GISS gets it wrong again, and again! (or, Russia's second October revolution)

NASA's  Goddard Institute  for Space Studies (GISS) provides one of the most relied upon indicators of global temperature, GISTEMP.  Each month, GISS releases key data in the form of maps and overall global average temperatures which are eagerly studied for evidence for or against global warming. So you might think that GISS would carry out some elementary checks on the correctness of their data before publishing it. But apparently this is not the case. Here is what happened with the publication of the October 2008 data:

November 10: GISS publishes data for October 2008. This reports a global temperature (land+ocean) anomaly of 0.78. This figure is much greater than in other recent months, and much greater than the figures obtained from satellite observations (0.17 from UAH and 0.18 from RSS). The corresponding temperature map looks like this:

Notice that the whole of Russia has turned red (reminiscent of the October revolution 91 years ago).  Notice also the size of the temperature anomaly in Russia, up to 13.7 degrees C (the temperature anomaly means that October temperatures in Russia were allegedly up to 13.7 degrees warmer than the average for October over the period 1951-1980). It appears that nobody at GISS thought to check whether this very surprising result was accurate or not before announcing it to the world. 

Almost immediately, people started to check the data, and found a very simple error. This was reported at the blogs Watts Up With That (see the comment by Chris, "The first station I click on in the “offending” dark orange area of Russia has a mean temperature of 8.1C for October….. the mean temperature for September was also 8.1C. Coincidence?") and Climate Audit.  Note the degree of technical expertise required by Chris to find the error.  Not only did he have to click on the GISS station data map in the doubtful region and select a weather station,  he also had to click on "Download monthly data as text" and look at two numbers.    The "October" temperature data was in fact just a duplication of the September data, and of course this lead to the huge (anomalous!) anomaly for October. According to Steve McIntyre, this error occurred for 90 different stations, mostly in Russia but also some in Tunisia, Finland and the UK.

November 11: GISS remove some of the incorrect data and graphs, and post the following comment on their website:

Most data posted yesterday were replaced by the data posted last month since it looks like some mishap might have occurred when NOAA updated their GHCN data. We will postpone updating this web site until we get confirmation from NOAA that their updating programs worked properly. Because today is a Federal Holiday, some pages are still showing yesterday's data. 

November 12:  The GISS website is down for much of the day (apparently unrelated). It comes back with the message:

It seems that one of the sources sent September data rather than October data. Corrected GHCN files were created by NOAA. Due to network maintenance, we were only able to download our basic file late today. We redid the analysis - thanks to the many people who noticed and informed us of that problem.

What is meant by 'one of the sources' is not clear.  On their Graphs page it is stated that:

Monthly graphs and maps were created with corrected NOAA/GHCN data. 

The revised GISS land+ocean anomaly is now 0.58, and the temperature map looks like this:

The peak temperature anomaly has fallen to 8 degrees, and the red army appears to be on the retreat - but they are showing signs of resistance in central Siberia.  GISS assures us that the data has been "corrected".    NASA's  Gavin Schmidt confidently announces on his blog: "The corrected data is up. Met station index = 0.68, Land-ocean index = 0.58, details here. Turns out Siberia was quite warm last month." So that's surely the end of the story. The data has been corrected, and surely having made such a basic error,  GISS and Gavin will have double checked that the new data is indeed correct?

November 13:  

At Climate Audit, doubts are raised about the "corrected" data, by Jared and others ("Siberia, the Arctic, and most of Asia is still considerably warmer than the satellites indicate").  I repeat the highly technical analysis carried out by Chris previously (three clicks of a mouse in the dark red area) and see that GISS is reporting a mean October temperature for Irkutsk of 9.9, compared with 2.1 last year. But according to Weather Underground, the figures for this year and last are both 1 degree, and Russia's Weather Server gives a mean temperature for Irkutsk of 0.85.  There is the same problem at nearby Bratsk. This confirms that the "corrected" GISS data is still wrong, acknowledged by Gavin Schmidt: "Groan… The NOAA fixes are not complete. There are still some stations where they have some Sep data in the Oct column. Stay tuned for more updates…" (the source of the second error is that the GISS "correction" did not include stations where for some reason the September figure was missing in their data).

GISS post the following unclear comment on their website:

NOAA combined the data it obtained from their various sources into another, slightly differing version of v2.mean. We ran our analysis again based on this new file; you may have to clear your buffer to see the updated results.

And on the Graphs page the rather clearer:

NOAA corrected GHCN data for October, 2008 again (second correction). Monthly figures have been corrected (third version).

The GISS land+ocean anomaly is now down to 0.55, and we  have a third version of the temperature map:

The dark red region has now retreated to a small region of Northern Siberia, greatly exaggerated in size by GISS's  misleading choice of the Mercator Projection.  Very different from what happened in October 1917 (or was it November?)

This episode raises many questions about the data provided by GISS:

  1. How many other errors, less obvious to the casual observer, are there in the GISS data? (recall that in September 2007, McIntyre found an error in the way US temperature was handled). Can the GISS data be trusted at all?
  2. Why does GISS not carry out any checks on the data before publishing it?
  3. Where and how did these errors arise? GISS appears to be blaming NOAA or GHCN.  But this is no excuse - it is the data from GISS that is used as a key indicator of temperature and they are responsible for it.
  4. Why are there gaps in the GISS data, when the "missing" data is readily available? (Without the gaps, the second error referred to above would probably not have occurred).
  5. Looking at the GISS data shows that many stations on their list are no longer used. Why has GISS's number of stations used dropped so dramatically in recent years? What can be gained from a comparison with 20 years ago when the set of stations used then was completely different?
  6. Why are so many of the remaining stations at airports? (Three quarters of the GISS stations in Australia are at airports.)
  7. Why does GISS keep adjusting past temperatures, as shown here?
  8. Why are GISS temperatures consistently higher than other sources? (Even after the second correction, the GISS October Irkutsk figure is 2.4, significantly greater than the other sources given above)