In this post the raw temperature records for 7 Icelandic climate stations as reported by the Iceland Meteorological Office (IMO) are compared with the equivalent records archived by GHCN as V2, presumed previously to be raw temperature records. What should have been a straight forward exercise is in fact a total mess. Out of a possible total of 1056 annual records (8 locations times 132 years):
- 220 records exist in V2 that do not have an IMO equivalent*
- 104 records that exist in the IMO archive do not exist in V2
- 182 IMO records are found to be exactly the same as V2**
- 334 V2 records have been changed from the IMO raw data
- 216 records do not exist in either archive, i.e. no data
* This includes the whole of the V2 Vestmannaeyja record that we cannot find on the IMO data base
** For 7 locations there are strings of data that are identical in IMO and V2 archives suggesting the latter did indeed start life as raw IMO data.
At face value, something appears to have gone catastrophically wrong with the GHCN V2 archive which cannot be viewed as raw data in its current form. One possibility is that the IMO provided records to GHCN that do not match their current archive and we are awaiting IMO to clarify this point. Alternatively, we may have made some mistakes in accessing the raw IMO records.
However, despite what appears to be mass deletion, addition and changing of records by GHCN or NASA GISS the net result is minimal. It is hard to spot the differences between the IMO and GHCN V2 in the mean dT time series plot. As was the case in Central Australia and Southern Africa [2, 3], everything and nothing has been changed. Clarification is required from both the IMO and NOAA / GHCN.
Figure 1 Map showing the location of six of the climate stations discussed in this post.
[Note added 17:30 on 13th March: Ducdorleans has informed me that there is an additional unofficial Iceland temperature series that includes some early unaudited data for Akureyri and Reykjavik and this might explain how GHCN have some early data for these stations that is absent from the IMO official data set.]
[Note added 14:00 on 14th March: I have had a detailed response from the IMO that included a file with data for Vestmannaeyja, 1877 to 1995. The response details some of the many challenges involved in archiving records and also highlights the possibility for misunderstanding through inconsistent use of nomenclature. For example, what exactly are “raw records” and what exactly are “official records”. At present it suffices to say that data reported by GHCN for Akureyri, Reykjavik and Vestmannaeyja does exist (in green on my Figure 2), but evidently not archived on the IMO web site.]
In the wake of my recent posts on Iceland , Australia  and Africa [3, 4] that have attempted to clarify the impacts of GHCN homogenisation of raw climate data, commenter Ducdorleans suggested that we should cross check the GHCN V2 data against the raw data archived by the Iceland Meteorological Office (IMO). This seemed like a good idea but it was not a top priority for me and I invited Ducdorleans to assist by compiling the IMO data onto my spread sheet which he duly did.
GHCN stands for Global Historical Climatology Network and is administered by NOAA (the US National Oceanic and Atmospheric Administration). GHCN V2 temperature records are reported by NASA GISS to be raw records and are accessed via the NASA GISS web platform here. V2 lists temperature records for 8 Icelandic stations and this GHCN list forms the basis of this comparison. Here’s how NASA GISS describe V2:
UK Press reports in January 2015 erroneously claimed that differences between the raw GHCN v2 station data (archived here) and the current final GISTEMP adjusted data were due to unjustified positive adjustments made in the GISTEMP analysis.
On this basis we have wrongly assumed that V2 should be raw temperature records as delivered by national organisations like IMO. This analysis shows this to be the case. Here’s what NOAA have to say:
The archive [V2] also contains homogeneity-adjusted data for a subset of this network (5,206 mean temperature stations and 3,647 maximum/minimum temperature stations). The homogeneity-adjusted network is somewhat smaller because at least 20 years of data were required to compute reliable discontinuity adjustments and the homogeneity of some isolated stations could not be adequately assessed.
But of course, the IMO data are already homogenised and should require no further adjustment. And NOAA also say this:
Both historical and near-real-time GHCN data undergo rigorous quality assurance reviews. These reviews include preprocessing checks on source data, time series checks that identify spurious changes in the mean and variance, spatial comparisons that verify the accuracy of the climatological mean and the seasonal cycle, and neighbor checks that identify outliers from both a serial and a spatial perspective.
Data for the Iceland Meteorological Office were accessed here. The records for Reykjavik, Stykkisholmur, Akureyri and Teigarhorn were taken from the “Longer series” tab. The records for Grimsey, Hofn I Hornafirdi and Keflavikurflu were taken from the “Monthly averages for selected stations tab”. We could not find records for Vestmannaeyja and we are therefore unsure what source GHCN has for this station. The Vestmannaeyja data were found to be spurious in any case and not included in our final analysis .
Both GHCN and IMO report monthly data. GHCN use what is called a Meteorological year that runs D, J, F, M, A, M, J, J, A, S, O, N where December is derived from the previous year. Ducdorleans computed annual averages (metANN) from the IMO data accordingly.
As a first step for this data comparison we deducted the GHCN V2 data from the IMO data. The results shown in Figure 2 immediately reveal some serious issues as already described in the summary up top. Perhaps one of the more important observations is that for 7 of the stations, there are strings of temperatures where the difference between IMO and GHCN is exactly zero (Figure 2 and 3). This proves that GHCN is using the same IMO files that we are using, at least in part. Since we had to compute our own metANN values that may be subject to rounding errors zero is defined as 0.00±0.01.
The difficult part is that GHCN appear to have; 1) “added” data that does not exist in the IMO files (green); 2) deleted data that does exist in the IMO files (yellow) and 3) changed data from that reported in the IMO files on what appears to be quite a grand scale (Figures 2 and 3).
Figure 2 Screen capture of our spread sheet were the GHCN V2 data are deducted from the IMO data.
- Green = GHCN V2 record exists where there is no IMO record
- Yellow = IMO record exists but do not appear on GHCN V2
- Zero in cell means IMO = GHCN V2 (0.00±0.01)
- Number in cell means IMO≠GHCN V2
- Empty cell means no data in either IMO or GHCN V2
The matrix is plotted in Figure 3.
Figure 3 Plot of the data shown in Figure 2. If GHCN V2 was plotting raw temperatures all of these data should fall on the zero line. A sufficient number of the data do fall on the zero line to demonstrate that the source data for GHCN V2 is the IMO. The variance away from zero has similar appearance to the homogenisation dT plots shown in earlier posts and confirms that the GHCN V2 data have been processed and are not raw records. Divergence from zero is particularly marked pre-1960. Note that Iceland temperature records are already adjusted by the IMO and no further adjustments should ever be necessary.
The Temperature History of Iceland
The temperature – time series for 7 climate stations as reported by the IMO are shown for reference purposes in Figure 4. The series are congruous in that they go up and down together as one would expect. Grimsey, a small island off the N coast stands out as having the coldest record, consistent with its location (Figure 1).
Figure 4 Temperature-Time series for 7 stations as reported by the IMO.
Figure 5 The temperature series shown in Figure 4 were converted to anomalies (dT) by subtracting the mean temperature for a station from the temperature series for that station. The arithmetic mean of the dT stack is shown above. Sharp eyes are required to see the difference between the IMO and GHCN V2 average plot shown below in Figure 6.
Figure 6 Temperature – time series for the GHCN V2 data. The main material difference is the fact that the IMO data show 2003 to be a clear winner for warmest year by 0.32˚C over 1939. The V2 data show 1939 and 2003 tying for warmest year, achieved by V2 warming the past! When you look at the scale of data adjustments that have been made (Figures 2 and 3) it is astonishing that this has so little impact, so astonishing in fact, I find it highly suspicious.
Figure 7 Finally, to complete the story, subtracting the dT curve for GHCN V2 from the IMO average shows how little difference there is between the two (plotted at same scale as Figures 5 and 6). Enigmatically, in this case the past has been warmed a little and the recent past cooled a little by GHCN with the net effect of equalising the 1930s and recent temperatures.
Previous posts [1, 2, 3] that compared V3 with V2 data revealed robotic style homogenisation where exact decimal fraction adjustments were made to strings of data. This pretty well proves that V3 adjustments were being made to V2 data. At the time I believed that V2 was raw data which turns out not to be the case. The Iceland data has had valid adjustments made to it by the IMO and should never be tampered with. And yet GHCN V2 has applied non-valid homogenisation adjustments to it and then they have adjusted V2 with even more adjustments in V3. This is totally unacceptable, non-scientific manipulation of raw temperature data.
I find it difficult to rationalise what the individuals proposing and enacting this gross data manipulation hope to achieve. The pattern of gross “data addition” and “data deletion” evident in Figure 2 awaits clarification and explanation from IMO and NOAA / GHCN or NASA GISS where the data was accessed. But if it is confirmed would suggest that data quality control at NOAA / GHCN or NASA GISS has undergone a catastrophic failure.
Figure 8 The equivalent graphic to Figure 2 where in this case V3 data is deducted from V2 data. The band of 1960s records that are deleted is in itself suspicious. And where did all that post 1971 V3 data (Green) come from? The answer there is straight forward. It comes from the IMO, but why then was it not present in the V2 data?
If we remind ourselves of the pattern of data addition and deletion when comparing V3 with V2 data we see that post-1971 there was extensive “data addition” and I was left wondering where this data, absent in V2, present in V3, came from. Well, the answer appears to be simple: it came from the IMO.
I conclude that I do not trust any of the GHCN data published by NOAA or NASA GISS and temperature indices based upon them, and nor should anyone else. This is not yet the end of this story. Ducdorleans is conducting further forensic analysis of the GHCN records.