Record Arctic Warmth – in 1937

In this post I take a look at the GHCN V2 records for 32 stations in the Arctic and sub-Arctic areas of Norway, Sweden, Finland and the adjacent part of Russia (Figure 3). The data throw up some surprising results.

  • Despite coming from 4 different countries, these are among the most congruous time temperature series I’ve seen, i.e. the temperatures go up and down together across the region, with quite substantial oscillations of ±2˚C (Figure 1).
  • Averaging the data reveals that 1937 (and 1938) were the warmest years in this area (Figure 7).
  • At the end of the Little Ice Age, 1881, 1888, 1893 and 1902 were anomalous cold years. This cold start to the time-temperature series creates the inevitability that a regression through the data will have a positive slope (Figure 3). Had the thermometers not been deployed until 1920, then the regression would have been virtually flat (Figure 9).
  • Once again we have essentially a flat rangebound series with three main legs to the data. Warming until 1938, cooling until 1968 and then warming to the present. It was just as warm in this part of the Arctic and sub-Arctic in the 1930s as it is today.
  • The structure of the data is similar to the North Atlantic Oscillation suggesting that this climatic feature has played a role in modulating temperatures in this part of the Arctic. The temperature response is often 1 and sometimes 2 years delayed.
  • The structure of the data is also quite similar to a compilation of 21 records centred on Olenek in East Siberia.

Figure 1 dT spaghetti for the 32 station records using the station average as a base for calculating dT. There is a high congruity between all records suggesting that this quite large area that extends to Bjørnøya in the Barents Sea responds as a single congruous climatic zone.

Data and methods

GHCN V2 records as downloaded from the NASA GISS web site have been used. I placed Karasjok (I once dated rocks from the Karasjok Greenstone Belt) at the centre of a GISS search that returned the records shown in Figure 2. The distribution of stations is shown in Figures 3 and 4.

The metANN (DJFMAMJJASON) annual averages were used. The temperatures were converted to anomalies by taking the mean temperature for each station and deducting that mean value from the time-temperature series of each station. It is preferable to use a fixed base period, for example 1963 to 1992 has been used before. For a number of data sets I have compared the fixed base with the station average base and derived very similar results. It is simply easier to use the station average.

The average gradient is determined by summing (averaging) the dT stack for all stations. This too raises methodological questions. 4 stations have long continuous records (1880-2011) and these are plotted separately in Figure 5. A regression has been run through each station and the gradients averaged. This gives an identical result to summing the dT stack and running a regression through that average (Figure 5).

Figure 2 The stations used in this study. There are two sizeable cities – Murmansk and Archangel but neither show signs of urban warming.

Figure 3 Station locations are spread across 4 countries stretching from Bjørnøya (Bear Island) in the Barents Sea to the North to Onega in Russia to the South.

Figure 4 This area has a large number of stations with old records, many of which have unfortunately closed since. In general there is excellent station cover.

Figure 5 There are 4 long continuous records in this group that run unbroken from 1880 to 2011 (Haparanda stops in 2009). The regressions through each record are shown. Averaging those yields +0.98˚C per century identical to the gradient through the sum of the dT stack.


The temperature spaghetti plot (Figure 6) shows how strongly congruous these records are. There is a strong sense of flat records going across the chart and not much sense of warming. Note that some of these stations have average annual temperatures below zero and some above zero.

The conversion to dT spaghetti is shown in Figure 1 which emphasises the flat and strongly congruous nature of these records. The average of the dT stack (Figure 1) is shown in Figure 7.

Figure 6 Temperature spaghetti for 32 Arctic and sub-Arctic stations.

Figure 7 Averaging the dT spaghetti (Figure 1) produces this summary. One feature of these data throughout is quite large inter annual variability with warm years interspersed with cold years and average temperatures varying by as much as 3˚C from warm to cold. There are three legs to the data. Warming to 1937. Cooling to about 1968 followed by warming to the present day with a quasi 80 year cycle. One surprising result is that 1937 and 1938 were the warmest years by some margin. 

Figure 8 Comparing dT with the North Atlantic Oscillation Index reveals a fair degree of co-variance although the correlation is far from perfect. The frequency and style of oscillation are very similar. Quite often the temperature and NAO index are offset with temperature lagging the NAO by one year and sometimes two years. On other occasions there is very tight alignment, for example 1941/42 but even that may be one year displaced. A regression through the NAO index has negative gradient and so this cannot be used to explain the quasi 80 year structure.

Figure 9 Starting the regression in 1920 produces a near flat trend. There has been no warming in this part of the Arctic for 95 years.

Figure 10 The overall pattern seen in Northern Scandinavia is very similar to that seen in east Siberia. The E Siberian records are centred on Olenek, 2980 km east of Murmansk. The climatic regime in E Siberia is more harsh with larger seasonal swings. It is both interesting and surprising that the patterns are so similar.


Arctic warming and the “disappearance of sea ice” seems very much to be in the news. It is difficult to understand the basis for this panic based on these data or the data from Olenek and Iceland. Sure, there has been some warming since 1966, but this seems little different to the warming that took place in the period 1902 to 1937/38.

Is there evidence for Arctic warming elsewhere? Looking to the north there are a couple of stations on Svalbard. The Svalbard airport record begins in 1977 and so only catches the recent warming leg seen in all the other records. Taking this record in isolation presents a false picture. Also on Svalbrad is Isfjord Radio. But that record stops in 1980 and therefore misses any recent warming. Going across the Atlantic to NE Greenland there is a useless record at a military base called Nord Ads. It begins in 1952 and also misses the warm 1930s that provide context.

Svalbard Airport
Isfjord Radio
Nord Ads

If we go to Iceland, using the V2 records we see a similar picture to N Scandinavia with equally warm temperatures in the 1930s. In Iceland, V3 homogenised records create an image of warming created in the mind of the homogenisation bot.

This entry was posted in Climate change and tagged , , , , , . Bookmark the permalink.

42 Responses to Record Arctic Warmth – in 1937

  1. Andrew Thickpenny says:

    Fascinating data again Euan. Does anybody know if there are data on the extent of sea ice in the arctic in the earlier part of the 20th century, including the warm period in the 1930s?

    • Euan Mearns says:

      There was a post on WUWT some time ago summarising data from old sea charts. But I can’t put my hand on it or on a summary of Arctic Sea ice 1880 to present that I believe. And I’m not sure how relevant this might be.

      There are two main areas of sea ice loss in the Arctic. One is the Barents Sea and the other is the Sea of Okhotsk. The former may be relevant to current N Scandinavia data set. The latter is a long, long way away.

      My own prejudice is that Arctic Sea ice decline is more likely linked to ocean currents – the strength and temperature of the Gulf Stream. That of course may in turn affect land air temperature and precipitation patterns.

      • ristvan says:

        There are two sets of summer sea ice observations. DMI has maps, unfortunately interupted from 1939 to after the war. And there is a series of Russian observations and papers that have been summarized in English. Detailed examples and references are given in essay Northwest Passage.

        • ristvan says:

          Forgot to add, this qualitative historical Arctic summer sea ice info more or less confirms Euan’s analysis here. Also fits the Iceland record. I also think ocean currents are very important to ice formation/dissolution. But oceans also have a strong influence on seacoast station temperatures when they are not frozen.

    • Ben Vorlich says:

      There is some information on arctic convoys in WW2 including a map of routes and approximate ice extent,here,

  2. A C Osborn says:

    Another nice piece of analysis.
    Totally wrong of course because you are using real data, only homogenised data will do you know. /Sarc
    Clive also has some very interesting data in his latest post on his “find” of old GHCN V1.
    What is fascinating is how the current GHCN output now pretty well matches the Jones’ CRU data where the original raw data got lost when Stev Mc asked for it so he could check Jones’ work.
    Looking at the original GHCN data you can see why he conveniently lost it.

  3. Euan Mearns says:

    Not exactly on topic, but Clive Best has found a version of GHCN V1 and has made this publicly available today. BEST raw and V1 recalculated as metANN. There is overall a very good match with a couple of interesting departures where BEST raw follows V2 and not V1.

  4. Are you aware of Tony Brown’s work on historical tempreatures?

  5. Dave Rutledge says:

    Hi Euan,

    Great work.

    It is interesting to compare this to the state temperature records in the US, which are not homgenized, as far as I know.

    24 were set in the 30s. 1 was set in this century.


  6. Greg Goodman says:

    Slightly off topic here ( SH temp ) but you posted a link on Roy Spencer’s blog with a link to Andrew’s 300 high quality stations vs UAH. This must be the now “old” UAH 5.6

    It would be interesting to see this as a difference plot and compare the UAH 60-UAH5.6 difference:

    By eye there appears to be some agreement ( supporting the new UAH release ) and some not so similar, like 2002-2012.

  7. Neil Hampshire says:

    Looks like a clear stadium wave trend to me.
    Have you compared the data to the AMO?

    • I can’t check this because the helpful people from Microsoft wiped all my spreadsheets out yesterday (hope to get them back today), but as I recollect Arctic temperatures are indeed quite closely correlated with the AMO. There is, however, a complication. Temperatures lead the AMO by 5-10 years, which makes it difficult to assert that the AMO caused the temperature changes.

      • Euan Mearns says:


        If you check out Clive’s latest post you will find a link to V1 raw data. And in comments Nick Stokes provides a link to zip file that contains V3 raw data for about 7000 stations.

    • Euan Mearns says:

      I found another version of the NAO index that provides a better fit and more significantly has a positive gradient that matches the positive temperature gradient.

      I did have a look at the AMO. There seems to be a reasonable match back to about 1900 ± time shifts but not beyond that. I think looking back towards the LIA other forces were also at work.

  8. Greg Goodman says:

    LAGGED arctic oscillation shows remarkable similarities CO2 measured at MLO in Hawaii. Supposed to be at least a NH “well mixed” average, if not global.

    AO leads change in CO2. The lead difference seems to drift over decades but there are some very characteristic wiggles in the post-2000 period that match very closely.

    It makes it rather hard to suggest that the “dominant” cause of climate change is human CO2 when it global CO2 seems closely related to Arctic atmospheric conditions three to five years earlier. 😉

    • JerryC says:

      CO2 works in mysterious wsys. 🙂

      • Greg Goodman says:

        It will be interesting to see whether, in the next 6mo or so, rate of change of co2 drops in line with what happened to CO about 3 years ago.

  9. Pingback: Arctic Temperature Trends | NOT A LOT OF PEOPLE KNOW THAT

  10. Ian C says:

    I hate to ask you to do more work but did you perhaps also make a spaghetti graph of the same stations with the version 3 homogenized data? it would be interesting to see if it loses much of its coherence.

    TIA if you choose to graph it, totally understandable if you decide to decline or delay yet another project.

  11. Greg says:

    Daily Mail says “Swiss pilots Bertrand Piccard and André Borschberg are having to wait to continue their groundbreaking flight”

    Groundbreaking flight? Is that a mixed metaphor or an oxymoron ?

    Oh well, the Mail usually gets a bit confused when it strays from it’s home ground covering tits and celebs and celebs’ tits or Prince Andrew’s bald patch.

    • Greg says:

      Strange, I got a email notification and clicked on ‘reply’ but the post from Roger Andrews that it was supposed to notify me of does not seem to be here. 😕

      • Euan Mearns says:

        I guess Roger may have hit “publish” accidentally on Blowout and then took it down again.

        • My computer is in hospital and I’m trying to put Blowout together using a backup machine and I’ve had to run a few tests to make sure the links are going to work. Apologies for the confusion.

  12. A C Osborn says:

    Euan, have you seen the 2 posts at NoTricksZone?

  13. William says:

    Out of interest I repeated the 4-station average using a temperature averaging (as opposed to anomaly averaging) after aligning the common sections of the station data (1880-2009). The result is at

    I also get a trend of around 1C per century. I plotted the min, max and average using 12-month running calculations. In this method 2007 just pips the late 30s to the hottest 12 months.

  14. halken says:

    There is temperaturedata going back to 1930s for Greenland. Thou I am not sure if all of them can be considered Arctic.

    • Greg says:

      That’s interesting. Copenhagen ( Kobenhavn ) is a very smooth rise of about 2K , Greenland is a lot more volatile.

Comments are closed.