Spinning Probabilities in GRL
April 7th, 2009Posted by: Roger Pielke, Jr.
Earlier this week Andy Revkin pointed to a new article in press with Geophysical Research Letters by David Easterling and Gerald Wehner, which Revkin summarizes as follows:
The paper shows, both in recent records and projections using computer simulations, how utterly normal it is to have decade-long vagaries in temperature, up and down, on the way to a warmer world.
The goal of the GRL paper is to show that the current period of no rise in temperature is, in Revkin’s words, “utterly normal,” and in the words of Easterling and Wehner, “entirely possible” and “likely.” Revkin was duped by the paper, and I suspect many people will be. What the paper is arguing is not that the current period of no-warming is “utterly normal” or “likely” but that such periods of no warming are “likely.” The difference is subtle but critical to understand. That such a paper would pass peer-review with such basic confusion and spin is remarkable in some sense, but probably is to be entirely expected.
The confusion can be illustrated as follows. Imagine that you are playing a game of poker, in which you are dealt 5 cards. You’ve never played poker before so you don’t know the odds for a particular hand. You look at your 5 cards and see that you’ve been dealt two pairs. You then ask your companion, a poker expert, whether or not your hand is “likely” so that you can evaluate it rigorously. Which of the two responses that follow would you consider to be a more straightforward response to your question?
Response #1 You can see over many hands that being dealt two pairs can and likely will occur. In fact, Joe had two pairs in a hand dealt 20 minutes ago and Tim had one 10 minutes before that. And if you simulate a game of poker you’ll find a hand dealt 25 minutes from now has 2 pairs and one 7 minutes later also has two pairs. So in conclusion, both observations and simulations show that two pairs can and are even likely to occur. Your hand, therefore is utterly normal and entirely possible.
Response #2 The odds of you being dealt two pair in any given hand in about 1 in 21, so it is a statistically rare event.
To be more explicit, Easterling and Wehner have (purposely?) confused/conflated two questions, both of which are fair to ask. But they are not the same question.
One is, what are the odds of seeing a decadal cooling trend in a long period of warming?
They answer this by saying that:
. . . it is reasonable to expect that the natural variability of the real climate system can and likely will produce multi-year periods of sustained “cooling” or at least periods with no real trend even in the presence of long-term anthropogenic forced warming.
They support this argument by pointing to historical periods where a decadal lack of warming occurred and also to model runs that show similar decadal periods. This is interesting but unremarkable, and certainly not a novel claim (e.g., you can find similar claims on any number of blogs). This observation certainly wouldn’t justify publication of this paper.
The second question is, how unusual is it to see the current period of lack of warming?
Easterling and Wehner focus our attention on the current period of warming by introducing the paper as follows:
Anthropogenic climate change is one of the most contentious scientific issues of our time. Not surprisingly the issue has generated numerous blogs and websites with a wide range of views on the subject. According to a number of these sources the climate is no longer warming, in fact, some claim the planet has been “cooling” since 1998.
Although Easterling and Wehner never answer the question explicitly about how rare the current period of observed warming is, they imply throughout that this is the question that they are addressing. The answer to this question is that the current period with a lack of warming is a pretty rare event. How rare? Easterling and Wehner allow us to answer this question by providing a distribution of 10-year trends from a set of climate model realizations:
. . . for the simulations of the entire 21st century there is still about a 5% chance of a negative decadal trend, even in the absence of any simulated volcanic eruptions. If we restrict the period to the first half of the 21st century the probability increases to about 10% revealing that the trend in surface air temperature has its own positive trend in the A2 emissions scenario.
So a negative decadal trend (though not statistically significant, so perhaps better called a period with a lack of warming) is according to the distribution from these models is a 1 in 10 event. In other words, if Easterling and Wehner were asked ten years ago what the odds of seeing a decade of no warming, they would have answered 10%. They further report that a statistically significant (>95%) negative decadal trend is, according to their analysis, a 1 in 100 event based on 20th century observations, and an impossibility in the 21st century, since it is not found in the realizations (Table 1 of their paper). Of course, all of this and much, much more has been done far more rigorously by Lucia Liljegren, but I digress.
So while it is fair to say that the current period of an extended lack of warming certainly does not disprove global warming over the longer term, it is not appropriate to say that such a period is “utterly normal” or, misleadingly, to imply that this specific occurrence is “likely.” Given that we are in the midst of a rare event, it is strange to see a peer reviewed paper claim that “misleading” to raise questions about model predictions or to question established theory in such a context. Are such politicized editorial comments the norm now in climate science?
If temperatures cool further or remain without warming for a few years, it could very well be the case that we do see a statistically significant cooling trend over a decade or longer. Then we would get to see the Easterling and Wehner paper cited again, but in that case by skeptics as evidence that global warming has indeed stopped. That argument would be misleading as well.
In the current decade, climate modelers may have gotten unlucky or there may be real issues with predictions from climate models. We don’t know the answer to this question. But the “analysis” of Easterling and Wehner gets us no closer to an answer. They do provide some ammunition for the political debate, but little insight to the science. If one wants to perform rigorous comparisons of climate forecasts and observations, there are far more robust approaches than found in GRL this week.
April 7th, 2009 at 4:22 pm
Roger,
You write: “That such a paper would pass peer-review with such basic confusion and spin is remarkable in some sense, but probably is to be entirely expected.”
I don’t mean to get off topic of the post, but could you clarify the last part of this–”but probably is to be entirely expected”? What do you mean by this? Why is this to be expected?
April 7th, 2009 at 5:13 pm
A much more thorough look (than Easterling and Wehner) of how observed trends fit into model expectations can be found in Figure 3 of Pat Michaels recent House testimony (http://www.worldclimatereport.com/index.php/2009/02/13/committee-on-energy-and-environment-testimony/)
Figure 3 shows the 95% confidence intervals of model trends (during the first 2 decades of the 21st century under SRES A1B) ranging in length from 5 years to 15 years. Also shown is the current observed value for each of these trends.
Bottom line…observations are pushing the lower bound of model expectations.
-Chip
April 7th, 2009 at 6:18 pm
This paper may be ambiguous with respect to what is likely or utterly normal but how does this paper stack up in the contect of science by press release. How often in the last 3 months have we seen a press release saying that the indications of catastrophic climate change are much worse than we expected? These press releases seemed timed to some legislative decision somewhere in the world. I realize that these “much worse than expected” declarations involves a lot of extrapolation and assumptions that all have to line up exactly to get some computer model to predict catastophe but to the average laymen, arguing a short term cooling trendings is normal or likely while your buddies are making “much worse than we expected” hyperboly defies logic for all but the most sophiticated rationaliser.
April 7th, 2009 at 6:24 pm
-1-Keith
There is a lot of subpar work to be found in climate science of late. For example:
http://sciencepolicy.colorado.edu/prometheus/aberson-on-hollandwebster-4965
http://sciencepolicy.colorado.edu/prometheus/here-we-go-again-more-cherry-picking-by-the-ccsp-4923
http://sciencepolicy.colorado.edu/prometheus/stern-on-extremes-5068
April 7th, 2009 at 7:56 pm
In case nobody has noticed, the sun is ***still*** awfully quiet.
http://www.spaceweather.co.za/LatestInfo.htm
It has now been almost 2 years since the solar cycle 24 was expected to begin:
http://www.spacetoday.org/SolSys/Sun/Sunspots.html
But of course, anyone who’s read the IPCC assessment reports knows that changes in the sun don’t affect earth’s temperature.
P.S. And of course, there’s nothing unusual about what’s happening with the sun:
http://science.nasa.gov/headlines/y2008/11jul_solarcycleupdate.htm
April 8th, 2009 at 3:52 am
It’s pretty easy these days to get half-baked stuff like this past cursory peer review because it attempts to prop up climate alarmism. If your name is Roy Spencer and you submit 2 good papers to GRL on negative feedback in climate models, and the relationship between the PDO and 20th century warming – they get bounced without proper review. That said, the 3 Tsonis et al GRL papers on climate shifts provide a mechanism for 20th centrury temprature trends without invoking CO2.
The fact remains that there is a lack of global warming since about 2002, despite rapidly rising CO2 emissions. Either CO2 is irrelevant or it has a small effect that is overwhelmed by natural variability. Either global warming has peaked, or it hasn’t. Time will tell.
April 8th, 2009 at 4:18 am
So, when the ‘warming’ starts again how long do we have to wait to be ensured that it is the true fingerprint of the signal of the great global crisis and not an ‘utterly normal’ happenstance?
April 8th, 2009 at 5:16 am
As I understand, it’s actually worse than that. Current CO2 levels are at the very highest end of those used in models, not the mean expected value, so the forcing should be even higher than the mean prediction of all models.
Another key matter that is always ignored: the downward variation from the past that is always cited to explain away the lack of warming is to be expected under normal, non-forced conditions. Under anthropogenic forcing, it would be expected that such downtrends would be less likely, so random downward movement would be less likely. Each uptick of the Mauna Loa CO2 trend should be expected to push probabilities of temperature increases up, and make any leveling or downtread less likely than in the past. So a comparison with past – even recent past – data is biased.
The proponents of apocalyptic predictions are hoisted on their own petard – the more they claim “It’s even worse than we thought before,” the less they can explain current data.
April 8th, 2009 at 6:58 am
If it is warm, in means the models are right. If it is cool, it just means that laymen and denialists lack the sophistication to appreciate natural variability.
The models really knew this recent non-warming would happen but because they only revealed themselves in simpler forms and straighter graphs (much as Yahweh revealed Himself in metaphor and simple allegory), lesser minds were likely confused and frightened when the warming did not appear.
Climate science is leading the way to a truly post-modern science, freed from the tyranny of empirical consistency and the tiresome messiness of nature. Soon there will be only The Model and nature will be relegated to its rightful status as nothing more than an irrelevant rounding error that would detract from the purity of The Model if we let it.
As for whether peer review applies to the paper in question, what need for peer review when a Consensus exists?
As for #6 EDaniel’s irreverent question, confirmation of the models can be instantaneous–weekend heat wave, a crack in an ice sheet, etc. but refutation requires at least one full ATEE. (Note: ATEE is a unit of time–”average tenured employment expentancy”)
April 8th, 2009 at 8:32 am
Actually, I made a mistake – it’s not politically correct to talk of ‘non-warming’ or ‘cooling’ – it is ’suppressed warming’ in Orwellian Climate Newspeak.
April 8th, 2009 at 9:11 am
The models do not make predictions / projections / forecasts. Instead they use EWAG scenarios in what-if investigations. The scenarios provide boundary conditions to the models. In the normal course of solutions of PDEs, the solutions interior to the domain of interest are determined by the known information supplied at the boundaries. The solutions are only as good as the supplied boundary conditions. It so happens that none of the EWAG what-if scenarios actually correspond to what has happened in the real world. So, how can any calculated numbers be assumed to be of any real value.
Plus, as I am wont to almost always do, note that the numerical solution methods and the coding of these have not yet been Verified for any GCM. That is, it is not known if the displayed numbers are in fact solutions to the discrete approximations to the continuous equations. We do know, however, that no GCM has been able to demonstrate convergence of the numerical solution methods. That is beyond dispute. In all other areas, and I mean every other one, if convergence cannot be demonstrated the numbers are rejected out of hand.
On what basis can the ‘realizations’ calculated by computer software that has not yet been Verified and Validated, comprised of numbers that are known to not be solutions of the model equations, using boundary conditions that do not correspond to actual states attained by the real system, be taken to represent what the real system has, or will, experience?
It is also interesting to note that we are frequently told that the only way to interpret the calculated numbers is by use of ensemble averages. Yet at times the individual ‘realizations’ are employed to illustrate some point. Additionally, not all ‘realizations’ are always used. The illustrator seems to have the freedom to pick-n-choose the ‘realizations’ employed. I have often wondered why the enormous number of ‘realizations’ produced by the ClimatePredictionNet process are always ignored. The ClimatePredictionNet leadership, by the way, have already decided that some results can be discarded while others are retained; pick-n-choose.
The approved, high-impact-factor, peer-reviewed journals in which the certified climatologists publish have no policies in place to filter out those papers that are clearly based on untested and untestable hypotheses and numerical solution methods that exhibit known fatal deficiencies. The journals are publishing papers that have no place in the archival literature. That would not be so bad if the papers were clearly identified to be based on software that is not of production-grade quality. I have published such papers. The research-based nature of the papers were clearly identified and none have ever been the basis of policy decisions that affect the health and safety of the public.
The focus of the response to increased levels of CO2 is based on the concept of a simple homogenous system at a new equilibrium state following the full imposition of the boundary condition specifying the increased level. Additionally, it is not clear what the correspondence is between the quantities calculated by the GCMs and the quantities associated with the simple concept. The Earth’s systems have never in the past attained an equilibrium state and will never attain such a state in the future. But, if the concept is simply accepted, the new equilibrium state, after the full imposition of the increased CO2 level, will require several hundreds-to-thousands of years to obtain; the oceans are enormous energy sinks.
In the mean time, while we’re waiting, the quantity that has been taken to be the sole metric that correctly and accurately reflect the concepts from the simple homogeneous approach bares no known relationships to the final equilibrium state. The average temperature of a thermodynamically heterogeneous system not at equilibrium is determined by the thermodynamic states of the materials that make up the system as the system experiences the thermodynamic processes driving them toward equilibrium. A rough zeroth-order cut at this average temperature would necessarily require accounting for both (1) the mass and (2) the energy content of all the sub-systems. The algebraic average of the near-surface temperature of the atmosphere does not properly make this accounting.
On what basis can the ‘realizations’ calculated by computer software that (1) has not yet been Verified and Validated, (2) generates numbers that are known to not be solutions of the model equations, (3) using boundary conditions that do not correspond to actual states attained by the real system, and (4) using a metric that does not correctly and accurately reflect the state of the real-world system, be taken to represent what the real system has, or will, experience?
Finally, exactly what is a proper temperature, in the real world system, that correctly and accurately characterizes a radiative-equilibrium temperature is a whole nother question. And the behavior of that quantity as a function of time under non-equilibrium states is an additional whole nother question.
All corrections will be appreciated.
April 8th, 2009 at 10:10 am
Hi,
I’ll probably have comments on EDaniel’s comments later on.
In the meantime, I happened to take a look at the solar cycle lengths at the website I referenced:
http://www.spacetoday.org/SolSys/Sun/Sunspots.html
I calculate a mean length for the first 22 cycles of 11.0 years, with a standard deviation of 1.2 years.
I’m not sure when solar cycle 24 is taken as having begun (i.e. what condition is required for the “clock” to start). But assuming that we’re still in solar cycle 23, that would mean we’ve been in it for almost 13 years (13 years in May). Only one of the previous 22 cycles was longer than 13 years…solar cycle 4 lasted 13.6 years.
Of course, it needs to again be pointed out that there’s nothing unusual going on with the sun…solar cycles of 13 years have happened before. And in any case, changes in the sun don’t affect global temperatures, as the IPCC’s Assessments clearly show.
Still…
April 8th, 2009 at 12:36 pm
Roger,
Even if it’s an unlikely event (1 in 10) at any particular point in time, I think the point of this paper is to argue that within a longer (centennial) timeframe, the existence of such decadal periods without a significant positive trend are to be expected (“entirely normal” and “likely”). Just as when you play poker long enough, the existence of a hand with two pairs is to be expected.
The fact that the absence of a significant trend over the past decade is so widely used to try to claim that global warming has stopped, shows that there is a need to show the falsehood in that premise. In a way it is a sad state of affairs that scientists need to use their time to discredit obvious falsehoods, but I guess that’s where we’re at.
Bart
April 8th, 2009 at 12:46 pm
Isn’t “peer review” fast becoming a pejorative?
Some argue that most published research is wrong — http://www.plosmedicine.org/article/info:doi/10.1371/journal.pmed.0020124
As I understand it, grad students are no longer assigned the task of replicating other studies. Such work doesn’t generate grants for the department. Thus, it would stand to reason that a lot more bad science is now flying under the radar. This is compounded by the fact that scientists have less reason to fear that their poor work will be exposed.
What we do know is that there is an incredibly cavalier attitude toward quality control by climate scientists. See e.g. Gavin Schmidt’s mind-boggling comments re: mistakes in October database; see e.g. satellite errors in ice data going undetected for over a month until amateurs noticed; see e.g. temperature monitor siting disasters; see e.g. Jones failure to provide critical study; see e.g. trend away from transparency and data archiving; see e.g. multiple instances by IPCC and other committees ignoring major studies when compiling assessments; and on and on.
Climate science quality is a bad joke.
April 8th, 2009 at 2:20 pm
[...] Spinning Probabilities in GRL April 7th, 2009 [...]
April 8th, 2009 at 4:58 pm
You wrote: “…if Easterling and Wehner were asked ten years ago what the odds of seeing a decade of no warming, they would have answered 10%.” The definition of a decadal “period of lack of warming” is one in which “least-squares trends to running 10-year periods in the global surface air temperature time series” is not positive.
Most of the decadal periods that terminated in the past ten years showed net warming, with a few intervals of very strong warming. For example the interval 1993-2002 showed an increase of .359 degrees.
The only full decadal period out of the last ten years that showed a net cooling was 1998-2007.
That’s one out of ten, or 10%.
I do not understand the basis for your criticism.
April 9th, 2009 at 12:37 am
Chip (#2),
The graph you refer to is very sensitive to the endpoint chosen and the dataset used. This has been pointed out to you, and you agreed with the points raised:
http://www.realclimate.org/index.php/archives/2009/03/michaels-new-graph/langswitch_lang/6o
So you know that the bottom line you state is false.
Bart
April 9th, 2009 at 2:00 am
(#17) Bart,
Lucia addresses Gavin’s spin here:
http://rankexploits.com/musings/2009/look-i-can-use-made-up-data-just-like-gavin/
Her technique is different from Chip’s but the conclusions are the same.
April 9th, 2009 at 6:01 am
Raven,
Upon a quick glance her criticism seems directed at the filling in/guessing of what 2009 temps may look like, and how that would affect the graph.
If that is correct, then the main point, that such a graph is very sensitive to the endpoint chosen and the dataset used, still stands, and the so-called bottom line from Chip above is not warranted.
Bart
April 9th, 2009 at 6:23 am
Bart, Raven, Chip-
Seems to me that you guys are really arguing about the future not the past.
Bart, Chip says that recent temperatures are at the bottom end of model distributions. This seems to me the exact same thing that Easterling and Wehner are saying. Would you say that they are in the middle of the distribution? Top end?
The numbers all seem clear. It is the meaning of those numbers that people disagree about, which is why I say this debate is about the future not the past. But we should be able to get agreement on what the numbers say.
April 9th, 2009 at 6:52 am
Roger,
I have not analysed recent temps myself, but from what I’ve read, many of the short term trends of the last decade or so are smaller than the longer term trend.
E&W argue that such periods are entirely normal within a longer timeframe. Your critique of them seems centered on stating that at any particular point in time, this is a relatively rare event. But for it to ‘randomly’ occur within a longer time frame is to be expected. Isn’t that their main point, and is it not valid?
Bart
April 9th, 2009 at 7:57 am
Bverheggen,
Certainly the trend at any point in time is based on the data that go into it and the variability of the trend over time is related to the length of the trend, shorter trends have greater variability than do longer trends. In Figure 3 of Pat’s testimony we show the current (ending in December 2008) value of trends ranging in length from 5 to 15 years.
Also in Figure 3 is the Holy Grail that everyone (including Roger) is looking for—it the 95% confidence bounds of the expected range of model trends for the first 20 years of the 21st century of any length you want from 5 to 15 years (under A1B). We only show the 95% confidence bounds, but we could just as easily show you and confidence bounds you’d like to see. Thus, we can calculate the probability of occurrence of any trend value of any length (from 5 to 15 years).
Armed with this information, you can then check to see how any observed value of any trend length (from 5 to 15 years) compares to model expectations.
This is what we did for observed trends ending in December of 2008 calculated from the HadCRUT3v dataset. If you want to see, say, how the 7-year trend value ending in February 2006 calculated from the GISS dataset compares to the model expectations, all you need to do is calculate the value from the observed data and see where it fits in the model expectations.
The whole point of developing the model range is to eliminate the dependence on start and stop dates from the comparison. This point seems to be missed by our critics.
You can choose any start and stop dates and any data set you want, but if you start to find some that produce observed trend values that fall outside the range of expected model trends, then you have a potential problem. Where the problem lies, is the next task to undertake. Is it with the observed data? The model internal variability? The model climate sensitivity? The actual vs. the projected forcings? Random chance? Any of these may be the case.
So, while the observed trends that ended in 2007 lie further towards the middle of the range than do the trends that end in 2008, this is of less interest than the observed trends that start to fall outside the 95% confidence range—because these trends potentially point to something being amiss somewhere—and this is what we are starting to see. The longer this period of reduced warming rate continues, the more the observed trends are going to start to fall outside the range of model expectations.
In case you want to see the time history of the observed values of trends of various lengths (the kind of things that Gavin and Bverheggen suggest we are hiding), they are available in a World Climate Report article we posted in December 2008 (this is prior to our model investigation):
http://www.worldclimatereport.com/index.php/2008/12/17/recent-temperature-trends-in-context/
Further, like I have repeated on many occasions, a group of us are actively involved in preparing our findings for submission to a journal, where we will have a much better opportunity to better describe what we have done as well as provide more analysis and results.
I am quite confident that what we are doing will provide a useful measure by which to gauge observed trends against model expectations.
-Chip