Homogenisation Adjustments to Temperature Records, Southern Hemisphere

Homogenisation adjustments (corrections) made to raw temperature records has grabbed my attention. Those waiting for my post on The Sun will have to wait a while. Roger sent me his spread sheet with about 900 carefully logged temperature records and I was eager to compare these records from 2006 with those being used today. In the past, GISS temp used raw temperature records from GHCN V2 but in 2011 changed to using homogenised records GHCN v3.1 as described in the NASA FAQ page (see the Q&A relating to recent “scandal” at the end of the page). That reply also points to the archived version of GHCN V2 and so it is possible to run many checks comparing V2 data with homogenised V3.1 data. An example of deducting V3.1 temperatures from V2 temperatures is shown below for Sydney. It should be clear why this grabbed my attention. Homogenisation has warmed the past by 1˚C and adjusted virtually all the data.

It has been argued that Paul Homewood got lucky stumbling upon a handful of adjusted records in S America and that there is no wider issue. Roger’s records tell a different story but I wanted to find out for myself how widespread the adjusted records are and spent the morning comparing V2 with V3.1 records selecting stations pretty well at random though I biased selection to stations with long records. In his earlier post Roger observed that about 600 N hemisphere raw temperature records were closely aligned with GISS temp that uses homogenised data. The raw southern hemisphere records, however, did not suggesting that homogenisation introduces more bias in the bottom half of the world. Hence, I have only looked at S hemisphere records in this post.  11 pairs of records are shown below the fold that show highly variable degrees of adjustment.

All stations have visible adjustments. I didn’t keep count, but estimate that about 50% of the stations I looked at had only small adjustments while the remainder which are presented below have significant adjustments. This is certainly not an issue isolated to a handful of stations in Central America. The first station below (Santa Cruz) is included as a station that shows no / only small adjustments. The other 10 stations all show significant adjustments. In each pair of charts the GHCN V2 raw record is shown above the homogenised GHCN V3.1 record.

Santa Cruz Ae. V2 top, version 3.1 bottom. This is about as good as it gets where the effect of homogenisation is hard to spot. But even here the spike down around 1940 has been homogenised out of the data. Unadjusted top, adjusted bottom.

Capetown S Africa. The longest record in V2 begins in 1880 and has two marked discontinuities around 1910 and 1960. The V3.1 record with the same reference number (141688160000) begins in 1943. The warm past has been eliminated and the pre-1960 data appreciably cooled which creates a warming trend. Unadjusted top, adjusted bottom.

Pamplemousses is on Mauritius. At first glance the V3.1 homogenisation has removed the warming trend in V2. But take a look at the Y-axis scale. V2 data ranges from 22.2 to 24.0˚C. The version 3.1 data ranges from 23 to 24.4˚C. A significant amount of warmth has been added. Unadjusted top, adjusted bottom.

Pretoria S Africa. The warming trend of V2 has been totally removed by homogenisation. This could be to correct for urban warming? This has been achieved by warming the past and not by cooling the present. Again look at the big difference on Y-axis scales. Unadjusted top, adjusted bottom.

Durban S Africa. The V2 record begins in 1880 and has a major cooling around 1940 that may be associated with a change in environment. The version 3.1 record has the same name but different ID number begins in 1950. Again, like Capetown the warm past is simply eliminated. Unadjusted top, adjusted bottom.

Buenos Aires, Argentina. The V2 data shows a clear warming trend that again may reflect urban warming that is partly removed by homogenisation. For some reason the pre-1900 data, that look perfectly reasonable, are dropped. Note Y-axis scale again, the past has been warmed. Unadjusted top, adjusted bottom.


Punta Tortuga Chile. The V2 and V3.1 data are not recognisable as being the same record. Enough said. Unadjusted top, adjusted bottom.

Punta Arenas, S Chile. The V2 data is flat to gently cooling. The homognenised data is gently warming.  Note how the post 1900 spike down has been moved to become a pre 1900 spike. Unadjusted top, adjusted bottom.

Faraday is in the deep south of S America. V2 shows a clear warming trend that is removed by homogenisation. But look at the Y axis scale.  The original is -1 to -8˚C, the V3.1 is 0 to -6˚C. It looks like a couple of degrees has been added. Again, these are not recognisable as the same records – different trend and different scale. But the spikes show they are based on the same data. Unadjusted top, adjusted bottom.

Hokitika airport, N Island New Zealand. V2 is basically a flat trend. V3.1 introduces a warming trend. Look at the pre-1910 data which has wholesale been moved down by about 1 degree. Unadjusted top, adjusted bottom.

Onslow, NW Australia. The original data shows a gentle warming. The homogenised data has appreciably cooled the past and introduced significant recent warming. Unadjusted top, adjusted bottom.

Summary observations

  • Most / all records are adjusted, some more than others
  • Sometimes a warming record is flattened. Other times a flat record turned into a warming record. I did not come across a flat or warming record turned into cooling.
  • A common adjustment is to cool the distant past.
  • Some records, for example Faraday have simply had temperature added – approaching 3˚C.
  • Durban and Capetown have had a warm distant past deleted (this may be valid).


It is maintained that homogenisation does not significantly distort the global temperature record although it has added about 0.3˚C warming, mainly through cooling the distant past. If homogenisation makes such little difference, why do it? This simply serves to sow suspicion in sceptical minds. Whilst at one level I can accept the need to correct records where a justifiable cause is identified and understood. I find it equally hard to accept that automated wholesale adjustment is justified. I would like to hear the physical science explanation for adjustments made to the Sydney record posted up top.

Source of chart.

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

29 Responses to Homogenisation Adjustments to Temperature Records, Southern Hemisphere

  1. Yvan Dutil says:

    0,3 °C is approximately the effect of different radiation shield used in the past.

    Homogenization is mostly used because you need representative LOCAL temperature series for many applications.

  2. Alfred says:

    Solar variation may seem a small factor, in relative terms. However, in absolute terms, this variation is around 50 times as great as all the energy used by mankind.

  3. Graeme No.3 says:


    this gives many (.gif) graphs of temperatures from around the World, which have been unadjusted/altered etc. since John Daly died in 2004.

  4. edhoskins says:

    An interesting example of the official adjustments being made to the land based temperature record, is a single correctly sited and continuously well-maintained, rural US weather station is situated at Dale Enterprise West Virginia.

    The un-adulterated record shows modest cooling of 0.29°C per century, if all other adjustments made by “climate scientists” are ignored.

    However the NASA GISS published “value added” temperatures for this same location. This shows a massive adjustment lowering of past temperatures before 1965 to give the impression of very substantial (+0.78ºC / century) warming at this station, by converting a negative trend of -0.29°C into a positive trend of +0.49°C

    Of particular interest is the apparent step-wise adjustment of the homogenized data. This effect is not apparent by simply plotting the data but only becomes particularly evident when the differences are plotted calculated. Perhaps those making these systematic adjustments always thought the no one would be bothered do the simple difference sums but just look at the adjusted line temperature plots.

    In the first diagram above for Sydney V2 – V3.1, it is interesting to see a similar adjustment of exactly 0.1°C every 10 years from 1880 -1960.

    This stepwise adjustment cools the past to emphasize warming.

    If this trivial technique has been repeated by meteorological agencies worldwide it would seem to be an entirely spurious policy based adjustment.


    and the more detailed analysis at WUWT


    • Euan Mearns says:

      Ed, thanks for that. What I’m trying to get to bottom of is Roger’s observation that his 300 or so records from S hemisphere do not match GISS or BEST for that matter. From what I see so far we have three stations with long records that show a sharp step down. Your example from Darwin shows this in 1940. If there is a physical reason for this – station move, new thermometer, then the correct way to adjust this would be to break the record and move the recent half up by about 1 degree and you would have a flat record. Adjusting this to show warming is clearly nonsense.

      In the records I selected Durban shows a similar feature around 1940 and Capetown same about 1960. If these are genuine biases then they should be quite straight forward to adjust. Simply chopping a warmer past from the record is not the right way to do this IMO.

    • Euan Mearns says:

      Here’s what they do to Darwin airport. A full 2.5˚C adjustment along the record. The V3.1 and V2 records have different serial numbers (501941200000) (501941200004) but I think my chart shows they have the same base input data.

      The original V2 record is complete – no 999.9 in the metANN column and runs from 1882 to 2011.

      The V3.1 edited version runs from 1896 to 2105 and has 7 years of data deleted in addition to the first 4 years being chopped.

      Here’s an example of data from my spread sheet – dT:

      Darwin apt
      1897 1.9
      1898 1.9
      1899 1.9
      1900 1.9
      1901 1.77
      1902 1.9
      1903 1.9
      1904 1.9
      1905 1.9
      1906 1.9
      1907 1.9
      1908 1.9
      1909 1.66
      1910 1.2
      1911 1.21
      1912 1.2
      1913 1.2
      1914 1.2
      1915 1.2
      1916 1.2
      1917 1.2
      1918 1.2
      1919 1.2
      1920 1.15
      1921 1.2
      1922 1.2
      1923 1.2
      1924 1.2
      1925 1.2
      1926 1.2
      1927 1.08
      1928 1.1
      1929 1.1
      1930 1.1
      1931 1
      1932 1
      1933 1
      1934 0.9
      1935 0.9
      1936 0.8

  5. Sam Taylor says:


    No mention of how homogenisation lowers the warming trend observed ocean temperatures? Or that when you combine corrected land/ocean temperature series you get a decreased warming trend (See Victor Venema’s post here: http://variable-variability.blogspot.co.uk/2015/02/homogenization-adjustments-reduce-global-warming.html ).

    In any large dataset it is appropriate and sensible to do some data processing to try to remove noise and systematic bias before you start using it in anger. If the dataset is sufficiently large, it is entirely sensible to do this using computers, because that’s basically what they were made for. I certainly wouldn’t suggest you went back to drawing your graphs of energy consumption by hand. In this case, the difference made by appropriately processing the dataappears to be a few tenths of a degree in trend, which is significant and which I would have thought is sufficient to swing measures of climate sensitivity a bit. Given that we’d like to know things like climate sensitivity relatively precisely, in order to best plan for our future (a sensitivity of 1.5 vs a sensitivity of 3 imply different things, as you’re aware) I would argue that we want to be using the cleanest data possible, and that with the messy history of much of this data some kind of processing would be necesarry.

    It’s also important to note that many of the stations have far from perfect metadata relating to their history. While some discontinuities from station moves and such are going to be recorded, there will be plenty which won’t be, which we would ideally like to correct for. Statistical inference methods allows these to be picked up with a relatively low false positive rate (around 5% I think) and corrected for.

    Ultimately, while this analysis is interesting to see the variations in correction, it doesn’t tell us all that much about what’s being done over a wider area. Instead of picking one single station free of local context, why not pick several nearby stations and see what the overall effect is? Perhaps also address the specific statistical techniques which were used on the data, and show why they’re inappropriate (homogenisation ultimately being a statistical process). It would also be useful to how independent methods (pair analysis and use of composite reference time series) both tend to give the same result. Specifically, I would be very interested to know how a technique such as pair analysis, which compares multiple pairs of unaltered time series to try to detect discontinuities, could ever end up adding spurious warming.

    • Euan Mearns says:

      Sam, thanks for the input, comments like this are much appreciated. But you know we have different outlooks on this. Great charts from Victor Venema. But I shudder when I see them and read some of what he writes.

      For the moment I’m doing this out of interest. I wasn’t aware that the temperature record had been subject to such large adjustment. The main source of my curiousity comes from Roger’s 600 N hemisphere records that match GISS and his 300 S hemisphere that don’t. Roger’s S hemisphere based on raw records shows zero warming. And so I’m left wondering if its possible the whole of that 0.3˚C somehow resides in the S hemisphere the records I’ve looked at so far have 1 to 3˚C warming added to them.

      At Real Climate it was pointed out to me that BEST use unadjusted records and this should be sufficient to put my mind at rest. Maybe.

      • A C Osborn says:

        Euan, you have just confirmed for yourself what many other bloggers have shown for many other places.
        As an ex MOD Metrologist, changing data without showing the Documented reason for each & every change is totally unacceptable.
        Creating a false reality for the area where the weather station is located is not Data, it is lies at best and -snip-.
        Sam Taylor says “In any large dataset it is appropriate and sensible to do some data processing to try to remove noise and systematic bias before you start using it in anger”
        But in this case the systematic Bias is being ADDED, not removed.
        In any other Research, Engineering, Production or Financial systems this would lead to serious repercussuions for the perpatrators

        • Euan Mearns says:

          AC – new blog policy is to avoid making careless allegations. Hence the snip. I understand how you feel. The V3.1 records I’ve looked at cannot IMO be regarded as a record of anything individually. The question remains if contiguous groups of records amount to anything that retains original signal. So far I can make the following observations of V3.1 compared with V2.

          V3.1 often cuts large amounts of old record and sometimes significant amounts of new record
          V3.1 often deletes lines of data (a year) for no apparent reason
          V3.1 sometimes creates data (i.e. inserts years) where that data is absent in the V2 raw record
          The corrections are on occasions huge – up to 3˚C
          The corrections are often robotic increments of 0.1, 0.2, 0.3, 1.0, 1.5˚C
          In the records I’ve compared so far (that could easily be non-representative) the overwhelming tendency is to create warming

          None of this can be regraded as a scientifically based correction for station performance at the individual station level.

    • No mention of how homogenisation lowers the warming trend observed ocean temperatures?

      Here’s an extended account. I wrote it a few years ago but it’s still current.


  6. A C Osborn says:

    No idea, it is supposedly from the horse’s twitter mouth.

  7. CTP says:

    Can anyone show an instance of temperatures since, say, 1970, being adjusted downwards?

    • Euan Mearns says:

      CTP – I listened to the comment by Sam Taylor, and realising that looking for stations that have been warmed, is not entirely objective. I spent the greater part of the last two days looking at all stations within 1000 kms of Alice Springs – 30 stations.

      Big post on Monday, if I get it done, will be both good news and bad news for GHCN and GISS.

      The good news will be that homogenisation does not appear to have introduced significant bias to the gross data set – i.e. the average of all 30 stations. The bad news will be that virtually all the GHCN V3.1 data is “cooked” to greater or lesser extent – in fact to quite an extraordinary extent. Warming added in some stations is cancelled by cooling added in others. The procedure is quite extraordinary IMO. One example where cooling is added is given below:




      and after


      There are more examples but you can’t easily spot it on the charts

  8. William says:

    In various graphs the top/bottom station IDs don’t match. For these stations, there are multiple records for the same station with overlapping date ranges.

    For example:

    Station Name Lat Lon ID Pop. Years
    Buenos Aires 34.6 S 58.5 W 301875850000 9,927,000 1880 – 1991
    Buenos Aires 34.6 S 58.5 W 301875850001 9,927,000 1950 – 1990
    Buenos Aires 34.6 S 58.5 W 301875850002 9,927,000 1961 – 1970
    Buenos Aires 34.6 S 58.5 W 301875850003 9,927,000 1987 – 2006

    You seem to have selected the option “after combining sources at the same location” for your V2 graphs. For the V3 graphs only a single station ID is available spanning the entire date range.

    I haven’t investigated what effect this has and how the various sources are combined, but these graphs are definitely NOT showing raw data as you say in:

    “In each pair of charts the GHCN V2 raw record is shown above the homogenised GHCN V3.1 record.”

    BTW a Radio 4 program “Costing the Earth” recently did an interesting program on Iceland. An part discusses changes to soil temperature of many degrees after a seismic event – another factor that can perhaps change recorded temps.

    • Euan Mearns says:

      William, this is a good spot and true. But I don’t think it has material significance. One thing I found after a couple of hours comparing pairs of charts and seeing no differences between then was that the V2 GISS site bounces you back to the V3.1 data – but that’s a side issue.

      In the post I have Monday (see other comment) I compare the V2 minus V3.1 data from 30 “random” sites” and in many cases the differences are integers or decimal fractions thereof suggesting that combining stations is rare. Many pairs of stations also have periods of data with 0.00 adjustment. I’d hope you would agree that adjusting a record by exactly 1˚C is unlikely to happen by chance.

      I try to not put claim to the truth and I try to correct mistakes when I make them. I wish you would channel your knowledge into constructive criticism.

      • William says:

        Pointing out another potential source of error in temperature records is not constructive? It was intended to be purely informative. It means that even if there were no apparent TObs changes or station moves or UHI there could still be factors that can add bias to records.

        As for pointing out that some of your V2 data has been combined from multiple overlapping records in unspecified ways, without plotting each of the data sets together and seeing how they differ it is impossible to say whether they influence whatever conclusions you draw. You clearly don’t trust the V3 processing, so why you would trust the processing by which the V2 data is combined is not clear to me. If that is not a constructive criticism, I am sorry. It was intended in good faith.

  9. Pingback: Temperature Adjustments in Australia | Energy Matters

  10. edhoskins says:

    repeat contribution

    Hi Euan

    What I am still fascinated with is that different Meteorological agencies from the USA (Dale Enterprise West Virginia) and Australia (Sydney) could appear to use the same technique for lowering past temperatures.

    Sydney Australia 0.1 °C steps from 1865 – 1950 in ~10 year jumps
    Dale Enterprise West Virginia 0.1 °C steps from 1895 – 1940 in ~6 year jumps

    I have become a suspicious sort of chap and it would seem to me that in order to get such consistent results it would perhaps appear that a form of international collusion has taken place in temperature adjustments of the past and the technique has been passed around. These agencies would not expect that those difficult people would ever bother to do the difference sums using their published data.

    Accordingly I was rather disappointed when I saw your diagram from
    Figure 3

    Sadly it appears there that the same consistent technique, giving mindless precision. has not been employed at other Australian locations.

    Certainly the cooler past about -1.0°C is well emphasised in your figure 3 but not with same sort of mechanistic approach.

    I do not know if this comparison is worth pursuing. Perhaps Roger Andrews and Paul Homewood might comment.

    I have updated the Dale Enterprise diagram at


    to clarify the technique a little more.

  11. Why adjust?

    If you find a broken station record, determined to be broken by the criteria used to say a station requires adjustments, what do you do?

    You either throw away the station, or, you adjust it and keep the station.

    If you keep the station, you get coverage. This is true in the Southern hemisphere, particularly in South American countries where large portions of the data before 1950 are not digitized or unavailable for various reasons. If you get coverage, your confidence limits narrow.

    If you don’t keep the stations, the broad trends in the global average would probably be the same but the error bounds for the earlier periods would be wider.

    The temperature agencies make adjustments because they can keep stations this way.

    it is impossible to say what cans of worms would be opened by constructing a high-fidelity global average in a manner opposite to what has been currently done.

    • Euan Mearns says:

      Why adjust?

      Station records and their amalgamation can be broken for 3 reasons:

      1) Changes to the physical environment
      2) Discontinuos records – you have to spend a lot of time deleting the 999.9s
      3) Discontinuous time series like this http://www.euanmearns.com/wp-content/uploads/2015/02/number_stations.png

      The way I would manage this is for:

      1) Assume that changes to the physical environment will cancel with enough records – some will have warmed, some will have cooled. I would simply not use records from large urban areas.
      2) I’d patch missing data with data from nearby similar stations, imperfect but better than holes.
      3) At regional level look for breaks in average temps that align with breaks in station numbers. If you look at my more recent Australia post you’ll see that 3 stations do not provide representative cover for Australia while more than 7 seems to work.

    • A C Osborn says:

      I am glad to see that you have looked at the actual data to verify what the “Experts” are saying, oh wait you haven’t.
      Cowtan’s response has already been taken apart by the original author to whom he was responding.
      I suggest that you find someone else to confirm your own bias.
      NCDC/GISS openly admit that their adjustments have added about 0.5 degrees C of the total 0.8 degrees C of the warming in the last Century and presumably you think that is OK?

Comments are closed.