Anthony has suggested that I post a paper of mine that was written in 2006 about unusual temperatures in Svalbard. It was published as a peer-reviewed submission in Energy and Environment, the journal that AGW supporters love to hate.
First, where the heck is Svalbard? It’s up at the top of the world, at 78° North, where it’s frozen most of the year.
Figure N1. Location of Svalbard. The islands are north of Norway, marked by a snowflake. The North Pole is shown as a red star, and Greenland is to the left.
In April 2006, there was an anomalously high temperature recorded at Svalbard, which was the subject of my paper.
How anomalous was April ’06? Here’s the month by month record for Svalbard, showing the temperatures from January (1) to December (12) since 1912
Figure N2. Monthly temperatures in Svalbard. The green circle is the anomalously high temperature in April 2006. Note also the orange circle (April 1917), blue circle (September 1990) and the yellow circle (October 1968). Photo is of Svalbard.
The most obvious feature of the Svalbard record is that there are huge year-to-year swings in the winter temperature. The months from November to March show a temperature range of about 20°C.
Is the April 2006 temperature (green circle) unusual? Yes, it is. It is 3.0 standard deviations (SD) from the average of all of the April data. However, it is not the most unusual. That would be the October 1968 (yellow circle) at 3.5 standard deviations from the October average. We also have September 1990 at 2.9 SD, and April 1917 at 2.8 SD out.
So with that as a prelude, here is my 2006 paper.
————————————————————–
PROBLEMS WITH PUBLISHING SCIENTIFIC INFORMATION ON THE WEB: HOW UNUSUAL WERE TEMPERATURES IN SVALBARD, NORWAY?
Willis Eschenbach
INTRODUCTION
Traditional science journals accept papers for publication following peer review of their contents that provides at least some independent assessment of the paper’s contents. As part of the scientific process, the content of such papers can be challenged in the journal that publishes them. However, with the advent of the web it is becoming common for some scientists to publish work on web sites. This practice has the advantage of rapid communication to a wide audience, but it by-passes the safeguards inherent in the traditional publication procedure. This increases the risk of flawed information becoming published, with inevitable lowering of scientific standards and possible harm to the reputations of individual scientists.
This paper discusses an example that raises the problems of ‘web publication’ for scientific information. The web site in this case is RealClimate.org. Michael Mann and Gavin Schmidt are listed as the first two principals of the site. Both are well known for their scientific research in support of the hypothesis that the we are currently experiencing dangerous anthropogenic (human-caused) global warming, and both have strongly defended this position in public as scientific truth.
The stated purpose of the RealClimate site is the presentation and discussion of factually accurate scientific information concerning climate change. The site is commonly known by the abbreviation ‘RC’. The example concerns publication on that web site of a claim concerning unusually warm temperatures in April 2006 at Svalbard, Norway.
This discussion should be understood in the context of the fierce debate about the causation of current Arctic temperatures, as well as the prediction of rapid further rises by many climate models. Both form a major platform in current energy policy debates world wide, and especially the UK and USA.
THE EVENTS
On May 22nd, 2006, the RealClimate web site published an article titled “More On The Arctic” by two very well-known climate scientists, Michael Mann and Phil Jones. The full text of the report published on the RealClimate web site is as follows:
Svalbard, an Arctic island in the Northern North Atlantic, is predicted to warm considerably more than most of the rest of the earth in many model-based scenarios [http://www.grida.no/climate/ipcc_tar/wg1/fig9-10.htm]. See for example the figure to the right, which represents a relatively high-end IPCC Third Assessment Report scenario for the projected surface temperature difference between the period 2071–2100 and 1961–1990. Svalbard is the island north of Norway at about 80N between 15-30E. The enhanced warming in this region is related to the issue of polar amplification that we have discussed previously on RC. It also happens that the Svalbard meteorological station is the 2nd station in the World Meteorological Organization (WMO) meteorological station list. This means that it tends to get noticed. The Climatic Research Unit (CRU) of the University of East Anglia maintains one prominent version of the global surface temperature data set and as part of its routine quality control, CRU flags any unusual (anomalous warm or cold) new measurements that come in. Svalbard has now been flagged consistently over the past several months, but the values have been confirmed as accurate by the Norwegian Met Service, which operates the Svalbard station. Here are the recent Svalbard monthly surface temperature measurements, the longterm (1961–1990) means (“ybar”) and standard deviations (“sd”), and associated anomalies i.e., departure from average (“delta”) for Dec 2005 through April 2006 (all in degrees C):
Month, Value, ybar, sd, Delta
Dec 05, -3.8, -13.3, 4.4, +9.5
Jan 06, -2.7, -15.3, 4.7, +12.6
Feb 06, -9.8, -16.3, 3.7, +6.5
Mar 06, -13.1, -15.8, 3.7, +2.7
Apr 06, 0.0, -12.4, 2.7, +12.4,
The numbers are fairly remarkable. April ‘06 was warmer than any previously recorded May, and January ‘06 was warmer than any previously recorded April. The previously warmest April was -7.0C (1996). There is currently an absence of sea ice off much of the coast of Svalbard, which is also unprecedented for so early in the year.
The April mean temperature is almost 5 standard deviations above the mean, a “5 sigma event” in statistical parlance. Under the assumption of stationary ‘normal’ statistics, such an event is considered astronomically improbable (< 1 in 10^6), and, like the summer heat wave in Europe in 2003 (which was a 5 sigma event in Switzerland, 3 sigma over Europe as a whole), deserves special attention. As we have nonetheless remarked before on RC, particular events, even seasonally-persistent anomalies as unusual as these, do not “prove” anthropogenic warming. But in a statistical sense, large outliers like this make it more probable that the underlying distributions are shifting and give us a glimpse into the types of anomalies we might expect to become more common in the decades ahead.
In their article on RC, a number of claims are made, including that the recent unusually warm April temperatures in Svalbard, Norway, represented a “five sigma event”.
“Sigma” is the distance of a data point from the average of a data set, as measured by the standard deviation. The higher the sigma of an event, the less likely it is that the event is the result of chance. The odds of a “five sigma event” happening by chance are less than one in a million. If this were actually a “five sigma event”, then the reported warm April 2006 Svalbard temperatures would be greatly outside the normal temperature range, and would thus represent a very significant occurrence with important implications.
This was an extraordinary claim which, if true, would have had great importance for detection of climate change in the region of Svalbard. Therefore, replication of the finding was warranted. This replication required use of the source data for the calculation conducted by Mann & Jones and reported in the article. The text quoted above states that this data is archived at CRU.
My investigation failed to reveal the data at CRU, but I found that the GISS temperature database does have a record of temperatures measured at Svalbard. However, although the data reported on RC was for the period “1961-1990”, the GISS Svalbard record contains no data prior to 1977. In an attempt to resolve this problem I tried to post a comment to the RC thread. The comment I tried to post said in full:
According to GISS, the Svalbard station only started reporting temperatures in mid 1977. How did you determine the “long-term (1961–1990) means (“ybar”) and standard deviations (“sd”)”? Thanks, w.
This seemed to be a reasonable question, but it was not published by RC and did not appear on the web site. I did not receive any answer to my question either in public or as personal correspondence. Therefore, I resent the question, and suggested that my original post might have been caught in the spam filter. This was answered on RC, but the entire answer said:
[Response: The comment was most likely deleted because it was already indicated above that the record is available back to 1911, and the source of the data (Climatic Research Unit, not GISS) was already indicated. —mike]
Presumably, the “Mike” who answered the question was Michael Mann. It seems that my question had lacked precision, and that I should have asked exactly where I could find their dataset. But their answer repeated that the data was from CRU, so I went again to look at the available data at CRU. Again I failed to find the relevant data. In the meantime, Hans Erren picked up the topic, in a posting that was accepted and posted by RC which said:
Fair enough, but Svalbard Luft started in October 1977 and Isfjord Radio stopped with continuous recording in June 1976. How was the homogenization obtained between the two stations. Moreover the nearest other station Bjornoya has the 50’s hotter than the 00’s, Which suggests an inhomogeneity in Svalbard in 1977. Here is the giss list:
0 km (*) Svalbard Luft 78.2 N 15.5 E 634010080002 rural area 1977–2006
47 km (*) Isfjord Radio 78.1 N 13.6 E 634010050010 rural area 1912–1980
425 km (*) Bjornoya 74.5 N 19.0 E 634010280003 rural area 1949–2006
RC posted a reply to that, saying in full:
[Response: For any further details, you and any other interested readers should refer to the linked CRU website for information and references leading to an extensive body of literature that describes how CRU forms representative composite 5 degree latitude ´ longitude gridbox estimates (which is what is referred to here) that account for time-dependent sampling variations and potential inhomogeneities associated with the individual recording stations that fall within the same grid cell. —mike]
Once again I examined the “linked CRU website”. Still finding nothing, I asked directly for the source of their data, saying in an attempted post to RC (in full):
Mike, thank you for your answer to Hans Erren. You say:
For any further details, you and any other interested readers should refer to the linked CRU website for information and references leading to an extensive body of literature that describes how CRU forms representative composite 5 degree latitude ´ longitude gridbox estimates (which is what is referred to here) that account for timedependent sampling variations and potential inhomogeneities associated with the individual recording stations that fall within the same grid cell.
Now I’m confused. You say that you are referring to the representative composite 5° ´ 5° gridbox temperatures … but in your original post, you say “Here are the recent Svalbard monthly surface temperature measurements …” and you compare them to the December ‘05 to April ‘06 Svalbard surface temperature measurements. You also refer to a check made with the Norwegian Met Service to make sure that your figures were correct, which clearly means that they are not gridbox temperatures.
Could you clarify which dataset you used for your calculations, the merged Svalbard/Isfjord Radio surface temperature dataset, or the 5° ´ 5° gridbox temperatures?
I ask in part because, despite an extensive search the CRU site, I cannot find any dataset for Svalbard or Isfjord Radio. Is the CRU the source of your dataset, and if so, where?
Many thanks for your clarification, w.
This comment was a simple request for the actual location of the dataset, but once again it was not published, and did not appear on RC. Again, I received no reply, either by email or on the web. I again tried to obtain an answer by attempting to post the following comment:
Thank you, Mike. Much appreciated. However, the “source of the data” (CRU) doesn’t make individual station records available, so I had to make do with what I had.
Upon further research, I found that the complete spliced dataset is available at http://www.unaami.noaa.gov/analyses/sat/#table.
In any case, the record is a spliced record, with a one year period between the two records without any data. I’m curious about the justification for the splice, since there is no overlap, and I’m curious that a spliced record with no overlap between the two station records would be used for this type of statistical procedure.
There is a larger problem with the claim, however, This is that you have taken the mean and standard deviation of a subset of the data (1961–1990), and are using these figures to derive sigma figures for a time period which is out of the subset (Dec ‘05 to April ‘06).
Mathematically, this is simply not correct. For example, could we take the mean and SD of a ten year subset, say 1983–1992, and compare that to April ‘06? If we do, we will get a very different answer. By that measure, the odds of April ‘06 are only about one in a hundred.
The only way to do a sigma calculation is to include all of the data when calculating the mean and standard deviation. Otherwise, the answer depends on which subset of the data is chosen, which is clearly wrong.
This is a very serious error, which materially affects the stated results. I expect that, in conformance with your stated policies, my message (being scientific in nature, not a troll, and not abusive in any way) will not be censored. Many thanks, w.
This posting was also not published, despite the reminder of the stated policy of RC. The censoring of this comment is in direct contravention of their stated policy which says in part:
1) Questions, clarifications and serious rebuttals and discussions are welcomed.
2) Only comments that are germane to the post will be approved. Posts that only contain links to inappropriate, irrelevant or commercial sites will be deleted.
3) No flames, profanity, ad hominen comments, or you said/he said type arguments.
4) As stated in the blog description, no discussions of non-scientific subjects will be allowed.
However, although my attempted post was a germane, scientific, and serious rebuttal of their claim. it was not “welcomed” as they claimed—instead, it was not published at all. However, soon afterwards a curious “update” from Michael Mann appeared on RC which said:
[Response:(update) I was a bit careless in my wording. The long-term series from 1911 to date currently maintained by the Norwegian Met Service (and to which we are referring to here) is based on the combined record for the Svalbard airport and Isfjord Radio sites. The latter started in 1911. The Norwegian Met Service combined the two records in one of the recent nordic projects, and any inhomogeneities were taken into account in the process. There is little doubt that the anomalies observed so far this year are unprecedented as far back as the measurements go (early 20th century). – mike]
This was very strange. In the original posting (quoted above), RC clearly said the data was station data from Svalbard maintained at CRU. Then in their first response to me, they again cited CRU as the source, saying “the source of the data (Climatic Research Unit, not GISS) was already indicated”. In their response to Hans Erren (quoted above), they said they were discussing, not CRU station data, but “composite 5 degree latitude ´ longitude gridbox estimates (which is what is referred to here)”.
Finally, they said they had used a totally different dataset, not maintained by CRU, but by the Norwegian Met Service. This is not “careless wording” — either it is three very large but honest errors, or it is an attempt to obscure the data source.
Then on the 28th of May, RC closed the thread without further comment or explanation. The thread had existed for only 6 days from posting to closure, which is very much shorter than other threads on RC. A number of very important questions remained unresolved. By censoring posts and closing down the thread, RC has made it impossible to resolve those questions on the thread, making it worthwhile to discuss them here.
The Problems With The Svalbard Record
I eventually found the location of the Svalbard dataset although Mann & Jones were unwilling to reveal it – and they never did – despite repeated requests for the exact location where the dataset could be found.
The Svalbard dataset is available here. The file, Nordklim_data_set_v1_0_2002.xls, is a 9.1 meg Excel file containing a variety of climate data for a number of Norwegian sites. The Svalbard station has the Norwegian station ID 99840, and the Svalbard monthly mean temperature data starts on line 4639 of the sheet called “101”.
As Mann finally admitted, the source of the Svalbard record is a “spliced” dataset, meaning it is a combination of the Svalbard and Isfjord Radio records. He does not mention, however, and may not have known, that the location of the Isfjord Radio station was changed three times. The station then fell into German hands during WWII, and subsequently was returned to Norwegian control at the end of WWII. The Norwegian scientists have done an admirable job combining the data from these four locations into one record. They have made a number of adjustments to each section of the Isfjord Radio dataset to try to make it have relevance to the final dataset, that of Svalbard. The complete merged Svalbard record is provided here as Figure 1.
Figure 1: The Svalbard temperature data set. The blue line is monthly temperatures. The black line is a 35 month (+/- 3 std dev) Gaussian average of the monthly temperatures.
There are two things of note about this record. One is that there is no sign of any recent “polar amplification” of warming that was claimed on RC. On the contrary, the warming occurred at the beginning of the record, not at the end. Second, the record shows no signs of any recent significant warming. For the period 1920–2006, there is no statistically significant trend in the data. The 1920–2006 trend is 0.04°/decade, with a two standard deviation error of 0.08° per decade, which means it is not a significant trend. The 1970–2006 trend is 0.04°/decade, with a two standard deviation error of 0.2° per decade, which again is not significant.
Finally, it must be remembered that this record is a spliced record from three Isfjord Radio locations in one area (with data collected by two different countries at different times), plus the Svalbard record from a location in a different climate zone 43 kilometres away. Since it is a spliced record from four different locations, any statistical conclusions from the record must be suspect. With that caution in mind, let us look at the claims made by Mann and Jones in their article published on RC.
THE ERRORS IN THE MANN-JONES ANALYSIS
Inflated Claims
According to their own figures, this was not a “five sigma event” with odds of less than one in 10^6. It was a 4.6 sigma event, which has a probability a full order of magnitude smaller, less than one in 10^5. (An “order of magnitude” means one number is ten times larger or smaller than another.)
Improper Choice of Statistical Methods
As they stated, the authors used statistics designed for stationary normal datasets. However, the temperature dataset they used is neither stationary (which in this context corresponds to trendless) nor normal. Both the standard deviation and the average change over time. Because of this, the results depend on the period chosen for the baseline. Figure 2 shows the variation in the results for different thirty year periods.
To illustrate the effect of the choice of the exact thirty year period, I calculated the “sigma” of the April 2006 temperature using all possible thirty year periods in the record. Using different thirty year periods, the odds of the April 2006 Svalbard temperatures range from a low of one in 28, to a high of one in 5,392,753. Hence, one could pick any period to prove any point. The “five sigma event” claimed by Mann & Jones is merely one of many possibilities.
In addition, the authors neglect to give error figures for their result. The green line in Figure 2 shows the error bar (+/- 2SD) for their results (using the ‘61=’90 data).
Incorrect Calculation of Standard Deviation
It appears that the Mann & Jones calculations for the 1961–1990 period contain an error. Although the averages they published on RC were exactly correct, their calculated standard deviations (sd) were not. While it is possible that this is from some corrections made to the data (removal of incorrect data points etc.), since they have not made their data available it is not possible to say whether this is actually an error or is an unrevealed alteration of the data. Here are their sd values, compared to the actual sd values for the 1961–1990 period.
Month, M&J sd, Actual sd
Dec 05, 4.4, 4.7
Jan 06, 4.7, 5.1
Feb 06, 3.7, 4.0
Mar 06, 3.7, 4.2
Apr 06, 2.7, 3.0
It is of interest that in all cases, the effect of their error was to reduce the calculated standard deviation, and thus increase the sigma. In an attempt to determine the source of their erroneous standard deviations, I calculated the standard deviation of all possible periods in the record. There are only three periods where the five months have a standard deviation that matches sd values published on RC by Mann & Jones. These periods are 53 years starting in 1938, 52 years starting in 1939, and 51 years starting in 1940. While this was probably an honest error, Mann & Jones chose one of the lowest periods in the record. For example, of the 44 fifty-one year periods in the record, only 9 of them give smaller standard deviations and thus larger sigmas. Once again, their error artificially increases the calculated sigma.
Figure 2: Average, standard deviation, and sigma for 30 year (trailing) periods of the Svalbard April temperature data set. The blue line is the thirty year trailing mean (right scale). The red line is the 30 year trailing standard deviation. The black line is the sigma of the April ‘06 temperature record, based on that mean and standard deviation. The green line is the error bar (+/-2SD) for the sigma of the author’s chosen period (61–90).
While the post 1990 average is rising, this rise is not statistically significant.
Lack of Correction For Autocorrelation
At the beginning of their statistical section (above), Mann & Jones say “Under the assumption of stationary ‘normal’ statistics, …”. A “normal” dataset is one that obeys the standard laws of random events (see “Further Reading” at the end for more information on normal distributions).
However, their assumption of stationary normal distribution is totally unwarranted. As professionals in the field, they know that temperature series are generally not “stationary ‘normal’” distributions. In particular, even a cursory examination shows that the April temperatures of this dataset are significantly non-stationary and nonnormal. Both the average and the standard deviation of the April dataset change over time (see Figure 2 above), showing that they are not stationary. The dataset is also what is called “fat-tailed”, which means that it contains an unusually large number of temperatures which are either higher or lower than the expected range. Figure 3 shows the Svalbard temperature data compared to a normal distribution. The Jarque-Bera Test for normality finds the Svalbard dataset non-normal with p<0.02.
Figure 3. Histogram of Svalbard Data. Blue dots show midpoints of bins for data. Red line shows normal distribution given the same bins.
Temperature series are also generally “autocorrelated”, meaning that a hot year is often followed by another hot year, and vice versa. Because of the non-normality and the autocorrelation of the Svalbard April dataset, standard statistical tests such as they have used give inaccurate results. In particular, they produce artificially inflated sigma values. They also often indicate significant trends where no such significant trend exists.
The True Sigma Value Of The Claimed ‘5 Sigma Event’
The errors of Mann & Jones all combine to artificially increase the apparent significance of the April temperature record. Let me correct them, one at a time.
a) The odds a five-sigma event occurring by chance are one in 1,744,000. This is the “less than one in 10^6” figure Mann & Jones claimed in their article.
b) However, according to their own figures, this was a 4.6 sigma event. The odds of this are one in 236,700, about an order of magnitude smaller than their claim.
c) In addition, they made an error in their figures. When their erroneous 1961–1990 standard deviation calculations are corrected, it becomes a 4.2 sigma event. The odds drop by another order of magnitude, to one in 37,465.
d) Next, we can get a more accurate idea of the sigma by using the average and standard deviation of the full dataset. Using the full dataset, April 2006 represents a 3.4 sigma event, with odds of one in 1,484.
e) Finally, when the calculations are corrected for autocorrelation, we get a sigma of 2.87. The odds of this are one in 244.
So correcting for the above errors in the calculations reduces the sigma of the April 2006 temperatures from their claimed “5 sigma event” to a more accurate 2.87 sigma event. Corresponding to the drop in the sigma, the odds of the April 2006 temperature occurring by chance are reduced from one in three million to one in 244. And these corrections do not include correction for the non-normality of the dataset in any measure except autocorrelation.
In other words, the April 2006 temperature in Svalbard is not statistically unusual in any way. It does not indicate “enhanced warming”. It is not “astronomically improbable”. It does not support the claimed “polar amplification”. It does not “make it more probable that the underlying distributions are shifting.” In fact, it doesn’t indicate anything at all—it’s just another April in Svalbard.
CONCLUSIONS
The ‘Svalbard Affair’ is a clear demonstration of the problems that can be posed to scientific standards by the increasing use of ‘web publication’ that by-passes the traditional method of scientific publication. Errors that could be expected to be detected by peer review can be published on web sites. In the case of the ‘Svalbard Affair’ the problems with the published work were so obvious that they were immediately seen and reported to the authors. But the response of the authors was to attempt to try to stop any question of the matter, and this would also have been difficult if traditional publication had been used. Other less obvious cases may go undetected before data published on the web have become used in related studies by other workers.
Indeed, the analysis of Svalbard temperatures published on the RC web site continues to be used after ‘Svalbard Affair’. For example, the UKCIP (UK Climate Impacts Programme) is based at the University of Oxford and is funded by the UK government to help organizations assess how they will be affected by climate change so they can prepare for its impacts. On 12 June 2006 (i.e. weeks after RC closed the Svalbard thread) UKCIP circulated an email newsletter saying;
“5. Artic study demonstrates temperature rise
Real Climate has published a paper detailing new climate observations in Svalbard, an Arctic island in the North Atlantic. Observed temperatures this year are far warmer than comparable data from previous years. In statistical terms, the April 2006 temperature difference from average is equivalent (if not greater) to the summer 2003 heat wave in NW Europe. You can read the article at http://www.realclimate.org/ index.php?p=309”
It is difficult to imagine a more clear example of how ‘web publication’ can result in effects that are not justified by the worth of the published work. Errors of Mann & Jones in the ‘Svalbard Affair’ would very probably have prevented their analysis passing through peer review for publication in the traditional manner.
The detected errors of the RC authors in the ‘Svalbard Affair’ were:
1) They exaggerated their own calculations by a factor of ten, from calculated odds of less than one in 10^5 to claimed odds of less than one in 10^6 (seemingly for shock effect).
2) They used an inaccurate procedure, calculating the sigma from a subset of the data rather than from the full data set.
3) They did not correctly calculate the standard deviation of the 1961–1990 subset. 4) Although they know that temperature records are rarely “stationary, ‘normal’” datasets, and if they had looked they would have known that this temperature dataset was not a “stationary, ‘normal’” dataset, they used statistical procedures designed solely for use with stationary normal datasets.
5) They were either “careless in their wording” in specifying three different incorrect locations as the source of their data (they twice claimed it was CRU station data, then said it was CRU 5° ´ 5° gridbox data), or they did not know the source of their data, or they were deliberately obscuring the source of their data.
6) Rather than choosing a complete record from a single station for their analysis, they used a spliced record from four different locations. This makes the result of any statistical analysis very questionable. They did not reveal this in their initial posting, and but for the prodding of Hans Erren and myself, the fact would have gone unremarked.
7) They claimed “enhanced warming in this region” and a “polar amplification” of warming. However, not only is there no polar amplification in the Svalbard record, there is no significant warming in most of the Svalbard record (1920–2006 trend is 0.04° per decade +/– 0.08°C per decade [2 SD]). Even if one were to take their numbers at face value and to ignore the lack of statistical significance, this is only 60% of the reported global warming for that period, which is the opposite of “amplification”.
8) Despite repeated requests, they never indicated the actual web address of the dataset.
9) When scientific questions were asked about some of these errors, they did not publish the questions. When the questions were asked again, they did not publish the new questions.
10) To shut off the scientific questions entirely, they closed down the thread after only six days.
The last three of these problems are the most disturbing. Everyone makes mistakes, and some scientists have been known to try various types of unusual or doubtful methods with the data and calculations, or to exaggerate their claims. It is part of science. But to refuse to publish important scientific questions, to obfuscate regarding the source of your data, and to close down discussion when someone points out an error, is the antithesis of science.
Science proceeds by trial and error, and the finding and subsequent publication of scientific mistakes is a critical part of that process. The participation of Drs. Mann and Jones in this affair clouds their scientific reputations, and for the RealClimate web site to claim it welcomes scientific discussion when in fact it shuts out scientific discussion is a grave misrepresentation of the actual situation that may not be known to many who use the site as an information resource.
I am not an opponent of the publishing of scientific information on the web. I think that it is an very good way to get new ideas out into the scientific community, and serves as an excellent complement to traditional scientific journals. However, as is the case with publication in scientific journals, if scientists choose to use this method they need to publicly answer scientific questions about their claims. To do otherwise brings all of their claims into question.
I would welcome any comments from Drs. Mann and Jones, or from any of the principals of RealClimate such as Gavin Schmidt, regarding any of these issues. Unfortunately, despite repeated attempts to get them to answer even one of these questions, to date Drs. Mann, Jones, and Schmidt have bluntly refused to address these matters that seem to be egregious, grave and serious errors of fact, method, and scientific conduct.
Willis Eschenbach 6 June 2006
FURTHER READING
There is an excellent explanation of statistics as applied to temperature series called “Statistical Issues Regarding Trends”, by Tom M. L. Wigley. It is available at Climatescience.gov/Library/sap/sap1-1/finalreport/sap1-1-final-appA.pdf. It is a short paper, and is well worth reading by anyone interested in the theoretical underpinnings of the statistical concepts I have discussed in this paper.
———————————————————————–
FINAL THOUGHTS:
1. It’s always interesting to re-read something I wrote a while ago. I’d write it different now, but that’s how life works.
2. The high April temperatures in 2006 have not been a harbinger that “the underlying distributions are shifting” as Mann and Jones claimed. Svalbard temperatures peaked in 2006, and have dropped since then:
Figure N3. Monthly and annual temperatures in Svalbard. Photo is of Longyearbyen, Svalbard, the town where the current temperature station is located.
3. When scientific studies are published on the Web, it is crucial that the authors stick around and answer all relevant scientific questions. It is also vital that scientific questions are not censored. Watts Up With That is a good example of a web site doing this the right way. Yes, some of the claims posted here are wrong. I’ve been wrong on WUWT before. But I’ve read and attempted to answer all scientific questions, and acknowledged my errors when they are pointed out. This is how science advances.
4. Given our current knowledge of the emails and the evasion of Freedom of Information requests at CRU, the Mann/Jones claim that “the source of the data (Climatic Research Unit, not GISS) was already indicated” is particularly risible.
Final Thought 4: Yep. Climategate was one big “Told ya so”!!!
The arrogance, cowardice, and disingenuity of the so-called scientists running RC never ceases to amaze.
You have something going until you hit this spot,
‘According to their own figures, this was not a “five sigma event” with odds of less than one in 10^6. It was a 4.6 sigma event, which has a probability a full order of magnitude smaller, less than one in 10^5. (An “order of magnitude” means one number is ten times larger or smaller than another.)’
One order of magnitude between 100000 and 10000 is a not a significant difference when talking about one event. Either way, 5 sigmas or 4.6 sigmas, it’s still an extraordinarily small chance of it happening.
Getting down to the 1 in 200 range is significant, but as you point out, it’s questionable how significant a ‘spliced’ set of data is in assessing the temperature changes that have occurred over the last 80 or so years.
Also,
‘It is of interest that in all cases, the effect of their error was to reduce the calculated standard deviation, and thus increase the sigma.’
is contradictory. Sigma is the standard deviation. If we are reducing the standard deviation, then we are also reducing sigma. Reducing the standard deviation will make a single point seem farther from the mean, thus increasing the number of sigmas that single point is away from the mean. That may have been the point you were trying to make, but you’re saying it in a roundabout and confusing way. It just needs to be cleared up.
As for,
‘While the post 1990 average is rising, this rise is not statistically significant.’
what significance test/s did you use to determine whether this trend is ‘significant’ or not?
I think that you’re correct in your assessment in the way that RC dealt with this particular post and the comments you brought up. It’s likely that they felt no one would call out that type of aspect of the analysis.
Thanks for the post and the place to comment.
Cheers.
Thanks, Willis!
I found it interesting that you pitched this as primarily a critique on web-publishing of scientific documents.
Reading it, it is also a detailed refutation of a web-published paper, a strong criticism of the site in question, and a criticism of the scientists involved. Did you consider publishing it as only a refutation of the web-published paper, and added the critique parts later, or were they an integral part of the development of this paper?
Re the author’s points 1-10 “… detected errors of the RC authors in the ‘Svalbard Affair…’
What is the probability that all of these errors could occur in one article, written by professional scientists, by chance?
10 May: Space and Science Research Center: Food and Ethanol Shortages Imminent as Earth Enters New Cold Climate Era
Press Release SSRC 2-2010 :
The Space and Science Research Center (SSRC), the leading independent research organization in the United States on the subject of the next climate change, issues today the following warning of imminent crop damage expected to produce food and ethanol shortages for the US and Canada:
Over the next 30 months, global temperatures are expected to make another dramatic drop even greater than that seen during the 2007-2008 period. As the Earth’s current El Nino dissipates, the planet will return to the long term temperature decline brought on by the Sun’s historic reduction in output, the on-going “solar hibernation.” In follow-up to the specific global temperature forecast posted in SSRC Press Release 4-2009, the SSRC advises that in order to return to the long term decline slope from the current El Nino induced high temperatures, a significant global cold weather re-direction must occur. According to SSRC Director John Casey, “The Earth typically makes adjustments in major temperature spikes within two to three years. In this case as we cool down from El Nino, we are dealing with the combined effects of this planetary thermodynamic normalization and the influence of the more powerful underlying global temperature downturn brought on by the solar hibernation. Both forces will present the first opportunity since the period of Sun-caused global warming period ended to witness obvious harmful agricultural impacts of the new cold climate. Analysis shows that food and crop derived fuel will for the first time, become threatened in the next two and a half years. Though the SSRC does not get involved with short term weather prediction, it would not be unusual to see these ill-effects this year much less within the next 30 months.”
The SSRC further adds that the severity of this projected near term decline may be on the order of 0.9 C to 1.1 C from present levels. Surprising cold weather fronts will adversely impact all northern grain crops including of course wheat and the corn used in ethanol for automotive fuel….ETC
http://www.spaceandscience.net/id16.html
So now they are hiding their OWN decline! LOL
They decline to answer questions. They decline to show their data. They decline to even allow further questions. They decline, indeed, to be scientists because they decline to follow the scientific method and they decline to admit to mistakes they’ve made while appearing to be scientists…
And so they continue to manufacture false premises and as they announce one falsified horror after another, their coffers fill up with government Grants fed by the fear they create.
And so the decline of Science continues into the abyss of Political venality…
I used statistical analysis for many years in an industrial experimentation context. I can appreciate the complexity of analysis when not dealing with a designed experiment but rather analyzing “raw” data. (As was often the case to determine the experimental “bite size” of the designed study to follow.)
Getting “corrections” after the fact and especially after recommendations were provided to management were particularly onerous. Care must be taken when straying from what you know by using methods and techniques that can be helpful when used correctly and extremely misleading when used improperly.
A cautionary tale with “significant” implications.
What are the odds that all of Mann’s errors were unintentional and unbiased? The answer is probably a six-sigma event.
Willis,
No disrespect but it seems like a strange thing to write about in a peer-reviewed paper. What interest to academia is the to-and-fro of emails and blog postings between you and Mann. Or even the incorrect postings of data on a blog. What is the academic value of your E&E paper?
“Real Climate Science by Real Climate Scientists”
In the light of recent events, this is unfortunately all too true.Which is a sad commentary of the state of “climate science”
With the current state of Peer Review this might have been published in Nature or Science with your conclusions also being ignored.
Some pretty dodgy stuff gets through peer review as well – web publishing as exemplified by this post might well bring a new layer of accountability because of greater transparency. It also circumvents some of the covert censorship inherent in peer review processes. Ultimately, the greater market freedom of the web (while more anarchic) may allow for frank and free discussions.
In a geological context, perhaps it’s worth mentioning that, 56 to 34 million years ago, during the Eocene, the island of Spitsbergen, which is in the Svalbard archipelago, had alligators.
And I know what you all are thinking. No, it wasn’t at a much lower latitude. The Arctic was just MUCH warmer back then.
HR says:
May 12, 2010 at 8:13 pm (Edit)
That’s an interesting question, HR. More and more science is being done in the context of the Web, from pre-publication of interesting ideas to the type of analysis published by RC about Svalbard. All of this is (or should be) of interest to academia.
More to the point, when scientists such as Mann and Jones misbehave in public on a very visible blog, it is damaging to all of science, academia included. Academics seem to think that what happens in the blogosphere is of no interest to them. But millions of people (including academics) read blogs such as this one, and form their opinions of both the science and the academics from what they read.
As such, it behooves us all to behave “scientifically” when posting about science. For me this means that the work (like all scientific work) needs to be transparent, well cited, with the data clearly identified and code and methods explained, and that objections to the work be answered or at least acknowledged. You know, the standard scientific stuff that Mann and Jones didn’t do.
If we don’t do that, all of science suffers … including the academics. The arrogance of the people who run and write head posts at RC is one of the reasons that people don’t trust climate scientists much, and that affects us all.
A-pril in Svalbard … Sig-mas in blos-som …
zzzMmmpf! Sorry, must have dozed for a moment. Once again, thanks for an excellent article.
This is yet more confirmation that Mann, Jones, RC & Co. are not doing science, they are desperately doing propaganda to keep a fundamentally ridiculous fraud alive — which has been obvious for over a decade and is made devastatingly clear in the Climategate emails and programmer’s notes.
maxwell says:
May 12, 2010 at 7:28 pm
My point exactly. Their claim that the event was extraordinary and highly significant was not true. It was unusual … but not unusually so.
I wonder why the Willis’ stymied search for responses and accurate information from Mann & Jones sounds so ummm, …familiar?
Willis,
Awsome piece of detective work, wow.
That said, I followed the link in your post to the article on polar amplification on RC. I noticed something rather odd in the article.
Fig. 1 in the article quotes a 1980 study predicting (modeling?) temperature increase by latitude and height for CO2 quadrupling. Not double, quadruple. Based on the last few decades of CO2 measurements that would take over 600 years. They seem to show (I think) that surface temps in the tropics will rise about 3 degrees and arctic temps about 9. But then… look at Fig 2 which shows the surface temperature warming in degrees PER CENTURY from a DIFFERENT study and shows the arctic at…. 9! Now if in 1980 some wild eyed climatologists thought we would be averaging 600 million barrels per day by now, I can roll my eyes and laugh. What’s the excuse for still using that number in 2010? Or am I reading this wrong?
BTW – nothing in the article about surface temps being much lower in the arctic in the first place and hence by Stefan Boltzman more sensitive to CO2 forcing, which I find equally odd in an article about why the polar regions should be expected to heat up faster!
What’s the excuse for still using that number in 2010?>>
I should be in bed already. The date on the RC article is 2006. Which doesn’t really change the question.
Graeme W says:
May 12, 2010 at 7:32 pm
It was written as all of those. I was upset by the bogus scientific claims, outraged that they would not allow even simple questions about their work, and astounded that they either didn’t know or wouldn’t say where they got the data … so I wrote about it all.
The biggest statistical error is not accounting for selection bias.
How many stations are there to choose from? If there are 100, then the odds of a 2 or 3 sigma event happening at one station are pretty high.
Now if they want to pick a SINGLE station this year to be analyzed for outliers NEXT year, then I might pay attention. Otherwise, it is just cherry picking.
I would not be surprised if after correcting for the number of stations available that this result becomes a 1 sigma event.
James
I note that your last graph shows the same pattern post 1990 that I’ve found in many other locations. A “hockey blade” rising with a ‘clipping’ of low going monthly data. Note that the tops run along about the 5 line, it is the bottoms that make the “trend” in the annual line. As near as I can tell, this is an artifact of the “QA” process that tosses out low going extremes more than high going ones. It is a serious flaw in the product of GHCN and GISS (and one presumes, CRU as well).
A very valuable project would be to gather truly raw daily data from some selected sites and compare it to the computed monthly means. Basically, audit the “QA” process (that despite it’s name, does more than simply check quality… For the USHCN it will also create and fill in ‘missing data’ with averages from surrounding locations; so if you toss low extreme events and fill them in with averages, what happens to the monthly mean you then calculate from those ‘data’?… ) I suspect a similar thing is done to the GHCN data.
You misspelled “heck” in the title. Touch typing is not for the inexperienced.
As one whose initials are ‘RC’ I take personal offense at the association.
That April 2006 outlier looks very suspicious. How does it compare with other stations in the Arctic region?
Maybe its another case of a missing M.
Svalbard looks to be on the tip end of the Gulf Stream, which suggests it should get a few odd warmer temperatures from time to time.
http://en.wikipedia.org/wiki/File:Ocean_currents_1943_(borderless)3.png
I had cause to look up Grimsey today, an island a bit north of Iceland right on the Arctic Circle. Yet their record max temperature is as high as 26 C, which is again linked to the North Atlantic Current.
pat says: (May 12, 2010 at 7:45 pm) 10 May: Space and Science Research Center: […] As the Earth’s current El Nino dissipates, the planet will return to the long term temperature decline brought on by the Sun’s historic reduction in output, the on-going “solar hibernation.” […] In this case as we cool down from El Nino, we are dealing with the combined effects of this planetary thermodynamic normalization and the influence of the more powerful underlying global temperature downturn brought on by the solar hibernation.
I hope someone will pick up the quotes above and provide some opinions. The attribution to the state of the sun seems far more definite to me than many commenting here on WUWT would accept. Would you accept this certainty, Leif?
FYI: “Academics urge new approach to climate change, CO2 should not be the prime target of climate policy, the report argues” http://bit.ly/aUw6GS
“The Svalbard dataset is available here. The file, Nordklim_data_set_v1_0_2002.xls, is a 9.1 meg Excel file containing a variety of climate data for a number of Norwegian sites. The Svalbard station has the Norwegian station ID 99840, and the Svalbard monthly mean temperature data starts on line 4639 of the sheet called “101”.”
*
*
The data set no longer exists at the URL you’ve listed.
Any suggestions as to where it may be obtained?
Thank you,
How ironic that the web publication (outside of peer review) that you use to demonstrate that the reader must beware of reliance upon, is an RC paper authored by of all people Mann and Jones.
As most WUWT readers have observed, even Obama is now ranting in graduation addresses against over reliance on i-this and i-that, that he said he did not indulge in (right). I don’t think he was taking aim at Apple’s technology as much as he was pointing out to a graduating audience that he was addressing that in his mind “we” have become over reliant on web-based information. I interpreted those remarks to caution the graduates to not stray too far from main stream media reports and public reports issued by his governmental agencies, and to stop thinking independently from the public media as a result.
Your paper analyzes a situation that occurred four years ago. That is before Climategate and the investigation of these very parties by various agencies for just this type of transgression in data manipulation at public cost. And the conclusion, that you have to be careful about what blog-based information is relied upon, is particularly astounding since it is based upon a vigorous peer reviewed analysis of an RC paper by the very scoundrels implicated by Climategate emails, that is relied upon and endorsed by Algore and Obama.
This article should be brought to the attention of the Virgina AG (if it hasn’t already). Some of this work may have been funded under the time frame of his investigation.
Keep it up!!
It can’t be true. Lord Oxburgh has come to the conclusion that the scientist chaps and chapesses at CRU are sterling characters who wouldn’t dream of trying to hoodwink people.
Oh wait…
“It was a 4.6 sigma event, which has a probability a full order of magnitude smaller, less than one in 10^5.”
I’m sorry but the wording here isn’t quite right. I think you mean a full order of magnitude larger not smaller. Yes, 10^5 is smaller than 10^6, but the “one in” (an the fact that these are probabilities) implies that the you are actually talking about probabilities of 10^-5 and 10^-6, in which case 10^-6 is smaller.
I assume you know this, of course, because your interpretation of the meaning of the probabilities was correct-the improbability of the event was exaggerated by RC, that is, they implied it was smaller than it really was even with their own numbers. However, when you say that they made the probability too small, that’s the opposite of what I think you mean.
A close look at the changes of the Geomagnetic field at Svalbard (and the nearby North Pole) may finally give us a clue what drives climate of the North Atlantic.
http://www.vukcevic.talktalk.net/LFC16.htm
It must be very difficult to splice the Isfjord Radio and Svalbard Lufthavn (“Svalbard Airport”) records. The sites are only about 40 km apart but have very different local climate. Isfjord Radio is a very exposed site on the outer coast, while the airport is deep inside Isfjorden, close to Longyearbyen (though probably not close enough for UHI). Most important, while there is often quite dense ice along the outer coast in winter, Isfjorden almost never freezes because of upwelling Atlantic waters.
899
SMHI has revised its website. The Nordklim dataset is now available at:
http://www.smhi.se/hfa_coord/nordklim/data/Nordklim_data_set_v1_0_2002.xls
Your record of the RC behaviour is unveiling. It should however be noted that later research has demonstrated that temperature events at Svalbard (Spitzbergen) in 2006 was a rather unique event, but not unprecedented.
Svalbard sits high north at the edge of the Atlantic current, resulting in a quite complex climatic situation. Usually, the Western Spitzbergen current carries warmer Atlantic water northwards along the west shelf of Svalbard. (And a gyre of cold polar water SW of this).
For a typical view see upper right corner :http://www.osdpd.noaa.gov/data/sst/anomaly/2010/anomnight.5.10.2010.gif
Around 2005, steady southernly winds had strengthened this current and accumulated unusually warm water north of Svalabrd. In 2006, dominating northernly winds flushed this warm water back and pushed it into the fjords of W Svalbard, filling the fjords with “warm water” rendering them ice-free.
This event had profounf impact on the average temperature at Svalbard. “Isfjord radio sits at the S shore of Isfjorden at the mouth of the fjord, Longyearbyen is at the bottom of one of the fjord’s arms.
Here is a popularized accont of the event (unfortunately in Norwegian).
http://www.forskning.no/artikler/2007/november/1193913609.81
Scroll down to the map to find the relevant picture.
The popularized paper descibes how wind not climate change was the best explanatioin for the event (and BTW also noted similar events in the 30ties).
I assume it should be possible to find the relevant research published in peer reviewd journals. Search terms: Frank Nilsen, Finlo Cottier, Svalbard, Spitzbergen .
Cassanders
In Cod we trust
@ Willis Essenbach. I am very impressed by the systematic, persistent and compelling way a scientific statement from fellow scientists critically questioned and their final hypothesis is refuted. This is the ultimate and rational debate that science needs to move forward and thus thought to serve humanity. Bravo and thanks.
However, I disagree with your prologue to the post stating “The ‘Svalbard Affair’ is a clear demonstration of the problems that can be posed to scientific standards by the increasing use of ‘web publication’ that by-passes the traditional method of scientific publication. Errors that could be expected to be detected by peer review can be published on web sites. In the case of the ‘Svalbard Affair’ the problems with the published work were so obvious that they were immediately seen and reported to the authors. But the response of the authors was to attempt to try to stop any question of the matter, and this would also have been difficult if traditional publication had been used. Other less obvious cases may go undetected before data published on the web have become used in related studies by other workers.”
In general the mind-set of the scientists should be by nature: open mind for critics and the active seek to it, patience in answering (critical) requests for explanation and justification by others and be creative in the design and testing of alternative explanations and theories. This is not only a question of having the right mind-set and personality for becoming a good scientist, but also of education and coaching.
Scientific procedures, like peer-reviews, could be of good help to perform according this high professional standards. But procedures will never guarantee quality, because even procedures could be overruled, mislead and manipulated. In general the Internet is a – thanks to Google c.s. – the most open community ever invented for the mankind for discussing, opinions and critics. And this give hope for a higher level of professionalism.
So in my opinion it is not the lack of peer review in the case of the 2006 RC posting “More On The Arctic”, but simple the lack of scientific professionalism of the authors.
And the internet could be a mighty mean to discover this kind of “fraud”, as long as the discussion is open, not under the direction of the government, and has the low entrance for the participants. In that regard is it not good that much of the scientific production – almost always fully financed by the government – is behind pay-walls of commercial publishers.
Complication, now a days, in the scientific discussion around climate, is that the discussion is always a political discussion. The research is financed by governmental money, who expect of course to be delivered by science with political correct results. These mechanism leads also to manipulate the peer review process which was proved in the Climategate case. Its thus also going about independency, especial form government and fundamentalist.
The scientist simply shouldn’t sell their soul to the devil for money and honor. How it works is clearly demonstrated by your remark: ” For example, the UKCIP (UK Climate Impacts Programme) is based at the University of Oxford and is funded by the UK government to help organizations assess how they will be affected by climate change so they can prepare for its impacts. On 12 June 2006 (i.e. weeks after RC closed the Svalbard thread) UKCIP circulated an email newsletter saying; 5. Artic study demonstrates temperature rise ……”.
There are still scientists, even in Oxford” , who will play a role in the propaganda scenery.
Jantar says:
May 12, 2010 at 10:11 pm
I haven’t compared it to other stations. I did, however, get the daily max and min data for the station for April 2006 from eKlima (free registration required), and calculated the monthly average from the daily data. The GISS monthly average was exactly equal to my calculation of the average of the daily data, so I think it is real.
REPLY: Willis, see the next story at WUWT, some news of interest. -A
Could not agree more. Especially:
899 says:
May 12, 2010 at 11:06 pm
Thanks, I’ll change the head post. Info on the dataset is here, and the dataset (in XLS form) is here.
The NORDKLIM dataset only runs to 2002, and is identical with the GISS dataset for recent years. The GISS dataset continues to the present.
899: You may find the Nordklim data set here: http://www.tiede.fi/keskustelut/geologia-maa-meri-ja-ilma-f7/maapallon-pintalampotilan-seuranta-t29467-705.html
The Svalbard airport record has been reconstructed from several other stations, not only Isfjord radio (which is in another climate zone: Isfjord radio remains relatively ice free throughout the year, Svalbard airport does not). Some years ago I asked the Norwegian meteorological institute about how the homogenisation was done, and they pointed me to this paper: “Homogenity analyses of temperature and precipitation series from Svalbard and Jan Mayen”, P.Ø. Nordli, I. Hanssen-Bauer, E.J. Førland, Report NO. 16/96 KLIMA. I found the paper listed in the Norwegian polar institute library in Tromsø and they kindly photocopied the article and mailed it to me. The reconstruction has been refined since this paper, which I haven’t been able to find anything on, but the changes don’t seem to be great.
According to the paper the temperature record was reconstructed from these stations:
1911-12 – 1916-10 Green Harbour
1916-11 – 1920-05 Longyearbyen
1920-06 – 1921-09 Green Harbour
1921-10 – 1923-08 Longyearbyen
1923-09 – 1930-08 Green Harbour
1930-09 – 1934-08 Longyearbyen
1934-09 – 1934-12 Barentsburg
1935-01 – 1935-09 Longyearbyen
1935-10 – 1936-10 Barentsburg
1936-11 – 1939-06 Longyearbyen
1939-07 – 1941-08 Barentsburg
1941-09 – 1941-11 Interpolations
1941-12 – 1942-06 Longyearbyen
1942-07 – 1945-08 Interpolations
1945-09 – 1946-08 Longyearbyen
1946-09 – 1947-12 Isfjord radio
1948-01 – 1956-12 Barentsburg
1957-01 – 1975-07 Longyearbyen
1975-08 – Svalbard lufthavn
Longyearbyen and Svalbard lufthavn are relatively close, about 4 km, but I would say that temperature between the two locations quite often differs quite a bit. Data from the individual stations are available at eklima.met.no (free, but registration is required).
Note the are interpolations during WWII. The record for these years were based on a reconstruction for Isfjord radio, which in turn was based on data not even from Svalbard. So there is a considerable amount of guesswork here.
I’m not saying that the guesswork is poor, but I think it’s unfortunate that the reconstructed Svalbard airport series is a reference climate station often used in climate studies. Much guesswork and uncertaincy are hidden in this series.
tty says:
May 12, 2010 at 11:57 pm
Agreed. That’s why I said that regardless of their findings, using statistics on a spliced dataset that I described as being “in a different climate zone 43 kilometres away” was a mistake.
# Cassanders says: May 13, 2010 at 12:17 am
>>>>>>The popularized paper descibes how wind not climate change was
the best explanation for the event (and BTW also noted similar events in
the 30ties).<<<<>Postscript: 8 Oct 03 << saying:
___“Willis Eschenbach did a closer examination of Vardø and also found the same discontinuity around 1920, amounting to 0.73°C. When that artificial discontinuity is discounted, the temperature rise is only +0.12°C per century, a tiny result for a region that according to the models should have undergone rampant warming in the last century.”
The Eschenbach-Figure is at: http://www.john-daly.com/stations/vardo2.gif , showing an increase from 1919 to 1939 !!
For comparison see here: http://www.arctic-heats-up.com/pdf/7_2.pdf )
Cassanders says:
May 13, 2010 at 12:17 am
Cassanders, that is a fascinating article. Here’s the translation of the first part:
Many thanks,
w.
PS – as a commercial fisherman, I loved your tagline, “In cod we trust” …
Re: Maxwell
Perhaps Willis could have phrased this better. Its 30 years since I last did any statistics so I might be wrong but it is my understanding that when you say a “5 sigma event” what you actually mean is that the event is 5 standard deviations from the normal. Thus if you decrease the standard deviation you increase the number of sigma.
Although Willis doesn’t explicitly say “and thus increase the number of sigma” it is clear from the context of the text that that is what he means.
Willis,
You can find some details about the reconstruction of the Svalbard airport series on this page (in Norwegian):
http://voksenlia.net/met/lyr/
I wrote the page a couple of years ago and intended to dig a bit deeper, but I didn’t finish.
You mean just like Alaska is north of the USA? 🙂
IN full lenght my previous comment (1:07, May 13) goes as follows:
# Cassanders says: May 13, 2010 at 12:17 am
>>>>>>The popularized paper descibes how wind not climate change was the best explanation for the event (and BTW also noted similar events in the 30ties).<<<<>Postscript: 8 Oct 03 << saying:
___“Willis Eschenbach did a closer examination of Vardø and also found the same discontinuity around 1920, amounting to 0.73°C. When that artificial discontinuity is discounted, the temperature rise is only +0.12°C per century, a tiny result for a region that according to the models should have undergone rampant warming in the last century.”
The Eschenbach-Figure is at: http://www.john-daly.com/stations/vardo2.gif , showing an increase from 1919 to 1939 !!
For comparison see here: http://www.arctic-heats-up.com/pdf/7_2.pdf )
Lies, damn lies and climate science.
Steinar Midtskogen says:
May 13, 2010 at 1:55 am (Edit)
Many thanks, Steinar. The page can be read in English by using the Google Translator, and entering the voksenlia URL in the top box. Select Norwegian to English, and it will return a URL below the boxes for a translated web page below. Good stuff.
w.
That is not the original port, it moves.
They shift it, that’s not real arsefault and airport is on sleds, just an arsephault carpet for visitors. It’s an ice port. Actually it’s an ice Island with an eigen rotation value of 3.7. The open water stays the same, the Island rotates.
The open water is subter fluggen rotchten flume slow emission volkanic lactactic effect.
Everyone knows that. Been doing it 1500 years.
The following TV program was shown on the Norwegian state TV channel NRK today. It is in english language and claims to show the “consequences of increased CO2 emission for the oceans”. From about 31 minutes into the program, there is a long segment from Svalbard, including conversations with british scientists at Ny Ålesund. It is a typical alarmist program.
Eit hav i endring (“A changing ocean”)
http://www.nrk.no/nett-tv/klipp/634248/
This stuff is pushed on our kids continuously. I saw it with my 12 year old son. Sceptical as he is, he didn’t quite believe me when I told him this was all untrue preudo-science……
Sorry, that should be “pseudo-science” of course.
lies, damn lies, statistics
No knows why the floater dont shift. But it don’t.
Some say Geodesic memory from the time it was the magnetic pole. I don’t know, the great Explorer Hans Barnstade, Christian sub Andersonian Wolfe-Smythe said it is enigmatic being this close Polaric and Messampiotical, to the North pole, Mythological, Astorogical, Meterological and Magnetical.
But the only roads to the airport are Ice tracks, there are 360.
The port stays the same for mariners but the truckers need a lot of road signs.
The wonderfully topsy-turvy world of skeptical maths:
The odds a five-sigma event occurring by chance are one in 1,744,000. This is the “less than one in 10^6” figure Mann & Jones claimed in their article.
> 1744000<10^6
[1] FALSE
Of course that port is also recognised as a cancer treatment centre, the Rooman Zen Centre for Holistic cancer treatment.
It has one hundred per cent success rate.
Cancer Victims are taught the treatment over the Lactactic Volcano.
THey swim out out 50 metres and perform the ceremonial arm wave. When they stop waving they are cured.
Not one patient has ever complained, no matter how severe.
Some however get ecstatic about the success and wave two arms.
Last thing we need maritime is [snip, funny, over the line ~ ctm]
Svalbard data from Nordklim only extends to 2001. Where do I find the 2001-2006 data?
TerryS,
‘Although Willis doesn’t explicitly say “and thus increase the number of sigma” it is clear from the context of the text that that is what he means.’
I understand statistics pretty well having to use it on a regular basis, but the way it was phrased was confusing to me. So I have a hard time believing the context was clear.
As even Willis points out, these types of web science publications have to take great care in preparing an argument that is well thought out because there is no peer review and, taken at their word, a poorly presented (or manipulative, as seems to be the case with this particular RC post) argument can cause wrong conclusions about the science.
Willis,
still wondering, what significance test did you use to determine that the trend since 1990 is not significant?
Thanks and cheers.
Richard Telford: “> 1744000 b, then 1/a < 1/b — no? It's the difference between "Less than one in X" and "One in more than X"
SORRY for trying the third (and last) last time to get the full text of my comment transmitted:
# Cassanders says: May 13, 2010 at 12:17 am
“The popularized paper descibes how wind not climate change was the best explanation for the event (and BTW also noted similar events in the 30ties).”
It was certainly not the wind that pushed the winter temperature to rise at Spitsbergen dramatically from winter 1918/19 to 1921/22, which continued until about 1940, but had been caused be the West Spitsbergen Current (see here : http://wattsupwiththat.com/2009/11/04/arctic-warming-goes-with-the-floe ; And here: http://www.arctic-heats-up.com/ )
A. W. Ahlmann ( 1946) called the event a ‘climatic revolution’, and B.J. Birkeland (1930) saw the temperature rise, as “probably be the greatest yet known on earth”. These assessments are correct until today. Never since meteorology was establish as a scientific discipline two centuries ago, such a dramatic temperature rise had been ever observed again. (http://www.arctic-warming.com/introduction-the-scope-of-this-investigation.php ).
That a climatic shift took place in the high North Atlantic at the end of WWI, is well demonstrated at: http://www.john-daly.com/press/press-03b.htm , Postscript: 8 Oct 03 saying:
___“Willis Eschenbach did a closer examination of Vardø and also found the same discontinuity around 1920, amounting to 0.73°C. When that artificial discontinuity is discounted, the temperature rise is only +0.12°C per century, a tiny result for a region that according to the models should have undergone rampant warming in the last century.”
The Eschenbach-Figure is at: http://www.john-daly.com/stations/vardo2.gif , showing an increase from 1919 to 1939 !!
For comparison see here: http://www.arctic-heats-up.com/pdf/7_2.pdf )
Leif Svalgaard says on Tips & Notes in regard http://www.spaceandscience.net/id16.html:
“Anna, Casey is a nut and a liar. Don’t pay any attention to this self-serving nonsense.”
Which I guess fully answers my question above ( Roger Carr — May 12, 2010 at 10:31 pm) “The attribution to the state of the sun seems far more definite to me than many commenting here on WUWT would accept. Would you accept this certainty, Leif?”
I have posted on Realclimate but the experience was frustrating and non productive.
I posted some comments on the subject of positive feedback without which CAGW is dead in the water. I got many replies, but when I tried to reply to them my comments were deleted and the moderator inserted a comment saying that one of the replies explained it all perfectly and if I disagreed with the reply that I was wrong.
The reply I was trying to reply to had gross errors which I pointed out.
Arguing on realclimate is a waste of energy because the censorship is so extreme. If you post a particularly good response it will never be published. [Probably to convince readers that you have been shut up by superior logic.]
Since Willis was writing for publication in a journal, I’m not entirely sure how much weight to put on “tone”, since he obviously would have realized there was only so far he could get away with on the “snark” front. But generally speaking, I still get a sense from this 2006 article of how far the obfuscation and stonewalling of the “lords of data” have frustrated and radicalized the skeptics between 2006 and 2010. In 2006, one gets a sense of genuine puzzlement and “wtf?” on data sharing issues. In 2010, the puzzlement is gone. . .
I agree with Jantar – that April 2006 outlier does look very suspicious – especially when you see that its value is zero (0.0) degrees.
And the green September outlier looks like it is 5.0 degrees, and the October 1968 outlier looks like it is -15.0 degrees – all whole numbers? Not sure about the orange outlier though.
Could these points all be errors in the spliced database – perhaps where a point has been left out or two points for the same month – and a nominal value entered?
Huh, the link you give to the supposedly correct data set goes to the swedish met. institute, not to norwegian met.
Why is that?
Or did the mann jones couple actually use a swedish managed data set but still managed to screw it up and refer to a norwegian one? That’d be hilarious. :-()
You do great work, Willis.
You root out the truth.
John Murphy says:
May 13, 2010 at 4:19 am (Edit)
John, the GISS data (available here) is identical to the recent Nordklim data. So I used it to extend the Nordklim data to the present.
Anthony and Willis:
This is the kind of posting that keeps me (and so many others) coming back to WUWT. My compliments to you both.
d.
“richard telford says:
[…]
The wonderfully topsy-turvy world of skeptical maths:
The odds a five-sigma event occurring by chance are one in 1,744,000. This is the “less than one in 10^6” figure Mann & Jones claimed in their article.
> 1744000<10^6
[1] FALSE
"
The mathematical prowess of expert climatologers (using a computer). Sigh.
PJB says:
May 12, 2010 at 7:49 pm
I used statistical analysis for many years in an industrial experimentation context. I can appreciate the complexity of analysis when not dealing with a designed experiment but rather analyzing “raw” data…..
Getting “corrections” after the fact and especially after recommendations were provided to management were particularly onerous. …
A cautionary tale with “significant” implications.”
____________________________________________________________________
I have watched various “flavor of the month” statistical toolboxes taught in industry and cringed. There is nothing more dangerous that handing a set of statistical computer programs to a scientist who has never learned the difference between discrete and continuous data much less gaussian, or exponential curves.
I have a real problem with these “sliced data sets” we keep seeing in the temperature record. The data points do not belong to the same distribution darn it . To me it is the same as mixing product from two different extruders at two different factories and then trying to do analysis on the data. You loose much of the information and you may draw some really bad conclusions especially if you were not aware the data was mixed in the first place. Good way to chase your tail looking for phantom problems.
899
I downloaded the .xls file last night. However, the data doesn’t go past 2001.
Willis
Ta.
Willis
If you look at the GISS graph for this data on their website, it commences only in 1980 and gives a very, very misleading picture of the temperature history there.
If NORDKLIM has the data, how come GISS doesn’t? These databases aren’t worth two bob, yet crooks like Mann and that ponce Jones expect us to shell out a kazillion dollars on the strength of their “interpretation” of the data they contain.
John Murphy says:
May 13, 2010 at 4:06 pm
Well, NORDKLIM has the early data, both have data in the middle, and only GISS has the recent data. This is not uncommon. There’s a host of stations out there where you can find information on the web right up until today, but GISS or GHCN don’t seem to be able to find them.
If I were in charge of the data, I’d pull it all out of circulation for a couple of months and go through every single station. I’d graph the distribution of every station’s data, I’d graph the amount and direction of all adjustments made to the data, I’d search every single station for more data on the web, I’d run every conceivable test to find bogus data points, and a whole host of other checks and error traps, I’d re-write old software in a modern language with clear commenting and version control, and in a month or so I’d have a clean dataset to show folks. Is there money to pay for that? We’ll, they’re all drawing salaries, so put them to work for a change, instead of spending company time posting at RC.
But then I’m from the old school, where when a man is hired for a job he’s expected to do that job really well before he branches out as a politician and a post-normal scientist and an anti-scientific blog owner and a civil disobedient …
@Willis Eschenbach (May 13, 2010 at 11:26 pm)
“I’d re-write old software in a modern language with clear commenting and version control”…
In the case of GISTEMP, Clear Climate Code have already rewritten it in a modern language (Python) with what we hope is clear commenting and of course we use version control.
It can always be made clearer of course; would you like to contribute?
Willis
I think when you use the term sigma, you are referring to the variance which is the square of the standard deviation which in turn is normally denoted by teh Greek sigma.
Willis
Sorry about the typos in my last.
If the April 06 number is a “six sigma” event (according to you know whom), I winder how come and why their much vaunted quality assurance systems didn’t reject it as spurious.
The best thing that could happen to RC would be if people with a POV different to the moderators stopped posting and it simply became an echo chamber.
drj11 says:
May 14, 2010 at 12:42 am
I do my work in R, Mathematica, and Visual Basic (and C if it has to run wickedfast), and my knowledge of Python is somewhere between zero and … well … zero.
But I definitely wish you well, and I wish you had done it in R … why Python, when R handles climate data so well and cleanly?
Thanks,
w.
John Murphy says:
May 14, 2010 at 3:06 am (Edit)
Typos are a fact of life.
It’s not a six sigma event. However, it is a real event, as the daily data shows.
@Willis: “why Python, when R handles climate data so well and cleanly?”. This article discusses some of the reasons why we chose Python. Basically it comes down to: we already knew it; it’s already promoted (by others) as a language for writing clear code in.
The actual bit of code that “handles climate data” (file input/output) is not actually very important.
drj11 says:
May 19, 2010 at 12:12 am
Good reasons all … but in my opinion, far from good enough, as I discuss below.
Sorry for my lack of clarity. By “handles climate data” I didn’t mean file in/out. I meant doing the actual calculations.
For example, suppose you have a large block of climate data called “climdata”. Suppose it’s a 60,000 row by 12 column dataset, where rows are days and columns are months. In most languages, to get the anomaly of that data, you need to say something like the following:
## Pseudocode##nrows = rows(climdata)
ncols = columns(climdata)
dmean = mean(climdata)
for n = 1 to nrows
for m = 1 to ncols
climdata[n,m] = climdata[n,m] - dmean
next m
next n
In the computer language “R”, on the other hand, you simply say
climdata = climdata - mean(climdata)Can’t get much simpler than that … in fact, I don’t think you can get any simpler than that.
It gets worse when (as is common in climate data) there is missing data. In R, missing data is represented by “NA”. NA propagates through calculations, yielding NA so you immediately know if there is missing data. If the climdata contains NAs, then you can get the anomaly in R using the code:
climdata = climdata - mean(climdata, na.rm=TRUE)Depending on what other language you are using, handling missing data can get quite complex.
Here’s another common situation. We need the annual means of our dataset, which contains missing data. In pseudocode, this would be:
## Pseudocode##nrows = rows(climdata)
ncols = columns(climdata)
dmean = mean(climdata)
for n = 1 to nrows
temp=0
monthcount=0
for m = 1 to ncols
if climdata[n,m] != NA then
temp = temp + climdata[n,m]
monthcount = monthcount + 1
next m
result[n] = temp/monthcount
next n
In R, this would be done with the following code
result = rowMeans(climdata, na.rm=TRUE)Now, suppose we decide that we need means by month rather than by year. While you are laboriously rewriting and re-debugging your Python code, I write
result = colMeans(climdata, na.rm=TRUE)I know which one I’d use … and I am very sure that the R code is easier to debug and less likely to contain hidden mistakes.
So yes, you can pick a language because you know it already … but tragically, that has not lead you to pick the best language for the job at hand.
Arctic warming is not due to a greenhouse effect, much less to an “amplified” greenhouse effect. It began suddenly at the turn of the twentieth century after a two thousand year cooling trend. It is highly likely that its cause is a rearrangement of the North Atlantic current system at the beginning of the century that directed the Gulf Stream unto its present northerly course. Today it enters the Arctic in a broad front between Iceland and the Skandinavian peninsula and keeps the Russian Arctic ports ice free in the summer. Svalbard, which includes Spitsbergen, is on its left flank, on the border between the ice free and permanently frozen Arctic, and any slight variation in the path taken by the Gulf Stream should have an exaggerated influence on Svalbard temperatures.