Guest post by Bob Tisdale
INTRODUCTION
Chapter 4-7 of If the IPCC was Selling Manmade Global Warming as a Product, Would the FTC Stop their deceptive Ads? included comparisons of the CRUTEM3 Land Surface Temperature anomalies to the multi-model mean of the CMIP3 climate models. For those who have purchased the book, see page 99. As you will recall, CMIP3 stands for Phase 3 of the Coupled Model Intercomparison Project, and CMIP3 served as the source of the climate models used by the Intergovernmental Panel on Climate Change (IPCC) for their 4th Assessment Report (AR4). CRUTEM3 is the land surface temperature data available from the Met Office Hadley Centre and the Climatic Research Unit at the University of East Anglia.
This post compares the new and improved CRUTEM4 land surface temperature anomaly data to the same CMIP3 multi-model mean. CRUTEM4 data was documented by the 2012 Jones et al paper Hemispheric and large-scale land surface air temperature variations: An extensive revision and an update to 2010. I’ve used the annual time-series data, specifically the data in the second column here, changing the base years for anomalies to 1901 to 1950 to be consistent with Figure 9.5 of the IPCC’s AR4.
And, as I had with the other 20th Century Model-Observations comparisons, the two datasets are broken down into the 4 periods that are acknowledged by the IPCC in AR4. These include the early “flat temperature” period from 1901 to 1917, the early warming period from 1917 to 1938, the mid-20th Century ‘flat temperature” period from 1938 to 1976, and the late warming period. The late warming period in the chapter 4-7 comparisons in If the IPCC was Selling Manmade Global Warming as a Product, Would the FTC Stop their deceptive Ads? ended in 2000. For the late warming period comparisons in this post, I’ve extended the model and CRUTEM4 data to 2010.
COMPARISONS
As shown in Figure 1, and as one would expect, the models do a good job of simulating the rate at which the CRUTEM4-based global land surface temperatures rose during the late warming period of 1976 to 2010.
Figure 1
But like CRUTEM3 data, that’s the only period when the IPCC’s climate models came close to matching the CRUTEM4-based observed linear trends.
According to the CMIP3 multi-model mean, land surface temperatures should have warmed at a rate of 0.043 deg C per decade from 1938 to 1976, but according to the CRUTEM4 data, global land surface temperature anomalies cooled at a rate of -0.05 deg C per decade, as shown in Figure 2.
Figure 2
Figure 3 compares the models to the global CRUTEM4 data during the early warming period of 1917 to 1938. The observed rate at which global land surface temperatures warmed is almost 5 times faster than simulated by the IPCC’s climate models. 5 times faster.
Figure 3
And in Figure 4, the models are shown to be unable to simulate the very slow rate at which land surface temperatures warmed during the early “flat temperature” period.
Figure 4
According to the models, the linear trend of the global land surface temperatures during the late warming period should be 6.6 times higher than during the early warming period. See Figure 5.
Figure 5
Yet according to the new and improved CRUTEM4 land surface temperature data, Figure 6, the land surface temperatures warmed during the late warming period at a rate that was only 40% higher than during the early warming period.
Figure 6
CLOSING
The models show no skill at being able to simulate the rates at which global land surface temperatures warmed and cooled over the period of 1901 to 2010. Why should we have any confidence in their being able to project global land surface temperatures into the future?
ABOUT: Bob Tisdale – Climate Observations
ebook (pdf and Kindle formats): If the IPCC was Selling Manmade Global Warming as a Product, Would the FTC Stop their deceptive Ads?
SOURCES
The CMIP3 multi-model mean data is available through the KNMI Climate Explorer
http://climexp.knmi.nl/selectfield_co2.cgi?id=someone@somewhere
And as noted in the post, the annual CRUTEM4 data is available through the Met Office website:
http://www.metoffice.gov.uk/hadobs/crutem4/data/diagnostics/global/nh+sh/index.html
Specifically:
http://www.metoffice.gov.uk/hadobs/crutem4/data/diagnostics/global/nh+sh/global_n+s
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Thanks, Anthony.
How does this CMIP3 performance using CRUTEM4 data compare to CMIP3 performance using CRUTEM3 data? Any indications that the changes from CRUTEM3 to CRUTEM4 have been hand-adjusted to improve CMIP3 backtesting (even if it is still pretty bad)?
Presumably somebody will answer that rather than the ensemble we could just pick the “best model” for each subperiod. Also they’ll point out on multidecadal scales the differences don’t appear to be much.
Anyway…if you could also show Version 3 alongside version 4, things will look very very suspicious especially at the right end…
I am with Alec on this one.
Are the adjustments making the the data fit the models better for the last 36 years?
Because that would really be a coincidence wouldn’t it?
Thanks, excellent article Bob!
I’m hoping WoodForTrees will soon have the new data base.
Models are laboriously tuned to match the latest warm AMO period. Period. They are not able to model even the 20th century, since they a) deny any natural variability in the best tradition of hockey-stick BS, b) they are based on unphysical background, or at least many times exaggerated radiative effect which has never been directly measured.
Once again; the warm AMO period 1910-1945 is INDISTINGUISHABLE from the warm AMO period 1975-2005. See how the models are absolutely unable to model the North Atlantic natural variability.
http://oi56.tinypic.com/wa6mia.jpg
The only period they match has already ended. North Atlantic SST peaked in 2006 and now it is heading down for another 30 years, exactly as it did in 1945-1975, dragging the whole global record with it. Models are alchemy, AGW is nonsense which can not withstand full information: deal with it.
The defense is that the differences between observation and models are due to “natural” variations that have higher amplitudes but shorter frequencies than expected from modelling. It is not an unreasonable defense. Not that I am a warmist, but I can justify the claim that, so far, observations are enough of a match that their position is not falsified. (The adjustment bias is the only reason that they both have a case for CAGW that holds some water, it is noted. We need the adjustments to stop!)
The important difference for me is the immediate excess warming the models show relative to observation. The models look like they need to be tamped down, which of course drops the projected 2100 outcome to the bottom of the CAGW narrative, and into the solar/natural regime at the same time, UNLESS there is a near-future, sudden climatic response to CO2.
I’m happy that the models are off right now, on the serious warming side. The time when the difference is too great to arm-wave away comes quicker the more potent the warmists think CO2 is. Although they may claim that the future holds sudden increases, their models don’t provide for it right now. They can’t have it settled and certain unless, very soon, there is an uptick in both global temperatures and sea-level.
The newspapers today are touting the global warming winter of today. In North America. I wonder what the European-by-themselves stats are? Of course the biggest believers and most dangerous politically are in the States, so Europe, being in a different universe, doesn’t count.
Did you MASK the output of the models to match the spatial coverage of the observations?
For example: in the early part of the record the entire globe is NOT sampled. So the average has some spatial bias in it. If you used the complete model output to come of with the average for this period, then you are doing the comparison wrong. First, you have to extract the MONTHLY observation mask from the observation dataset. That will tell you which GRIDS there are samples for. Since CRU do NOT interpolate or extrapolate U must extract this data.
Then you have to take the monthly grids from all the models. You have to decimate those grids to match the observed grids.
If you do that you have a valid comparison. This is the method used in attribution studies and one which must be used in evaluating model fit to observations.
Ok, so when the data shows we have entered a cooling cycle we need to fix the data to show even more warmth so it can match the models that cannot even back predict over the last 100 years. Got it. Why is this farce/fraud not shouted from the highest rooftops……?
Why dont you put in the 3 horizontal lines (#1. before 1920, #2. 1940-80, #3. 2000 continueing
the present temp. plateau ? These mark the stepwise temp increase with the tipping
points 1822-1881-1941-2000, when temp plateaus always set in according to the
Scafetta 60-year cycle…..
Please put the lines in and the graph will be highly enhanced in heuristic value….
JS
The new Crutemp4 increase over Crutemp3 on a monthly basis. Not so much difference except for the early 1850s and then the last 15 years which rises to +0.1C or so.
http://img28.imageshack.us/img28/3845/crutemp4vscrutemp31851t.png
HadSST3, of course, has more of a change compared to HadSST2 – particularly in the 1883 to 1886 period (cooling this period), 1936 to 1944 period (cooling it) and the 1945 to 1971 period (warming it). It looks a little strange right now so I will wait till they release the final.
Mr. Mosher, as much as I appreciate your devotion to detail I must confess the explanations are appearing more and more to be long reaches to justify the unjustifiable. I am weary of the constant adjustment to exsisting data in order to make it fit the false models. The real observations do not back the modeling, but when that occurs we hear weather isn’t climate unless of course the weather might seem to portray AGW then it is acceptable data.
Did they really think that no-one would notice that they had their thumb on the scales?
here you can see the CMIP5
crutem4 is adjusted for it 🙂
http://www.climate-lab-book.ac.uk/2012/on-comparing-models-and-observations/
Steven Mosher says:
March 20, 2012 at 11:37 am
Did you MASK the output of the models to match the spatial coverage of the observations?
For example: in the early part of the record the entire globe is NOT sampled. So the average has some spatial bias in it. If you used the complete model output to come of with the average for this period, then you are doing the comparison wrong. First, you have to extract the MONTHLY observation mask from the observation dataset. That will tell you which GRIDS there are samples for. Since CRU do NOT interpolate or extrapolate U must extract this data.
Then you have to take the monthly grids from all the models. You have to decimate those grids to match the observed grids.
Alternatively we cold just admit that prior to 30-40 years ago we dont know what global temperatures were.
The new ‘meme’ is that only 30-year trends can detect climate changes. On my web site I’ve plotted 30-year trends for the old CRU temperature, GISS, NCDC and a 23-model ensemble. They show that with all three data sets the models failed to simulated the rate of change in temperature.
http://www.climatedata.info/Discussions/Discussions/opinions.php
Mosher’s comment contains a truth I wish would be observed more often when discussing climate model purporting to address “global” warming. Truth is that we have observed global temperatures only for about 30 years and no amount of adjustments, interpolations, or corrections can make the current kludged together data be physically meaningful on a global scale. No amount of matching models to data for the last 30 years can give a meaningful confidence that the model will match data for the next 30 (every time we get a mismatch a new physical factor is cited to be included such as dust particles or clouds). No amount of corrections can give you an absolute temperature that a sensor failed to measure correctly nor can any massage of proxy data give an accurate monthly temperature measurement from before thermometers were invented, nor can we ever recover the spatial distributions of temperatures in the southern hemisphere that we didn’t gather early in the 20th century.
The real truth is that some climate scientists are pretending to certainty in a way that constitutes scientific fraud. We don’t know that the model has failed in rates in the early 20th century because we don’t know what the early 20th global measurement are. But to pretend that we can use a model to predict the future is a great fraud because we can’t test it over a well documented period to have confidence that it accounts for the significant climate factors.
Steven Mosher says: “Did you MASK the output of the models to match the spatial coverage of the observations?”
We’ve been through this before, Steven. I’ve shown you that there was little to no difference between a replication of the IPCC’s Figure 9.5, cell a, with and without the spatial adjustments. Don’t you recall? I could find my reply to you from 4 to 6 months ago, if you’d like. It’s like comparing infilled GISS and NCDC data to HADCRUT data, which, of course, hasn’t been infilled. The differences are so small they have no impact on simple comparions like this.
Steve Mosher says…
Steve: I enjoy your posts, primarily because you’re a straight shooter and also because I learn from them (although the details can be slightly over my head). I think most of us agree with Juaj V. who says “Models are laboriously tuned to match the latest warm AMO period. Period.” and johnmcguire who says “I am weary of the constant adjustment to exsisting data in order to make it fit the false models.”
What’s your opinion on the constant ‘tuning’ of the models & ‘adjusting’ of the data until they finally match?
Frankly, eyeballing the two seems to give quite a good correlation … a faked one of course. Most importantly Hadcrut4 does away with Jonesy’s most embarrassing admission to parliament that there’s been no warming for the past 15 years.
Is data accessible enough for some climate realists to create a new version ?
THe question remains.
Were the models tuned to match CRUTEM4, or was CRUTEM4 tuned to match the models.
I wonder how the graphs would look if the horizontal component were stretched out so that the increment was one year per vertical bar. I’d bet the graph would look very dissimilar.
Alec Rawls says: “How does this CMIP3 performance using CRUTEM4 data compare to CMIP3 performance using CRUTEM3 data? Any indications that the changes from CRUTEM3 to CRUTEM4 have been hand-adjusted to improve CMIP3 backtesting (even if it is still pretty bad)?”
And A. C. Osborn says: “I am with Alec on this one.
Are the adjustments making the the data fit the models better for the last 36 years?
Because that would really be a coincidence wouldn’t it?”
I had a funny feeling people would ask. The differences between the models and observations improved (lessened) with CRUTEM4 during the “flat temperature” periods, but worsened (increased) during the warming periods, especially during the early warming period.
http://i40.tinypic.com/k1fkeg.jpg
Ron Manley says:
March 20, 2012 at 12:31 pm
On my web site I’ve plotted 30-year trends for the old CRU temperature, GISS, NCDC and a 23-model ensemble. They show that with all three data sets the models failed to simulated the rate of change in temperature.
_____________________________
I like your analysis, it does not rely on noise in source data like the original article does. The difference between models and reality is not big – 1C/century for a few years is really very little – but it indicates there might be some real-world periodicity missing in models. That might be even more important than minute differences in trends. It’s just hard to say if it’s not just an artefact of the method combined with some shorter period. It kinda reminds me of Scafetta’s work.
Note that CRUTEM4 seems to end at the peak of the 2010 El Nino, so the match looks better than it actually is. If a few years of extra data are added, the divergence becomes more apparent. Obviously this won’t be an issue if temperatures shoot up again shortly. If that doesn’t happen, it will be impossible to not notice the divergence. The bottom line is that even with these adjustments always occurring in the same direction (increased warming) it is still a relatively trivial amount when compared to what models project should be happening. Interesting times.