The Hunt For Global Warming: Southern Africa

Time – temperature series for 26 selected climate stations in southern Africa are presented.  The stations are spread from Capetown to Zanzibar. The mean result for GHCN V2 (unadjusted) is a flat temperature record from 1880 to 2011. The data seem to record the mid-1970s cooling recognised in Central Australian records and overall the results are closely aligned with Central Australia where station selection criteria were objective.

The GHCN v3.1* homogenised records have been subject to wholesale data deletion, data addition and data change. And yet, as was the case in Central Australia, the net result on the homogenised time temperature series is minimal, has tended to reduce variance and in part erase structure that may be due to natural climatic cycles. In southern Africa, homogenisation does not appear to be the source of warming present in climate reconstructions but evidently absent in the raw records I selected.

(* note that Roger advises that there is uncertainty about what generation of records are archived at this web link. They are GHCN homogenised records of uncertain provenance. They are evidently the station records used by GISS temp.)

Temperature anomalies for 26 southern African stations. The trend line, rising perhaps 0.1˚C in 130 years is effectively flat. The data display cyclicality with an approximate 85 year cycle. A cold period, 1890-93 appears to repeat in 1974-76. The latter cold period is also seen in Australian records. These mark the low points. The high points are in the 1920s and recent decades. 2005 is a high outlier caused in part by a low number of operational stations (9). The amplitude of variation is roughly ±0.7˚C. The data suggest that a regional temperature change of 1.4˚C in 43 years is not unusual.

This is my third detailed post hunting for global warming. The first two looked at central Australia and Iceland where no evidence for warming in the last 100 years was found. Southern Africa makes it 3 out of 3. In central Australia I used all 30 records returned by the NASA GISS web platform and in Iceland I used all 8 Icelandic records that exist. Objectivity was assured. In this post necessity has required a less objective selection of climate stations which I am fully aware is open to criticism. If I have subjectively de-seleceted urban stations that show warming I’m more than happy to discuss this in comments. But beware, a station that shows warming needs to be surrounded by stations that also show warming for the data to be valid (see Appendix 1).

Station Selection Criteria

In Australia I used the NASA GISS web platform, clicked on Alice Springs, and this produced a list of 30 climate stations within a 1000 km radius, covering a huge part of that continent, and I used them all. In Iceland I used the data from all 8 climate stations. In southern Africa, objective station selection has been more challenging.

In southern Africa clicking on the NASA GISS map returns very large numbers of rural records that span 1969-1991. This makes it difficult to objectively select stations with long records that are not affected by urban warming and so I clicked around looking for long records. Urban records that showed warming were rejected while urban records that did not show warming were selected. My eye was also drawn to the large number of records that showed mid-1970s cooling and I regarded that as a hallmark of station quality.

Within congruous climatic regions, temperature trends are also expected to be congruous. In other words they should all be pretty much the same (Appendix 2). Inspection of Figure 1 shows that the selected records are all trending flat, and so if a record shows warming my rationale is that there is likely something wrong with it. See for example Pretoria (Appendix 1).

I am fully aware that this subjective station selection is wide open to criticism. But commenters are invited to click around and see if they can find geographic clusters of warming records. Lourenco Marques/Countinho is a good place to start. It has been warming since 1910, but the nearest rural record, Skukuza is completely flat as are the majority of rural records in a 600 km radius. On this list, check out Inhambane and Chipinge, two long records that I missed on my selection survey. Both completely flat and record 1970s cooling.

GHCN V2 Unadjusted Data

Figure 1 Raw temperature records for the 26 selected stations. The range in temperature between localities is enormous from 15˚C in Bethal to 26 ˚C in Zanzibar. Bethal is a small farming township in S Africa, 26˚S while  Zanzibar is an island off the east coast 6˚S of the equator. Bethal is at an altitude of 5,518 ft (higher than Ben Nevis) hence the cool climate. There is a paucity of long, old records and also a paucity of recent records owing to station closures. Sharp eyes will see that there is no overall warming trend, but a tendency for saucer shaped records with a warm past and a warm present.

Figure 2 How to make sense of the spaghetti in Figure 1? Each station record was converted to an anomaly by deducting the mean temperature for that station from the temperature – time series. Taking the arithmetic mean of the stack provides this dT anomaly trend. There are a number of key observations:

  • A linear regression through the data is effectively a flat line. There is no warming in these southern African stations.
  • There are two short cold periods, 1890-93 appears to repeat in 1974-76. The latter cold period is also seen in Australian records where the cold period spans 1974 to 1977 (one year longer).
  • The cold periods mark the low points of a circa 83 year cycle with gradual warming and cooling in between.
  • There is a high temperature outlier in 2005 brought about in part by a low number of operational stations (9).

Mid 1970s Cooling?

Figure 3 In scanning a large number of records a mid-1970s feature caught my eye that I had also seen in Australia. The period 1974-1976 appears to have experienced regional cooling.

GHCN V3.1 Adjusted Data

Figure 4 The GHCN V3.1 data for the same stations. 

Figure 5 The difference between V2 and V 3.1 data. These are the changes made to “raw records” under the banner of correcting for non-climatic artefacts. It needs to be noted that the changes are concentrated in a ±1˚C band and records are both warmed and cooled. But then some huge corrections are applied to certain records, almost 5˚C in the case of Windhoek and over 3˚C in the case of Keetmanshoop.

Figure 6 Screen shot of my spread sheet used to make Figure 5. Making Figure 5 was a lot of work since the V3.1 has so many edits compared to V2. The coding above is as follows:

Empty cell = no data V2 and V3.1; number in cell V2-V3.1; zero in cell V2=V3.1; yellow cell V2 data deleted in V3.1; green in cell V3.1 data exists where V2 data does not.

A huge amount of data has been deleted in V3.1 with a strong tendency to chop the old part of long records, arguably some of the most valuable data. And there is a tendency to generate new data at the young end of records where no V2 data exists. This combined with Figure 5 looks like shocking, mass manipulation of temperature records.

Figure 7 The V2 and V3.1 record count shows the true scale of data deletion and how around 1990 data deletion gives way to data creation. The chart also shows how, in keeping with other parts of the world, station closures has resulted in the recent past being represented by meagre numbers of stations.

Figure 8 But then making a dT plot for the remaining V3.1 records, we find that, like Australia, the impact of all the editing is minimal. Note that the pre-1894 data were erratic and they are not shown. A slight warming gradient is added to what was a flat record. But the warming is spread across the whole record and not post-1970. The 1970s cool period is even preserved. What has happened is that the variance is reduced and the clear cycles seen in the raw records have been straightened. Removing variance is I guess what homogenisation is supposed to do but here it is likely removing natural climate cycles from the data. And that gives me cause for concern.

I really don’t understand what GHCN are up to. The wholesale manipulation of the raw data displayed in Figures 5 and 6 is appalling. And yet, the net impact is minimal. Why do it? I do not believe that it is homogenisation of data that has resulted in flat temperature records being turned into warming records, but as yet I do not know what the cause of “spurious warming” is.

Figure 9 To conclude this section, the chart shows dT_V2 minus dT_V3.1 which, by definition, is an aggregate view from GHCN of how non-climatic artefacts have affected temperature records in the 26 stations selected.


I began this line of enquiry into temperature records with the aim of testing the veracity of GHCN homogenisation methodology. What I have found in Australia and southern Africa is that homogenisation results in the mass deletion of data, the addition of data and wholesale overwriting of original records but the technique does not greatly distort the overall temperature history. The extensive data editing shown in Figures 5 and 6 has essentially a zero outcome. So why do it?

The big surprise has been that central Australia and southern Africa show little evidence of warming. This is more than a little perplexing since GISS, NCDC, Hadcrut4 and BEST all show significant southern hemisphere warming. BEST shows warming in South Africa that I would assert is absent in raw, unpolluted data.

Figure 10 Time temperature trend for South Africa from BEST. Includes data from 193 stations. BEST’s focus has been on quantity where the focus should be on quality. How do flat raw records get turned into this?

Roger Andrews has pointed out in Homogenizing the World that these 4 reconstructions do not agree with each other in the southern hemisphere and that the land temperature record in the southern hemisphere, when area weighted, accounts for only 10% of the global picture.

My approach is biased towards avoiding heavily populated prosperous areas. I suspect that moving into SW Australia and using warming urban records from Africa would produce some warming that would appear attributable to human activity.  I can’t help but feel that there is either some flaw in my methodology and logic or in the methodology employed by everyone else (apart from Roger).

Appendix 1: Bad Records

Figure 11 Hunting for long continuous records, these ones caught my eye. Durban, Port Elizabeth and Diego Suarez each have a step down of between 1 to 2˚C. Had these features been aligned they may have meant something. But they are not. Step changes like this are perhaps linked to a station move, but it’s a lot of work to find out. Rather than make assumptions and try to correct records such as these I believe it is simply better to leave them out. There are more than enough “good records” to build the story.

Pretoria is an example of an urban record that shows significant warming. Pretoria University has a flat to cooling record that records the 1970s cooling and probably provides the real picture. Urban records that show warming like Pretoria should always be rejected in my opinion unless supported by the majority of surrounding records. I have used neither of the Pretoria records.

Appendix 2: Congruous Temperature Trends

The UK provides a good example of congruous temperature trends. ALL records display the same structure over a large area. Lerwick on the Shetland Islands is 1200 kms north of Southampton on the south coast of England. If a UK climate record did not conform to this regional trend there would most likely be something wrong with it.

Figure 12 Tmax, 5y running averages for 23 UK stations.

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

26 Responses to The Hunt For Global Warming: Southern Africa

  1. Euan Mearns says:

    Clicking through a large number of South African records I see either flat records or warming records. Both cannot be correct IMO. S Africa is either warming or not. I believe the warming records are correlated with urban areas, although not all urban areas show warming. I’m travelling for a week with regular wifi contact, but limited time to engage.

    • Euan Mearns says:

      In great haste I’ve made this chart for some of the rural stations around Laurenco Marques. Masvingo is not rural. Key observations, mid-70s cooling is there, there’s a temp spike in 1983, the overall trends are flat, and all records end in 1990 – which I think is likely the most significant observation.

  2. Euan, I don’t know if you’ve seen this, but Warwick Hughes & Robert Balling actually did a study of UHI in South Africa in the mid-1990s:;2-V
    Hughes & Balling, 1996

    You might find it of relevance for your post here?

    P.S. I just tried and failed *twice(!!)* to post a comment on your “My Enquiry to GHCN” post. It kept saying “Duplicate comment detected; it looks as though you’ve already said that!”

    • A C Osborn says:

      Ronan, it did post your comment OK, I have read both papers and comented on the “My Enquiry to GHCN” post.

      • A C, yes I saw that – thanks!
        I still keep getting the same “Duplicate comment detected” notification whenever I comment here. Does anyone else have that problem, or is it just me?

        • A C Osborn says:

          I think everybody has it, you just have to select cancel.

        • Euan Mearns says:

          Ronan, I think you posted an interesting link. I’m supposed to be on holiday. Didn’t have time to read it. Can you post raw link again please. Sorry to hear about commenting problems. A few folks have reported this.

          • Euan,
            No problem! Here’s the comment I posted on your “My Enquiry to GHCN” post:

            “In case you’re interested, in February 2014, we actually did a very detailed review & assessment of the homogenization algorithms used by NOAA on the GHCN and also the one used by NASA GISS. We’ve written two papers on them (one each) as part of our series on studying the urbanization bias problem, which we uploaded to our Open Peer Review Journal forum.

            They’re both quite long, but if you (or your readers) have the chance, they might provide some insight… FWIW, Paul Homewood mentioned he found them useful in his analysis which you refer to above.

            1) For NASA GISS’s “urbanization bias” adjustments, see here:

            2) For NOAA NCDC’s GHCN homogenization (& a discussion of the extent of UHI in the GHCN), see here:

  3. Some clarifications and comments:


    The raw records used in this post are from the old GHCN v2 data set, as were the raw records used in Euan’s previous posts on Australia and Iceland. GHCN v2 has since been replaced by the GHCN v3.2 raw data set but the differences between the two are not large.

    The adjusted records are from the GISS homogeneity-adjusted data set, as also were the adjusted records used in the Australia and Iceland posts. GISS homogeneity-adjusted is an independent data set produced by GISS, not a GHCN data set. GHCN v3.1 adjusted is no longer used. It was superseded by GHCN v3.2 adjusted in August 2012.


    … a station that shows warming needs to be surrounded by stations that also show warming for the data to be valid … commenters are invited to click around and see if they can find geographic clusters of warming records.

    Here are twenty stations, all showing net warming and covering an area approaching 3 million sq km. (Only 18 stations show up because of overstrikes around Johannesburg). Temperature anomalies for the 20 stations are plotted in the graph below the map.

    How do flat raw records get turned into (the BEST series)? Because BEST uses these stations. The mean of these stations (blue line) matches the BEST mean shown in Figure 10 quite closely:

    A linear regression through the data is effectively a flat line. There is no warming in these southern African stations.

    Here is my surface air temperature series for Africa south of the Equator, based on 67 unadjusted records that contain 3,740 years of data. It shows about 0.6C of net warming since 1900 and certainly isn’t a flat line.

    Southern Africa Bad Records: If Durban and Port Elizabeth are bad records then so is Cape Town. But If we use “congruity” as the criterion for acceptance these three coastal records plus East London fit together about as well as any other group of records in Southern Africa.

    There are more than enough “good records” to build the story.

    The Africa south of the Equator records are in fact among the worst in the world, and certainly the most difficult to make sense of.

    • Euan Mearns says:

      Roger, thanks for this. I see 2 kinds of records in S Africa. Flat ones and warming ones, geographically interwoven. Both cannot be right. There is either a natural warming or not. Here is a random selection of some records on your map. I’m travelling today and guess I will have to spend some time chatting to my sister this evening 😉 But I’ll do another post on urban v rural records. Happy to yield on the “bad records” no axe to grind there. But why is the step down staggered and do you have a climatic theory to explain it.

      Francistown 22,000
      Jo burg 1.4 million
      Pietersburg 27,000
      Jan smutts 1.4 million
      Bethal 30,000
      Potcheftroom 57,000
      Bloemfontein 182,000
      Windhoek 61,000

  4. Sam Taylor says:


    I seem to recall reading somewhere that the infilling of records was done to make it easier to calculate an absolute temperature record over time for an area, without the network composition jumping all over the place. I assume that it’ll just be some sort of spatial interpolation method that they use, so shouldn’t alter results significantly. We do similar things in seismic processing, because for certain processes having large holes in your data coverage can really mess things up. I would assume that in the master data file, wherever that is, all the interpolated records should be flagged so that they can be removed if desired. Frankly I wonder whether it’s been worth all the trouble it’s caused.

    Similarly the deletions are probably from an automated QC of some kind, and should be flagged somewhere, ideally giving a reason for why they were deleted.

  5. Nial says:

    “The big surprise has been that central Australia and southern Africa show little evidence of warming.”

    Do we know if any GMCs say there will be regions of the world where no warming is seen?

    • Euan Mearns says:

      Don’t know Nial. But I personally find it helpful to build a regional picture. Radiative forcing by CO2 doesn’t have to be uniform since it is in part dependent upon the temperature structure of the upper troposphere / stratosphere.

  6. manicbeancounter says:

    Thank you for all your hard work in producing this data.
    My preliminary observations are that
    1. The important part is not that the trend is flat before and after, but that the homogenisation removes the natural variations. A flat unform trend without fluctuations is easier to explain in terms of climate models than many different trends that cancel out. It also gives the models greater data fit.
    2. The uniform dip in temperatures in the early 1970s was similar in magnitude to that in Paraguay in the late 1970s.

    • Euan Mearns says:

      It also gives the models greater data fit.

      This worries me Kevin. Homogenise the homogenised and before you know it, temperature history might fit CO2 exactly.

  7. Roger Andrews says:

    Euan: If you want to contemplate the erratic nature of the SA records, try this:

    As you will have discovered there aren’t many long-term records in SA, but here are eight of the longest. They all show warming before 1930, cooling between 1930 and 1970 and warming after 1970, just like many of the NH records (they could in fact be NH records). Yet the records in between generally look nothing like them. What’s going on?

    • Euan Mearns says:

      Roger, many thanks for this great graphic. I suspect that many of the rural records in between do in fact conform to this, or are you trying to trick me 🙁 The very large number of 1960 to 1990 records tend to be flat, but many warm towards the end.

      The structure is not a lot different to my average.

      • Euan: Nope. No tricks, no smoke and mirrors, nothing up my sleeve, It’s just something that has genuinely perplexed me ever since I came across it.

        GISS of course adjusts the hell out of these eight records to make them show warming and/or deletes the front end of the record altogether.

        I suggest you don’t spend too much time on urban vs. rural. My results, based on GISS’s GHCNv2 data set, suggest a negative relationship between 1969-1999 temperature change (actually the difference between the 1963-1974 and 1993-2004 means) and population in Southern Africa. Or in other words, you have urban cooling there. 😉

  8. A C Osborn says:

    First of all I have to disagree with Euan’s statement that there is not much change by homogenising the data. I call a 100% to 150% increase in trend “Significant”.
    Ok, it may not be much numerically, ie from around 0.2 to to around 0.5, but it still changes things in the right direction for the AGW story.
    The other point is that you are comparing the current data to Roger’s data that ended in 2011, what has happend to the temperatures at any of those stations during 2012-2014 when the satellite data says nothing should be happening?

    One thing that we do know from Ronan’s work is that those Urban Stations with UHI will not have had their bias corrected, so they will increase Euan’s 100% trend change to an even bigger trend for the whole of the African continent.

    • Euan Mearns says:

      AC, I think one important point is that homogenisation does not lead to wholesale warming as initially alleged. But I agree that turning flat rural records into slightly warming records and then adding in warming urban records gets you to a goal of a warming S hemisphere.

      I will spend some time on Sunday I hope doing a systematic check of urban v rural records in S Africa. I really hope to squeeze an explanation out of BEST. Roger had an interesting graphic comparing urban records with BEST that he shared with me by email. Roger?

      • Euan. Homogenization usually adds warming in cases where the raw records show minimal warming, no warming, or cooling, but usually not in cases where the raw records already show warming.

        Southern Africa is an example. The plot of BEST against Southern Africa raw data I presented earlier shows that few if any adjustments are applied to raw records that already show warming. Here it is again.

        Now here is what the GISS adjustments do the 8 records I presented in the graphic above above that don’t show warming:

        This is why you don’t get “wholesale warming”.

  9. manicbeancounter says:


    I have been trying to replicate your work, and roughly come back to a similar result. However, there are some of points about presentation, that help enhance your results.
    1. The scaling of the graphs should be less than -3 to +3.
    2. I find it helps to have centered moving averages to help visualize the trends. I prefer 5 year centered moving averages. It does make a difference. For instance the GCHNv2 “raw data” shows 2006? as a clear record, but the years either side are >0.6C lower. This is clearer an outlier, but visually (with the large blobs) makes it seems that the middle of the last decade was the warmest since 1880. In the late 1920s there are a cluster of years all around 0.7C. Moving averages shift the emphasis from the outlier years to the trend.
    3. It would be useful to show the moving averages for “raw data”, and GCHNv3 on the same anomaly graphs. I have bashed out 5 year moving average graphs for these two along with the GISS Homogenised used in the global surface temperature graphs. My “raw data” graph is not dissimilar to that of Roger Andrews spagetti graph for 8 long record stations on March 6, 2015 at 4:18 pm – though with a lower peak of 0.9 in 1927-9. GHCNv3 and Giss homogenised are almost identical, totally eradicating the early twentieth century warm period. From the mid-1970s all three series are virtually identical.


    • Euan Mearns says:


      1) Agreed. I think I did this first in Central Oz where there is not a lrage range in absolute temps and I scaled the dT charts to the natural range. Expanding Y-axis scale is no problem.
      2) I like 5y centred moving average. But 18 months ago presenting the work I did with Clive on UK sunshine and temperature (where my UK Appendix chart comes from) we got hammered on Climate Etc for using it. Its probably best to show raw data and 5Y.
      3) Yes, homogenisation removes structure from the data that is likely natural and important to retain.

  10. Pingback: The Hunt for Global Warming: Southern Africa Part 2 | Energy Matters

  11. Pingback: The Iceland Meteorological Office Versus GHCN V2 | Energy Matters

  12. Pingback: Understanding GISS Temperature Adjustments | ManicBeancounter

Comments are closed.