On occasion, comments posted on WUWT are backed up with data or graphs from the commenter, and are so germane that they merit their own post for discussion. This is one of those cases. Bill Illis has done a couple of guest posts on WUWT, the most recent about the “Trade Winds Drive the ENSO“. In the comments about the story on “When you can’t believe the model” he posted a significant comment on his work with NASA GISS model E (Global Climate Model) backed up with his own research graphs. For those brave enough to slog through it, here is the manual for Model E. I thought Bill’s comments were worth sharing. – Anthony
Image above is from my stock imagery and for illustration only, not from Bill Illis.
Guest Comments by Bill Illis
Awhile ago, I pulled apart the components of GISS’s Model-E and then extended the forecast it would have provided from 2003 (the end date of the data provided by GISS) to 2013, ten years.
The model would be off by about 0.15C in the first five years.
The more detailed version of this extension is here:
Click for a larger image
The simpler version is below.
Click for a larger image
Another way to look at is they have huge GHG temperature impacts built in (no way to get to +3.0C without it) but they need to build in almost as big negative temperature impacts from other sources to keep the hindcast close to the actual temperatures we have seen so far.
One could conclude they are just plugging the big negative numbers into the hindcast after the fact to make it work.
Which is close to the point Leland Teschler was trying to make in this article. (seen here)
Click for a larger image
Without a large uptick in temperatures in the next few years, the modelers really have to go back to the drawing board (or they need to discover another “negative forcing” to keep the models on track to reality).
I will once more repeat here that the deviation of “projections” to reality are inevitable for all the GCM models.
These models grid the globe with an arbitrary grid, take solutions of nonlinear coupled differential equations and use the first order terms ( means) as an approximation to the true. They have numerous unknown parameters to hindcast and force their view to the model. What happens though after a number of time steps into the future is that the true solutions that are highly divergent, and cannot be approximated by the first order terms ( and actually not really calculable,) diverge from reality more and more as time goes on.
I think this is an inherent flaw in the logic of GCMs and modelers have to sit down with their thinking caps on ( if they have them). I believe the way Tsonis et al are approaching modeling is much better fitted to the chaotic problem of climate behavior. http://www.nosams.whoi.edu/PDFs/papers/tsonis-grl_newtheoryforclimateshifts.pdf
It is worth giving the abstract here:
A new dynamical mechanism for major climate shifts
Anastasios A. Tsonis,1 Kyle Swanson,1 and Sergey Kravtsov1
Received 5 April 2007; revised 16 May 2007; accepted 15 June 2007; published 12 July 2007.
[1] We construct a network of observed climate indices in
the period 1900–2000 and investigate their collective
behavior. The results indicate that this network
synchronized several times in this period. We find that in
those cases where the synchronous state was followed by a
steady increase in the coupling strength between the indices,
the synchronous state was destroyed, after which a new
climate state emerged. These shifts are associated with
significant changes in global temperature trend and in
ENSO variability. The latest such event is known as the
great climate shift of the 1970s. We also find the evidence
for such type of behavior in two climate simulations using a
state-of-the-art model. This is the first time that this
mechanism, which appears consistent with the theory of
synchronized chaos, is discovered in a physical system of
the size and complexity of the climate system.
Citation: Tsonis, A. A., K. Swanson, and S. Kravtsov (2007),
A new dynamical mechanism for major climate shifts, GeophysRes. Lett., 34, L13705, doi:10.1029/2007GL030288
They use “the Pacific Decadal Oscillation (PDO), the North
Atlantic Oscillation (NAO), the El Nin˜o/Southern Oscillation
(ENSO), and the North Pacific Oscillation (NPO)”
Thank you – It (and the attached manual) begin to address several fundemental questions on GCM data I brougt up a few nights ago.
It’s a fairy-story isn’t it!
The Chinese have increased their coal consumption from 1 to 3 bn tons in the last 10 years (a lot of coal trains..). (Global fossil fuel consumption is close to 10 bn tons of carbon per year). The Chinese coal is said to be rich in sulfur content, which you may realize just by looking at photographs from China. If the Chinese decided to use scrubbers in all their power stations, and steel mills, and in their cement industry, which has moved there from Europe and other places, ….
then with all the sulfur dioxide gone from the atmosphere….
Model E would predict an unprecedented raise of global temperatures.
Bill Illis, tell me that I am wrong!
Why else would Dr. Hansen not speak up to that problem: Stop the Chinese coal trains…
It’s mindboggling that there are intelligent people out there that don’t see how stupid this is.
Makes me a little sad.
Could I suggest the strong negative forcing they have missed is:
“The cold winds of change”
lol.
It’s always going to 5-10 years in the future that the models will be “proved right”, and any current discrepancy is down to “weather”.
This kind of reminds me of the HIV-AIDS fiasco (which the media pretend does not exist) – for several decades the World Health Organisation (another pseudo-scientific political organisation, sort of like the IPCC) has been predicting massive increases in AIDS in the West due to a not-yet-happened HIV epidemic. It’s only recently they’ve (quietly) admitted it’s not going to happen. Not so surprising when you realise the HIV-AIDS models they’ve used are just plain wrong.
So, the non-GHG hindcast is a data set provided by GISS to make the result appear like GHG + non-GHG matches the actual temperature record… And you are saying that the non-GHG data set is generated after the fact by someone to make it appear that the model’s predictive ability is better than it actually is?
It is very important that you explain exactly what you are describing here in very clear terms. This is going to generate controversy and speculation since you seem to be implying something untoward. Please fill in the blanks:
I am saying, in no uncertain terms________________
I am NOT saying_________________
Sorry for other topic, but I’ve been following AMSU temperatures and have discovered an astonishing silmilarity for January and February of 2006 and 2009. Some similarity with 2007 is also noticable but the correlation between 2006 and 2009 is really incredible. The same forcings have to be involved…?
http://discover.itsc.uah.edu/amsutemps/
You have to choose 5 km, 2006 and 2009
What’s up?
Hi,
aren’t you supposed to show the +- 1 sigma (at least) boundaries of the model and the experimental error bars to check if the two trends are consistent or not?
Anyway, as you said, we’ll see in a few years if that specific model was right or not. I just hope someone will remember…
By the way: from AMSU data, it looks like Feb 09 will be very close to Feb 05 or 06
It all comes down to the validity of the individual components in the complete GISS Model-E. Since neither the components not the complete model have been validated, it means the results are worthless. GIGO.
Bill Illis: Thanks again. I always find your posts and comments informative.
I took a look at the GISS Model E hindcasting data in these two posts:
http://bobtisdale.blogspot.com/2008/09/giss-model-e-climate-simulations.html
http://bobtisdale.blogspot.com/2008/09/giss-model-e-climate-simulations-part-2.html
The first curiosity I found was the Model E adds a slope to the volcanic aerosol data that doesn’t appear in the Sato Index Data used as the forcing.
Model E Output for Volcanic Aerosols:
http://i34.tinypic.com/inc32t.jpg
Model E Input of Volcanic Aerosols:
http://i38.tinypic.com/2i6copc.jpg
There’s a logical explanation for that, but I don’t know what it is.
The second was GISS uses the Lean et al (2000) TSI Reconstruction (Data with Background) as their solar forcing, after acknowledging that it’s outdated. Why would they use outdated TSI data other than to add to the slope of the hindcast?
I am very interested in this. Where exactly does GISS Model E fit in the galaxy of climate models?
Is it so different from others that these problems woudl be unique to it?
Re my previous post, I should have added, well done Bill and keep up the good work. I wonder if this will be headline news in the MSM!!!!!
If AGW has actually bought us roughly +0.5C and stopped inevitable cooling, I guess I will be thankful — as I prefer becoming too warm to becoming too cold.
It could be truth that AGW is offsetting declining temperatures. Something to think about. Still, this result seems to fuel my skepticism.
US considers a national climate service
Programme would merge climate-change data from multiple agencies.
Published online 19 February 2009
Nature | doi:10.1038/news.2009.108
http://www.nature.com/news/2009/090219/full/news.2009.108.html
Thank you, Mr. Illis.
“Without a large uptick in temperatures in the next few years, the modelers really have to go back to the drawing board (or they need to discover another “negative forcing” to keep the models on track to reality).”
If the models are periodically tweaked to match actual past past data, that’s just curve fitting. So by 2100, the models should give very accurate results for what has happened up until then. Or in 2100 you could look out the window and see where the shoreline is or if there are any glaciers in sight.
I hope no one is going to work on Model E for the next 91 years, but it’s nice work if you can get it.
Let’s hope there isn’t a large uptick in temperatures in the next few years, then.
Bill,
What have you actually done here? It’s very unclear. Did you compile and run the complete Model E? Your green line seems much smoother than what GISS people report.
Maybe the IPCC should put in a request for a new model. The current one they are using seems to suffer from a really bad Y2K bug.
Good work Bill.
One problem with saying they have a 0.15C difference in five years is that the GISS monthly data is at best ±0.2C and may be worse than that. It would take 7 years jut to drift away from the monthly data scatter.
For any curve fit model, extrapolation is fraught with danger. I think my old science teachers used to say “interpolate data with caution never extrapolate”.
For some additional clarity, I’ll put up the other non-GHG components individually.
Aerosols,
http://img175.imageshack.us/img175/6919/modeleaerosolshb4.png
Volcanoes,
http://img101.imageshack.us/img101/7802/modelevolcanoesmr4.png
Land-Use,
http://img127.imageshack.us/img127/7097/modelelandusexr5.png
Solar,
http://img355.imageshack.us/img355/3111/modelesolarej8.png
Net Other forcings,
http://img127.imageshack.us/img127/5221/modeleothertw0.png
Actual Model E hindcast versus GISS’ temperature anomaly, R^2 0.544
http://img355.imageshack.us/img355/9043/modelehindcastoz1.png
You can compare Model E’s hindcast to the simple model I built (R^2 0.713) using the ENSO, AMO and only about half the GHG impact they have built in. (There are no aerosols or volcanoe impacts built into this model.)
http://img301.imageshack.us/img301/7785/finalgissmodelns3.png
Hi Bill and Anthony,
Thanks for the article – very interesting but, alas, not surprising.
“For those brave enough to slog through it, here is the manual for Model E. ”
Anthony – you’re being very charitable calling the material you linked to a manual. No equations, no detailed description of the algorithms, just what you see at the link plus a few papers. Other organizations do a much better job with their documentation (e.g. NCAR, GFDL, etc.).
Bill – did you modify any of Model E’s subroutines, or did you just do some runs using the model as it exists today with the available input data? Thanks.
AnnaV
You are right and I am surprised that Tsonis is not more frequently analysed on blogs dedicated to the climate because I am convinced that he is on the right track .
D.Koutsoyiannis who also has a similar statistical approach (power laws and scaling) comes independently to similar insights .
This :
This is the first time that this
mechanism, which appears consistent with the theory of
synchronized chaos, is discovered in a physical system of
the size and complexity of the climate system.
says it all .
However I think that you have not the right idea about what the GCM do .
They do NOT solve any coupled non linear PEDs (Navier Stokes) .
With the too coarse gridding they are obviously unable to do that .
What they do instead is merely conserve energy , mass and momentum IN AVERAGE what is much easier .
As the system is cut up in cells , f.ex 100kmx100kmx1km , the bulk of the cell is actually represented by only 1 point with 1 velocity , 1 pressure , 1 density and 1 temperature .
So the only thing you need is to write the boundary conditions with the 6 cells surrounding the one you look at (6 other points with different parameters) and then go one temporal step farther by conserving energy , mass and momentum .
All the complexity is concentrated in the equations describing the boundary conditions and the way how the characteristic parameters (subgrid parametrization) change .
And then , of course , as it is a numerical exercice , you get rounding errors and such that make you violate conservation laws so you must make sure that those deviations are “absorbed”/”redistributed” numerically over the whole system before going a step farther .
So the right picture of a GCM is more a simulation of a discrete system with N points (like N molecules) that obeys conservation laws and some boundary conditions rather than solving fluid dynamic and thermodynamic equations for a continuum .
And you are saying that the non-GHG data set is generated after the fact by someone to make it appear that the model’s predictive ability is better than it actually is?
No. What happens is the forcings resulting from all the Non-GHG climate drivers (such as sulphate aerosols) are tuned to produce a fit to the historical temperature record.
Since the non-GHG forcings are fictional and bear no resemblance to the actual forcings, they have no predictive value. So even if the forcing for CO2 is exactly right, the model will never produce accurate predictions (except by chance or by replicating the same trend).
Of course, those who follow the progress of Anthony’s Surface Stations Project will know the surface temperature record has serious problems and tuning the models to the flawed historical data makes the models even more wrong.