Arguments
Did Phil Jones really say global warming ended in 1995?
Posted on 16 February 2010 by John Cook
A headline in the Daily Mail has spread like wildfire, claiming that Phil Jones, ex-director of the University of East Anglia’s Climatic Research Unit, said "there has been no global warming since 1995". Not only did Phil Jones not say these words, this interpretation shows a poor understanding of the scientific concepts behind his words. To fully understand what Phil Jones was saying, one needs to read his actual words and understand the science discussed. Here is the relevant excerpt from the BBC interview:
BBC: Do you agree that from 1995 to the present there has been no statistically-significant global warming
Phil Jones: Yes, but only just. I also calculated the trend for the period 1995 to 2009. This trend (0.12C per decade) is positive, but not significant at the 95% significance level. The positive trend is quite close to the significance level. Achieving statistical significance in scientific terms is much more likely for longer periods, and much less likely for shorter periods.
BBC: How confident are you that warming has taken place and that humans are mainly responsible?
Phil Jones: I'm 100% confident that the climate has warmed. As to the second question, I would go along with IPCC Chapter 9 - there's evidence that most of the warming since the 1950s is due to human activity.
Phil Jones is saying there is a warming trend but it's not statistically significant. He's not talking about whether warming is actually happening. He's discussing our ability to detect that warming trend in a noisy signal over a short period. To demonstrate this, look at the HadCRUT temperature record from 1995 to 2009. The linear trend is that of warming. However, the temperature record is very noisy with lots of short term variability. The noisy signal means that over a short period, the uncertainty of the warming trend is almost as large as the actual trend. Hence it's considered statistically insignificant. Over longer time periods, the uncertainty is less and the trend is more statistically significant.
Figure 1: HadCRUT global temperature change in degrees Celsius. Blue is yearly average. Red is linear trend (HadCRUT).
It bears remembering that the HadCRUT record only covers around 80% of the globe. Analysis by European Centre for Medium-Range Weather Forecasts (ECMWF) and NASA GISS (Hansen 2006) find that the areas omitted by HadCRUT are some of the fastest warming regions in the world. Consequently, the HadCRUT record underestimates the warming trend, as demonstrated by the NASA GISS record which covers the whole globe:
Figure 2: NASA GISS Global temperature change in degrees Celsius. Blue is yearly average. Red is linear trend (NASA GISS).
However, even this doesn't give you the full picture. Surface temperature is only a small fraction of our climate with most of global warming going into the oceans. When all the heat accumulating in the oceans, warming the land and atmosphere and melting ice is tallied up, we see that global warming is still happening.
Figure 3: Change in total Earth heat content from 1950 (Murphy 2009).
Anyway, check out the full article here:
http://www.abc.net.au/mediawatch/transcripts/s2820429.htm
One of the comments most highly rated"(1510 votes) include: ...and so the scam of AGW begins to unravel!
One of the comments most lowly rated has 1079 votes against containing the phrase: ... humans ARE having an impact on climate change.
Any measured change or trend that is not statistically significant DOES NOT EXIST scientifically.
What "statistically significant" means is that any apparent change is smaller than the accuracy of the means for measuring the change.
Thus, scientifically, a trend that is not statistically significant DOES NOT EXIST, and must be treated as ZERO for the purposes of valid science.
Furthermore, the observed results are in conflict with the predictions of the computer models.
Therefore, the computer models are wrong.
Since we have no experimental (i.e., scientific) data whatsoever to support the idea of man-made global warming, and are relying only on the computer models, the invalidity of the computer models destroys the entire global warming alarmist argument.
Can I suggest you read the article to the end to see experimental data besides the HadCRUT record. Then I recommend Empirical evidence that humans are causing global warming.
Although this is not Phil Jones' fault, I do think he should have been much more careful with his wording. Even though what he said is completely accurate and not controversial, it is confusing to those with little scientific knowledge and easy to twist by those who want to get a good headline. Perhaps he should have gone into a bit more detail to avoid such confusion. He appears to have underestimated the quote mining ability of others.
Yes, but only just. I also calculated the trend for the period 1995 to 2009. This trend (0.12C per decade) is positive, but not significant at the 95% significance level. The positive trend is quite close to the significance level. Achieving statistical significance in scientific terms is much more likely for longer periods, and much less likely for shorter periods.
C - Do you agree that from January 2002 to the present there has been statistically significant global cooling?
No. This period is even shorter than 1995-2009. The trend this time is negative (-0.12C per decade), but this trend is not statistically significant.
in (B) - is positive for an upward temp increase and there is 'almost' statistical significance with 1995 to 2009. a +0.12 up trend.
in (C) - the 2002 to present shows a -0.12 down trend...but is not statistically significant because it is a shorter time frame.
now, i am not a mathematician but i think if you add +0.12 with -0.12 the answer would be zero. no net gain or net loss...exactly zero.
thus, how can things be warming +0.12 per decade and then cooling -0.12 per decade be interpreted as temperature gain?
remember that:
E - How confident are you that warming has taken place and that humans are mainly responsible?
I'm 100% confident that the climate has warmed. As to the second question, I would go along with IPCC Chapter 9 - there's evidence that most of the warming since the 1950s is due to human activity.
remember that 100% confident is still a 100% opinion. there is evidence that smoking can cause cancer but it is not 100% of time. is smoking good for any body? absolutely not. this is just an example that NOTHING in science is 100%. to say so is 100% subjective opinion.
The whole idea of measuring the "temperature" of the earth by averaging (area weighted) air temperatures 6ft off the ground (a proxy for surface temperature) with (area weighted) sea surface temperatures is just such a crazy idea.
I realize the OHC has been hard to come by and even now deep ocean heat is a problem.. but still.
The heat capacity of the atmosphere is 1/1000th of the heat capacity of the ocean.
are making too many errors in reporting climate science.
Also, I refuse to believe their errors are bad luck, it seems
more like harassment by an overactive anti IPCC lobby
group. In the Netherlands it is the Telegraaf and Elsevier
that show this behavior. My policy is to avoid those
media, moreover if you need to say something then insist
that they start with a written press release and that you
get to see their text prior to publication.
Two other points worth mentioning: 1. the significance level chosen for rejecting the null hypothesis is somewhat arbitrary; 2) choosing a particular null hypothesis entails making assumptions. Regarding the first point, Dr. Jones used the conventional 95 % threshold. What if he had selected 90 %? The trend might well be significant at that level. One the second point, choosing to test for whether the climate has warmed over the past 15 years (null hypothesis: there has been no change in global temperatures over the past 15 years) does not seem like a particularly useful test, given that climate is typically defined as the average conditions over 30 years.
Finally, many branches of science (e.g. applied ecology) are moving away from hypothesis testing, in large part because of the aforementioned arbitrariness, and increasingly relying on Information theory, predictive ability, and related approaches.
Just so that I'm clear on this....the HADCRUT data, and the data Jones refer to, are measurements of both land and marine surface areas. The chart you have on the bottom includes energy accumulated in deeper areas of the oceans, which are not included in the other measurements.
Is this correct? I was a little unclear on this point. If anyone else knows the answer, feel free to confirm.
Also, what is the reason that temperature of the oceans is not included along with the rest. Is this because of inadequate monitoring, or for some other reason?
Thanks,
The global temperature indices are only measuring surface temperature of land and sea. The top few mm of the ocean.
So the HADCRUT, GISS, etc don't measure Energy (Let's see if html tags work..)
Which is a problem because that's really what everyone is interested in (everyone in climate). Obviously surface temperature is a function of energy but in very derived and convoluted ways.
For example, during ENSO events, colder water gets pushed up to the surface and we say "global temperatures have gone down".
In fact, probably the total ocean energy is the same, and more likely to be going up for a few months or years because less heat is radiated out from the oceans now the surface is colder.
The reverse is true.
Because we only measure the surface temperature we are measuring quite a random or unpredictable variable. Especially as the oceans are 70% of surface area and therefore weighted to 70% of global temperatures.
But it's hard to measure OHC (ocean heat content). I read a climate guru recently (can't remember his name) saying that they had more confidence in global temperature measurement than OHC measurement. Although he implied but didn't really say whether consequently it was more use.
The Argo project now has about 3000 sensors around the oceans collecting temperature and salinity data down to 2km. They started being deployed late 90's but really have only fully been in place for a few years.
And they don't measure below 2km.
In the past there were lots of measurements of vertical temperature profile in the oceans done using XBTs but then someone found out that they had some biases (don't really know the story on that one) and consequently a lot of the old numbers that had been collected have issues.
I think that once we have our hands around OHC much more of the climate will be understandable. This should also be the measurement that is the headline number and graphic everywhere, not the "surface temperature". Just my opinion.
This is a general statement, and completely true, because the uncertainty in the rate will be higher for shorter periods. He did not make any claim about what the significance for 2002- beyond 2009 would be if there were more data recorded. But if you took earlier data, say 1980-2009, then the data would show a warming trend at a very high level of confidence, easily surpassing the 95% confidence level that is usually used.
Another consideration is the issue of whether or not one is testing a one-tailed or two-tailed null hypothesis. The link above (hopefully it works) shows a two-tailed confidence interval. I don't know if Jones' test was one- or two-tailed, but really it should have been one-tailed. A one-tailed test is more powerful (more able to reject the null hypothesis).
Almost finally, independent from the issue of signal versus noise, there is the issue of what's special about 15 years for this particular test. Hypothesis testing requires that the test be constructed a priori. It's invalid to watch as data are added, going forward in time, and then to make simple conclusions (ignoring the non-independent tests that have already occurred) when a calculated p-value finally goes above or below 0.05. That is, you can't cherry-pick the time frame.
Finally, JonMoseley in #3 might do well to construct another hypothesis test: Ho = the rate of warming does not differ significantly from a rate of 0.2 C/decade. Failing to find significance there means, if you apply Moseley's funny interpretations of statistical rules, that differences with consensus expectations of AGW do not exist.
If you want to clarify the recent apparent slow down of temperature increase a better question would be:
"is the recent (apparent) slow down of the temperature increase statistically compatible with the long term trend?"
The answer is a sound yes.
In turn, it is wrong to use the 1998-2009 trend - without taking into account the impact of EN(LA)SO.
2. Marcus repeats: the sun "expire" ... and the warming continues. However, the present warming is consistent with the millennium cycle.
I recall a diagram: http://www.rni.helsinki.fi/research/info/sizer/fig2big.jpg; and I am adding a new: http://mclean.ch/climate/figures/GISP_to_4Kbp.gif. Present Warming is greater in NH than the SH. Rahmstorf has repeatedly said, that the Millennium cycle is dependent on the cycles of the sun - but not directly. So far, created a few theories to explain this relationship.
"Any measured change or trend that is not statistically significant DOES NOT EXIST scientifically."
No, that is incorrect (but a very common fallacy). If a measurement is not statistically significant it means there is insufficient evidence to reject the null hypothesis. In this case it means the trend is too small (w.r.t. the noise) to be confident that the trend is not zero. That is not at all the same thing as "the trend does not exist scientifically".
Of course one would not build an argument solely on a non-statistically significant trend, but then again AGW theory has a much broader base of evidence and nobody (who understands statistics) would claim that a short term trend is proof of AGW.
"Furthermore, the observed results are in conflict with the predictions of the computer models.
Therefore, the computer models are wrong."
Actually that is not correct either, see the paper by Easterling and Wehner, which shows that the internal variability of the climate means that there will be occasional decadal trends showing no warming or even cooling, and that these are seen in the output of the models.
http://www.agu.org/pubs/crossref/2009/2009GL037810.shtml
A common demonstration of the Dunning-Kruger effect in discussions of climate is a lack of understanding of what the models actually say.
As I understand it, your precision may be near perfect, but the system may be intrinsically 'noisy' (eg El Nino is a real physical phenomenon, but in the context of looking for warming it contributes to 'noise').
So your precision may be enough to detect a trend, but without knowing how to fully account for all of the noise, you must treat it as such and determine based on the known high frequency noise in the data whether it is statistically significant or not.
So Jones is right; there is a trend, but there is a 5%+ chance that it could have occurred from noise. Which is unsurprising, given that an ENSO switch is 0.4-0.7C in a few months and 'global warming' is a trend of 0.1-0.2C/decade atm and so people tend to talk about longer time periods.
Please correct me if I'm wrong and point me towards where I can fix my understanding.
The reason a shorter time frame is less statistically significant for the same trend is that noise tends to average to zero at longer time scales. this also demonstrates the effect.
Also, it's less likely that noise will always 'add up' to a trend over 20 years than it will do so for a shorter time period. If you think of it as flipping a coin each time with heads meaning noise goes warmer, tails cooler, then it's clearly less likely and you can claim higher statistical significance.
As for +0.12 -0.12 = 0, that's not how it works in trends. If you take 1995-2009 you get +0.12K/dec and this includes the 'cooling' period from 2002. It just so happens that 1995-2002 is a trend of +0.26K/decade.
And you can't add trends anyway unless they're for equal time periods. Eg if you raise temperatures at +0.2C/decade for 50 years you end up 1C warmer than you were before. If you then cool at 0.2C/decade for 10 years, you'll only cool by 0.2C and you'll still be 0.8C warmer than you were 60yrs ago. The 60 year trend will remain positive, even though +0.2 - 0.2 = 0.
"To be valid science, one must state that a measurement that is not statistically significant is equivalent to ZERO."
The thing to remember about statistical tests is that "absence of evidence is not evidence of absence". If you treat the trend as equivalent to zero you are saying that the lack of statistically significant evidence for a positive trend is evidence that the trend is zero, which is not correct. This is precisely the assumption you make when you criticized the models
"Furthermore, the observed results are in conflict with the predictions of the computer models."
For there to be a model-data conflict, the error bars of the observed trends need to be outside the spread of the modelled trends. If you did the comparison properly there is no conflict.
Would you have a link to either a Non Paywall version or a discussion of either or both of Murphy 2009 and Von Schuckman 2009.
Many thanks
40 Shades
If 15 years without statistically significant warming is not enough to give one pause, then how many years would be required. 20, 25, 30, 50, 100?
It's like watching the last 14 waves that came in on the beach and trying to work out whether the tide is coming in or out. The variation in height of the last 14 waves does not give one enough information to produce a statistically significant trend (even less so the last 8 waves). The noise (waves) hides the trend (tide). After ten minutes and a hundred waves the direction of the tide becomes very apparent.
I wonder why Phil Jones didn't think to use such a clear analogy? The lay audience would have got it in one and those with a basic scientific understanding would/could not have questioned the way he expressed himself.
there's no a priori time span. You need to perform statistical analysis on the actual data and determine it.
Here's a draft of Von Shuckman et al. 2009..
Basically, the theory is that cyclical circulation of colder deep ocean water and warmer surface waters has a strong impact on climate. So, if the PDO, as one of the larger cycles observed, is pushing alot of cold water to the surface that is going to have a cooling effect on the atmosphere. Such is often used to explain cooling trends from 1880-1910 and 1940-1970. Then the cycle reverses and ocean heat is transferred to the atmosphere, causing warming.
However, looking at the OHC graph it seems like there was a drop from 1960-1970... the same time the ocean was supposedly 'sucking up heat' from the atmosphere. Then from 1970-2000, when the ocean was supposed to be releasing heat into the atmosphere, OHC went through the roof.
Does this disprove oceanic heat transfer as a major driver of climate? I suppose greenhouse warming could have overwhelmed ocean cooling from 1970-2000, but why did ocean heat decline from 1960-1970? Was there a solar decline in that timeframe which would explain both the atmospheric and oceanic heat declines?
I think you missed the point of '40 Shades of Green's' question. He's (she's?)asking if someone with the required knowledge has done the statistical analysis on the actual data, and what the result was? A fair question I would say.
As a lay response I would say there's no magic level of data that must reveal a trend hidden in noise. Imagine looking at a newspaper photograph at a high level of magnification under a microscope: with a few dozen dots (data) visible you wouldn't have a clue what the picture was about. Now reduce the magnification. At a certain point -- and number of dots (data) -- one can start to make out the image and have a stab at what's being revealed. The lower the magnification, the more dots (data) and the more certain you can be of what you're looking at.
I hope that helps.
DIkran's description:
http://sciences.blogs.liberation.fr/files/dix-ans-de-froid-dans-un-si%C3%A8cle-chaud.pdf
Now you can delve into the weeds about significance levels, time periods, datasets, overall trends, caveats, blah, blah, blah ... but the text speaks for itself:
BBC: Do you agree that from 1995 to the present there has been no statistically-significant global warming
Phil Jones: Yes
Now if you want to start adding caveats to ALL global warming headlines ... that would be fine. Like the IPCC in its 2007 report said: "Eleven of the last 12 years rank among the 12 hottest years on record (since 1850, when sufficient worldwide temperature measurements began)."
That last parenthetical statement seems to go by the wayside a lot. And maybe a mention of the Medieval Warm Period, and the Holocene maximum would be helpful. And satellite temps versus ground records that omit quite a bit of the earth's surface.
Somehow the IPCC infactuation with 12 years of data was important and significant ... but the same years found to be not statistically significant don't really mean much of anything now.
Any questions?
"One final point, though not statistically significant, temperatures rose by 0.012 degrees per year between 2000 & 2009, yet average sunspot numbers fell by around 14.5 per year."
O.K.
Cyclical changes in solar activity do not (simple) coincide with the cycle of the Millennium. Agreed. But now we have a typical growth phase temperature in the Millennium cycles in a typical (for him) periods. It was not me saying that this cycle in some way depends on solar activity, and Rahmstorf. In many, many works and commentaries.
What you're confirming is that Phil Jones is a straightforward, uncomplicated (and perhaps politically-naive) scientist, rather than a politician with a lawyer's brain.
Given that it's agreed by the vast majority of scientists that the world has been on a warming trend since temperature recording began, would it not be totally illogical for Phil Jones to mean, "there has been no warming since 1995"? Clearly, if he thought like a politician, he should NOT have said, 'Yes'; but should have said something like, "It has not been possible to separate a trend from the noise over that period".
Surely logic suggests that we must assume that the long term trend continues, given that the noise over the last fifteen years has obscured any statistically-meaningful variation from that temperature trend, either higher or lower?
If one of my students said he had collected some temperature readings but the location of the the readings was uncertain then I would have to disregard them.
It should now be obvious that the disputed Chinese readings have to be disregarded.
#41, oracle2world. Just one question. First, read post #21. If the warming trend from 1995-2009 is not staistically significant, but the trend from 1994-2009 IS statistically significant, as is the trend from any year prior to 1995 through 2009, what do you think the headline should be?
#43, John Russell, I got a similar impression of Jones from the transcript. He was spin-free. Certainly it would have been better for him to have said, "the trend from 1995-2009 is not significant, but the trend from X-2009 is. The reason for the difference is that 1995-2009 is too short a time span." But we were not the ones sitting there in front of the TV cameras -- it's easy to second-guess.
I have an intersting graph based on current GISS data and a paper by Hansen et al, 2006, which I would like to post. This shows that we are currently tracking below the zero-emissions trajectory for global warming.
I Hope that you understand why I remain sceptical about the disputed Chinese readings.
No doubt the issue will be examined by each of the two inquiries.
Is it accurate?
'There has been no warming since 1995' is a very different statement from 'there has been no statistically significant warming since 1995'.
In Jones' set there is a warming trend, and every year after 2001 has been warmer than 1995. There has been warming.
It is not statistically significant because a certain combination of known noise could recreate that trend in 5%+ of cases.
Of course, if you think my stats understanding is wrong, point me to where I can read up and fix it.
Nature online 16/2/2010