Guest post by Verity Jones
A Tamino rant aimed at Joe D’Aleo’s Arctic ice refreezing after falling short of 2007 record (also at ICECAP) has had me smiling. Tamino’s accusation against Joe of cherry picking are centred on one of the graphs originally posted here at DITC.
“D’Aleo tries so hard to blame Arctic climate change on ocean oscillations. Part of his dissertation includes a plot of “Arctic Region Temperatures”:”
“Do you suspect that these six stations were “hand-picked” to give the impression he wanted to give? Do you think maybe they were cherry-picked? If so, you’d be right.”
Well excuse me but of course they were cherry-picked, but not for the reasons Tamino suggests. If you really want to spit cherry stones, Tamino, chew on them first.
The graph was originally posted here: http://diggingintheclay.wordpress.com/2010/09/01/in-search-of-cooling-trends/ and here’s what I said about it then:
“Tony had found many climate stations all over the world with a cooling trend in temperatures over at least the last thirty years…
…
We were concerned that this could be seen as ‘cherrypicking’ … In many cases it was not just cherrypicking the stations, but also the start dates of each cooling trend.”
However, the story the post revealed wasn’t the one Tony wanted to tell from the original reason why the stations were chosen – the story that came out of the work was the unexpected (to us) cyclical pattern exhibited by so many of the stations across the world. The pattern matched more closely in regional stations – hence the closely grouped Arctic set in the graph above. So no, the stations in the graph weren’t meant to represent the whole of the Arctic (the original presentation of the graph is here).
But while we’re at it let’s look at a few more stations.
One of the reasons for choosing the stations we did in the graph above was the longevity of the record. This was something I had a look at in the Canadian Arctic too when comparing GHCN/GISS data and that of Environment Canada.
Tamino also berates Joe for not averaging/spatially weighting the data:
“He wants you to think that Arctic regional temperature was just as hot in the 1930s-1940s as it is today.”
If we want a simple comparison of the 1930s with the present we need stations that cover both time periods. In GHCN v2 for Canada, only one station (Fort Smith) has data in 2009 and also has data prior to 1943. Now Fort Smith is more than 1°C warmer on average in the five years 1998-2002 than in 1938-1942, but if we look at Mayo in the Environment Canada data set, it is only 0.275°C warmer in recent times when comparing averages of the two periods.
These are just two stations but such differences intrigue me. If you don’t compare like with like, how can you be sure there is no inadvertent bias? Are we comparing apples in the 1930s/40s with oranges in the 1990s and 2000s?
If we plot all the (GHCN/GISS) data (yes it’s another one of those ‘awful’ spaghetti graphs ;-0) – look at that big white gap under the plots from1937-1946.
Those years look pretty warm compared to 2000-2010, but unfortunately the data for Hay River, Mayo and Dawson does not extend to recent times. To do a comparison, you need to plot GISS and Environment Canada data together, and (as I showed here) there is a bit of a mismatch that needs to be overcome. In Dawson the 1940s are warmer; Hay River shows a slow continuous upward trend.
If you want to compare the two periods in Canada, unfortunately you mostly have to rely on combining stations, and methods for this are well documented (I’ll not go into detail here). What is still debated though is the magnitude of correction (if any) for urban warming.
Much as scientists are required to be objective, there is a need for subjectivity in looking at the surface temperature records. What has changed around this station? Why is one station producing a cyclical signal while another gives a near linear trend? Like I’ve said before, I’m a fan of a parallax view.
Canadian Arctic stations are mostly rural and very small settlements; they’re labelled as <10,000 population by GISS. Analysis by Roy Spencer showed the greatest warming bias associated with population density increases at low population density. Ed Caryl in A Light in Siberia compared “isolated stations” with “urban” where there was a possible influence from human activity. He found distinct warming trend in temperatures of “urban” stations where there were increasing evidence of manmade structures or heat sources. In contrast, he found little or no trend in “isolated stations”. Normalising the data for the isolated stations, he too produced a ‘spaghetti’ graph, which, lo and behold, also shows up that cyclical variation. Not only that, but for a lot of these stations (go ahead – call them cherry picked if you wish), the 1940s are clearly warmer than recent times.
Graph from: http://notrickszone.com/2010/09/24/the-calculations-behind-a-light-in-siberia/
So it is very clear to me that, in comparing station data, we’re dealing with cherries, apples, oranges, and probably a whole fruit bowl. Banana anyone? The problem is that Tamino and others insist on mixing it all up to make a smoothie. Now that’s OK as long as you like bananas.
============================================================
Stay tuned – Anthony
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
I wonder how hard it would be to document a warming bias in cold climates due to anthropogenic movement. Such as traffic. I have noticed on nights where it goes below -20F, that is is colder at my house than at the road, and colder at the road than at the highway, and colder at the highway than in town. It is also warmer at the top of a tall bridge than on the road surface on either side.
These kind of clear, wind still nights happen maybe once or twice a winter where I live,but I bet they are more common in the arctic, by far.
Well, tomatoes are botanically a fruit, but I would be glad if somebody who’d pick fruits for a fruit salad would leave them out of the bowl – he might be scientifically right, but he is not going to win any favours showing up with a fruit salad that contains tomatoes. I guess the team’s answer would be to put the whole fruit basket through a grinder and hope that nobody notices? And I couldn’t say if the picked fruits were any good, once they are in a slimy pulp, now, can I?
Theo Goodwin Thanks. I don’t suppose you know anyone who could put in a good word with Exxon do you?
Anthony – vivisectioning? Well I suppose the copious production of bile is a give away that were are talking about a living organism.
Doug in Seattle Perhaps Tamino’s invective is the only thing that keeps his blog stats up these days.
Kim ??? Oh ye of few words can be so cryptic
Juraj V. Thank you for the graphs. I still find the sine wave pattern amazing.
ClimateForAll said:
“Lets concentrate on improving the understanding of the current science and focus less on the detractors. Lets these alarmists continue to rant and rave. But let them rant to themselves. I would like to see the lot of us clean house and work towards defining the science”
My thoughts too. But sometimes I give into temptation.
Tonyb (Climatereason) Thanks – it was hard work at the time but a pleasure working with you nontheless. As ever I am in awe of your literary archaeology (if I may be allowed to term it so).
Bernie McCune I remember you mentioned this before and I’d still like to see what you did sometime.
A paragraph by D’Aleo is worth more than Tamino’s life’s work so far.
Robert says:
September 21, 2011 at 8:11 am
Verity Jones,
Tamino has responded
http://tamino.wordpress.com/2011/09/21/fruit-loops/
“Normalising the data for the isolated stations, he too produced a ‘spaghetti’ graph, which, lo and behold, also shows up that cyclical variation. Not only that, but for a lot of these stations (go ahead – call them cherry picked if you wish), the 1940s are clearly warmer than recent times.”
On what basis does Tamino claim that the graph by Joe D’Aleo is not credible? the curve shown above of Arctic temperature stations matches very closely the graph published by Levitus et al, 2009 (pdf here) showing Barents sea temperatures in the upper 150m. This correspondence gives strong credibiity to the data and graph by Joe. Tamino would by contrast need to show why the Arctic temperatures should depart significantly from Arctic sea surface temperatures (upper layer OHC).
See the WUWT article here:
http://wattsupwiththat.com/2009/10/08/new-paper-barents-sea-temperature-correlated-to-the-amo-as-much-as-4%C2%B0c/
Thanks Verity. You are reminding me of the piece I did that was published here, that evidently really got up Tamino’s nose because he honoured me with two rebuttals. Nick Stokes was jumping in there too, to try and bat me down. Hasn’t changed. Then I realized Tamino had handed me something I could use to strengthen my own argument.
Yes most of the Arctic records show it as warmer in the 1940’s than recently. And the Greenland ice record shows it was warmer in medieval times, warmer still in Roman times, warmer still in the Holocene optimum. Click my name and then click on the section labelled “-distorted” to see this.
IMHO it’s really important to NOT mush records together, but simply use reliable stations, that have (a) a really long record and (b) no real UHI issues – as did my mentor John Daly. However, as Verity suggests and Illarionov demonstrates, UHI effects can be most serious in the most sparsely-populated “rural” areas.
Lucy – I linked to your UHI post – “Bananas”.
We need to be careful with data and the way it is analysed. If comparing analyses, an essential question is: were the hypothesis being tested the same.
I looked at just the Mayo data, which is the longest of the time series. I did not transform the data, or calculate anomalies (which can introduce bias). After filling some missing months, which made no difference to the long-term series monthly averages, I tested for step-changes.
Step changes are not incremental changes, such as those usually investigated by regression approaches (although we can incorporate “steps” into regression); steps essentially expose external “shocks” to the time series. These could be station moves for example; changes in instruments or observation times; or changes in atmospheric circulation, such as summarised by indices like ENSO, DPO etc. Step-changes reveal inhomogeneities in the base series, and for a valid apples vs. apples comparison, data need to be homogeneous through time.
Clearly, if monthly values move in opposite directions, it is possible for annual data to remain the same. So I analysed month by month as well as annual data, and there is some nifty software available now to do this easily. I checked the variance – variability in period data contributing to the mean; as well as the mean; and when I finished, I checked the residuals after step-corrections, and they showed no trends.
I found statistically significant (P<0.01) steps in all months except September and November, and they were months that showed no linear trend over time either (slope not significantly different to zero).
Fitting any model that assumes data are serially coherent leads to false conclusions when in fact data are "stepped". In the case of Mayo, the same years ring a bell as they do in Australian data – there was a major climate shift in 1948 in both places; again in 1974 and 1996. 1948 was the most pronounced shift. For Australia, temperature changes were lagged relative to the rainfall shifts. It is also clear in Australian data that relationships between maximum and minimum temperature exhibit hysteresis – their trajectory is different depending on whether temperature is in a falling or rising phase.
For Mayo, the eyeball average temperature from 1938-1947 (3.72 deg C) may not be statistically different to that from 1997 to 2010 (3.99 deg C), and if it were, the difference (0.27 deg. C) would be immaterial. It is not a difference due to a trend, it arises because of a step. The later step corresponds to the 1998 ElNino.
My view, is that climate is a continuum; it has a bandwidth of values that it moves between, and that changes in its relative trajectory reflect passing interventions (eg. station moves; ElNino), rather than trends. That the residuals for Mayo showed no pattern after step-change effects were removed debunks the "trend" hypothesis, at least for the period of the data (1929-2010).
The final issue that I'd note is that for measured data there is an obvious limitation (bias?) resulting from the length of record. Most data ends post the 1998 ElNino in 2010-2011 (hot, dry); so that "end" of the data spectrum is fixed. The start date (year) is variable. Data may start for example on a low time step; thus giving a strong but misleading "trend" to 2010. It may start on a high time-step, that results in no trend at all. If we had comparable data going back far enough, we would probably be less likely to be able to support hypotheses relating to contemporary warming.
Bill Johnston
I’ve been aware of many of these issues, even if I’ve not had the facility to do such analyses myself, and your conclusions are very reasuring. I am intrigued by the dates you assert for the major climate shifts. We had previously used 1939 and 1969 in mapping trends here: http://diggingintheclay.files.wordpress.com/2010/08/multiple-maps.png?w=610&h=782. I have always assumed that, while at the time there was a data handling necessity for this, it was not ideal. My perception was always that such shifts would not necessarily manifest themselves in all regions simultaneously. Is this correct? Is there any reading on this that is readily accessible?
Instead of apologising for spaghetti graphs, how about stretching out the x axis a bit so that the wriggly lines are not so tightly jammed together? Then they would be a lot easier to understand.
phlogiston
Sorry, I overlooked your question earlier. Thanks for that link.
Tamino’s problem with the first graph in (this) post (and in fact all of them) seems to that the stations are cherrypicked and not geographically representative of the Artic as a whole (therein he insists they are wrong). In this he is not dealing with the issue I have raised in asking why these stations show the strongly cyclical pattern while other stations do not. In fact I suggest it is an issue that he would rather avoid.
Rattus:
St. Roch made 2 trips through the NW passage, 1943 and 1944 I believe. One trip tokk a grand toatal of 86 days PORT TO PORT. Light speed for the day. They encountered very little ice.
Verity Jones
There are several regime shift detection packages available that you could find. There are also interesting papers in the general field of “Nonsense regressions due to neglected time-varying means” mainly in econometric literature, but its an equally valid issue for climate time-series. (My long-ago training was in biometry, not mathematical statistics, so I’m not big on the maths part of this, but conceptually it is easily understood.)
An issue that I mentioned was that an “average” may mask what is going on within data. This is because across datasets, months (or stations) may dull or even cancel out step-changes that on a “unit” (month, or particular station’s data) basis could be real. This was why I analysed for individual months, as well as for step-changes in yearly means.
Obviously, the more values that contribute to a data average results in a smoother average. This increases the risk of masking small differences within stations, and stations within zones, that may be real and important.
For instance, in Australian data, if we averaged a parameter for stations in different climatic zones, say on the east and western side of the Great Dividing Range, we may find no time-steps in the average; but if we looked within zones we may find large steps. It would be cheating to claim an overall trend, when its components act to cancel-out real variation.
The other issue that may have slipped by is that statistical significance is not a reward in itself. The difference that it detects must also be relevant and important. In my mind, it matters little, for a station that is always hot; say 45 deg.C that its temperature increases or decreases by a statistically significant point-something of a degree. The same for a station that is very cold. So after a statistical test, there needs to be interpretation, and part of that should consider the relevance or otherwise of the data change/difference. In blogspace, there seems to be much more excitement about something being “significant” rather than if it makes any real difference. It would be a challenge for anybody to walk outside and detect a temperature change of much less than a couple of degrees.
As for spaghetti graphs, it is possible to standardize across multi-data, calculate an error estimate for the series at each time; and plot just one line with error bars. It is also possible to test for steps in combined data. I’ve seen this recommended as the 1st step in climatic time series analysis.
If there are no steps, then it would be OK to assume something about trends. Linear detrending of non-linear data can also result in spurious results. There is also the issue of data transformation to handle mean-variance correlation; serial correlation and so on. A full-blown analysis is somewhat more complex than just fitting a few trend lines in EXCEL and getting excited about a high R^2.
A point of clarification:
I was only looking at maximum temperatures for Mayo; and only to illustrate an important oft-neglected point. The question that I missed from Verity Jones, was:
“I am intrigued by the dates you assert for the major climate shifts. We had previously used 1939 and 1969 in mapping trends here: http://diggingintheclay.files.wordpress.com/2010/08/multiple-maps.png?w=610&h=782. I have always assumed that, while at the time there was a data handling necessity for this, it was not ideal.”
The real answer is: The data should tell you where the shifts occur.
It is easy to test; I have 2 excel add-ins that do it in seconds; and there are more sophisticated stand-alone packages; as well, most full-blown stats-programs can be programmed do it, but they are more complicated to drive.
If you test between stations, the similarity of detected “dates” should tell you something about how well they cluster around an average date; if they are scattered all over the place they could be nonsensical/not real. If they are not, then you can go off and look for reasons for changes at those times, either in the climate literature or using other data.
Cheers
Verity Jones says:
September 21, 2011 at 5:26 pm
phlogiston
Sorry, I overlooked your question earlier. Thanks for that link.
Tamino’s problem with the first graph in (this) post (and in fact all of them) seems to that the stations are cherrypicked and not geographically representative of the Artic as a whole (therein he insists they are wrong). In this he is not dealing with the issue I have raised in asking why these stations show the strongly cyclical pattern while other stations do not. In fact I suggest it is an issue that he would rather avoid.
How much scope is there even for cherry-picking of Arctic temperatures, even if you tried? How much variation is there in temporal trends from one location to another? In the Antarctic – a much larger frozen region – the western peninsula seems to behave differently to the main continent. But at the Arctic does the variation in temporal pattern even exist to provide the potential for cherry picking?
@ur momisugly Juraj V. – “Oh another academic here”…… What’s that supposed to mean? Personal remarks are seldom effective when debating a position.
@ur momisugly Disco Troop. -” Prof A. Net melting is not unarguable evidence for net energy input into the system. It can just as easily be evidence of the movement of the same energy from somewhere else, with the total unchanged.”
Yeah. I agree. Very good point. In an isolated system energy can move from one place to another. It’s the First Law. Are you suggesting planet Earth is isolated from the rest of the universe? Leaving that aside, if there is this movement of energy, thermal energy transfer occurs as a result of temperature difference and would result in a counterbalancing cooling somewhere else on the planet. Bearing in mind that the enthalpy of melting (latent heat) of ice is huge, one should see a warmer zone cooling substantially. One would also need to show how this energy flow from one part of the planet to the other has actually increased over time, giving rise to the progressive melting.
Of course if you agree that planet earth is an open system then we need a mechanism to account for the increased energy flow into the planet that is giving rise to melting, even if, as is being asserted by some, there is no temperature increase. As I said before, there is a mechanism on the table that many of us find very uncomfortable.
Verity Jones says:
September 21, 2011 at 5:26 pm
Tamino’s problem with the first graph in (this) post (and in fact all of them) seems to that the stations are cherrypicked and not geographically representative of the Artic as a whole (therein he insists they are wrong).
Quite so, since he was addressing d’Aleo’s post concerning the Arctic (not one quadrant of it), which used that figure, he is quite right to object to such cherry picking.
In this he is not dealing with the issue I have raised in asking why these stations show the strongly cyclical pattern while other stations do not. In fact I suggest it is an issue that he would rather avoid.</em.
You're right he's not addressing your issue because he's not referring to your post.
Tamino is an ignorant hick who thought Mannian Statistics was real mathematics; who thinks the Jones-Briffa ‘extensions’ to the ‘new field of Climate Math’ are real mathematics.
The fact everyone under the sun’s taken Mann’s algorithms and made calibration noise turn into graphing error using it means nothing to him. He’s so stupid he actually BELIEVES that the men who were FURIOUSLY CALCULATING DOOMSDAY with SCRIBBLES THEY COULDN’T TELL W.E.R.E.
N.O.T.
R.E.A.L.
M.A.T.H.
have nothing to be ashamed of, intellectually.
Tamino is the one who thinks the – and I mean every syllable of this I’ve followed this scam for 18 years – the ignorant Mann, referred to as a “Charlatan of Statistics” – who is actually nothing more than a geologist: not an atmospheric scientist, not a statistician of A.N.Y. note –
Tamino thinks
the HEAD
of the ROYAL STATISTICAL SOCIETY
was W.R.O.N.G.
when he said that Mann’s statistics “aren’t statistics” and that “for that matter no ‘Climate Math’ is mathematics of any kind.”
and that Mann, mediocre intellect without a talent for much but SCAM and statistical EPIC PHALE,
knows more of statistics than he.
So whenever you ask yourself how much sense Tamino has, R.E.M.E.M.B.E.R:
HE STILL THINKS AN ALGORITHM PROVEN REPEATEDLY to PRODUCE ERROR
is REAL MATHEMATICS,
and that the men caught red faced FURIOUSLY teaching it to ANYONE WHO AGREED not to REVEAL what they were doing,
are prophets of CREATIONIST Al Gore’s SCIENCE foray into doom and gloom.
Tamino: the CREATIONIST Al Gore’s,
mathematics speshULisT.
This is a man who thinks photons dragging entire gas molecules VIOLENTLY upward, emit DOWN if they emit from said gas. Have any one of those clowns explain that mechanical event to you. You’ll be laughing so hard you won’t be able to type.
This is a man who believes that James Hansen is an actual scientifically relevent voice, after his LEGENDARY propensity to be wrong if he so much as tells you a traffic light’s gonna cycle Red-Green-Yellow.
So – always be aware: an ignorant mathematician who can’t count is among the stupidest things to ever be cursed with capacity to wrap a paw around a pen.
Bill Johnston says:
September 21, 2011 at 6:42 pm
I appreciate your explanations and agree with you. My own foray into all of this began more than two years ago and the Arctic spaghetti graph must be some 18 months old. I do not believe the station data is in a state where we can confidently treat each single time series in the same way without the detailed examinations you describe (over and above simple corrections for time of observation bias and station history adjustments). I have since moved on to other analyses but rarely post on it these days due to lack of time (now) for the rigour of checking and completion, but I have learned a lot about the state of the data. I will look into “nonsense regressions…”. I see there is plenty readily available to investigate.
Despite previously bemoaning my lack of programming ability I actually now bless it because it has forced me to go the other way and, instead of dealing with large numbers of data sets and requireing to treat them all in the same way, look at individual stations and very small groups or pairs – in Excel. So I am very receptive to all that you say.
phlogiston says:
September 21, 2011 at 7:33 pm
How much scope is there even for cherry-picking of Arctic temperatures, even if you tried?
Good question. There are stations with no or small overall trend and plenty showing substantial warming. Why? – that’s an issue. Proximity to water (frozen, recently less so) is likely to be a factor. For example in Canada, Alert had data in GHCNv2/GISS only up to 1991, whereas Eureka and Resolute remained in the dataset. Alert, however, was frozen in for much nore of the year and showed less warming. See: http://diggingintheclay.wordpress.com/2010/04/25/canada-3-comparing-eureka/
How much variation is there in temporal trends from one location to another?
In the Arctic, offf the top of my head I’d imagine the onset of warming shows up soonest where currents are strongest. I don’t think there is much temporal variation say by longitude in the Arctic, but there is some varaition by latitude, can’t say exactly without a detailed look. IIRC there is a hemispheric see-saw (teeter-totter) going between the regimes at the poles in certain periods.
Verity Jones says:
September 22, 2011 at 2:14 pm
For example in Canada, Alert had data in GHCNv2/GISS only up to 1991, whereas Eureka and Resolute remained in the dataset.
Thanks for this insight – the paucity of data around the Arctic with the big fall-off of stations – and the related issue of stretched extrapolation of station data, narrows even thinner the possibilities for cherry-picking along with the limited “raw material” of trend variability. Seems Granny Foster chose a strange battle to fight here – no wonder he’s being evasive!