How Hemispheric Homogenization Hikes Global Warming

In last week’s post on the Horrors of Homogenization I presented examples of the large distortions caused by the adjustments applied by NOAA/NCDC to individual raw surface air temperature records. In this follow-up post I analyze the equally large distortions that similar  adjustments applied by GISS introduce at the hemispheric and global scale. The analysis is performed by comparing the adjusted GISS series with unadjusted series I constructed from scratch some years ago using the same set of raw records as GISS and procedures which I describe briefly below.

Record selection: Over a period of months I went through thousands of GISTEMP raw records one by one, selecting those which I could verify by comparison with adjacent records and throwing out those that didn’t fit. At the end of this process I had selected 800 raw records, about 500 in the Northern Hemisphere and 300 in the Southern Hemisphere.

Projection distance: I found that there was no one-size fits-all answer, so I segregated the records into areas in which temperature trends were similar but different to the trends in adjoining areas, ultimately blocking out 64 areas ranging in size from a few hundred thousand to several million square miles.

Averaging: I then averaged the records in each area and area-weighted the averages to construct global and hemispheric temperature time series.

Finally I compared my global series with the published GISS “meteorological station data only” global surface air temperature series, which was constructed from homogeneity-adjusted versions of the same set of GISTEMP raw records that I had used. I don’t have the original GISS series to hand so in Figure 1 I compare it with the current GISS global series, which is not significantly different (the data are expressed as anomalies relative to 1890-1910 means so that the series start off around zero. My series on this and other Figures are labeled “RA”):

Figure 1: GISS vs RA surface air temperature series, Global

The peaks and troughs match up well but GISS shows about 0.3C more overall warming, which as illustrated by the GISS-minus-RA difference plot is added quite regularly. Since both GISS and I used the same raw data set we can reasonably assume that this added warming was a product of the GISS homogeneity adjustments.

The question now becomes, is the added warming real or was it manufactured by the GISS homogenization algorithm? Comparing the hemispheric series gives the answer.

Figure 2 compares my Northern Hemisphere series with the GISS Northern Hemisphere series. The two are very similar, showing a close peak-to-trough match and the same amount of overall warming. (The match is in fact close enough to allow my results to be considered an independent verification of GISS’s.) Clearly the GISS homogenization process has not added any significant amount of warming in the Northern Hemisphere:

Figure 2: GISS vs RA surface air temperature series, North Hemisphere

Which of course means that it must have added it in the Southern Hemisphere. And indeed it has. About six-tenths of a degree C since 1900:

Figure 3: GISS vs RA surface air temperature series, South Hemisphere

Is there any chance this added warming is real? None. Figure 4, which plots 1970-2000 warming calculated from raw GHCN v2 surface air temperature records at 495 stations against latitude, confirms a large warming differential between the hemispheres. I spent a considerable amount of time looking into this differential and confirmed that there is no way the Southern Hemisphere raw records can legitimately be adjusted to the point where they show as much warming as the Northern Hemisphere records:

Figure 4:  1970-2000 warming from 495 raw surface air temperature records plotted against latitude

Yet this is what the GISS adjustments have done. The GISS Southern Hemisphere series shows the same amount of warming since 1890 as the GISS Northern Hemisphere series (Figure 5). The trend lines in fact have almost identical gradients. If this is a coincidence it’s a very remarkable one:

Figure 5:  GISS North and South Hemisphere surface air temperature time series

Not even climate models can match GISS’s Southern Hemisphere series. I downloaded the multi-model surface air temperature (tas) means from the CMIP5 suite of models the IPCC used in the AR5 from KNMI Climate Explorer and plotted them against the GISS and RA series. There’s a good match with both in the Northern Hemisphere:

Figure 6: GISS (red) & RA (blue) surface air temperature series vs. CMIP5 climate model means (black), North Hemisphere

But the climate models fall far short of matching the warming shown by the GISS series in the Southern Hemisphere. If anything they are a closer match to mine:

Figure 7: GISS (red) & RA (blue) surface air temperature series vs. CMIP5 climate model means (black), South Hemisphere

How did GISS achieve this result? As Euan pointed out to me in correspondence: “it should be obvious that the S hemisphere has been homogenised to match a nearby set of stations – the N hemisphere.” And if one starts in an area of the N. Hemisphere that shows significant warming, homogenizes the surrounding raw records with it, and then moves progressively south homogenizing raw records with the already-homogenized records to the north it’s possible to see how something like this could happen. Confirming that it did happen, however, would take a lot of work.

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

59 Responses to How Hemispheric Homogenization Hikes Global Warming

  1. John Reid says:

    Great work, Roger. Perhaps “confirming that it did happen” could now be done by some of GISS’s 150 employees.

  2. Graeme No.3 says:

    There was some controversy in NZ re temperatures, leading to the exit of a UEA trained researcher. The adjusted temperature rise was approx. 0.91℃ over a century versus 0.28℃ calculated from the raw readings. This would seem to confirm your results.

    A Reanalysis of Long-Term Surface Air Temperature Trends in New Zealand – by C. R. de Freitas & M. O. Dedekind & B. E. Brill.

    P.S. Great work, Roger.

  3. Sam Taylor says:


    I’ve a few criticisms I’d like to make. Firstly, I’d like to know what your criteria were for records that did or did not fit, it seems somewhat arbitrary, could you explain further? Did you analyse station metadata for information like altitude, location, station moves?

    Secondly, while CIMP5 is indeed pretty far from the GISS southern hemisphere record, it’s also pretty far from yours. As a criticism that works both ways, by eyeball it looks like about 1.5C away from both of them, (the errors on measurements would also be nice to see).

    Thirdly, the GISS warming for the northern and southern hemisphere’s is roughly congruent (1.4 and 1.2). You’ve now got a massive differential in hemispheric warming which you need to provide a mechanism to explain, which is a pretty big new ask.

    Fourthly, if the homogenisation algorithms really do provide a warming trend, you need to explain why they do so in the south and not the north. Euan’s explanation really doesn’t cut the mustard here, the onus would be on you to show why the statistical techniques (pair analysis or using sussets of nearby stations) used are flawed and would introduce a bias, and why independent methods would both produce the same false trend.

    Fifthly, you would need to explain how it is that the algorithms don’t produce any spurious warming trends when benchmarked on syntyhetic data ( yet do so when applied to data from the real world. Is there something special about this particular dataset that causes false trends to be introduced? Is it a conspiracy by the coders?

    • Euan Mearns says:

      Sam, lets imagine that what Roger shows is correct. What then is your opinion about GISS and the GISS temperature record? What theory do you have to explain that what Roger has done is invalid? He is simply comparing raw data with processed data. And why would you expect temperature evolution in the N hemisphere to match that in the South given the huge difference in land sea distribution, distribution of land masses and ocean currents. I consider it miraculous that the temperature evolution is deemed to be the same.

      • Sam Taylor says:

        Euan, that is begging the question. I don’t assume what Roger has done is right, which is the whole point. All he has done is produce a different result with a different method, which doesn’t invalidate anything just yet. As far as I can tell, he hasn’t analysed station metadata to correct for location, altitude or investigate any other sources of systemic error, though feel free to correct me if I’m wrong in that regard.

        If we wanted to see if Roger’s method measures up, then the best thing to do would be to benchmark it against a synthetic dataset with a known trend and known errors inserted into the data, and compare the accuracy of his result with that of the algorithms (though understandably I doubt he’d want to spend his time doing this). Because as I mentioned earlier, none of the homogenisation algorithms have produced false trends when benchmarked on synthetic data, a fact which you would have to adequately explain when alleging that homogneisation algorithms distort data trends.

        • Euan Mearns says:

          When you say homogenisation does not distort when applied to synthetic data, can you provide a link to the source.

          • Sam Taylor says:

            I did, in the earlier post which you replied to.

          • Euan Mearns says:

            So you did.

            23 Authors working on algorithms that adjust raw climatic records. Why?

            Training the users on homogenization software was found to be very important.

            The study showed that automatic algorithms can perform as well as manual ones.

            So there is user dependency on outcome and they can work suggesting that they don’t necessarily always work.

            Sam, you’ve had four comments on various threads in the space of 30 minutes. I quite simply cannot cope with trying to respond to this deluge. The non-greens make their point and leave. Green Trolls just keep banging away in the hope that the host gives up.

            I put you back on comment moderation.

    • I’d like to know what your criteria were for records that did or did not fit,

      Given that these criteria resulted in a very close match to GISS in the NH – when I was trying to prove GISS wrong, incidentally – I would say they were pretty good ones.

      Secondly, while CIMP5 is indeed pretty far from the GISS southern hemisphere record, it’s also pretty far from yours.

      The idea was to demonstrate that models show a lot less warming in the SH than GISS. Whether they fit my series is irrelevant. There are a lot of things climate models don’t fit.

      You’ve now got a massive differential in hemispheric warming which you need to provide a mechanism to explain,

      No I don’t. I’m simply documenting a data bust. What causes the warming differential is a separate issue.

      Fourthly, if the homogenisation algorithms really do provide a warming trend, you need to explain why they do so in the south and not the north.

      Well, I did make an attempt to explain it, but this is something GISS is better positioned to investigate. Do you have a theory?

      Fifthly, you would need to explain how it is that the algorithms don’t produce any spurious warming trends when benchmarked on syntyhetic data yet do so when applied to data from the real world.

      See my previous comment.

      Is it a conspiracy by the coders?

      I think it’s more likely a problem with the way the homogenization algorithm works.

      • Sam Taylor says:

        > when I was trying to prove GISS wrong, incidentally

        So you weren’t trying to get the best result, just obtain a different one? And you expect me to believe there was no bias in your data selection? I know that accusations of cherry picking run rampant, but come on.

        > Do you have a theory?

        No, because I don’t accept your assertion that the warming is introduced by the algorithms and believe your analysis is fairly unsophisticated and probably faulty.

        As it stands, most of the homogenisation algorithms have been put through the wringer much more thoroughly than your method, and tested on many more trial datasets, I mean they don’t just do these things blind. Until you can claim a similar level of robustness, and demonstrate that your analysis is free of sources of bias and systemic error, I’m firmly in their corner.

      • Sam Taylor says:

        Additionally, since your method is another form of data processing, how about putting the entire, uneditied, raw data series on those graphs for comparison with both time series?

        • Sam: In reply to this and your other comments, all I’ve done here is document a data bust. I speculate briefly on what might have caused it but freely admit that I don’t know what really did cause it.

          Now I could adopt your suggestions and spend many hours putting the entire unedited raw data series on graphs, running synthetic series through homogenization algorithms and demonstrating to you that my selection procedures really were unbiased etc etc, but by the time I got through the bust would still be there and you would still be arguing that it wasn’t.

          So I’m not going to do that. Instead I’m going to suggest that you go through the data record-by-record yourself, construct your own set of hemispheric time series and show me where I went wrong. See you in a year or two.

  4. A C Osborn says:

    Roger and Euan, if as I suspect Roger has used GISS V2 and his NCDC files from 2010 then I am sorry to say that as interesting and damning as it is, this study is only of academic value as it has ben consigned to history. GISS V3 and the latest NCDC files have moved the “Adjustment” game on a lot further than it was in 2010.
    In fact last year GISS Data was changing on a MONTHLY basis, each iteration of their program was re-writing history. I am not sure if it is still as bad as that.
    NCDC has recently been using very large amounts of Estimated Data , both where there is NO data at all and also where there is perfectly good Raw Data.
    Combining the 2 is giving far larger adjustment even to fairly recent data.
    If you check out the work of Steve Goddard, Paul Homewood and Sunshine Hours during 2014/15 you will get an idea of what I mean.
    I personally checked Station data in the USA and Raw and Final NCDC data show major changes and the dramatic use of “E” Estimated data. For GHCN Stations in the USA NCDC have used up to 40 years of estimated data for historic (1890-1940) data where none existed.
    Latest years have perfectly good Raw data replaced by estimated data.
    You can imagine how accurate that data is.

    • AC: The GHCN v2 and v3 (unadjusted) data sets I use as raw data are pretty much the same (although not identical) and haven’t changed much over the years – I can demonstrate that from my files. What get changed are the GHCN v3 (adjusted) data set and the GISS homogeneity adjusted set, and I’ve used the latest versions. So my results will be current until they change them again.

      • A C Osborn says:

        Roger, then I am at a loss to explain how you find no warming adjustment in the Northern Hemisphere when others are finding it.
        Can you list some of the station numbers that you used?

        • AC:

          What’s happening is that Paul Homewood et al are picking on individual NH records or small groups of records and not looking at the big picture. When you look at the big picture you find that the NH adjustments pretty much cancel out.

          The graph below shows part of the big picture. It compares the mean of 49 raw surface air temp records in the Arctic (they end in 2008 so I would have downloaded them five or six years ago) with the current GISS series for 64N-90N. I haven’t checked exactly how many of the raw records have been adjusted but I know some of them have (like the Iceland records). But when you average all the adjustments they manufacture a few wiggles but add no warming (Red=adjusted, blue= unadjusted, gray= difference).

          • Euan Mearns says:

            Roger, so it seems what you are saying is that in N Hemisphere records get adjusted up or down with no net effect. In Southern hemisphere there is a strong bias. Do any southern hemisphere records show cooling or is it always warming?

            And why are they using an algorithm in the first place? Surely with all their resources, NASA could have easily done this by hand decades ago.

          • Do any southern hemisphere records show cooling or is it always warming?

            There are records in both hemispheres that show cooling, but relatively a lot more in the SH than in the NH:

            And why are they using an algorithm in the first place?

            I think it’s because it delivers the “expected results” 😉

          • Euan Mearns says:

            I guess I phrased my question badly which was about the sense of the adjustments. Do all adjustments in the S hemisphere lead to an amplification of warming trend?

          • Do all adjustments in the S hemisphere lead to an amplification of warming trend?


            Here’s the pic from the “Horrors” post that shows S. American records.

            11 records receive cooling corrections (blue) and 40 receive warming corrections (red).

            19 records receive warming corrections that amplify existing warming (top right quadrant)

            3 records that show no warming receive warming corrections (along X=0)

            18 records that show cooling receive warming corrections (top left quadrant)

            One record that shows cooling receives a (tiny) cooling correction (bottom left quadrant)

            10 records that show warming receive cooling corrections (bottom right quadrant)

            Bear in mind that the points on this graph were estimated by drawing linear trend lines through some pretty erratic data. If I had a month or two to do a more detailed analysis I could probably smooth them out a good bit.

          • Sam Taylor says:


            Why on earth would anyone, in the computer age, do this by hand? Should they use slide rules to calculate the sums too? Most of the temperature series contain thousands, if not tens of thousands of stations, and doing it by hand would be a tedious, laborious, and probably error-strewn process which could easily be automated by a computer. Computers also allow much more powerful statistical techniques, such as pair analysis of nultiple time series, to be used. Techniques which would take centuries to perform by hand.

            It’s no different than in seismic processing in the oil industry. Nobody pick velocities or does migrations by hand any more, for perfectly sensible reasons. One could never run a tomography or inversion by hand, but the oil industry seems happy enough with the results regardless.

  5. A C Osborn says:

    As an aside the latest Telegraph article by Chris Booker on Paul Homewood’s work has now had 15,000 comments and the majority of them are not at all complimentary to NASA and Climate Scientists.
    THe word “Fraud” is now being used a lot and I am hopeful that this is going to lead to Congressional Inquiry.

  6. A C Osborn says:

    What is also very enlightening is that the other major Forums are avoiding the issue like the plague with nothing on WUWT or Climate etc even though it is gaining in Press coverage.

  7. Rob Slightam says:

    does the the relative amounts of ocean North vs South make a difference?

    • Euan Mearns says:

      The chart shows how cloud cover has evolved with time since 1983. X-axis is time passing from left to right, annual cycles should be obvious, the Y-axis is latitude. So all you have to do is compare the top half with the bottom half and ask should these two atmospheric systems evolve identically with time. Well we know the answer. The data shows that they don’t and that is no big surprise to me.

      Furthermore, the raw data shows little warming in S Hemisphere and in fact at high southern latitude actual cooling and this ties in with expanding Antarctic sea ice. It seems that there are simple explanations.

    • I think probably yes. The decrease in surface air temperature from north to south tracks the decrease in the proportion of land relative to ocean quite closely.

  8. A C Osborn says:

    The latest Case in point.

    Adjsutment up to 4 degrees C.
    Newspaper article to provethat the temperature really was as warm as the Thermometer record showed it to be.

  9. Yvan Dutil says:

    Can you clarify is it only land or global (ocean+land)?

    • My series and GISS’s series are global surface air temperature series that take measurements from land stations and project them over the ocean. GISS claims 80% global coverage after 1960; I’m more conservative at 60%.

      They are not the same as the Had/CRU series we usually get to see. HadCRUT4 is a combined land-ocean series that estimates global temperatures by weight-averaging air temperatures over the 30% of the Earth that is land (CRUTEM4) and SSTs over the 70% of the Earth that is ocean (HadSST3). SH land makes up only about 10% of the Earth’s land surface area, so if 0.6C of spurious warming has been added to to CRUTEM4 in the SH then this will add only 0.06C of warming to the HadCRUT4 global series.

      There are a lot of different published “surface temperature” series and it’s important to understand the difference between them.

    • I think the coherency is more likely to be with this:

    • Here it is with the mean temps from Figure 4 superimposed, after much stretching:

      • Euan Mearns says:

        This looks good. Adjust land surface for albedo and the Antarctic anomaly disappears. I think this may lend some support to the idea that the Oceans are swallowing some heat – the more static surface in the north warming the air above it more rapidly than the more fluid surface in the South. Or do you have another idea?

        It could just be the distribution of stations. What do the satellites show between N and S hemispheres?

        • UAH shows the NH warming by ~0.3C relative to the SH between 1979 and the 1998 El Niño and both series going flat thereafter.

          I merely note the correlation between air temps and the land/ocean area ratio. I’ll leave the question of what might cause it up to the atmospheric physicists 😉 But I don’t think it has anything to do with the distribution of stations.

  10. Pingback: Climategate II? | Scottish Sceptic

  11. Sam Taylor says:


    You might also be interested to know that when applied to to sea surface temperature, the homogenization adjustments tend to decrease the warming trend ( which shows HADSST3), largely I would imagine due to the bucket correction. Would you still rather use raw data, or do these corrections fit more closely with your notion of how they should look?

  12. Javier says:

    Roger, I was reading an essay by Freeman Dyson and he has an interesting explanation.

    An increase in CO2 has very little effect on humid air, because the water vapor is already blocking the thermal radiation. CO2 therefore is expected to have more effect when it is increased in dry air. But the air is dryer the colder it is, so one would expect CO2 to have a bigger warming effect in the Arctic, high mountains, during winter and at night more than in the opposite conditions. In essence increase in CO2 should make cold places warmer, but not hot places hotter.

    This fits really well with a lot of observations, including why there is more warming in the Northern Hemisphere, given that the Southern Hemisphere is mostly sea and therefore there is little dry air.

    I suppose it could also fit with Euan’s observations on snow, as the warming effect of the CO2 would only kick in after the winter is set and the air gets dry, and its effect would be to make it shorter, hence the snow melts earlier. Once the snow melts, no more effect.

    It does definitely fit with the observation that the temperature effect appears to be to increase the nocturnal minima, rather than the diurnal maxima.

    I suppose also that this hypothesis is scientifically testable if it hasn’t been done already, as the air humidity is measured in many places.

    One of the unavoidable conclusions is that as the Earth warms, the increase in humidity in the air should reduce the effect of further increases in CO2. No runaway hothouse effect.

    • Javier (and Euan): This graph may be of interest. It segregates Figure 4 into blue “ocean” stations (mostly on islands with a few on exposed coastlines) and red “land” stations (the rest) outside the Antarctic. The trend lines show little difference between surface air warming over the land and surface air warming over the oceans:

      Does this tell us anything?

      • Javier says:

        It shows a 20% difference at high latitudes versus a 0% difference at low latitudes, wether significative or not, given the spread of the data. If real, it could definitively contribute to the difference between hemispheres.

        ¿Should we expect more difference that what it shows? Maybe not. We have the effect of winds averaging temperatures on a hemispheric basis and I don’t know how strong is their effect.

        That graph could conceivably be picking the effect of the sea moderating temperatures.

        According to theory an interesting graph would be to compare night-winter average anomaly to day-summer average anomaly by latitude. I don’t know if it can be done simply. I would expect that the night-winter graph would show both most of the warming and latitude difference. Again, I know little of climatology, so maybe what I am saying is already well known and studied.

    • And climate models, which are driven pretty much entirely by CO2 forcing, agree that there should be more warming in the NH than in the SH.

  13. This one is gross:

    Five stations in and around the Philippines (Ninoy Aquino, Aparri, Iloilo, Legaspi, NWSO Agana). First the unadjusted GHCN v3 records. Note that all of them have a 4 to 5-year gap during WWII:

    Now the adjustments. You would expect that they would have been applied during the WWII GAP to bring the records more into line before 1940, but they’re applied everywhere except during the WWII gap:

    And the result is this:

    The NCDC homogenization algorithm at work again.

  14. A C Osborn says:

    Roger, Madagascar has a nice one too.

  15. Nigel says:

    There seems to be some somewhat informed comment/analysis on this topic here:

    I make no judgement of the worth of any of this, not having studied the data or the physics at all.


    • Euan Mearns says:

      Thanks Nigel, I just posted this comment to Real Climate – awaiting moderation.

      Roger Andrews has a post on my (our) blog that summarises 800 raw (un homogenised) temperature records.

      How Hemispheric Homogenization Hikes Global Warming

      This painstaking work has been done over many years. Roger finds that in N hemisphere his selected subset of raw records matches GISS very well. But in the S hemisphere that match is not there at all. There seems to be a large number of S hemisphere records where the distant past has been cooled or the recent past warmed relative to the raw records.

      Furthermore Roger finds that the temperature gradient through GISS N hemisphere is IDENTICAL to the gradient through the S hemisphere which I find highly suspicious / bordering on physical impossible I’d expect N and S hemispheres to respond quite differently to both natural and Man forced warming.

      I sent email to Gavin Schmidt yesterday asking for an explanation, awaiting a reply. We are simply seeking truth and will readily accept an acceptable explanation 😉

      PS cross posted this comment to my own blog.

    • Neither the Telegraph article nor RealClimate’s response add anything to our store of knowledge on the subject.

      • A C Osborn says:

        RealClimates response has already been shredded by Paul Homewood.
        Plus the commentors on Climate etc are giving the BEST trio a hard time as well.

  16. Mikky says:

    Sam Taylor, when you go to your doctor do you insist on being diagnosed by a set of computer algorithms? Suppose some people did, and several of them died following treatment, would the others continue to insist that algorithms are better?

    I find it breathtaking that you would prefer algorithms to expert meteorologists with local knowledge. The brain is very good at assessing plots with hundreds of data points, it would only take a few months for (say) Icelandic experts to produce their best answer for Iceland, if they’ve not already done so.

  17. Pingback: Justeringer eller «juksteringer av historiske temperaturer? | «Klimagrasrota

  18. Jeffrey says:

    Here’s GISS’s map of temperature changes. This maps shows substantially more warming in the Northern Hemisphere than in the Southern Hemisphere, and they even provide a graph highlighting this fact.

    Has GISS fixed their map in the last few days?

  19. Pingback: AWED Energy & Environmental Newsletter: February 17, 2015 - Master Resource

  20. Pingback: Recent Energy And Environmental News – February 17th 2015 | PA Pundits - International

  21. Pingback: Homogenisation Adjustments to Temperature Records, Southern Hemisphere | Energy Matters

Comments are closed.