GISTemp just published their April Temperature anomaly (0.50C). As many of you know, I like to show trends since 2001 (because that’s the first January when the SRES for the AR4 were frozen). I thought some you you might like to see the current temperature anomalies superimposed on the multi-model mean projection and ±1 standard deviation and the best fit trend all:
I thought you all might also like to see where the observed trend is relative to 55 runs from the AR4:
When the current trend falls below all but 2.75 runs, it will be below all but 5% of runs (that would constitute a 1 sided test using the actual population of runs.)
Given the appearance of the first figure, I’m expecting the trend since 2001 to continue to fall a bit before it rises. (The reason is that the most recent monthly data fall below the trend line. For the computed trend to rise next month, next month’s data has to come in above the current trend line. It’s just math.) I guess we’ll see what happens.
More graphs when HadCrut and NOAA report. 🙂
Lucia, the graph uses SKS-style trickery by showing the model line in thick black and the actual real-world observations in thin barely-visible green! Lol.
Well, I say real-world observations – hmm, it’s GISS after all.
Other than that, the story tells itself really. Thanks!
Humm… Makes me wonder how long it will be before climate scientists will be ‘forced’ ( 😉 ) to admit the models are way too sensitive. The best known in the field continue to publicly insist the models are correct, that rapid warming will resume any day now, and that temperatures will be >2C warmer than today before 2100 (and sea levels a meter higher!).
.
It would be funny to see their discomfort in having to deal with reality diverging from projected outcomes, save for that there is still danger of real economic harm being done worldwide based on mistaken climate model projections and hyped claims of catastrophic consequences. I hope only for political gridlock everywhere over the next decade, so that reality can continue to impose itself before too many bad decisions are made. Rational and prudent public policy choices on energy can only be made after the worst of the hype and frenzy in climate science has passed. Better policy choices will be made later rather than sooner.
AR5 forecasts are on more-or-less on the same track as AR4 so they are in the uncomfortable position of starting their forecast well above current temps.
Might have to be some massaging before the AR5 forecasts are published. (Technically AR4 let in many changes even up to the spring of 2006).
havent you learned that 12 years years of data cant show you anything, except that observations fall outside the …
errr.. wait.
Bill Illis–
That’s why we don’t compare to the AR5 projections. We can’t be sure what they are until the fat lady sings.
Any comment on David Appell’s article?
http://www.yaleclimatemediaforum.org/2013/05/wither-global-warming-has-it-slowed-down/
As I commented at the article, it seems to spend a lot of time and effort trying to convince that the sensitivity estimates don’t need to come down because of the “pause” – something that Bayesian statistics should say is impossible.
MikeR–
I don’t have any particular comments. Clearly people are being forced to explain why temperatures can be flat or rising more slowly than projected.
The only thing I really found interesting about that piece was the idea that we should make less of the pause because it could be predicted, retrospectively. Or as the paper referred to says:
That’s a new one on me. I wonder what else we can retrospectively predict.
“I wonder what else we can retrospectively predict.”
The NY Mets will win the world series in 1969. Shhhh, don’t tell anyone, but this is sooooo easy.
Lucia – after years of prevarication and resistance I’m attempting to teach myself R at home. Do you have a link to the code that calculates the trends, autocorrelation etc and produces the graphs? I have a feeling you did so once but I can’t find it.
Grant B–
I’ve periodically put up scripts, but never for the one that plots the graphs in this post.
You people are a funny lot. It’s clear that global warming has just shifted from atmospheric warming to deep ocean warming… you know… down below 700m… where we cannot measure it with precision. It’s all very scientific really. 😉
“The ability to predict retrospectively this slowdown not only strengthens our confidence in the robustness of our climate models…” Brandon, Lucia, I understand that this sounds ludicrous, and maybe it is. But can anyone explain to me what in the world they think they are saying?
‘But can anyone explain to me what in the world they think they are saying?’
simple. re run the model with better inputs and different assumptions. standard practice
So I hear that it’s good to adjust your model as you gain new information. But I don’t understand how that can help to “strengthen our confidence in the robustness of our climate models”. It means that the last model didn’t work well, and now we need to test our new model against future data to find out whether it’s better. In the meantime, we don’t have confidence, we need to find out if it will work. How can it strengthen confidence?
Mosher
Standard in many fields. But there is an important and thorny issue in climate modeling. I’ll explain by resorting to analogy.
Suppose someone create the perfect vehicle dynamics code (i.e. model based on physics). You can predict some aspect of vehicle performance (e.g. time to accelerate from 0-50 mph) perfectly provided you are given certain specifications. (Engine performance features, car mass, rolling resistance etc.)
Then some tells you a car is equipped with certain properties: V8 model blah, blah, X lbs, blah, blah. You predict time to accelerate from 0-50 mph.
Turns out you are wrong.
It would be very natural for the modeler to say: Let me inspect the car. If he discovers a different engine was installed and every one agrees “Yep. That’s the V6! Not the V8!” He reruns is code with the correct engine. Then he compares his answer. If it matches, the problem was not his code or model, the error was due to his being provided the wrong input– in this case wrong engine model.
In this standard practice situation, there is generally very little disagreement that one can determine which engine was used by inspecting the car.
The difficulty with climate models is that even after temperature have been observed, there is still uncertainty in the forcings. And the “correct” forcings are– to some extent– “discovered” with knowledge of what temperatures were projected with a previous set. If observed temperatures were lower than projected, a lower forcing will likely do the trick. And– not withstanding any objections by a modelers– one can do a back of the envelope calculation to estimate the magnitude of the effect of a % difference in forcing. (To some extent, getting an estimate of the effect of a forcing is that’s the goal of modeling in the first place!)
So, with climate, it’s all a bit circular. (Not that this is unique to climate. But it is thornier than in some other fields.)
Can someone explain the 3 or 4 models that predict more CO2 will result in a lower temperature? And why aren’t models creators nominated for a Nobel Prize?
Bruce,
None of the models predict CO2 will result in lower temperature. Some models have very large “weather noise”. In models with really huge “weather noise”, individual runs will have negative trends over some time spans with temperature eventually rising.
Lucia,
“So, with climate, it’s all a bit circular.”
.
More than a bit circular. The field is void of meaningful tests of projections… and the goal posts are always moved ‘retrospectively’. I can appreciate the difficulties involved, but I object to the certitude placed on projections of warming, and to the seeming resistance to allowing new data to impact old projections. A few honest comments from leaders in the field like, “Wow, we really missed that projection; I guess the models are less accurate than we thought” would be both appropriate and welcomed. I won’t hold my breath.
I’m wondering how they are planning to produce model runs that do a reasonable hindcast, predicting a slowdown in the last decade, while still projecting lots of warming going forward.
> “The ability to predict retrospectively this slowdown not only strengthens our confidence in the robustness of our… models…â€
This sort of claim seems to get nods all around the room, at a climate science get-together.
It’s worth repeating how the success or failure of a pivotal clinical trial is determined. The key endpoints are specified prior to the start of the trial, as are the methods by which those predefined endpoints are to be analyzed.
It’s hard for me to reconcile these disparate approaches.
MikeN,
Don’t worry, the process has already started; “reanalysis” (a lightly constrained climate model output which conflicts with measurements) is already being used to explain the slower warming. Count on aerosols, ocean heat, and vigorous arm waves to explain everything. It seems politically unacceptable for estimates of future warming to go down; the data matter not at all in the cultural calculus.
AMac,
The flexibility the climate scientists give themselves is truly appalling. But don’t worry about it too much it’s not like they are using the information for anything important like raising a whole generation to believe humanity is screwing up the planet or asking for the global economy to be re-arranged.
Here’s another example of an argument for global warming that seems… strange:
Skeptical Science is saying ice sheets and global sea ice absorb heat without warming. How in the world does that make sense?* When things absorb heat, they get warmer. Everyone knows that.
Originally I thought it was just a matter of semantics, and “warming” meant something like “warming the surface.” Then I read the next sentence:
They cleary distinguish between “warming” and “warming the surface” here, so that couldn’t be the case. Warming must be taken to mean… warming. As in, the ice is melting without ever getting warmer.
And that’s not even touching on the fact ice melting doesn’t automatically indicate warming. It could just indicate a steady, warmer state.
*Absorbing energy can cause a change in matter state without raising temperatures, but the amount of energy covered by that is negligible.
Brandon, they probably mean the latent heat (enthalpy of fusion).
Edim, I looked up the equations, and maybe I was wrong that the energy needed to change states (what you call enthalpy of fusion) is negligible. It looks like it takes as much energy to change states in pure water as to increase its temperature by 80 degrees. That’s a lot higher than I thought. I’m not sure what it’d be for sea ice/ice sheets since though. They have salt in them, as well as air bubbles, and I’m sure there are other factors.
Anyway, I’m not sure how much energy goes into that with ice like they mentioned, so maybe I’m wrong. I’m just not seeing how any meaningful amount of energy would be absorbed like that. And while changing states may not cause warming, I’d imagine the water still warms afterward.
I approach this from a different perspective. I ask what kind of physical red noise, derived from an Ornstein-Uhlenbeck process, can generate similar fluctuations to what are observed in the global temperature anomaly records:
http://theoilconundrum.blogspot.com/2013/03/stochastic-analysis-of-log-sensitivity.html
WebHubTelescope–
Could you clarify how your perspecritive ‘differs’? I show the uncertainty due to red noise in my figure. You use red noise. That would seem to be “the same”. The only relevant difference I see is you are trying to the thermometer record from 1880-2000 and my graph looks at data after 2001 and you are trying to look at response to CO2 using a very simple semi-empirical, while I am comparing observations climate models projections from the AR4.
Brandon (#112924) –
Heat of fusion for water is 334 J/g = 3.34E17 J/Gt. The highest ice loss rate I can recall is 373 Gt/yr, from Velicogna 2009. [More recent estimates are lower than that one.] The product gives about 0.12 ZJ/yr. By comparison, Levitus et al. 2012 gives an OHC rate (0-2000m) of 4.3 ZJ/yr. That’s over 1955-2010; by eye the Argo-era rate is slightly less than that. [~70%?]
So melting ice accounts for a few % of the total energy imbalance.
HaroldW, I looked into it last night, and I think ice loss only accounts for a couple percent of the energy imbalance. I think the IPCC puts it at ~1%. Even if the change of state took upxall that energy (it doesn’t), I still think it’s a negligible value.
But it certainly wasn’t as wrong as it sounded to me. It’s small, even compared to atmospheric energy absorption, but its not nothing.
By the way, ice loss isn’t just caused by melting. I’m not sure how one accounts for the various other factors when doing energy content analyses.
lucia
Another great post summerizing model performance.
At what point (in your opinion) should individual model’s be dropped from further consideration for use in AR5?
Brandon (#112931) –
I agree, small, much less than OHC, but not a completely irrelevant quantity.
Interestingly, while looking up ZJ (to be sure I didn’t mess up), I noticed that Wikipedia says that world energy consumption is approx. 0.5 ZJ/yr. While small, that’s not at all negligible, and it’s all dissipated over land.
Rob
Not my call. 🙂
Anyway, it’s quite likely models were tweaked? We’ll see what we end up getting in the AR5. I think the final draft was submitted. I presume whatever the projections are won’t change between the draft submission and final version. It’s tweaking by gov’t people that might happen now… right?
Tweaked! I thought I was the only one who had doubts as to the robustness of the claims that the ‘models’ are based on basic physics and are not ‘trained fitting programs’.
Probably the same. Physicists call the model for red noise an Ornstein-Uhlenbeck process. Statisicians call the model for red noise a member of the auto-regressive model family. The big difference is that statisticians normally don’t derive a physical interpretation to the noise, whereas that is what physicists are supposed to do.
The physical interpretation is that the temperature is the response of random walk energy fluctuations in a shallow energy well. The characteristic property of the O-U process is that it has a reversion-to-the-mean pull, which is essentially towards the minimum of the energy well.
When we see a trend in the long-term temperature profile, we have to make an educated guess whether this is just a slowly varying fluctuation from the mean, or whether the mean itself is changing and the reversion-to-the-mean is just following an upward (or downward) trend.
Statisticians may have the upper hand in discerning whether the trend is real or random, but physicists will have the upper hand in trying to pull all the supporting evidence together, and make a judgement based on the bigger picture.
WebHubTelescope,
Ok. But I don’t understand your point.
I know what red noise is. I know what the Ornstein-Uhlenbeck process is. I know the physical interpretation or the Ornstein-Uhlenbeck process and have for at least 30 years. My ph.d. is in Mechanical Engineering in the area of fluid dynamics, and I understand diffusion of particles by brownian motion.
But I just don’t know what your point is. You seemed to be suggesting we were using “different” approaches because you used “red noise” from “Ornstein-Uhlenbeck”. It’s the same dang red noise. The only difference is you are typing out “Ornstein-Uhlenbeck” and saying “physics”. Wedging these words into your analysis isn’t material to the results or the approach.
OK, Good to see that there are PhD’s in MechE working the problem.
It definitely needs a multidisciplinary approach to crack the nut.
NOAA have reported for April now, but only in the Global State of the Climate Report, not the data files, so the latter could be different:
http://www.ncdc.noaa.gov/sotc/global/2013/4
God knows when Hadley will report, they didn’t publish March until early May.
NCDC/NOAA data files now updated for NH & SH. but for some reason not global.
Hemispheric figures tie in with Global SOC report.
Global anomaly = 0.52c
Hoorah!
Global data now updated for April.
lucia, I think part of what WebHubTelescope is saying is the difference between a purely statistical and a physical approach is with the physical approach you have a physical model that constrains the statistics.
With statistics you can produce a model that e.g. violation of energy, add physics and you can constrain your statistical models so they don’t.
Of course I know you know this so…
Of course you know I know this. After all: When Tamino advanced the claim his two box model was “physical”, I pointed out his result violated the 2nd law of thermo because given the parameters he found, heat flowed from the cold box to the hot box. There would be no problem with that if the two box model was just a mathematical model or complicated curve fit.
Also:I agree that physical models constrain. That’s both advantageous and useful if you can advance a physical reason why your physical model applies. Otherwise, there is no meaningful difference between the mathematical and physical model.