This post follows up on the hornet’s nest stirred by Paul Homewood and Christopher Booker and my recent post on Temperature Adjustments in Australia comparing raw temperature records (GHCN V2) with homogenised (adjusted) temperature records (GHCN v3.1). The latter is currently used by NASA GISS and NOAA in global temperature reconstructions.
In this post I examine the records of eight climate stations on Iceland and find the following:
- There is wholesale over writing and adjustment of raw temperature records, especially pre-1970 with an overwhelming tendency to cool the past that makes the present appear to be anomalously warm.
- In the 1960s, Iceland (and the whole N Atlantic) experienced a run of very cold years caused by extreme atmospheric pressure differentials linked to the North Atlantic Oscillation (NAO). Many of these cold records appear to have been systematically deleted in V3.1 with the effect of all but removing this well-documented event from Icelandic climate history.
- Following the end of the Little Ice Age, Iceland experienced rapid warming in the 1920s reaching near “record warmth” in 1939. This near record warmth has also been written out of Icelandic climatic history by adjusting the temperature records down, leaving the false impression that 2003 was an anomalously warm year.
- In addition to wide-spread deletion of records, large amounts of temperature data that does not exist in V2 appears to have been created in V3.1. It is difficult to understand why this should be done since it is quite straight forward to manipulate data without apparently having to make it up.
In central Australia I found that wholesale manipulation of records did not bias the outcome. The V2 and V3.1 averages for 30 stations were the same. This cannot be said for Iceland. I dare say if one looked at a larger number of stations any bias may disappear but that is beside the point. Iceland holds a strategic geographic position in the N Atlantic and the public have the right to unadulterated information about that island’s climate history. There is little evidence of warming since 1917 which just happens to be the same conclusion I reached for Central Australia.
[Erratum 4th March: Commenter Sam Taylor, setting out to replicate my results, discovered an error in my spread sheet. The V3.1 record for Teigarhorn that should have begun in 1884 was accidentally pasted into 1880, an offset of 4 years. It was noted in the post that Teigarhorn was chaotic, and now we have an explanation. Fixing this actually removes noise from a number of charts. And for example, this completes the zone of data deletion in the 1960s (Figure 8). It does not materially alter the conclusions of this post.
In important Figures 6, 7 and 10 I’ve left the original figures there and added updated ones so that readers can see the impact of this error. In Figure 8, the corrections are not material and that figure has simply been replaced.]
First, for a little bit of background I suggest reading Paul Homewood’s post The Sea Ice Years that provides a perspective of Icelandic climate change through the lens of peer reviewed literature. And so what is GHCN and V2 and V3.1 and homogenisation all about?
GHCN is the Global Historical Climatology Network (GHCN). It is a database of temperature, precipitation and pressure records managed by the National Climatic Data Center, Arizona State University and the Carbon Dioxide Information Analysis Center (Wikipedia).
GHCN V2 temperature records are supposed to be raw temperature records as measured by thermometers from around the world. Raw temperature records are subject to a multitude of possible spurious variables over decades and centuries: 1) movement of measurement site, 2) change to the physical environment of the site – growth or removal of trees, urban encroachment, 3) change to the time of day measurements were made and so on.
There are two approaches as to how to manage this spurious input to temperature records. The first is to assume that in aggregate, some changes will result in spurious warming and some will result in spurious cooling and that on average most of this will come out in the wash. This is the approach I would follow and was in fact the one preferred by NASA GISS.
To recap, from 2001 to 2011, GISS based its analysis on NOAA/NDCD’s temperature collection GHCN v2, the unadjusted version. That collection contained for many locations several records, and GISS used an automatic procedure to combine them into a single record, provided the various pieces had a big enough overlap to estimate the respective offsets; non-overlapping pieces were combined if it did not create discontinuities. In cases of a documented station move, the appropriate offset was applied. No attempt was made to automatically detect and correct inhomogeneities, assuming that because of their random nature they would have little effect on the global mean.
But NASA GISS, who use the GHCN records have had “homogenisation” thrust upon them. Homogenisation is a computer robot (bots) based procedure that purports to detect these occasional site specific incidents and to correct for them using what is called the Pairwise Homogenization Algorithm (PHA). So how can we be sure that these bots can discriminate between temperature changes specific to a site and those that might be a change in the regional temperature trend?
This post is about examining the justification of the temperature changes made by bots to the raw temperature records of Iceland. Here, we need to first consider the possibility that the Icelanders may have already corrected their raw records for station moves etc. Do the bots know this or not? Can any of the massive and systematic data manipulation presented here really be put down to changes in the physical conditions of 8 climate stations in Iceland? GHCN V2 = raw temperature records, GHCN V3.1 = records adjusted by bots for supposed non-climatic variance.
GHCN V2 raw temperature records
Figure 1 The first 8 stations on this list are the 8 Icelandic stations analysed. This is the printout from V3.1 with Akureyri placed at the centre. The V2 stations where I used the selection “after combining sources at same location” have the same ID numbers apart from Akureyri and Keflavikurflu where the last digits differ. The next nearest station is then Kap Tobin in Greenland and then Thorshavn on the Faroes and Jan Mayan that is a small island in mid Atlantic, 699km N of Iceland. Thorshaven and Jan Mayen are interesting because they have the same temperature structure as the Iceland Stations.
Figure 2 This chart shows the basic structure for all 8 stations, V2 unadjusted. Flicking through the individual station records its possible to identify some features common to most records 1) pre-1917 there were periodically cold years, 2) 1917 to the 1930s was a period of marked warming 3) warming was then followed by cooling and then in the 1960s there was a run of very cold years 4) from about 1980, modern warming began. Further to this, Vestmannaeyja, an island off the S coast, is an outlier with a warmer past than the other stations. The warmest annual record ever recorded on Iceland was on Vestmannaeyja in 1928.
Figure 3 This chart shows the simple arithmetic mean of the stack shown in Figure 2. Roger Andrews has cautioned that this is not best practice since discontinuous and incomplete records can impart structure to the data. The distribution of stations with time is shown in Figure 4. For the first two years there are only 2 stations and this may provide biased non-representative data, but to be honest any structure that links to the coming and going of records is not obvious. It will undoubtedly impact the fine structure, but not the big picture. One reason for this is that most stations are similar, with a range in temperatures any given year of only 2˚C. Hence I believe this chart is a fairly good average rendition of what Icelandic thermometers show. The main features are as described for Figure 2 and as annotated on the chart.
Figure 4 The history of opening and closing climate stations on Iceland. For most of the period there have been at least 5 stations. They all show pretty much the same thing and so there is no major issue about being representative. Note how V3.1 has deleted many station records, especially in the 1960s cold period and how V3.1 manages to have more records post-1968 than there were operational stations, a point discussed at length below.
GHCN V3.1 adjusted data
Figure 5 Moving on to look at the adjusted data, comparing this chart with Figure 2, changes that have been made may not appear obvious.
Figure 6 Looking at the arithmetic mean of the data shown in Figure 5, sharp eyes are required to see the difference with the V2 data (Figure 3). The cool past is still there, the 1920s warming is still there. But the 1939 “record warmth” has gone completely. The 1964 to 1971 cool period is still there but much reduced. And the recent peak warmth in 2003 stands out imperiously as the warmest year by far. The V2 data was effectively a flat record for 100 years has been turned into one of steady warming. Its easier to see what is going on by looking at the difference between the V2 and V3.1 data (Figure 7).
Figure 6a Correcting for the Teigarhorn error (see Erratum up top) the main effects are the amplification of the temperature high in 1964 and an amplification of the temperature low in 1881. The record 1939 warmth is still gone.
Figure 7 Subtracting the V3.1 data from the V2 data reveals the structure of adjustments made by the “homogenisation” process. There are three significant features. A steady warming of the record has been achieved by cooling the past. The 1939 “record warmth” has been removed by adjusting data in a very narrow band 1936 to 1943 and the marked 1960s cooling is achieved by adjusting data in another narrow band 1965 to 1971. How has this been achieved?
Figure 7a Correcting for the Teigarhorn error (see Erratum up top) this chart has become more robot like with straight lines and right angles. It has not changed the picture for 1939. But it has subtly changed the picture for the 1960s cooling. While there appears to be systematic deletion of records 1963 to 1966 (Figure 8) it becomes more clear to see that the suppression of cooling has been achieved more by cooling the pre-1964 records than actually deleting the cold records.
Figure 8a Correcting for the Teigarhorn error (see Erratum up top) has no major material effect on this graphic.
This is a screen capture of my spread sheet for the V2 minus V3.1 data. You’re not supposed to be able to read the numbers. The first column is year and the next 8 columns are the delta temperature series for the 8 stations. 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.
The main effect is to complete the zone of data deletion in the 1960s. I don’t why this has been done.
In green are segments of data that do not exist in V2. I really don’t know how to explain this. It is not necessary to add data in order to adjust it. This requires an explanation from GHCN. But none of this explains how they got rid of the 1930 warm peak.
Figure 9 These screen captures show segments from V2 (top) and V3.1 (bottom). Scan along 1939, 1940 and 1941 and you’ll see that the temperatures have quite simply been reduced and a peak has been turned into a trough. The finishing touch is to simply delete a segment of high T data in the final column (Keflavikurflu).
Figure 10 For completeness, here is the dT plot for the individual stations which has a number of interesting features. The robotic decimal fraction additions and subtractions seen for central Australia are also present here although in this case there is a heavy bias towards cooling the past. There is a major discontinuity across the deleted band of 1960s data. Since then, all adjustments have been small. Have the Icelanders stopped moving their weather stations? The adjustments to the 1960s cold spell are not visible because the data has been deleted 🙂 The adjustments to the 1939 warm peak are visible but not obviously affecting many stations. That is because pre-1939 there were only 5 operating stations. One has not been adjusted at all (zero line for Stykkisholmur). Three have been adjusted by exactly 0.6˚C and are therefore superimposed (Akureyri (way up north), Reykjavik and Vestmannaeyja (way down south)). And one station, Teigarhorn, is chaotic throughout. I’ve double checked this and the V2 and V3.1 records have nothing in common. It is hard to believe that V3.1 record for Teigarhorn is a derivative of V2.
Figure 10a Correcting for the Teigarhorn error (see Erratum up top) perhaps has the greatest impact on this chart. The unexplained chaos in Teigarhorn is gone and replaced by robotic adjustments to temperature. Its quite surprising to see what an accidental 4 year displacement in a single record can do.
Whilst I am a confirmed lukewarm climate sceptic, I have until now accepted without question that Earth’s lower troposphere was warming through a combination of natural and anthropogenic causes where the division between these two has yet to be worked out and confirmed. Now I’m not so sure. Every time I scratch the surface of climate data I find something wrong. It shouldn’t be like this!
GHCN V3.1 homogenisation of temperature records is supposed to remove non-climatic artefacts from the data. Is it possible that the homogenisation algorithims and robots have accidentally removed real climate signal in Iceland through a combination of deleting records, changing records and adding records? And is it a coincidence that the outcome is to create a picture of warming from one that was largely flat for the last 100 years (Figure 11 below)? If this is the case then GHCN need to tear up their V3 algorithims post haste because they are clearly rubbish.
Or is it possible that there is a darker side to this where a hand of human intervention has played a role? It should be abundantly clear that the US government needs to urgently conduct an enquiry into this agency and those related to it who seem all too willing to accept the data produced by GHCN.
The conclusions here are exactly the same as for my previous post on central Australia. GHCN V3.1 homogenisation corrections appear to have deleted real, regional climatic signal whilst creating a warming signal that does not appear to exist in the raw records. In Iceland it is worse since in Australia the manufacture of warming was at the station level. In Iceland, it is at the island / country level. In fact there is a geological argument for Iceland being a micro-continent.
To conclude, let’s take a look at the post-1917 data, when the cold influence of The Little Ice Age seemed to end in Iceland. A simple arithmetic mean of unadjusted temperature records is close to flat for the last 98 years. There is little evidence for warming in Iceland for the last 98 years, similar to the conclusion reached for central Australia.
Figure 11 Since the beginning of 1920s warming, the temperature history of Iceland is close to flat though warming slowly.
Added Sunday 1st March.
In light of the discussion in comments I have produced an anomaly chart for 7 of the stations, excluding Vestmannaeyja that is an outlier.
Figure 12 Temperature anomaly chart where each station is normalised relative to the mean temperature for that station. Vestmannaeyja is excluded. 1939 has become the warmest year on record with dT=1.73˚C compared with 2003 dT=1.70˚C.
Added Wednesday 4th March
I guess I need to apologise for the Erratum added today that makes reading this more difficult. One careless mistake pasting one column of data offset by 4 years. But the real problem here is the chaos in the GHCN archives. It should be a very straight forward exercise to compare one generation of data to the other. The actual raw data should be archived and never tampered with. Comparing V2 (alleged raw data) with V3.1 (processed data) is a minefield to navigate. And now I’m embarking on comparing the Icelandic Meteorological Office (IMO) data with the raw GHCN data. There seems to be mines everywhere.