One of the most well-known graphics from Skeptical Science is the escalator. It illustrates how contrarians tend to cherry-pick short time intervals so as to argue that there’s been no warming, while “realists” recognise the reality of long-term warming.
A classic example of the former was the so-called pause, which was based on claims that there had been no warming since the strong El Nino in 1998. Of course, as illustrated by the escalator, there really hadn’t been a pause in global warming.
The Skeptical Science escalator is, however, now slightly dated, with the latest version ending in 2015. I also noticed that Robert Rohde had presented an updated version on Twitter, which he called the staircase of denial.
This motivated me to produce an updated version for Skeptical Science, which I’ve also posted below.
...and Then There's Physics 
Thanks, ATTP. That’s an excellent example of Simpson’s Paradox.
A more interesting example, however, is looking at the same data but instead using an actual structural breakpoint analysis.
Since this is not done by “cherry picking” but by mathematical analysis, this brings up the interesting question … what is causing the jumps?
Other than the relationship of some of these to El Nino events, I have no idea about the answer.
Thanks for your interesting blog. I often disagree with the ideas propounded therein, but they are almost always thought-provoking.
Best regards,
w.
Willis,
Even if there are “jumps” this doesn’t somehow challenge global warming. The key point (as I suspect has been pointed out to you before) is that the entire climate system is warming – atmosphere, oceans, cryosphere, land. This is because the increase in greenhouse gases in the atmosphere is reducing the outgoing long-wavelength flux, causing the system to warm in order to return to energy balance. That this may include – for example – the energy being stored in one part of the system for a while before being “released” to suddenly warm other parts (producing jumps) doesn’t really challenge this basic understanding.
Mine’s not animated, but it does draw from the four major temperature anomaly series.
I coded this both for the “Monckton method” of working backward to find the longest non-positive slope. Interestingly, for GISS LOTI going backward from the beginning of the most recent step, there is no period of negative slope.
Also notable is that UAH is the preferred series for finding long pauses.
…and Then There’s Physics says: February 2, 2023 at 8:10 pm
===
“Willis,
Even if there are “jumps” this doesn’t somehow challenge global warming.”
===
Please point to where I said that the jumps somehow “challenge global warming”.
Oh, wait, you can’t, because I SAID NOTHING EVEN REMOTELY RESEMBLING THAT.
Not sure what you are fantasizing about, but it’s not about what I said.
I identified the issue in question as being an example of Simpson’s Paradox. Then I put up a non-cherrypicked graph and asked a question.
That’s all.
w.
Willis,
Past experience. If you’ve suddenly recognised the reality of anthropogenically-driven global warming, then that would be a significant step forward.
I said nothing either way about “anthropogenically-driven global warming”. And it has nothing to do with what I wrote. Please stop introducing your fantasies and irrelevancies and deal with WHAT I ACTUALLY SAID.
Or not, you can keep making things up, trying to change the subject, and avoiding what I posted.
Your choice.
w.
“Since this is not done by “cherry picking” but by mathematical analysis, . . . ”
The two are not mutually exclusive.
Willis,
My point is that in terms of the big picture, it doesn’t really matter what’s possibly causing the jumps. Given that the surface temperature measurements represent a region that makes up only a small fraction of the total climate system, it isn’t surprising that there is quite a lot of variability (you don’t see anything like this in the ocean heat content). The whole context of this is the idea that variability in the surface temperature dataset doesn’t challenge that basic point that global warming is happening and that it is predominantly driven by human emissions of greenhouse gases.
What is also evident is that ‘piecewise linear plus occasional jumps’ is terrible as a predictive model, just like ‘fit a high order polynomial would be’. It has very little ‘skill’ even if you can figure out how to make a prediction.
If you were using it to make bets on future climate, you would tend to lose quite a lot.
Just fitting random curves to data tends to have that effect, which is why having some idea of the underlying physical process is helpful…
Willis asks “what is causing the jumps?”
Good question. If the non-continuous line segments of Willis’ “mathematical analysis” are correct, then global temperature is capable of huge instantaneous jumps. Considering the heat capacity of the atmosphere – and the heat capacity of the ocean mixed layer that it is linked to over period of a decade or so – this can only happen from a huge instantaneous input of heat.
I think we would have noticed such an input. Unless Scotty beamed it in?
The explanation is that Willis’ “mathematical analysis” and “actual structural breakpoint analysis” has nothing to do with the real world. The non-continuous line segments are not representative of any real physical mechanism or process or environmental condition. There is nothing that could possibly change global mean temperature (anomaly) from 0.75 to 1.0 in an instant (at 2014.9+/-0.1 on Willis’ graph), so the “mathematical analysis” is producing line segments that cannot possibly represent an actual pathway of global temperature over time.
Now, if Willis wanted to do an analysis that required all line segments to connect into a continuous set – i.e. the end points of each segment meet – he might be able to claim some sort of real physical phenomenon. Until then, he’s just deluding himself.
The ups and downs – the variability – is driven by real climate relevant processes. Sea surface temperatures, notably ENSO related, features high and there are examples of temperature series adjusted for known variables in order to better identify underlying trends.
Rahmstorf, Foster and Cahill, 2017, in reference to the supposed “hiatus/pause” post 1998 –
“By physical arguments, by model simulations, or by correlation analyses with additional data (e.g. El Niño/Southern Oscillation indices or solar forcing data) it is possible to identify specific physical causes of temperature fluctuations, and this is a fruitful topic of ongoing climate research (Foster and Rahmstorf 2011, Kosaka and Xie 2013, England et al 2014, Suckling et al2016) which helps us to understand natural climate variability. However, this is distinct from the question of whether a significant trend change has occurred in the temperature data as such. That is not the case. It is unfortunate that a major public and media discussion has revolved around an alleged significant and unexpected slowdown in the rate of global warming, for which there never was a statistical basis in the measured global surface temperature data.”
Of course “The Pause” diminishes a lot just by choosing 1997 or 1999 as the start point – and would be less disingenously misleading for
notchoosing something midway between the known up and down fluctuations rather than the (then) record warmest year as the start; ordinary fluctuations soon after (unsurprisingly) look like cooling as a consequnce.As the argument went it was like choosing a record warmest day in Winter and declaring the axial tilt theory of seasons false and predicting there will be no Summer because of a week in Spring that was cooler than that day.
ClimateAdam haz a relevant hypothesis:
Willis –
Please point to WHERE ANDERS SAID THAT YOU SAID THAT THE JUMPS SOMEHOW “CHALLENGE GLOBAL WARMING.”
Oh, wait, you can’t, because HE SAID NOTHING EVEN REMOTELY RESEMBLING THAT.
Not sure what you are fantasizing about, but it’s not about what he said.
He identified the issue in question as that “skeptics” cherry-pick time periods to make it seem as if there hasn’t been a continuous warming when in fact there has been.
Then he put up a non-cherrypicked graph and made a point.
That’s all.
Joshua says: February 3, 2023 at 3:58 am
Joshua, I repeat my post:
Joshua, my problem was simple. I said NOTHING about “global warming”. So why is ATTP claiming that my discussion of “jumps” doesn’t challenge global warming? It’s a blatant attempt to sidetrack the discussion.
Also, I have no clue who “Anders” is. I was responding to ATTP, whoever that might be. Is he “Anders”? And if so, why doesn’t he sign his own name?
Regards,
w.
Willis –
My problems is simple. Anders didn’t say that you challenged global warming. So why did you imply that he said that you did? You claiming that he said that is a blatant attempt to sidetrack the discussion.
But that’s not even your only attempt to “sidetrack the discussion.” This post was about how some “skeptics” cherry-pick dates in order to make it seem like warming isn’t happening when in fact if you don’t cherry pick dates it’s clear that warming is taking place.
Why did you repeatedly blatantly try to sidetrack the discussion?
And Willis –
Maybe instead of wasting your time pearl-clutching, you could tell me whether you agree that what appear to be effectively sudden “jumps” in surface temps are relatively irrelevant with respect to global warming, as they reflect variability in just a small fraction of the earth’s climate system, and if you look at the larger system there aren’t any such effectively sudden “jumps.”
I assume you agree with that, am I right?
I ask that question because Judith Curry, in Congressional testimony, referred to a “pause” in global warming.
That seems too me too be highly misleading.
Rather, what she was calling a “pause” in global warming was actually time-limited variability in a small fraction of the earth’s climate system within a longer-term, ongoing and steady trend of warming within that small fraction of the climate system, as well as a longer-term, ongoing and steady tend of increase in the larger system as a whole.
Do you agree that her testimony was misleading in that sense? I’ve brought this question up many times at her blog and haven’t been able to get any “skeptics” to respond. I’m hoping that maybe you’ll respond.
Experience suggests the inertia of Wills disbelief in the calculus, integration, and the ramifications of scale in geophysical systems may prevail over any inconvenient facts adduced here, witness his preference for quackery over pharmacology in the Vaxx Wars.
The incredulity he inspires is a feature rather than a bug, as naively taking his advice may be perilous, or life-threatening:
https://vvattsupwiththat.blogspot.com/2020/04/of-quinine-and-chloroquine-willis.html
Regarding a graph showing short term mini-trends, I think rather than ask “what causes that” it would be interesting to ask “What are the odds of NOT seeing that kind of pattern on a triple-point water planet with continents, volcanism, surface plants and a star with short-term activity cycles?”
The different factors aren’t synchronized so they’re all waving the temperature up or down all the time, on their independent schedules and strengths. It’s no surprise that factors might reinforce for a bit, cancel for a bit, and so on. That’s a big reason why “trends” in surface temperature < 30 years are pretty much ignored as uninformative and not worth paying too much attention to (unless you're talking about weather, not climate).
When Hansen gave his June 1988 testimony in Congress it would have been a lot easier if all the different factors operated with zero variability – in that case, even "Structural Breakpoint Analysis" with minimum of 7 years would show a straight line with nothing but greenhouse signal changing it. Congress would have said "OK, that's very clear!"
But he said this: "In all of these cases, the signal is at best just beginning to emerge, and we need more data. Some of these details, such as the northern hemisphere high latitude temperature trends, do not look exactly like the greenhouse effect, but that is expected. There are certainly other climate factors involved in addition to the greenhouse effect."
So to answer "what causes these breakpoint lines" my guess is "normal climate noise, don't try to make much of it". But do note that the long-term trend is up.
Back in the old days with just a few TV channels that shut off at night, I'd stare closely at the black and white static to try and "see" something. Other than COBE, it was really pretty much just random bits of black and white. I would have been surprised if it were just randomly a flat gray instead.
Is Izen in the house? If so, do you remember this “prediction” of yours from 2016?
Willis – As you are well aware, I have been persona non grata at WUWT for many years. Perhaps you could pop in there on my behalf, and ask Chris Monckton what he makes of your structural breakpoint analysis?
Willis
“Please stop introducing your fantasies and irrelevancies and deal with WHAT I ACTUALLY SAID.”
he did answer your question, but you chose to ignore it.
Here’s his answer:
“. The key point (as I suspect has been pointed out to you before) is that the entire climate system is warming – atmosphere, oceans, cryosphere, land. This is because the increase in greenhouse gases in the atmosphere is reducing the outgoing long-wavelength flux, causing the system to warm in order to return to energy balance. That this may include – for example – the energy being stored in one part of the system for a while before being “released” to suddenly warm other parts (producing jumps) doesn’t really challenge this basic understanding.”
It’s pretty clear…
Taking an 11-year running average, as suggested by Hanson, smooths most of the variation
Thanks AT.
Willis: “Since this is not done by “cherry picking” but by mathematical analysis, this brings up the interesting question … what is causing the jumps?”
Since a dataset with a random nose added to a linear trend would exhibit the exact same results according to your “mathematical analysis”, what is causing the “jumps”?
Your imagination Willis.
Either that, or you’ve just disproved the First Law of Thermodynamics.
Nathan says: February 3, 2023 at 11:10 am
Here’s his answer:
Thanks, Nathan. That’s just a complex way of saying “It’s natural processes”. It doesn’t tell us what the processes might be, where to look for them, or what evidence exists that might identify them.
So yes, it’s natural processes … but then, everything about climate involves natural processes of some kind. So it’s an answer that answers nothing.
Thanks,
w.
Willis,
Huh? The key point is that the reason the climate is warming is because of human-caused emissions. That there is variability in a dataset that measures the temperature in one part of the climate system doesn’t really challenge that conclusion.
Willis, you have used monthly data with a non-negligible autocorrelation, which means that breakpoint analyses will tend to produce a lot of false-positives (identifying spurious breakpoints). What did you do to avoid that problem?
“what is causing the jumps?”
ENSO (and other forms of internal variability), volcanic forcing and incorrect statistical assumptions about the noise would be my (educated) guess. The first two of which we have known about for over a decade, so I am somewhat surprised that Willis doesn’t know.
…and Then There’s Physics says: February 3, 2023 at 6:50 pm
Last time I looked, humans were part of nature.
But you miss my point. The answer is still a non-answer until the actual processes causing the breakpoints are identified. Yes, it may just be the nature of random fluctuations … but if that’s the case, why do some of the breakpoints line up with the El Nino variations?
Regards,
w.
Willis,
Who said random? It’s not surprising that the “breakpoints” line up with El Nino variations. They do indeed produce variations in surface temperatures.
Thanks, Dikran. I’m using the same data as used in the head post. And yes, including autocorrelation means that all of the shorter sections are not statistically significant.
I discuss this question over on Christopher Monckton’s WUWT post here.
w.
Grrr … the link didn’t work. Here it is again.
w.
“Last time I looked, humans were part of nature.”
not that old canard?
“The answer is still a non-answer until the actual processes causing the breakpoints are identified. ”
That is assuming the breakpoints are meaningful and not spurious. As a statistician, I would not make that assumption. The reliability hierarchy is
physics > statistics >= chimps pulling numbers from a bucket
note the >= we should take statistical findings with more than a pinch of salt.
Willis,
That is just a not very subtle way to argue by ignorance.
The point you are missing is that the onus is on you to explain what you found, and more importantly to make sure that your answer is not an artefact of your method.
If humans are part of nature, unicorns are too.
Willis, you did not answer my question. What did you do to deal with the autocorrelation (note it will lead to spurious statistically significant breakpoints because the autocorrelation means that there is less information in the data that you would estimate from the number of datapoints). If the answer is “nothing”, then just say so and we will have identified the likely cause of the jumps.
Double grrr … still not working … it will only accept the link to Monckton’s post and not to my comment on the post. Go figure.
Search for “Willis”, my comment is right near the top.
w.
BTW I am not going to read posts at WUWT, that particular echo chamber has made it very clear I am not welcome there, including a post about me resulting from criticising Moncktons silly “no warming since [cherry picked start point], that totally failed to understand the criticism. Sorry, life is too short.
BTW the easiest solution is to find the breakpoints in the annual time series, which has far less autocorrelation. Oddly enough, whenever I have tried it, it has far fewer breakpoints [generally around zero IIRC].
dikranmarsupial says: February 3, 2023 at 7:25 pm
Sorry, I thought I was being clear. In my comments here I did nothing about autocorrelation, just like ATTP did in the head post.
However, I discussed it in my comments over at Moncktons post, q.v.
Thanks,
w.
Willis,
You seem to be illustrating the whole point of the escalator graphic. Short term variability doesn’t somehow suggest that there’s “no warming”.
dikranmarsupial says: February 3, 2023 at 7:28 pm
I didn’t invite you to read Monckton’s post. I invited you to read my comment on his post.
Me, I do my best to read all sides of each question, including regularly reading but rarely commenting here at ATTP … but hey, you do you.
w.
WE sorry, I am not interested in pedantry, I am not visiting WUWT, for the reasons I gave. That includes comments. I used to visit WUWT and comment there, so I know from experience there is nothing of value there that I can’t get from other places without the rude behaviour.
Willis “Sorry, I thought I was being clear. In my comments here I did nothing about autocorrelation, just like ATTP did in the head post.”
Thank you for being so clear. In that case, I can tell you that the results are essentially meaningless because the autocorrelation isn’t taken into account in assessing the statistical significance (or otherwise) of the breakpoints. I have in the past set this as a mini project for MSc statistics students (as it is a fun problem), if you apply standard breakpoint detection algorithms to this sort of monthly data, you will get lots of spurious breakpoints. So you can either look for breakpoints in annual data where autocorrelation is less of a problem, or develop a breakpoint detection algorithm that reliably deals with the autocorrelation (which is not straightforward).
If anyone is interested, I could find the comments.
Please beware your wishes.
BTW I suspect Tamino has covered this several times with great clarity.
“Willis – As you are well aware, I have been persona non grata at WUWT for many years. Perhaps you could pop in there on my behalf, and ask Chris Monckton what he makes of your structural breakpoint analysis?”
Jim Hunt might instead ask the Editors of Teen Vogue to put the question to Nigella Lawson or Rosa Monckton on his behalf.
Don’t underestimate the hard-hitting journalism going on at Teen Vogue.
https://www.teenvogue.com/story/donald-rumsfeld-secretary-defense-accused-war-criminal-dead-88
It isn’t a one-off, either, they are regularly doing a better job than big-name newspapers.
Hm:
Hmmm…
Oh, my. Willis sure is insistent in continuing what Willard has called “argue by ignorance”.
Willis continues to focus on the “jumps”: “…until the actual processes causing the breakpoints are identified.”
The actual processes causing the breakpoints is bleeding obvious: they are artifacts of the ignorant application of a statistical model that has no physical meaning.
Earlier, upthread, I pointed out that in the real world the jumps would need huge energy inputs to create an instantaneous “jump”. Let’s pursue that a bit further. What are the typical month-to-month changes in the BEST data set that is being “analyzed”?
The following graph shows the month-to-month change for the BEST temperature (anomaly) series. As we can see, it is possible for monthly averages to change by a fair amount, but most of the time we only see small changes. Although the graph looks like random noise there is a strong serial autocorrelation in the anomalies themselves – the r^2 at a one-month lag is 0.88, and an anomaly is most likely followed by another anomaly of similar magnitude.
We can see more if we look at the distribution of the month-to-month change in the anomalies. A clear, strong central tendency, with large changes being quite uncommon.
I won’t graph it here, but if you look at how this month’s change in anomaly compares to last month’s (i.e., this month minus last month, versus last month minus the previous month), you find a negative correlation – if last month’s change was negative, then this month’s change is more likely to be positive (and vice versa). That’s why the first graph looks so noisy: it’s constantly flipping back and forth across zero.
So, what does this have to do with Willis’ breakpoints? Well, Willis sees the breakpoints because his model has breakpoints. Yes, Willis, you’re looking at a model, not reality. You may call it a “structural analysis”, but it is a statistical model. Just because your model has breakpoints does not mean that the real world has breakpoints.
So, we need to examine the assumptions that went into your model. What are they?
1) When the model finds a breakpoint, there is no requirement that the line segments connect. The model can have “jumps”.
2) The breakpoints need to be at least 7 years apart.
3) In between breakpoints, the model can only change slowly, at a constant rate (the slopes of your linear segments).
What do the real-world data show? Well, they certainly do not show that there is any tendency for the month-to-month change to remain constant for any significant length of time. In my first graph, can you see any signs of periods of constant month-to-month change? No, you cannot.
Does the data (first graph) show any propensity for very slow, small changes over time, interspersed with sudden jumps? Not if you look at the second graph, which clearly shows a typical bell-shaped curve. If Willis’ model of “steady small change with jumps” was really in the data, we’d expect to see some sort of signs of that tendency in the distribution of month-to-month changes. We don’t.
The line segments in Willis’ graph also show a variety of slopes, from -0.12 to +0.07. And they vary from 7.1 years in length (just above the assumed minimum) to 13.1 years. What is the physical cause of these periods of slow, steady change in Willis’ “structural analysis”? There isn’t one. In fact, given that the slopes change and the lengths change, you’d have to come up with six different “physical processes” – one for each different line segment.
So, the breakpoints are a statistical artifact – and so are the line segments.
As you search, Willis (and I’m sure you will), don’t forget to try to come up with a physical process that explains the “jumps” and the cause of your model’s line segments. Because they both come out of the same “structural analysis”. They are both essential parts of your model. You need to prove an explanation of both before your model has any credibility.
As is shown in The Escalator, when you look at the entire time series – not segments – you do get a consistent change in the anomalies that has a physical explanation – the radiative effects of rising greenhouse gases.
> The actual processes causing the breakpoints is bleeding obvious: they are artifacts of the ignorant application of a statistical model that has no physical meaning.
>> Earlier, upthread, I pointed out that in the real world the jumps would need huge energy inputs to create an instantaneous “jump”.
Me? When I see a direct reponse like that to something I’ve written, I respond in good faith. But that’s just me. Willis may well avoid answering in one fashion or another, maybe with some passive aggressive pearl-clutching thrown in for good measure.
‘Cause Willis will do Willis.
I must be psychic Willis:
https://skepticalscience.com/escalator_2022.html#140220
I’m also pleased to see that you followed my suggestion, and what’s more your heresy hasn’t got you “banned” yet!
Russell/Ben,
I typed the term “Arctic” into the teenVogue search box. The most relevant headline offered was “North West’s Special FX Makeup Is So Good That It Even Has a TikTok Warning”.
I don’t suppose either of you know Nigella’s mobile number do you?
Jim, why is “Arctic your go-to search” term for fashions in denial ?
Bob, Teen Vogue has been on the ClimateBall radar since the day they signed Meghan Markle as Guest Editor:
https://vvattsupwiththat.blogspot.com/2019/07/new-conde-nast-climatogy-journal-for.html
Uh-oh:
Big hmmm…
DM,
Tamino has covered this most recently in 2017 …
Global temperature evolution: recent trends and some pitfalls
https://iopscience.iop.org/article/10.1088/1748-9326/aa6825
“3.2. Broken trends problem
The discussion has so far used broken (i.e. discontinuous) trend lines. This is a further problem of many past analyses, also tending to enhance the (in this case false) impression of a significant slowdown.
Figure 1(a) shows a model with broken trends applied to HadCRUT4 data. A naive statistical analysis suggests that the change is real because the two linear segments have significantly different slopes.”
So, as far as I know, the use of a 7 year minimum, or any minimum length of reasonably short durations (say less than ~15 Years to as small as say a few years) with discontinuous steps, will produce staircase like steps (if the long term trend is positive, as is the case here).
Both WE and ATTP’s are indeed staircases to a very 1st approximation.
What is worse, in this case, is that there are no statistical tests to show if the individual steps have significance or not (they do not and that is their very point, to show a series of flatter lines with discontinuities in between).
So, other than a rehash of standard denier talking points (e. g. the paws), there is nothing new here.
In fact, several of Tamino’s posts (can’t find published papers at this moment) have looked at multivariable analyses with indices such as volcanic activity and El Nino which tend to better explain the underlying long term warming trend caused by us humans.
> Setting aside the question of whether these estimates are hopelessly contaminated by urban warming (most probably they are),….
The “jumps” what about the “jumps?” Did urban warming cause the “jumps?”
Can’t oblige, Jim, but her father’s climate secretary is at 0207 340 6038
Apropos:
Good morning Russell (UTC).
Thanks for the info. However I am already in possession of the contact details of the “Clerk to His Lordship”. I was politely informed many moons ago that:
“Lord Monckton has asked me to state that his private correspondence is not for public circulation (being contrary to European human-rights law)”.
Arctic sea ice decline is my tallest soap box, and to be fair to teenVogue I should point out that their allegedly second most relevant “Arctic” article does mention that:
“The market for the natural products and cures that claim to bring us closer to nature by removing toxins from our bodies or keeping them from ever entering in the first place, is an understandable reaction to the fact that we live in a deeply contaminated world. Microplastics lurk in the deepest oceans and even Arctic ice melt can be too contaminated to drink.”
I suspect Willis may have moved on, but this isn’t an example of Simpson’s paradox either AFAICS as the “groupings” are just spurious interpretations of the noise.
It is more of a reverse-Simpson’s paradox as some (e.g. many denizens of WUWT) mistake the spurious groupings as meaningful physics, and downplay or ignore the meaningful physics is in the long term trend to claim things like “no global warming since [insert cherry picked start date]”.
But Da Paws:
Source: https://wattsupwiththat.com/2023/02/03/the-new-pause-lengthens-again-101-months-and-counting/#comment-3675400
Last edited 19 hours ago.
Willard, the odd time period made me look ;o)
In that case it is a synthetic time series, which by construction doesn’t have any genuine breakpoints in it, and yet the algorithm finds them anyway, which rather made my point for me. ;o)
I am not at all surprised that WUWT still has not learned that a lack of statistically significant trend does not mean there is no trend. It is because it is an echo chamber and they won’t listen to those willing to explain it to them, it’s hardly rocket science:
https://skepticalscience.com/statisticalsignificance.html
I think this article is a pretty good example of Teen Vogue climate coverage:
https://www.teenvogue.com/story/climate-change-mitigation
It also neatly illustrates how different the discourse is between the generations in mainstream media. Really not pulling any punches about, for example, who is responsible. Sneer at them for being woke, but they’ll just wear it as a badge of pride.
Willis said: “Since this is not done by “cherry picking” but by mathematical analysis”
Getting mathematical tools to do the cherry picking for you is still cherry picking. Optimisation is the root of all evil in statistics. Whenever you optimise some criterion you introduce the possibility of over-fitting the sample of data over which the criterion is evaluated (i.e. exploiting the properties of the noise in the sample, rather than it’s underlying structure). The breakpoint detection algorithm is picking cherries, it is searching for the breakpoint positions that give the lowest loss (i.e. the ripest, tastiest cherry).
Similarly Monckton has an algorithm for cherry picking the start point, but it is still cherry picking. His algorithm is selecting the start point that maximises the strength of his argument (at least for a lay audience that doesn’t understand the pitfalls).
“In that case it is a synthetic time series, which by construction doesn’t have any genuine breakpoints in it, and yet the algorithm finds them anyway, which rather made my point for me.”
Note that the minimum has changed from 7 years to 7.2 years (why) and we have no idea how little trend WE added (why), for example. Arbitrary? You betcha! Just as Monkers choice of dataset is not arbitrary, which time series gives the longest current zero trend? It use to be RSS but now it is UAH.
Three La Nina’s in a row, bahahahahahahahahahah!
“Getting mathematical tools to do the cherry picking for you is still cherry picking.”
Its simply a case of cherry-picking the tool. Out of the many different tools to choose from, Willis has chosen one that has his “conclusion” (that breakpoints exist) as one of the key assumptions.
His admission that he has no idea why the breakpoints are there is an admission that he has no idea why there may or may not be breakpoints to begin with. Since he started the process with no physical hypothesis that would cause breakpoints, he has no justification for using a statistical tool that assumes there must be breakpoints.
Willis chooses the tool because he’s convinced (without previous evidence or physical reason) that breakpoints must be there. He’s assumed his conclusion. For Willis, there just absolutely has to be a pony in that pile of data.
Everett – you said “In fact, several of Tamino’s posts (can’t find published papers at this moment) have looked at multivariable analyses with indices such as volcanic activity and El Nino which tend to better explain the underlying long term warming trend caused by us humans.”
The following paper does a pretty good job of evaluating some of the natural causes of bumps and boosts that create the short term swings. It’s in Geophysical Research Letters and figure 2 is a very nice graph separating out the time-series for ENSO, volcanos, solar cycles and anthropogenic factors.
“How natural and anthropogenic influences alter global and regional surface temperatures: 1889 to 2006”
Judith L. Lean, David H. Rind
First published: 16 September 2008 https://doi.org/10.1029/2008GL034864
There’s also the climate impact of economic recessions, but I can’t remember where I’d seen that one.
Another Year of Record Heat for the Oceans
https://link.springer.com/article/10.1007/s00376-023-2385-2
https://www.dailymail.co.uk/sciencetech/article-11619553/Ocean-temperatures-hit-record-high-2022-data-shows.html
Their I go changing the subject again! 😀
b fagan, the bit I found interesting was this:
If I am reading Fig 3d right, it is suggesting that anthropogenic forcing is having almost no effect beyond 60-70 degrees North/South, and yet that is the region that has been observed to be warming most quickly. I’ve only skimmed the paper but AFAICS it doesn’ say what is causing the warming there.
It is somewhat ironic that they attribute the disparity between their statistical model and the physical models used by the IPCC as being due to a lack of modelling of fundamental processes in the physical models! Of course it is not conceivable that it may be due to inadequacies of their very simple statistical approach!
I don’t really like the idea of independent models at each grid point, in practice the parameters of the models at neighbouring grid points should be very similar, and that probably ought to be taken into account in a joint fitting of all of the models together.
It is also important to remember that regression can tell you that Y can be explained by X, but it can’t tell you that Y is caused by X. So it is not encouraging to see a preference for a statistical model over a physical model (which also has to explain the strength of the effect).
Everrett said:
Fluid dynamics often seeks staircases and stratification as lowest energy configurations. The plateau or linear riser of a stairstep means that that the 2nd derivative is zero and thus more stable. What’s happening at the steps themselves is related to the seasonal predictability barriers of ENSO — similar to what happens with lake turnover, density differentials are at a minimum twice per year and this is when metastable mixing is triggered, transiently bringing stratified cold water closer to the surface. That’s essentially what causes the huge La Nina or El Nino temperature variations.
So why are the large steps seemingly at 7 years instead of annually? Since the triggering is metastably sensitive, it would seem that other geophysics behaviors are at play. For example, the Mf tidal periods when aliased against an annual cycle will reveal 3.8 year periods. The other strong tidal cycle Mm will alias to 3.9 year periods.
Take another look at those breakpoints and you can see how ~2 steps can squeeze in between the markers. But that’s only qualitative — to get at a quantitative view, a stairstep pattern can be generated by calibrating to LOD variations.
Middle stairstep pattern in the figure below:
However, this is just random fractional Gaussian noise plus a linear trend.
Does that mean what I think it means?.
is this a significant “pause”?
Maybe someone can help me out. Even in the “skept-o-sphere,” does anyone offer an explanation for how the underlying physical causes of a long term trend of warming might “pause?”
I suppose there might be an argument that there is no trend. Or an argument might be that there are temporarily countervailing forces that swamp the long term trend. But I don’t see how, in any reasonable way, that could be described as “pause.”
Joshua, I think it is a synthetically generated time series designed to be superficially similar (in a statistical sense) to the actual observations. So noise plus a trend, i.e. there are no genuine breakpoints in the data by construction.
I suspect by “is this a significant ‘pause'” Willis may have been indicating some skepticism about the validity of the breakpoint detection algorithm as it gives spurious breakpoints in artificial data and so is likely to do the same thing with the observations (I like to be as charitable*).
* however, that is inconsistent with Willis’ question upthread “what is causing the jumps?”, so YMMV on that.
Joshua, ISTR that some climatologists were using “the pause” to refer to a discontinuity between models and observations, presumably without intending it to mean that there had been a fundamental change in the physical processes. So it could be an element of misunderstanding/misappropriation of a technical usage of the term? Rather unfortunate, but there is a history of unfortunate naming, e.g. “greenhouse effect” ;o)
Jim , Nigella Lawson is not a Monckton , albeit her brother Dominic married one.
As Rosa is generally in the Grenadines this time of year, you might ring Basil’s Bar around sunset, Atlantic time.
Hi Dikran, I’m not really qualified to critique their math or modeling approach, but I think part of the reason behind the low Arctic impact in their analysis is that they aren’t looking at the specific feedbacks that add to arctic amplification – so the contribution of those they did review could well be diminished in higher northern latitudes.
What I felt was helpful to the discussion here of short-term variance was the individual breakouts of year-by-year impacts of each of the effects they did study, so the various pieces that answer Willis’ “Gee, how could a signal be noisy” question are all there, accentuating or cancelling each other out as the years moved along.
Earlier I’d put in a quote from Hansen’s 1988 testimony, and I’d left out the sentences that he immediately followed with: “Some of these details, such as the northern hemisphere high latitude temperature trends, do not look exactly like the greenhouse effect, but that is expected. There are certainly other climate factors involved in addition to the greenhouse effect.”
So I can’t critique Lean and Rind’s conclusions vs. all those they claim got it wrong, but the high northern latitude warming is accentuated beyond GHG by lots of other feedbacks.
b fagan
There isn’t anything particularly wrong with their model AFAICS, it was just the headline in the abstract that I thought was interesting, the rest of it was mostly that I would have taken a different approach, but that doesn’t mean it wouldn’t produce the same result!
dikran –
> (I like to be as charitable*).
Being charitable is good, so let’s go with that.
> * however, that is inconsistent with Willis’ question upthread “what is causing the jumps?”, so YMMV on that.
Yes. This is what’s galling.
I started to list all the physical questions that would remain to be answered if there really were “pauses” and/or “jumps.” But they are so numerous! And I’m quite sure that despite great certainty that “pauses” and “jumps” can be seen in the data answers to those questions are not offered.
I’m certainly no statistician but it seems to me that without plausible physical explanations for what’s causing warming to “pause” or “jump,” it’s most parsimonious to assume that “pauses” and “jumps” are artifacts of statistical methodology.
For example, Willis seems to think that urbanization explains some significant portion of a warming trend (if we can assume he really does think there’s a warming trend). Well, is there any evidence of “pauses” or “jumps” in urbanization? If there isn’t, is there any reason to think that the other cause(s) of warming would magically actually go negative during the “pause” period, enough to compensate for the ongoing warming from the urbanization?
If I didn’t know better, I’d say that “skeptics” are conveniently inventing stories to “sidetrack” from the evidence that shows an ongoing trend of warming, with a known physical explanation.
Can anyone make an ‘escalator’ that shows short periods of rapid warming with discontinuous jumps between those periods of sudden cooling ?
Would it have any more ‘validity’ than the alternatives shown here of cooling or little warming with discontinuous jumps of rapid warming between them ?
izen,
That’s an interesting idea. I’ll try and have a go.
Joshua,
I think part of the problem is that the internal climate variability (the “noise” from a statistical trend perspective) is also caused by real physical processes which are also intrinsically interesting and worth studying. But these are largely superimposed on the long term trend. All statistics can tell you is that the hypothesis that the underlying rate of warming has changed is plausible (given some assumptions), but not confirm that it actually has. So largely people are not thinking about the physics of the noise (e.g. ENSO is redistributing energy) and being “all-or-nothing” about whether the underlying rate of warming has changed. Generally evidence isn’t “all-or-nothing”, mostly it just alters the balance of plausibility.
So it is reasonable to say that there has been a pause in GMST if the redistribution of energy due to ENSO means that the additional energy has been transferred to the oceans rather than being visible at the surface. However, that doesn’t mean that there has been a change in the underlying rate of global warming (as that energy is still accumulating in the climate system, and ENSO will eventually redistribute it to the atmosphere at some point). The “no rise in GMST” interpretation doesn’t require unrealistic transfers of energy, or that GHGs have stopped being GMSTs, and it is not just a statistical artefact (as there is an identifiable physical cause – ENSO). But then again, under this interpretation, it isn’t telling us anything much about global warming or what we should do, which is the general reason for bringing up these questions.
Talking about NHSTs though gives the discussion an air of “scienceiness” that you wouldn’t get by saying “look at the data”, but in actuality it is not really doing much more than that in this case.
Tamino discussed this effectively, but careful breakpoint analysis (i.e. not WE’s chart junk), and general curve fitting, directly addresses the idea of “parsimony” using “information criteria” like AIC and BIC.
The idea is to penalise “excessively complex models” that have too many parameters (and are thus prone to overfitting), because they will always fit the data better than simple models with fewer parameters, even if “what is really going on” is just a linear trend+noise.
Autocorrelation plays a big role here because you need to know how much ‘information’ is really in the data, and not mistake a smooth lump of noise for an actual change to the underlying trend.
Continuing an examination of Willis’ “structural analysis”, and considering Ben’s points about parsimony and information criteria… just how much of an improvement of fit does Willis’ six-line-segments-with-breakpoints give us?
This won’t be an exact duplicate of Willis’ values, because I am just breaking the BEST data into 6 segments using the years/periods in Willis’ graph,and then doing regressions, but here are the results of looking at his six segments:
End year Standard Error P-value
1976.8 0.131 0.07
1986.9 0.141 0.764
1994.7 0.141 0.250
2001.8 0.137 0.376
2014.9 0.119 0.005
2022.9 0.137 0.126
The standard error is, of course, the spread of the residuals around each regression line segment. It give a good idea of how much remaining “unexplained” variance there is after performing the regressions. The second last segment is a slightly better fit than the others (it is also longer), but overall we are seeing roughly 0.14 C as an overall standard error. Only one segment shows a significant result – the second last one that shows 0.07C/decade warming.
Note that the P-values above make no account for serial autocorrelation in the data, so take them with a grain of salt.
How does this compare to doing a single regression on the entire period? That is, the linear fit that represents “Global Warming” in The Escalator? Well, for Willis’ period from 1969 to current, the standard error for a single regression is 0.145. The result is highly significant.
That’s right. Willis’ six-segment “structural analysis” reduces the standard error by roughly one one-hundredth of a Celsius degree, and gives us five segments that are not significant as individual regressions.
How much complexity is involved in Willis’ “structural analysis”?
* six slopes
* six intercepts
* five breakpoint years.
…for a total of 17 parameters.
What is the complexity in the single “Global Warming” regression? Two parameters: the slope and intercept.
With 15 extra parameters, Willis improved the “explanation” by about 0.01C – or less than 0.001C per parameter.
The fact is that Willis’ analysis provides literally no improvement on a simple regression, at greatly increased complexity. Nearly all the signal in the data is the result of the overall gradual warming, and breaking it into several “jumps” has virtually no explanatory power.
Hmm. WordPress seems to have merrily reduced the horizontal spacing in my simple text table to a single space between columns. Let’s see if a paste direct from a spreadsheet looks any better.
End Year Standard error P-value
1976.8 0.131 0.07
1986.9 0.141 0.764
1994.7 0.141 0.25
2001.8 0.137 0.376
2014.9 0.119 0.005
2022.9 0.137 0.126
Nope. No better…
WP’s parser interprets any number of spaces as one. This is strange to us frogs who are used to two spaces between sentences. The consensus is now one space.
As for tables, the TABLE markup is now deprecated because it leads to accessibility issues. Not sure it’s possible to use CSS calls in a comment.
Let’s try TABLE nevertheless:
I did not type it by hand, but used a converter:
https://www.tech-spy.co.uk/webtools/text_to_html_table/index.php
Thanks, Willard. Actually had tabs in there, which WordPress obviously treats as spaces. It’s all white space, in the end.
Autocorrelation is one of those terms that has different ramifications (and even definitions) depending on the discipline. In physics, scientists spend careers trying to find an autocorrelation in a phase-change, indicating long-range order. In other disciplines it has almost a bad connotation and you often find a statistician requested to remove autocorrelation in a time-series. For me I need to know exactly what the person is after when they request to reduce or remove autocorrelation. For example, a signal could be a sinusoid of a given frequency and after removing the autocorrelation you are left with nothing — as the sinusoid is perfectly autocorrelated with a lag equal to the period. Were they after the sine wave in the first place? Give it to a statistician and he will remove it, maybe no questions asked, and you may be left clueless.
Same can be said for cross-correlation or pair-correlation. Could be that two time-series have a common factor (not truly independent) and removing that is critical to determining some other relationship. Yet, removing it blindly when there could be a causative effect is not what you want either.
As an example, for my chart of ENSO against a tidal forcing model posted above, there is an excellent cross-correlation (with a low-DOF out-of-band cross-validation) and a complicated autocorrelation. The cross-correlation is encoded by a nonlinear solution of the fluid dynamics while the auto-correlation is scrambled to some extent by the same nonlinearity. You would not want to give this to a statistician and suggest them to remove an autocorrelation in the time-series because that would be the end of the road to finding anything interesting.
Perhaps that’s why machine learning often works — it’s not removing relationships based on preconceived notions but discovering the patterns, only leaving you to decode the nonlinear physics leading to the complexity. Therein lies the rub — infinitely many more nonlinear possibilities exist than linear and so there’s lots of exploring yet to do if you want to understand why the ML worked so well. I know what works for fluid dynamics so am one step ahead.
There will be legitimate uses for finding statistical step changes – for example where there are actual physical causes to account for – but in the presence of ongoing global warming any shortish period (under 1 decade it seems to me) with an el Nino near the start and a La Nina near the end will probably look flat or possibly cooling. It looks like that describes most of those “warming stopped” periods. And the other way around will look like rapid warming, a big, fast “step” up that the deniers don’t want to draw attention to, not when there is a flat or cooling period to draw attention to.
Without global warming the same el Nino to La Nina picks will appear to be cooling, in about equivalent amounts to the other way around – between a La Nina and an el Nino. Because of global warming they are not in balance.
Of course warming or cooling in this is about Surface Air Temperatures, that are highly responsive to sea surface temperatures. Ocean Heat Content would be my preferred global temperature change indicator – showing ongoing heat gain within the climate system – and as pointed out in a comment above there is a lot less variability from year to year than SAT’s. Having two years running without hitting a new record high is exceptional. Having temperature records that go back to the 1800’s may have made SAT’s the go to data set to find evidence of change but there is enough variability to allow the determined to mislead the uninformed with.
I note that ATTP’s Skeptical Science example isn’t demonstrating the legitimate use of step change but demonstrating the flawed use.
Easy-peasy dm
Willis is Just Asking Questions.
Lack of self-consistency, failure to posit a testable hypothesis, and ignoring the laws of physics are features, not bugs.
Dave_Geologist well, he seems to have disappeared and done nothing with the answer to his question, which is indeed consistent with JAQing.
For what it’s worth, I ran a number of temperature anomaly series through the Foster/Rahmstorf 2011 code (period 1980-present) to nominally remove effects of El Nino, Aerosols, and Solar Irradiance effects. The “adjusted” time series were then run through the “escalator” code to produce the plot (hopefully) linked below.
Unsurprisingly this gives a trending series with lower variance and autocorrelation resulting in more and smaller steps on the escalator and supports the idea that the “pauses” are largely a function of these “natural variation” components.
The longest remaining “pause” is in the UAH data set from about 1997-2008 which some have suggested is residual from the satellite orbital drift correction on the long-lived NOAA15 vehicle.
For comparison, here are the “unadjusted” time series. The blue “steps” are running the “longest negative trend” algorithm backwards from present and the green are running it forward from 1980.
Mark,
Very interesting. Thanks.
Bob L: that’s a neat worked example, thanks.
I like Mark B’s graphs, too. A couple of things to note.
The use of “adjusted” data sets – and having that lead to more, smaller steps – could also be used to argue “see, I told you something was causing those jumps!” Although the adjustment was based on solar variability and El Nino as known processes, it’s still open to the argument that “oh, not, those are the wrong processes – there is still something else all the climate scientists are missing – my {insert favorite crank theory here}. Beauty is in the eye of the beholder.
More importantly, if a method is actually detecting something real, there should be some robustness to the method. You should not be getting different results when seeing slightly different inputs. Focusing on the unadjusted data sets, there are two obvious features:
1) Although the different data sets are part of the same climate/weather system, the “breaks” that are identified can be quite different in timing and size. The longest “pause” is in the UAH data set, after the late 1990s spike. That spike is a spike – within a couple of years, the anomaly has gone very negative, and there is no indication that the spike is showing a long, slow decay that persists over more than a decade. The “pause”, shown by the linear segment, also starts well before the spike. To claim that “something” starts the pause first, then causes a spike that dissipates rapidly, and the pause continues for another 15 years is to create a complex explanation tat looks like a just-so story.
2) Running the algorithm forward and backwards can give quite different results. The change in direction means that noise (variations in the data, away from the trend) gets introduced at different stages – either early in the “I think I’m in a pause” part of the process, or near the end. A strong indication that the steps/pauses are an artifact of the noise, not a real feature. Of course, to the True Believer in “jumps”, the argument will be “that’s not my analysis method”. (I would not wait for them to actually show that their method does not suffer from the same problem, though. They’d actually have to understand the criticism, first.)
These SSW events might be an artifact of increased global warming. The frequency and strength of these atmospheric disruptions might mark inflection points in the escalator pattern.
SSW event forecast to happen by Feb. 17th.
Somebody should launch a giant weather balloon with all kinds of sensors to collect this rare and important data.
https://www.severe-weather.eu/global-weather/sudden-stratospheric-warming-polar-vortex-collapse-effect-forecast-february-march-united-states-europe-fa/
I wouldn’t waste my time with artificially constructed escalator time-series when nature creates them without our help. Consider the tidal forcing assisted up and down escalator that likely contributes to the long-term variation in AMO. Highly accurate long-period tidal calibrations are available from length of day (LOD) measurements since 1962 and when plugged into the fluid dynamics formula, the 60-year cycle in AMO can be accounted for.
As nobody else seems to have done it, and just for fun….
izen,
Very nice 🙂
Increasing signs that the 3-year Nina is ending. Likely that Nino returns producing another escalator step-up this year. Per chart, should get close to 1.5C, perhaps a tic shy.