why Met Office science advice to inform policy is based on multiple models (e.g.: CMIP3, CMIP5) and not just our own - when uncertainties are so large, it's unwise to rely on a single model or single projection.
Indeed, this is the reason the Met Office was involved in setting up the Coupled Model Intercomparison Project (CMIP) back in the 1990s, so as to have a systematic way of comparing model projections with each other and with observations.
Lance Wallace
" The graphic, although correct, is misleading because it states nowhere that the temperature increase is from (I think) the pre-industrial level rather than the present-day temperature."
As you note, that fact is stated in the caption, just below the graph, so there is no question of readers being misled. It is common practice to put such information in the caption. If you look at the Met Office graph, which mine is a corrected, more informative version of, you will see that I have (deliberately) used exactly the same y-axis label as they did. Showing warming from pre-industrial times, as the Met Office did, is also pretty standard practice in climate science, although I agree that it would be helpful also to show warming from the present day as well.
" Also, Nic Lewis states that his graphic shows the same 5.5-degree increase for the HADGEM-ES model as the Walport graph, but I read the median value as closer to about 4.4."
The red bar represents HadGEM2-ES warming (not HadGEM1-ES - a typo). It looks to me pretty clearly to show a rise of between 5.5 and 5.6°C, not 4.4°C. And it doesn't represent a median - it is just for HadGEM2-ES. You may be confusing the red bar with the white bar, which shows the median warming for CMIP5 models as (mis)calculated by the Met Office at 4.5°C. The correct value for CMIP5 of 4.0°C is shown by the black bar.
Hi Richard
Thanks for your comment. A few observations on what you say:
" This is exactly why Met Office science advice to inform policy is based on multiple models"
Maybe so, although that isn't exactly the message one gets from the Met Office's statement about HadGEM2-ES: "The new model is being used to provide a more comprehensive analysis, building on previous work, to address science questions that affect policy."
In any event, almost all the complex climate models exhibit substantially higher climate sensitivity (ECS and TCR) than that which seems to be indicated by the latest observational evidence. May I ask if the Met Office has ever considered basing science advice on what the observational evidence implies rather than models, or on models adjusted to conform to the latest observational evidence, and if not then why not?
"Indeed, this is the reason the Met Office was involved in setting up the Coupled Model Intercomparison Project (CMIP) back in the 1990s, so as to have a systematic way of comparing model projections with each other and with observations."
I certainly see the sense in doing so, and I have no doubt that the Met Office has performed valuable work in this regard. However, there seems to me not to be enough emphasis on the comparing model projections with observations part.
I don't imagine that, as a Met Office scientist, you can be very happy that your senior management has published a report (the July report referenced in my post) that is replete with misrepresentations (of Otto et al 2013), misleading statements and elementary errors. (I have only mentioned a few of them in this post.) Do you know if it is planned to publish a corrigendum?
Nic Lewis - Sep 5, 2013 at 10:03 PM
I wonder who, if anybody, will be authorised to answer? Doubt that it will happen at 10.15 am on Wednesday 11 September 2013 Thatcher Room, Portcullis House
Edit, double post
I would like to ask Richard Betts to explain the difference between a "projection" and a prediction.
I would also like to ask your assessment of the difference between recent MO projections, predictions etc., and observations.
It was interesting to view the other replies, all very erudite but there's one flaw. If you look at the problem from basic physics, including the irreversible thermodynamics completely missed by the IPCC and the failure to understand that there can be no thermalisation of IR from a higher temperature source in the atmosphere at local thermodynamic equilibrium, there can be no significant CO2-AGW!
This is because it's the working fluid of the heat engine that keeps SW IN = LW OUT whilst minimising radiation entropy production rate.A group at Brookhaven is working on this as well. The models have to be changed considerably.
Nic Lewis to Richard Betts:
By his forthright answer, and corresponding efforts to bring the Met back from the brink behind the scenes, Dr Betts will we trust be judged by history as not in the same bracket as those most responsible for this disaster for UK science. Our Andrei Sakharov, not our Yuri Zhdanov.
Yes - see, for example, the paper The upper end of climate model temperature projections is inconsistent with past warming by Peter Stott et al.
The key results from this have been presented directly to Ed Davey and DECC.
Can't stick around to chat today, I'm afraid - got to finish my IPCC text for Working Group 2.
Hi again Nic
Actually I meant to add earlier - please can you elaborate on your statement about the July report:
ie: what are your other concerns beyond those mentioned in your post?
If you can explain further here, I'll check the thread again later on.
Thanks!
Richard
Hi Richard
Thanks for your comments.
"The key results from this [Stott et al] have been presented directly to Ed Davey and DECC."
I'm glad to hear it! But the message from that paper about RCP4.5 and RCP8.5 projections by HadGEM2-ES being at or above the 95% percentile of projections constrained by observed past warming, on both bases considered, and so extremely unlikely to be valid, doesn't seem to have reached Mark Walport - or maybe he doesn't care what the observational evidence is implying.
Does the Met Office think it is right to continue to sanction government use of HadGEM2-ES warming projections when it knows, from its own scientists' study, that they are almost certainly unrealistically high?
" what are your other concerns beyond those mentioned in your post"
I'll make details of all the things that concern me about the July report available to you, and others who are interested, as soon as I've finalised my write up.
Richard Betts
Apologies for I realise that you are under pressure at the moment, but the thread includes and debates current goverment forecasts for global climatic projections through 2100. As you are well aware, many on this website are highly curious about the impacts of natural climate variability and less concerned about putative anthropogenic influences. As a geologist, I am more curious about our climatic destiny on slightly longer time scales. I would like to know when our currrent climate research expertise anticipates the onset of the next full glacial advance, and its impacts. Please park that request for response at your convenience, please.
I looked at the Stott et al paper referenced by Richard Betts. In section 2.2, we find:
"... forcing for one climate model may be estimated using the forcing derived from another, using a linear scaling factor."
The absurdity of this methodology is breathtaking.