The Hunt for Global Warming: South America

Ever since Roger Andrews sent me his spread sheet showing that 300 or so land based climate stations in the Southern hemisphere recorded little warming I have been trying to find out what exactly is going on. Does CO2 not force temperature down under? At this point, I should confess that at the moment I am looking for global warming where I least expect to find it and that means avoiding areas where there are a lot of people and where the Earth’s surface has been completely reworked by human beings.

I have also been examining the impact that GHCN V3 homogenisation has on the less processed V2 records. This is very time consuming and this is the last time that I will perform this exercise.

In summary, GHCN V2 records from 25 climate stations selected by the NASA GISS platform from southern Argentina, Chile and the Falkland Islands produce a completely flat time-temperature anomaly trend. I have succeeded again in not finding evidence for global warming in the southern hemisphere. The GHCN v3 homogenisation adjustments vary individual stations by up to ±2˚C and follow the same robotic style of exact decimal fraction adjustments seen elsewhere. In this case, these adjustments do add warming of the order 0.4˚C since 1888. GHCN V3 records and temperature reconstructions based upon them are to be avoided at all costs.

Figure 1 The beautiful Lago Argentine in Patagonia.

Station Selection

I clicked on southern Argentina, placed the beautiful Lago Argentine at the centre and the NASA GISS platform returned 25 stations within a 1200 km radius (Figures 2, 3, 4). Most of the stations are classified as rural although there are 7 larger towns. There is no evidence for urban warming in this remote and wild corner of the world and so all 25 stations are included in this summary.

Figure 2 Station list with Lago Argentin at the centre.

Figure 3 Station locations. Four stations are located on the Pacific coast, west of the Andes and two stations on the Falkland Islands.

Figure 4 At no point were all 25 stations simultaneously operational. In 1901, station number increased to 6, prior to then, results are based on a very low number of localities. In 1992 station number plunged to 5, presumably on the back of political woes in Argentina.

The V2 “Unadjusted” Data

Clicking through the records and charts it is easy to spot three main kinds of record. Flat, slowly warming and a couple show marked cooling in the 1960s. These are Valdivia and Puerto Mont, both on the Pacific Coast of Chile. While there may be a case for looking at the Pacific and Atlantic coasts separately, everything is lumped together in this summary.

The individual time temperature series (Figure 5) were converted to anomalies by deducting the mean temperature for a station from that station and then the average anomaly for the stack was calculated as shown in Figure 6*. Once again we have a totally flat line. No evidence for global warming in this remote part of South America.

[* I have had growing unease about my normalisation procedure. I believe it is the most correct option available and I did some checks before adopting it. I have now done a more thorough check on the central Australian data where I normalised to the means of the period 1965 to 1974 as detailed in this comment on Climate Etc. It makes no material difference to the outcome.]

Figure 5 Temperature spaghetti plot for 25  S American stations.

Figure 6 Trends don’t come much flatter than this. There is absolutely zero evidence for global warming in this remote corner of S America. The 1998 temperature top corresponds to the global top associated with the big El Niño that year. This pattern seems to repeat about every 50 years.

The V3 Adjusted Data

The exercise was repeated for the V3 homogenised records. In central Australia homogenisation did not significantly alter the average temperature. In Southern Africa it added a little warming. In South America it has added a significant amount of warming, about 0.5˚C since 1888 (Figure 7).

Figure 7 GHCN homogenisation has added about 0.4˚C warming since 1888 to this group of records.

Figure 8 In my workbook I lay the V3 spread sheet on top of the V2 spread sheet and create a new dT spread sheet by deducting the V3 matrix from the V2 matrix. The data then need to be cleaned since there are a large number of years where there is no data, there are a number of years where there is V2 data and no V3 and vice versa. The pattern of data cleaning and consistency is shown in Figure 9.

The resultant plot describes the pattern of data modification that V3 has done to V2. This is supposed to describe non-climatic trends identified in the data which is clearly nonsense. In this group of records there are in fact long strings of data that have not been modified (zero in cell in Figure 9).

Figure 9 This screen capture of my dT spread sheet shows the pattern of data modification and editing between V2 and V3. Empty cell = no data; zero in cell V2=V3; number in cell = dT adjustment; yellow cell V2 data exist V3 not; green in cell V3 data exist where V2 data do not. This latter category is a puzzle since V3 is supposed to be a derivative of V2. The scale of adjustment and edits in this data set is not as extensive as seen in Australia, Africa and Iceland. It is therefore somewhat ironic that the V3 adjustments have in this case biased the data towards a warming trend.


I have now looked at Central Australia, Southern Africa and this remote corner of South America and am struggling to find evidence for global warming in the Southern hemisphere. What’s going on?

I’ve already completed analysis of Antarctica and E Siberia. N Scandinavia is almost done. What I’m seeing in the data is providing me with plenty of motivation to continue. I hope readers bear with me for another few weeks. I’ll keep trying to have at least one energy related post per week.

I’d like to conclude with this nice chart from commenter William that shows temperature anomalies for 9 stations from Chile that seem to corroborate and extend the conclusions drawn in this post.

Figure 10 dT trends for 9 Chilean stations from this comment by William.

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

54 Responses to The Hunt for Global Warming: South America

  1. A C Osborn says:

    Euan, you can’t just take the data and look at it.
    Nick, Mosher, Zeke & ATTP have told you that you are doing it wrong. You have to find a suitable method of averaging, filtering or any other method to show warming.
    I would really like to get the advice of an unbiased mathematician on this subject.
    Notice that none of them have answered the IMO question of how can it be correct to Adjust already adjusted data, basically they are saying that in the whole world they are the only ones who know how to do the job correctly, aboslutely insulting all the people who have previously worked with the data.
    They also do not really address the continual changes that GISS & NCDC are making to even recent data let alone historical data.
    I am not sure if I posted a link to the Roy Spencer post which has now been followed up by Paul Homewood at

    • mbe11 says:

      AC, actually you and the rest are simply wrong, Every time you massage the data to correct it you add bias of unknown amounts. If you simply throw it in the pot and average it out each year you will get a long term trend. That trend will of course have an error in it . Since the error is unknown there is no way to actually filter it out. You, your fellow bloggers , Hanson etc. can pretend to filter it out by introducing bias into the data but you are simply fooling yourselves. Why not just do the simple thing like ever statistical sample where the average is unknown, take a large sample and use that to calculate the average. In the case of temperatures take them all unadjusted and a period of the whole period and than calculate the anomalies from that. You will have introduce no bias into the data and because of the large sample for both the anomalies base line and the data smoothed out a lot of the noise. The fewer samples in periods will have a larger uncertainty but since that is 70 to 100 years in the past it would not really matter. I looked at the 1979 to 2014 temperature data from Giss and the mauna loa CO2 for correlation. Depending on the period the correlation varied a lot from ..2 to .5 which means that the adjusting is simply throwing noise into the calculations.

      • Euan Mearns says:

        AC was being satirical / sarcastic 🙂

        I, and most of my commenters are absolutely in favour of minimising the corrections made to data. Normalisation to dT is an essential evil if one is to look at an average. The method I use places the average of all the data at the centre.

  2. Zeke Hausfather says:

    You may find the actual GHCN datafiles from NCDC much easier to work with than the v2 vs. v3 GISS station data, especially since GHCN v3 added a fair amount of new station data in recent years not contained in v2.

    The data can be found here: is the raw data (with only QC but no adjustments) is the file that contains TOBs-corrected homogenized data.

    If you want to look at specific homogenization adjustments in detail, I’d strongly suggest looking at max and min files separately, as the tavg file will tend to blend different adjustments to each (e.g. something like a transition from a mercury thermometer to an electric thermistor will result in different biases to min and max temperature, as I discussed here:

    • Euan Mearns says:

      Zeke, thanks for the links. The second two are the same. And when I try to download it downloads a 12 Mb file but the decompression fails. I’m using a Mac.

      Why don’t you guys come and join in the conversation. Over on Climate Etc Mosh’s main line of defence / argument seemed to be this:

      The notion that a global product should be used for local issues OR evaluated by looking at local issues fundamentally misunderstands what the product is meant to do.

      I’m afraid that ain’t going to “stand up in court”. For starters, here’s a stack of 23 rural and urban stations for southern Africa where urban stations that show clear signs of recent warming are excluded:

      And here’s the 10 stations that were excluded:

      These stations are spatially mixed. Southern Africa is then either warming or it is not and you have to decide which set of stations are telling “the truth.”

      Another point of philosophical difference is that you don’t normally need a lot of data to define the temperature trend of an area, especially where temperatures are congruous which they normally are over large area. Data quality trumps data quantity.

  3. Euan Mearns says:

    I’ve become a bit concerned about my method for data normalisation – calculating dT from temperature. I have been deducting station average for a station from the temperature series for that station.The conventional way would be to take the average for a reference period and deduct that from the temperature series. I have run some prior checks and everything seemed to be more or less OK. I’ve now run some more checks using this data for E Siberia. Everything still seems OK.

    However, I’m pretty astonished at what this chart is telling me and posting it on its own in hope of some feedback. The conventional way to see this is to run a regression that clearly shows a warming trend. However, the alternative view is that it shows no warming at all.

    • mbe11 says:

      I bet you do not have 21 stations back in 1880 nor continuous records for a large part of the period. so the data prior to 1940 is suspect as to accuracy.

      • Euan Mearns says:

        I think what you mean is that back in time uncertainty increases as it always does and always will do. The more important observation is that in 1943, 20 stations were operational.

  4. Euan Mearns says:

    Same chart with 1964-1968 base period:

    And with 1942-1946 base period:

    The gradient of the regression on these tow charts ia about +1.8˚C since 1880. Using the station mean its about +1.7˚C.

    Deducting the station mean base from the 1964 to 68 base period:

    And deducting the station mean base from the 1942 to 1946 base:

    And finally deducting 64-68 from 42-46:

    What we see is that the 1964-68 base has a similar mean to the station average and these two give similar results. The 1942-46 base has an average different to the station mean. We can conclude that choice of base period affects the results you get, but the regressions have same gradient in this case. I suspect neither method is correct and normalisation may impart some structure to data, but it does not appear to be significant for the task at hand. The issues become greater back in time as data continuity drops.

    • Sam Taylor says:

      The length of your base period is probably also important. Shorter periods are going to be more likely to be abnormally warm or cold than longer periods. If you were to end up using an abnormally warm/cold period as your baseline, then you risk introducing a bias. If you’d chosen a period of, say, 15 years instead of 4 (especially since one of those periods coincides with WW2) then I suspect that the differences you’re seeing might diminish somewhat.

      The first difference method can also run into this issue if your first datapoint is abnormally high or low.

      • A C Osborn says:

        I would have thought that the baseline should be the whole dataset, not some arbritrary period.

        • Sam Taylor says:

          Not really. Imagine if you have two stations in a similar location, one active from say 10 years before another. If those first ten years were somewhat warmer than the period in which both were active then using the entire station lifetime to normalise will introduce a bias that wouldn’t be there if you just used the period that both stations were active for.

      • Euan Mearns says:

        Sam, I deliberately chose short periods and extreme values since I was running a test. Using a longer base period will reduce any effect described here and the base period should ideally be central in the data. If I were wanting to produce a temperature index to be used to inform multi-billion pound investment decisions I’d want to nail this issue with greater precision. But I think my existing methodology is fit for current purpose.

      • William says:

        Nick Stokes and many others have evidently done a lot of work on this sort of question, for example see and the references therein. Simply using a station average to normalize that station (as I did in the graph of mine you included) is wrong I now realize.

        • Euan Mearns says:

          Its not wrong and in fact may be more right than using a fixed time period base.

          What is important here is the dT a station has relative to the average of the stack when it comes or goes. If the station value is close to the mean value it has little to no impact. If it is different to the mean it will impart structure, especially when there are a small number of stations in the stack. What is important to the gross result is the distribution of these differences to the mean as stations come and go. If the distribution is normal then there will be no net bias. Back in time, when we may only have a handful of stations the probability of a non-normal distribution will increase. That would normally be pre-1930.

          Using the station average to normalise places the average of all the data at the centre of the distribution. That is why the anomaly for S America is exactly zero. This increases the probability of not introducing a bias to the data. Choosing a base period instead will introduce a bias if the data have a trend.

          I still need to check out Roger’s first difference method. What Nick Stokes is doing is simply torturing the data to produce the result he wants. This is the dT spaghetti for Central Australia that Nick has managed to turn into a warming trend.

          • William says:

            simply torturing the data to produce the result he wants

            That is insulting to Nick Stokes. His work over the years to understand the issues and to demonstrate his results has been impressive. Take a look at his own indices and charts. The data he used that you say he “tortured” is available from the link he provided and you can plot an average yourself.

            This is what you get using 1971-1992 as the base period:

            It is not so very different from his plot. Did I “torture” the data too?

          • Euan Mearns says:

            I don’t mean to be insulting to Nick who seems a decent enough guy. But the fact remains he has applied a number of statistical techniques to the data to produce a warming trend that I’m quite sure he feels should be there.

            Australian data starts in 1880 and the V2 data ends in 2011. So I don’t know what data you are using. And since you lectured Roger the other day on “cherry picking” a start period I’d have hoped for better.

          • William says:

            Did you read the link above – the one for which you said Nick was torturing the data? It says:

            For all calculations I have used annual averages, discarding years with fewer than nine months data. The annual data (unadjusted GHCN V3 monthly from here) that I used is in CSV form here.

            where ‘here’ and ‘here’ are links to the data. It clearly says “unadjusted GHCN V3”, which is not V2 and doesn’t end in 2011.

            The CSV data file starts at 1901. I’m sure he could have gone back to 1880 but as your own graphs show, there are less than 3 stations at that time: (
            You chose to plot post 1907 raw data in figure 5 of your Oz post:
            where you said:

            Pre 1907 there were only three operating stations and this imparts bias to the record.

            What’s good for the goose…

            I just did a simple average using a 20 year base period and it gave me Nick’s curve, more or less. I tried Roger’s method and it gave more extreme warming – maybe I did it wrongly. Here’s the Excel, simple averaging is sheet 2, Roger’s method sheet 3:

            If you post your data, I’ll check that too – without torturing it.

          • William says:

            As you haven’t posted my last post yet, I have done some extra work. I downloaded your 30 stations and created a simple average with the base period 1971-1991, dropping 4 stations that had no data in that period. I also plotted Nicks data in the same way. The v2 and v3 curves are here:

            To me they look consistent and it seems like there’s about a degree of warming in both. They start in 1907 because before then there are so few stations.

            Just to answer your criticisms of me: Nick’s post (my link to his site, above) says that the data is “unadjusted GHCN V3 monthly” data. That is what I used as “Nicks data”. v3 extends to 2014, hence the extra years. The CSV data file starts at 1901. As your own graphs show, there are less than 3 stations at that time: ( and you also chose to plot post 1907 raw data in figure 6 of your Oz post:

            As you said there, “Pre 1907 there were only three operating stations and this imparts bias to the record”.

    • Euan: I don’t see much of a problem here. Your series all look pretty much the same to me, although I agree with Sam Taylor that a four-year base period is too short for an area where temperatures jump around from year to year as much as they do in Siberia

      But if you are still concerned I would suggest you try the following:

      1. According to GISS all your stations have data between 1940 and 1989. Recalculate using this as your base period.

      2. Then recalculate using the first difference method, which is independent of the base period.

      3. Compare the two and see what you get. 🙂

  5. A C Osborn says:

    Euan, do you know any local mathematicians or stataticians?

  6. William says:

    Can you post a link to your spreadsheet please. That saves me extracting the source data again.

  7. concernclub says:

    please have a look at the arctic sea ice data.
    All the data must be manipulated no?

  8. Nick Stokes says:

    I made an active map here which is designed to show unadjusted (or adjusted) GHCN station trends over various periods of time. Here is a montage of trends since 1967 of the areas you have been looking at here. You can find other places that don’t have much warming. And plenty that do. It varies.

  9. Sam Taylor says:


    It is trivial to show that it’s wrong.

    Start by taking a station of your choosing (I went with alice springs). Make 2 identical copies of the temperature series, then chop some data out so that maybe one starts earlier than the other, and the other finishes later, but they both still have a period of overlap in the middle.

    Now, normalise both of these time series in the following ways. Firstly, normalise them both using an average over their entire history. Secondly, normalise them both over the period where they both have data.

    Calculate the average anomalies over the whole time period for both of your normalised series.

    Now calculate the slope of a regression over the entire time extent for 1) The initial unchopped data 2) The average anomaly normalised for the entire length of the each dataset 3) the average anomaly for which had a common normalisation period.

    You will find, as I did, that the slopes you calculate for 1 & 3 match exactly, and that the slope of 2 is significantly different. Since you have used exactly the same data twice, result 2 is clearly nonsense and shows that the method is flawed.

    • Euan Mearns says:

      It is trivial to show that it’s wrong.</blockquote.

      Yes it will be trivial if you only use two records. The reliability of this whole approach is dependent on having a sufficient number of records so that when stations come and go they do not significantly move the mean. Repeat you exercise with 20 replicants.

      And you seem to have forgotten that I've calculated a dT stack for Australia using 1965 to 1974 as the base period in addition to my default method of using the station average.

      Looks pretty identical to me.

    • Nick Stokes says:

      Sam’s right. Not fixing anomaly base (or equivalent precaution) will diminish the trend. It’s true that here the effect is limited. But not nothing. I get with varying base, a 1920-2014 trend of 0.92 C/cen, vs 1.26 with what I think is a correct treatment of anomaly.

      There may be a minor discrepancy in our calcs – when I was eliminating duplicates from the v2 list, as you have to do to use v3 data, I eliminated Farina, which has the same WMO number as Marree. But it turns out it does have independent data in V3. I’ll fix that.

      • Euan Mearns says:

        Nick, you are using V3 “unadjusted” while I am using V2 “unadjusted”. The V2 data are actually adjusted. And there is heavy data editing between V2 and V3 with data both removed and added in V3. So that accounts for some difference, perhaps the most material being that I start my series in 1880. I see 1880 to 1906 data that are pretty well bang in line with the rest and so i’m happy to include it. Had it looked different I would have excluded it on basis of small sample size.

        The NOAA archive provides instruction for accessing data with Windows and Unix but not for a Mac. And my Mac will not un pack the files. Its curious that NASA GISS say that they began to use the V3 adjusted files when NOAA stopped updating the V2 unadjusted giving me the impression that V3 unadjusted were not available and as you know on the GISS web platform you can get V2 unadjusted or V3 adjusted.

        I agree that using a variable base period will act to suppress differences while choosing a fixed base at one end of a data series that has a gradient will act to amplify differences. Since the Central Australia data are in my mind largely flat you could choose a 10 to 20 year base at any point and it would make no difference to the outcome. In the case of the Australian data I don’t think using a variable base makes any difference at all as my chart comparing variable base with fixed base shows.

        At issue here is the notion that my methodology is taking a pile of records that show clear warming and somehow turns them into a flat summary. Well I just can’t see that I’m afraid. Take for example the Patagonia records for this post. I look at the temperature spaghetti and see flat records. I add them all together and get a flat trend. When I go else where to data that has structure, the structure is preserved. Sure different methods may yield slightly different results, but the differnces are likely of the order 0.1˚C over a hundred years.

        This chart has variable and fixed base period data on it. As I said over at Climate Etc, if this were a chart of the stock market everyone would conclude that it had traded side ways for a hundred years ± short term gains and losses. I don’t think my normalisation is producing an abstract picture and I don’t think anyone is going to convince me that this trend or the trend for Patagonia is rising.

        • William says:

          Euan, I am puzzled at your insistence that “Central Australia data are in my mind largely flat. You can see that it is not in the graph of mine that I posted earlier comparing v2 and v3. There is a clear ~1 degree rise over the century+ to 2014. You can see it too in your own data – if you plot it suitably, for example, this is a plot from your own spreadsheet showing the rise over the century using your two base period methods – 0.5 and 0.8 degrees for the two methods with the 65-74 base period giving 0.8. This is your data, your spreadsheet, I just fiddled with the way it is presented (colours, start year, trends). Maybe you could inline it.

          You might complain that I started at 1907 and not 1880. But as your own data shows there were only 1, 2 or for a few years 3 stations prior to 1907. You have two graphs in the “all T v2” sheet of your spreadsheet that show the sparsity of pre-1907 data (“pre-1906 < 3 records" it says). You say elsewhere that "I see 1880 to 1906 data that are pretty well bang in line with the rest and so i’m happy to include it." but that conflicts with the obvious 2 degree jump on your own charts (again from your own spreadsheet, sheet "all T v2"):

          • Euan Mearns says:

            William, in my first pass looking at the Australia data I made the mistake of averaging temperatures. I was keen to see what the average was and unlucky for me the distribution of the data was amenable to producing a sensible looking average post 1906 and it was clear that pre-1906 averaging temps gave false result because of low station number. A few days later I fixed up an anomaly sheet calculated the anomalies and all were in line from 1880 – bingo. If they were off I’d have excluded them on basis of low station number but they were not off.

            When I run a regression through the whole anomaly stack I get about 0.2˚C warming. If I start in 1920 I get closer to 0.5C – which zeroes in on your chart. On your chart I’m surprised to see such a large difference in gradient between the 1965 to 74 base and the station average base since the data appear to be pretty identical. When I compared the two methods the gradients came out close to identical – in each case about +0.2˚C since 1880. If this were a chart of the stock market, would you judge it to be rising or range bound moving side ways?

            Thanks for the XL tutorial. I’m self taught. Some of the things you mention I’m familiar with. But I am stuck in my ways. So long as there are no fundamental errors. You are posting from two different email addresses – right? I blocked a comment the other day from another commenter called William – I was unsure if it was an attempt at impersonation.

        • Nick Stokes says:

          “The NOAA archive provides instruction for accessing data with Windows and Unix but not for a Mac.”
          Euan, I have made a gadget here which may help. It gives access to individual station annual averages from GHCN V3.

  10. Euan Mearns says:

    Got to admit that this looks so good so as to be suspicious. I double checked my spread sheet. It is a matter of coincidence that the 1965 to 1974 means are the same as the station means. This is of course also an artefact of the data being flat 😉

  11. Dan Pangburn says:

    Proof that CO2 has no significant effect on climate and identification of the two factors that do are disclosed at

  12. Jim says:

    I have done quite a lot of similar work using GISS data and find that the BMOA data is heavily manipulated to prove their AGW point.

  13. Euan Mearns says:

    My spread sheet for Alice Springs / Central Australia. I decided to simply publish it as is, warts and all. There is a tab for each station with the GHCN data. V3 first followed by V2. Then there are 5 coloured tabs:

    All TV2 is a summary of the metANN temperatures for V2
    All TV3.1 is a summary of the metANN temperatures for V3
    all dT is TV2 minus TV3
    anom is anomalies calculated based on average for station
    anom 2 is anomalies calculated for the 1965 to 1974 base period

    Central Australia Data

  14. I’d guess that BMO Australia temperature record manipulations must be quite extensive judging by the efforts that BMO goes to to prevent queries and not release data. Would be good for all to see your analyses and conclusions by posting here.

  15. Doug Proctor says:

    If only stations within 100km of the coast were selected, would you see an enhanced “global” signal? If so, it suggests the oceans are releasing heat OR CO2 enhancement is only meaningful over water, i.e. in places where water vapour feedback is direct and immediate.

    The Arctic warming shows similar edge-of-water effects for Alaska.

    When I was in my early twenties and single, in any group of four ladies in a bar, on average one would talk to me. That’s the math. In reality, I had to find four groups of ladies before ANY would talk to me, and then the politeness of the group lead them ALL to talk to me. In other words, I was a scoreless nerd. But the magic of math makes it seem that in general I did okay.

    That’s what I’m hearing these days. Statistically, the world is getting warm all over as an average. Actually, that is only true near the coasts. The landmasses are like me during my dating years, left “out in the cold”.

    • Euan Mearns says:

      Its an interesting thought. But I was thinking about this in a rather different way. This little bit of S America is surrounded by Ocean. If the air masses over the oceans were warming I’d expect to see it throughout this area since air will be moving across land from the oceans all the time.

      • A C Osborn says:

        I have seen a study of coastal versus inland temperatures and if I remember correctly the coastal temps had much less variation and warming than the inland temps, especially when the land was hilly.
        I can’t remember whose forum I saw the study on.

  16. William says:

    I looked at the south America data and plotted this for post-1900:
    As you can see there is a trend of around 0.3C for the period.

    As with the Oz data where there were only 1, 2 and for a few years 3 stations reporting before 1907, there is a single station reporting before 1901. Starting the graph at 1901 seems reasonable. After all if a single station was all that was necessary, we’d forgo all this averaging and just pick one station to represent south America. The power of averaging is it helps reduce the noise – with one station only that doesn’t happen. I guess you will disagree though.

    • Euan Mearns says:

      William, its good that you’re taking time to run these checks. This for me its what blogging is about.

      The 1896 peak is defined by only 2 stations. Absolutely right that this reduces confidence. So how to handle this? I am working from the position that I know when I come to merge these data sets I’ll have several long records that will be less easily discounted.

      And you need to be careful in establishing a precedent since in other areas cutting out the old data may work another way.

      And one further suggestion, your chart has a y-axis scale of ±0.8˚C. I think using ±2˚C may be more appropriate. The raw temperature records vary from 6 to 14˚C.

      I presume you have seen my lengthy comment below on Australia. I’m presuming that the sensitivity to normalisation method and start date reduces the more data you have.

      So which normalisation procedure do you think is most appropriate?

      This has all been highly educational for me and I think the rabbit hole is going to run deep in the coming weeks.

      • William says:

        I don’t know enough about the subject to have a preferred normalization procedure. But having looked at many stations it is clear that much of the data is of poor quality and that it gets worse as we go back in time. Even recent data can be poor, as we know, but there is less chance of correcting biases the further back we go. This implies that we should emphasize recent data over older data and use recent data to base-line the whole. But there are various ways of normalizing, as Nick pointed out in his Oz article, that I know nothing about.

        As far as removing data, a principle is involved (precedents don’t come into it): removing a small amount of the overall data-set should not alter the conclusion one draws about that data. If it does you don’t have enough data for averaging to do its magic. You can test this if your spreadsheets are all equation-based, no hard-coded constants and protected against empty cells (as I discussed with you before). Delete a few stations and nothing much should change (if you had enough in the first place) – although if nothing changes at all, suspect the spreadsheet.

        Note that I have no expertise in this field, so take what I say with a spoon full of salt.

        • Euan Mearns says:

          William, I know how to “hard constant” cells in XL. For what i’m doing i don’t find it appropriate. The spread sheet is actually very simple, so long as I keep the formatting constant. You may recall that in Iceland I made a mistake, pasting 1 column of data 4 years out that Sam picked up.

          I double checked, the XL =average(yy:zz) ignores empty cells, so there is no issue there.

          For me the main lesson learned over last few weeks is that regressions run through these time temperature series are incredibly sensitive to minor variables like start date and are hence largely BS. I will continue to do it but its a subjective “art”. There is no right way of choosing a base period. I have done no reading around this and so make this up as I go along from my first principles. I suspect my way of using station average may be the best. This will tend to reduce differences and hence will not be liked by warmists. From now on I will use both Roger’s 63 to 92 base and my station average base.

        • Euan Mearns says:

          William, this comment sat in moderation a while because I don’t sit here 24/7 moderating comments. I know you would like to be off comment moderation. In the recent past you have patrolled blogs doing what you can to discredit me. Also claiming that you were banned here which is transparently false. Now I’ll admit that in the huge volume of stuff I write I may now and then write something inappropriate. For the record I sent an email of apology to Gavin. And so I will also concede here that having you around has pressed me to sharpen up my act – which is a good thing.

          And I genuinely appreciate that you are “auditing” what I do . I work at a crazy speed and make mistakes. But these are blog articles where the purpose of the comments is to pick up on and discuss and if necessary rectify these mistakes. Throwing swing balls with polynomial fits to data wastes my time and makes me think that this is part of your objective.

          Your technically instructive comments are most welcome. But ones like this where you seem to be trying to undermine my ability to use XL are not. You have a long long way to go to gain my trust.


  17. Euan Mearns says:

    There was a lot of discussion around normalisation and regression of the Central Australia data. Yesterday Roger checked things out by using a longer 1963 to 1992 base period and using his first difference method.

    We can conclude that 1963 to 1992 base and first difference give same result and 1963 to 1992 fixed base period is perhaps the preferred way forward.

    The following charts show regressions worked out 3 ways: 1) using station average as the base, 2) using 1965 to 1974 as the base 3) using 1963 to 1992 as the base. The “tables” below each chart shows the warming over the period. The first chart is 1880-2011, the second 1907 – 2011.

    Station average base +0.2˚C
    1965 to 1974 base +0.45˚C
    1963 to 1992 base +0.2˚C

    Station average base +0.45˚C
    1965 to 1974 base +0.8˚C
    1963 to 1992 base +0.7˚C

    For 1880 to 2011 it makes no difference using station average or 1963 to 1992.
    For 1907 to 2011 it makes a big difference which base method is used.
    The biggest difference however comes from selection of when the regression is begun. The fact that there are only 3 stations prior to 1907 and the data may not be representative is a reasonable argument, that should not be ignored. On the other hand are there good reasons for not using 27 years of data? Are the three stations with early records – Alice Springs, Conclurry and Farina – somehow faulty or biased?

    Station average base +0.3˚C
    1965 to 1974 base +0.3˚C
    1963 to 1992 base +0.35˚C

    I will admit that I may have been subject to confirmation bias. And in one of the comparisons I made between the 1965 to 74 base with the station average I made a mistake reading the gradient of the line – as noted above 65 to 74 does normally give a different result.

    Confirmation bias comes from not seeing rising tops and bottoms that would in my opinion be characteristic / diagnostic of a rising trend. I still see this as a flat range bound data set. And I do not see good reason for excluding the pre 1907 data, although I can understand that others may want to do so. It is unfortunate that those 27 years of data make such a big difference.

  18. Pingback: AWED Energy & Environmental Newsletter: March 30, 2015 - Master Resource

  19. Pingback: The Hunt For Global Warming: Antarctica | Energy Matters

  20. Pingback: Recent Energy And Environmental News – March 30th 2015 | PA Pundits - International

  21. Pingback: The Hunt for Global Warming: Moscow and Urban Heating | Energy Matters

Comments are closed.