Recently, I’ve been downloading “model data” from GCM runs that form the basis for projections in the IPCC AR4. I’m planning to compare some of the features to see how well the models reproduce the statistical properties of “weather noise”. But, first, I’m just plotting and looking at observations of earth data.
To simplify initial comparison, I’m merging NOAA, GISS and HadCrut measurements for Land/Ocean data, rebaselining relative to the average for 1970-1999 (inclusive), and rebaselining each model to 1970-1999 (inclusive) for the model.
Here’s how the five model runs for CGCM 3.1, forced using SRES A1B after 2001 compare to observations of earth data:
Notice the observations of real earth temperatures (in gold) match the average for CGCM during the 80s– that match is forced by my choice of baseline. Well… at least the data matches runs 1, 2, 3 & 4 pretty well. Run 0 is low compared to everything even in the 80s.
Recently, the data are dropping down from the “upper” range predicted by runs 1-4 of CGCM 3.1; the earth’s temperature are now approaching the levels ‘projected’ by run 0 (which was cooler than the earth during the baseline period.)
So, on the one hand, one might think model CGCM is “bracketing” the earth data. But, on the other hand. . .
Are you wondering about run 0?
Notice run 0 remains about 0.5C below the other runs for… oh…. 350 years!
(Yes, you also get to notice the model overpredicts warming during the 20th century.)
In case you think that 0.5C difference is due to something I did, here is how it looks at The Climate Explorer:
What does this mean? Is the 0.5C temperature lag lasting for 350 years due to some physical mechanism that exists on the real earth? Is it a numerical “glitch”?
Your guess is as good as mine! But these sorts of “weirdnesses” are things I’m pondering as I try to figure out how to compare the properties of “model weather” to “earth weather”!
I respect the restraint you show by not jumping to conclusions.
If only I were blessed with such restraint, I would have never decided to spend time in the kitchen whipping up some yummies. It was rather a Homer Simpson moment when I thought, “mmmm fudge. I prefer peanut butter fudge.” So off I go to toil over a hot stove top, destined to curse my lack of restraint.
Lucia,
You wrote: “…and rebaselining each model to 1970-1999 (inclusive) for the model…”
Do you mean you adjusted the outputs of each model to average “zero” anomaly over that period. If so, then maybe run 0 simply did not get adjusted. Because, if adjusted, there is no way run 0 could not be coincident with the others at some point in that period.
Things that come to mind are data with the “0” subscript not being “seen” by the adjustment routine (depending on the software used) — of course, that is merely a guess for the purposes of illustration. Or maybe NONE of the models were actually adjusted — so, run 0 remained below the others.
Or, maybe, “rebaselining” doesn’t mean what I think it does in this case.
Hmmm. Edit did not work. Whoops. It did work.
I did similar calculations for 1900-2000 using 20c3m to see how they compare to yours. i found that ensemble member 0 isn’t consistently on the lower end of the ensemble series. I thought I would get similar results because I thought all the models projecting from 2000 and beyond use the model results of the 20th century experiment to start with. Well, I suspected it was because I used a different baseline (1950-1980). I went to climate explorer and found that when I changed the baseline, the noticeable difference disappeared.
Allen– There are three possible ways of rebaselining.
Model 3CGCM has 5 runs.
* I rebaselined to the average of 1970-1999 for the 5 runs collectively. So if run 1 is “hotter” than run 2, that still shows. However, by this method, the average of the 5 runs for CGCM 3.1 between 1970-1999 matches the measured data.
* An alternate way is to rebaseline run 1 to to its own average for 1970-1999, then rebaseline run 2 to its average for 1970-1999 etc. If I did it that way, run 0 gets shifted so it is forced to match run 1 during 1970-1999.
* A third alternative is to not rebaseline at all, and just use predictions kelvins. In that case, the models overall, on average are “cold” compared to the real data.
I don’t do the third method because the claim by modelers is that the models predict the trend, even though they are cold on average. To compare trends, you need to rebaseline.
The first method shows that even with the same physics run 0 of CGCM has a shift compared to runs 1-4 even though all parameterizations in all runs are the same.
This is odd.
Chad– I don’t know if model 0 of CGCM is high or low compared to other runs from other models. I’m not sure what you are saying about noticable differences disappearing. If you look at model run in Kelvin, it is consistently different from runs 1-4 for the same model (i.e. CGCM). The “shift” lasts 350 years.
This is quite odd, particularly if we expect “weather noise” to somehow average out over 30 years as is often claimed at other climate blogs.
In your second figure from Excel and third figure from climate explorer, you can see run 0 is noticeably at the bottom. That is the noticeable difference that disappears that I was referring to. Also, I clicked “leave in K” but it somehow still spits out tas in °C. Strange.
Chad– The noticeable difference disappear when you do what?? And do whatever you do, where?
I downloaded these from climate explorer in degrees C. I used the default temperatures.
What did you do? There are a couple of models that do this, and I’m totally puzzled.
Lucia- Go back to climate explorer and when it displays the graph like the third one in this post, change the baseline to 1950-1980. I’ve uploaded a before and after shot. Yes, it is puzzling.
Chad–
I think I know what you did.
What I did was download the “raw data” from the top figure. So, that included the annual cycle variation. If scroll down and rebaseline, then the difference disappears. That figure, and raw data is not re-baselined. It’s in C. ( I can’t get it in K either.)
If you use the bottom portion of the screen and request a rebaseline, it appears the climate explorer rebaselines each run to the average for temperature for that run during the rebaseline period. So, yes, the difference disappears.
Chad– I think I understand what you did. Yes– if you rebaseline run 1 based on the average for run 1 during any period, and then run 2 based on the average for run 2 etc, then the 0.5C difference goes away.
But the fact that it’s there before you rebaseline, or if you rebaseline based on the average for all five runs together is puzzling. Why should run 0 be 0.5 C lower forever? That is, in terms of real temperatures, not anomalies?
This is quite odd.
Lucia, did all the model runs start at the same temperature? From what I remember of some other graphs, models were not necessarily initialized to the same start values. I don’t know if that’s what you’re seeing, but the intent of the model runs was always trend not absolute values, or so I thought.
BarryW–
The runs start at different points from a control run. So, no, they don’t start with the same temperature. Based on phenomenology, (conservation of mass, momentum and energy), the effect of the initial condition should vanish over time.
In fact, part of the purpose of the control run is to correct poor guesses for the initial condition.
The intent of the models is both to get the absolute values and the trend. If you read the AR4, you’ll find the section discussing the absolute values for the models. There is a known “cold bias”– and it’s a few degrees C. The absolute value at the poles is known to be off quite a bit, as is the temperatures in some equitorial basins of oceans.
Anomalies are used for two reasons:
1) to hide the fact that the average temperatures for the planets are actually incorrect.
2) Because of reason 1, the climatologists have made up an unproven conjecture that you can get the trends correct even if you get the absolute values wrong.
Whether or not (2) is correct is something that ought to require proof. It’s generally untrue for other models applying conservation of mass, momentum and energy to model transport phenomema.
Lucia – If the data they have archived breaks temperature into maximum and minimum temperatures, and you compare with the comparable observations of these two temperatures, it would provide additional information with which to contrast the models with reality. Our research shows that we should expect significant differences in long term trends between minimum and maximum temperatures; e.g. see
Lin, X., R.A. Pielke Sr., K.G. Hubbard, K.C. Crawford, M. A. Shafer, and T. Matsui, 2007: An examination of 1997-2007 surface layer temperature trends at two heights in Oklahoma. Geophys. Res. Letts., 34, L24705, doi:10.1029/2007GL031652.
http://climatesci.colorado.edu/publications/pdf/R-333.pdf
I did the same thing you did but manually adjusted all more recent temperatures upward, filtered for the the average thickness of bristlecone pine tree rings in years that are prime numbers, assumed that all temps before 1970 are wrong and need to be adjusted downward (but with an algorithm that only works on my desktop) and added some random stuff to add more spikes and bristles.
I pretty sure I have proved you wrong and that not only is the projected trend higher but it got warmer right now all around me when I ran the program. I would share my stuff but it is proprietary as well as unintelligible. Lastly, I should point out that on my graph, the growing divergence between model and reality results in reality being invalidated about 11.5 years sooner than on yours.
Nevertheless, you have again done some impressive work. Thank you.
Cheers.
Send said warmth here. I’m wearing sweats to ward off the unseasonal coolth and my tomatoes are taking too long to ripen. (It’s supposedly 76F. I like the 80s.)
Lucia- I just want to commend your instinct to look at the actual data. Time and again in life I have seen intelligent people miss important things because the focused on only “the final number”/average/etc.
Nothing beats the smell of data in the morning.
erik, Ah how I love the allusion to Kilgore.
Smell that? Do you smell that? Data son. Nothing else in the world smells like that. I love the smell of data in the morning. You know one time, we examined a theory’s data for 12 hours. When it was over, I looked at the theory. We didn’t find one of ’em. Not one stinking incongruity. The smell. You know that data smell. The whole theory. It smelled like… Victory. Someday this confusion’s gonna end.