E.M. Smith over at the blog Musings from the Chiefio earlier this month posted an analysis comparing versions 1 and 3 of the GHCN (Global Historical Climate Network) data set. WUWT readers may remember a discussion about GHCN version 3 here. He described why the GHCN data set is important:
There are folks who will assert that there are several sets of data, each independent and each showing the same thing, warming on the order of 1/2 C to 1 C. The Hadley CRUtemp, NASA GIStemp, and NCDC. Yet each of these is, in reality, a ‘variation on a theme’ in the processing done to the single global data set, the GHCN. If that data has an inherent bias in it, by accident or by design, that bias will be reflected in each of the products that do variations on how to adjust that data for various things like population growth ( UHI or Urban Heat Island effect) or for the frequent loss of data in some areas (or loss of whole masses of thermometer records, sometimes the majority all at once).
He goes on to discuss the relative lack of methodological analysis and discussion of the data set and the socio-economic consequences of relying on it. He then poses an interesting question:
What if “the story” of Global Warming were in fact, just that? A story? Based on a set of data that are not “fit for purpose” and simply, despite the best efforts possible, can not be “cleaned up enough” to remove shifts of trend and “warming” from data set changes, of a size sufficient to account for all “Global Warming”; yet known not to be caused by Carbon Dioxide, but rather by the way in which the data are gathered and tabulated?…
…Suppose there were a simple way to view a historical change of the data that is of the same scale as the reputed “Global Warming” but was clearly caused simply by changes of processing of that data.
Suppose this were demonstrable for the GHCN data on which all of NCDC, GISS with GIStemp, and Hadley CRU with HadCRUT depend? Suppose the nature of the change were such that it is highly likely to escape complete removal in the kinds of processing done by those temperature series processing programs?….
He then discusses how to examine the question:
…we will look at how the data change between Version 1 and Version 3 by using the same method on both sets of data. As the Version 1 data end in 1990, the Version 3 data will also be truncated at that point in time. In this way we will be looking at the same period of time, for the same GHCN data set. Just two different versions with somewhat different thermometer records being in and out, of each. Basically, these are supposedly the same places and the same history, so any changes are a result of the thermometer selection done on the set and the differences in how the data were processed or adjusted. The expectation would be that they ought to show fairly similar trends of warming or cooling for any given place. To the extent the two sets diverge, it argues for data processing being the factor we are measuring, not real changes in the global climate..The method used is a variation on a Peer Reviewed method called “First Differences”…
…The code I used to make these audit graphs avoid making splice artifacts in the creation of the “anomaly records” for each thermometer history. Any given thermometer is compared only to itself, so there is little opportunity for a splice artifact in making the anomalies. It then averages those anomalies together for variable sized regions….
What Is Found
What is found is a degree of “shift” of the input data of roughly the same order of scale as the reputed Global Warming.
The inevitable conclusion of this is that we are depending on the various climate codes to be nearly 100% perfect in removing this warming shift, of being insensitive to it, for the assertions about global warming to be real.
Simple changes of composition of the GHCN data set between Version 1 and Version 3 can account for the observed “Global Warming”; and the assertion that those biases in the adjustments are valid, or are adequately removed via the various codes are just that: Assertions….
Smith then walks the reader through a series of comparisons, both global and regional and comes to the conclusion:
Looking at the GHCN data set as it stands today, I’d hold it “not fit for purpose” even just for forecasting crop planting weather. I certainly would not play “Bet The Economy” on it. I also would not bet my reputation and my career on the infallibility of a handful of Global Warming researchers whose income depends on finding global warming; and on a similar handful of computer programmers who’s code has not been benchmarked nor subjected to a validation suite. If we can do it for a new aspirin, can’t we do it for the U.S. Economy writ large?
The article is somewhat technical but well worth the read and can be found here.
h/t to commenters aashfield, Ian W, and rilfeld
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
GHCN V3 produces two data sets – a qcu file of unadjusted data and a qca file of adjusted data. I couldn’t see stated, either here or at Chiefio’s, which obe he is talking about. AFAIK V1 only had unadjusted, and V3 unadjusted has changed very little. It looks to me as if the comparison is between v1 and v3 adjusted.
It’s the unadjusted file that is the common base used by BEST etc, although Gistemp has only recently started using the adjusted file (instead of their own homogenization).
@Mike Jonas:
A very long time ago I looked at GHCN airport percentage over time. IIRC it ended at about 80-90% of current stations at airports. They are often called “rural”…
So, as you point out, comparing grass fields in 1800 to tarmac today surrounded by cars and airplane engines running is considered “rural” comparison… Yeah, right…
@Barry:
The data series used are also posted, so you can just grab the data and do whatever you want in the way of comparisons. In the article here:
http://chiefio.wordpress.com/2012/06/01/ghcn-v1-vs-v3-1990-to-date-anomalies/
but down in comments I post the same data with a sign inversion that makes the comparison a bit easier. (That is, you don’t need to invert the sign yourself to get cooling in the past to be a negative value).
Like I’ve said: The data, methods, code, etc. are all published.
Even has the individual monthly anomaly values so you can do ‘by month’ trends if you like.
Oh, and the reports have a thermometer count in them too, so you can see when the numbers are in the single digits and the results are less reliable.
E.M.Smith says (June 21, 2012 at 7:33 pm): “FWIW, I’ve got a version where I cleaned up a typo or two and got the bolding right posted here:
http://chiefio.wordpress.com/2012/06/21/response-to-paul-bain/
with a couple of ‘lumpy’ sentences cleaned up a bit. It’s a bit of a ‘rant’, but I stand by it.”
Wow. Epic!
I don’t visit that site enough…
Reblogged this on luvsiesous and commented:
Anthony Watts and E. M. Smith point out how data is re-imaged, or re-imagined, in order to give it a semblance of integrity.
It should be that the data is original, not that it is re-duplicated.
Wayne
@Benfrommo:
That is exactly why I put all of it on line and available. So folks can replicate and find “issues”.
My point being that the changes in the versions introduce bias, and we are depending 100% on “other methods” to remove that bias. That the tool used is prone to responding to that bias is then used as an attack? Rather funny, really. So I’m saying “The other method looks like it is about 1/2 effective” and the complaint is “my approach finds 100% of the issue”… Go figure…
FWIW, I’m happy to soldier on and make yet more methods of comparing v1 to v2 to v3, and fully expect they might find ‘less’ difference. To the extent I’m perfect at it, they would find zero difference. But my whole point is just this: To the extent other methods are in any degree less than perfect, they will have SOME of the difference leak through. As that difference is a whole degree C in places like Europe, even a 50% of perfection method will have 1/2 C of variation from the changes in the data set. (In the case of Europe, making the warming trend LESS by 1/2 C). As the overall variation is 1.5 C between about 1984 and 1888 in aggregate, even a 75% effective method of defect removal will find 1/2 C of “warming” over that time period based only on leakage of data change artifacts.
The purpose of this analysis is NOT to say that the 1.5 C of difference is evidence of malfeasance, nor even to say it can not be mitigated, nor to say that the method that identifies it is the One Valid Way to calculate a Global Average Temperature trend. It is just to say “That bias exists” and then raise the question of “What percentage of it can reasonably be removed?” Near as I can tell, the “Warmers” wish us to just accept by assertion that the answer is 100%. I’ve seen no evidence that the other programs are perfect at removing such data bias. I have seen evidence that GIStemp is about 50% effective. (But even that is not as well proven as I’d like – GIStemp broke on attempts at better testing.)
For some reason this “measuring the data” mind set seems to throw a lot of the folks who are not familiar with it. They seem to jump to the conclusion that the goal is to show an actual value for the increase in the Global Average Temperature. As I’ve said a few times, the whole concept of a GAT is really a bit silly due to the fact that averaging intrinsic properties is fundamentally meaningless.
At any rate, one can show how ‘trend’ can be shifting in the data even if one doesn’t believe that the calculated ‘trend’ has much meaning. So I do.
I ran the software QA group at a compiler company for a few product cycles and I guess it changes how you think about things. Demonstrating repeatable and reliable outcomes is important… when the outcome changes based on the “same” input, it is cause for concern that maybe the “same” input isn’t really the same…
@Leonard Lane:
I do feel compelled to point out again that this does NOT show that individual data items in individual months are changed. It shows that the collection of instrument records has an overall shift that changes the trend. (Though there may well be some individual data items that also change).
As First Differences is particularly sensitive to the “join” point between two records, it will be particularly sensitive to changes in the first few years (near that 1990 start of time point) and will highlight the change of processing that happened about 1987-1990. I think that is mostly the move to electronic thermometers and largely at airports.
So this isn’t so much saying that lots of individual data items have been changed as it is saying that a “shift” in thermometers and processing at about that time causes a ‘splice artifact’ that shows up as a ‘warming trend’ when homogenized and over averaged. At least, that’s what I speculate is the cause of the shift.
It could simply be that the other codes, like CRUTEMP and GISStemp are so complicated that they do a very good job of hiding that splice artifact, and a less than stellar job of removing it, and that the folks who installed the changed thermometers and the folks running those codes truly believe that what they are doing is close enough to perfect that the result is valid.
Basically, it need not be malice. It can simply be pridefulness and error…
@jorge:
I especially like the way some posted ‘critiques’ in less time than it takes to actually read the article and look at the graphs 😉
But it is hard to tell a ‘Troll’ from a ‘Zealous True Believer’ sometimes. Still, the distractors of “go do what I tell you to do” and / or “but look at this orthogonal link” can be a bit funny…
@ur momisugly EMSmith
absolutely – as a comparative analysis of datasets, of course you couldn’t have studied every station data! But that was kinda my point – who has? – and how/what have they done to give us these different datasets?
@ur momisuglyMosh – come on! – FD analysis is hardly rocket science but it is seen as a sensible first representation of potential flaws in data/adjustments? Are you not annoyed that EMS has shown the data to be suspect? Never mind your personal stance on the issue – as a data user – would you not expect the data to be clean and reliable? (and ideally raw!)
Chefio’s analysis is, as usual, thorough and absolutely excellent. Beyond learning a lot though, watching as he eviscerated one troll after another was as entertaining a read as I’ve seen in some time. He even goaded Mosher into a comment using multiple paragraphs instead of a one line drive by snark. Wow!
The icing on the cake however were the grammar and spelling snobs, which Chefio crushed as badly as the technical trolls. C’mon Chefio, can’t you leave one or two trolls or snobs standing so that the rest of us can do some of the pummeling?
Spelling aint my strong point, but you’d think I’d at least get “Chiefio” right?
Best I not try and pummel any trolls tonight, to sleep, to tired….
unlss th spllng snbs shw agn, in whch cse I wnt to knw if spllng is so mprtnt why I cn tke mst of the vwls out and stll be ndrstd?
The original article is a longish read and I would recommend that most people ignore the early sections and pick it up from ‘The dP or Delta Past method’ which is about 15% of the way through. The remaining material on the quality of the data is fascinating. It’s written as a series of sections which can be digested independently. The meta message is inescapable and is summarised neatly by Smith in his penultimate paragraph.
Chiefio is strongest in this kind of analysis – detailed, documented and relatively easy to replicate. Would that more of the climate debate were based on this kind of material. If you find that his conclusions on the data series are interesting then go back and read what you missed at the beginning.
Marvellous work, Chiefio. You have set out the big picture with great thoroughness, demonstrating that the Global Warming story is based on dodgy data.
In support of your work, this link focusses on some specific Arctic stations, showing where the fraudsters’ fingerprints can be detected: http://endisnighnot.blogspot.com/2012/03/giss-strange-anomalies.html
EM,
thanks for the link to data. I did a linear regression on version1 and 3 using dT/yr data column, 1900 to 1990. Surprisingly both are negative! So I mayn’t have understood your labelling. dT/yr is just the average of monthly anomalies, isn’t it? Anyway, here are the results.
V1 trend is -0.0623 for the whole period
V3 trend is -0.0542 for the whole period
If I’m using the right metric, the trend difference is insignificant – but the trends are in the wrong sign, let alone barely sloping.
What have I done wrong?
@Nick Stokes:
I use ghcn v3 unadjusted. Don’t know where all I said it, but at least one place was during a (kind of silly really) rant about data quality in a non-adjusted data set. This was while I was developing the code and it crashed on ‘bad data’, so I complained in essence [that] non-adjusted data was not corrected for bad data. Still, it is a bit “over the top” to have rather “insane” values show up in the ‘unadjusted data’… but on the other hand, folks who want “really raw data” ought to expect it to be, well, raw… pimples, warts, insane values, and all:
http://chiefio.wordpress.com/2012/05/24/ghcn-v3-a-question-of-quality/
@Barry:
Oh, I see, you are talking about in the linked article, not about comments here. The estimate is from eyeballing the graph where about 1987 is around + 1/2 C and around 1888 is about -1/2 C and allowing about 1/4 C of “maybe I’m off”. But you can graph the data yourself as it is posted and do whatever you like with it. Here is the ‘difference’ pre-calculated back to 1880 (presuming you are not interested in prior to that time):
You can see that near the start, it’s about +1/4 C and near the end (but not AT the end) there are values of about -1/2 C. As with all such graphs, the exact start and end points you choose to ‘cut off time’ can shift the trend and this set leaves out some of the largest ‘cooling of the past’ back prior to 1880, but I didn’t want too long a set of numbers in a comment…
Hmm, I guess I have to go and read it.
But the first thing I notice is that he bangs on about unproven assertions, but makes plenty of unproven assertions himself. This makes what EM Smith says hard to take seriously.
It’s seems any thing he doesn’t like is an unproven assertion and anything he just makes up is a proven fact.
So is there any substance behind this fog. Well using an analysis based on differences has to be done very carefully, since differences exaggerate noise. That is very well known. Numerical Maths 101. Even a dumb LT knows that.
So let’s have a look.
Leonard Lane says:
June 21, 2012 at 10:03 pm
We hear the statement “chimes against humanity” all the time in a political context…
A fortuitous typo, since much of the AGW handwringing seems to be based on noise.
@davidmhoffer:
Love the spelling point 😉 FWIW, studies have shown that as long as the first and last letter are correct, all the other letters can be in random order and most folks can read the word. Strange, that. I find it particularly easy to do ECC on things so just read your ‘compressed’ versions quite easily. (Then again, I read ‘mirror writing’ and upside down print and don’t always notice…)
Per Chefio: As I’m fond of cooking, too, I’m Fine With That 😉 Just got a new “smoker” and made a simple brined smoked salmon. Yum! 1 qt water, 1/2 cup each salt and brown sugar. Soak for an hour. Then coat with soy sauce and let sit for another 20 minutes or so as a bit of ‘pellicle’ forms. Put in the smoker on ‘very low’ about 200 to 220 F, for about an hour and a half… Less for thinner chunks. Also did a chicken that was marinaded in about 1/3 lemonade and 2/3 soysauce, then slow smoked for a few hours… Hickory chips in the chip box…
(Can’t say I don’t aim to please 😉
Per dealing with Trolls: While I appreciate the compliment, mostly I’m just trying to answer questions in as open and honest a way as possible. “The truth just is. -E.M.Smith”. So I don’t know if they are Trolls or just folks with questions based on a slightly akimbo (mis?) understanding. All I do in any case is state what is.
At any rate, as it is nearing 2 am where I am, bed will be calling me soon, too, and there will be plenty of time for others to chime in…
@John Diffenthal:
Well, while I may be prolix, at least I can’t be faulted for lacking thoroughness 😉
And yes, I do prefer things that can be intuitively grasped and don’t have a lot of artifice and “puff” / complexity between the starting point and the result. So the actual dP/dt code is very short and very understandable. The “method” fits in one simple paragraph (see above).
@Brent Hargreaves :
Yes, GIStemp basically invents a completely fictional Arctic temperature. Though I’m loath to call it ‘fraud’ since it can simply be that they are ‘sucking their own exhaust’ and ‘believe their own BS’ (to quote two common aphorisms for the tendency to believe what you want to believe, especially about your own abilities, that are commonly used in Silicon Valley programmer circles…)
“Never attribute to malice that which can be adequately explained by stupidity” can cover a lot of ground as “Intelligence is limited, but stupidity knows no bounds. -E.M.Smith” 😉
Put another way: I’ve looked at code that I’d swear was perfect, run it, and had horrific bugs show up. Things you learn doing software QA… And WHY I stress that until there has been a professionally done and complete independent software audit and benchmark of any “Climate Codes” (such as GIStemp) they can not be trusted for anything more than publishing papers for meeting academic quota…
There is a reason that the FDA requires a “Qualified Installation” for even simple things like a file server and complete submission of ALL software and data used in any drug trial. They must be able to 100% reproduce exactly what you did or you get no drug approval. Even something as simple as a new coating on an aspirin requires that kind of rigor. Yet for “Climate Science” we take code that has never been properly tested, nor benchmarked, and feed it “Data Du Jour” that mutates from month to month, and play “Bet Your Economy” on it… Just crazy, IMHO. But what do I know about software, I’m just a professional in the field with a few decades of experience… I’ve even DONE qualified installs and signed the paperwork for the FDA… ( The “Qualified Installation” documents must be written such that a robot following the statements would get exactly the same result. Say, for example, “push the red power button” and the vendor changes to an orange button, and you will fail. So wording has to be careful with things like “push the power button to the on position”… No way the “Climate Codes” could even think of applying for the process. The data archival requirements alone would cause GHCN to be tossed out as too unstable. But hey, an aspirin is far more important than the global economy… /sarcoff>; )
The World Climate Data and Monitoring Programme of the WMO have been scratching their bonces about metadata and homogenisation:
http://www.wmo.int/pages/prog/wcp/wcdmp/wcdmp_series/index_en.html
Phil Jones was involved in a case study which included:
“…………..variations are related to non-climatic factors, such as the introduction of new instrumentation, relocation of weather stations, changes in exposure of instruments or in observing practices, modification of the environment surrounding the meteorological stations, etc.
At the same time, wrong or aberrant observations are common in most observational systems. All these factors reduce the quality of original data and compromise their homogeneity.”
http://www.wmo.int/pages/prog/wcp/wcdmp/wcdmp_series/documents/WCDMP_Spain_case_study-cor_ver6March.pdf
This pdf gives the WMO guidelines:
http://www.wmo.int/pages/prog/wcp/wcdmp/wcdmp_series/documents/WCDMP-53.pdf
In a cited paper where a certain PD Jones was involved (pg 40) it states:
“Judgement by an experienced climatologist has been an important tool in many adjustment methodologies because it can modify the weight given to various inputs based on a myriad of factors too laborious to program.”
Phil Jones: “I’m not adept enough (totally inept) with excel to do this now as no-one who knows how to is here.”
I expect the CRU have got their copy of Excel for Dummies by now.
Warmist papers should come with a warning: “May contain nuts.”
EM,
yes, I was talking about the post on your blog that this one links to.
my last post seems not to have got through. I did a trend comparison between v1 and v3 per the tables in the post you linked (thanks).
v1 trend 1900 to 1990 = -0.623 for the period
v3 trend 1900 to 1990 = -0.542 for the period
(v3 is from memory – I’d shut Exel down and couldn’t be bothered doing it again)
I used the data from the column dT/yr, which is the average of the monthly anomalies for each year. Assuming I’ve used the correct data, three things are apparent.
There is hardly any slope for the period.
The slope is negative.
The difference between V1 and V3 global trend is insignificant – less than a tenth of a degree over 91 years.
I find the negative result surprising. Am I doing something wrong?
Thanks for the difference table upthread, but it was easier to copy’n’paste from the other tables, as they had the years already marked.
Chiefio;
Per dealing with Trolls: While I appreciate the compliment, mostly I’m just trying to answer questions in as open and honest a way as possible.>>>>
I know! That’s what made it so darn amusing! (Many of the questions were of course legit, they weren’t all troll comments of course)
There’s a pet peeve of mine that I’m wondering if you’d comment on? Trending “average” temperatures has never made much sense to me. We’re trying to understand if increased CO2 results in an energy imbalance at earth surface that raises temperatures. If that is the case, why would we average temperatures and then trend them? The relationship between temperature and w/m2 is not linear.
Taking for example some cooling in Africa that is balance by some warming in say Canada. Do they balance each other out? Very possible that they do when one averages “degrees” but averaging “w/m2” give a whole different result. If equatorial Africa cools from +30 to +29, that’s a change of -6.3 w/m2. But if temps in northern Canada in an equal size area rise from -30 to -29, that’s an increase of only 3.3 w/m2. So, the “average” temperature based on those two data points would be a change of zero, but in terms of energy balance, the earth would be cooler by 3 w/m2.
One thing I would like to see is the temperature data converted by SB Law to w/m2 and THEN trended. By averaging temps alone, we’re over representing changes in energy balance at high latitudes, and under representing changes in equatorial regions. From a w/m2 perspective, a degree of cooling in equatorial Africa would wipe out several degrees of warming in Antarctica and may well show a completely different trend than temperature.
I’ve always wondered why, if the theory is that doubling of CO2 increases forcing by 3.7 w/m2, we would try to measure it through temperature readings which amount to a proxy that is KNOWN to not have a linear relationship to w/m2! If we want to see of CO2 is increasing forcing by some amount measured in w/m2, would it not make thousands of times more sense to MEASURE in w/m2 in the first place?
@Barry:
A trend line through “all data” differences has the following values:
f(x)= -0.0030733007x + 0.2706459054
R^2 = 0.6412526237
per the least squares fit line in Open Office.
The trend line hits the vertical axis at about 0.25 ( I believe that is the 0.27 value in the formula above) and crosses -0.6 at the start of time. It is at about -0.5 around 1740 or so.
As I understand things, that gives a formal LSF line of about 0.75 increase by using the data from the present back to about 1740.
Hopefully that satisfies your need.
(Though again, I do want to stress that it is the variation in the ‘trend’ between different regions that I think is more important. It shows that the trend is very non-uniform around the world, and that it varies dramatically in how it changes from place to place as v1 changes to v3. It is that non-uniformity and that the changes are orthogonal to CO2 changes that make it ‘suspect’ that CO2 has any impact on the data and that changes in the data set are more important to any ‘trend’ found.)
To quote the article:
What is found is a degree of “shift” of the input data of roughly the same order of scale as the reputed Global Warming.
The inevitable conclusion of this is that we are depending on the various climate codes to be nearly 100% perfect in removing this warming shift, of being insensitive to it, for the assertions about global warming to be real.
——————————————————————————————————————-
Aah – I LOVE the smell of facts in the morning!
@Pouncer
At the risk of being terribly pedantic, if the average temperature gets to 500K we are all already in a lot of trouble 🙂
500K is +227C…warm enough to bake bread. Gas Mark 8 in old money.
You might wish to rephrase it at about 290K (+17C). Cheers.
@Barry:
dT/yr is the change for THAT year compared to the next, not the running total that you want to use. It is the volatility, if you will. It is not the field you want for trend.
As I noted earlier, there are two sets of reports posted for the “all data” version. The one in the article itself has the opposite sign to that in the comments. ( That is, the dT/dt version of the code tells you how you must change things to get to the present, while the dP/dt version tells you directly what the past looked like compared to today. Or “it is 1 C warmer now than then” vs “it was 1 C cooler then than now”. This is detailed in the posting.
(The reason for this is pretty simple. The original dT/dt code was for a different purpose and was to show “how much warming was needed to get to the present”. So if it was cooler by 1 C in 1816 the code showed “You need to add 1 C to get to the present warmth”. This was ‘inconvenient’ for making the comparison graphs, not to mention counter intuitive if you are showing trend from the past into the present on a graph, so the dP/dt version was made that basically just inverts the sign. That shows “It was colder by 1 C in 1880” as -1 C.)
So, to get the rising trend, scroll on down to the comments and pick up the dP/dt version of the data (or change which is subtracted from what 😉
To get the comparison of changes between the two data sets, subtract the v1 data for dP from the v3 data for dP. Thus, if it was -1 C in v3 and -0.25 C in v1, you get that it is now -0.75 C cooler in that point between the two. A trend line plotted through those variations show the trend of the differences between the two sets of data. As noted above, that’s about 0.75 C of difference between 1740 and the present. I don’t know what it is at 1900 to the present as that isn’t a start date used by any of the climate codes, so I’ve not inspected it. As any given ‘cherry pick’ of a start date on a trend line data set can change the trend, YMMV as you try different start and end dates. (That is also one of the “issues” I have with the whole “warming” assertion, BTW. I can give you any warming or cooling trend you want as long as I can pick the start date. It was far warmer 6000 years ago. Colder by far in 1816 (the “Year without a summer”). About the same in 1934. Colder in the mid-’60s. The Younger Dryas froze things hard. A few hundred years later it was way hot. etc. etc.)
Hopefully I can be done now on the “100 ways to measure the change of trend in the differences”…
KR at 8:46 pm
You were looking for “well distributed, rural raw data, a temperature estimate that significantly varies from the GHCN estimate”
This deals with NCDC data for the contiguous US, but otherwise may suffice: Long, E.R, “Contiguous U.S. Temperature Trends Using NCDC Raw and Adjusted Data for One-Per-State Rural and Urban Station Sets,” available at
http://scienceandpublicpolicy.org/originals/temperature_trends.html.
For the rural stations in the study, the raw data showed a linear trend of 0.13º C per century, while for urban stations the raw data showed a trend of 0.79º C per century. Id. at p. 8-9. The long term trends were very similar until about 1965, when the trend in the urban raw data increases faster than in the rural data. Id. at 9- 10.
NCDC’s adjusted data for rural stations show a trend of 0.64 º C per century, compared to 0.13 º C per century for the raw data. In other words, the NCDC adjustment increased the rural trend by nearly five times. Id. at 11. The adjusted data for urban stations show a trend of 0.77º C per century, compared to a raw urban trend of 0.79º C. per century. Id. “Thus, the adjustments to the data have increased the rural rate of increase by a factor of 5 and slightly decreased the urban rate, from that of the raw data.” Id.
E.R. Long is a retired NASA physicist.
The other curiosity is that a large proportion of the graphs show the 1970s temperature dip.
No explanation why splicing errors produce such a consistent result.