GHCN Temperature Adjustments Affect 40% Of The Arctic
By Paul Homewood
Before After
There has been much discussion recently about temperature adjustments made by GHCN in Iceland and Greenland, which have had the effect of reducing historic temperature levels, thereby creating an artificial warming trend. These can easily be checked at the GISS website, where both the old and new datasets can be viewed as graph and table data, here and here.
It has now been identified that similar adjustments have been made at nearly every station close to the Arctic Circle, between Greenland and, going East,via Norway to Siberia, i.e 56 Degrees West to 86 Degrees East, about 40% of the circumference.
So it is perhaps time to recap where we are now.
Background
The NCDC has produced the Global Historical Climatology Network (GHCN), a dataset of monthly mean temperatures, since the 1990’s. Version 2 was introduced in 1997 and included “Methods for removing inhomogeneities from the data record associated with non-climatic influences such as changes in instrumentation, station environment, and observing practices that occur over time “. The GHCN datasets are used by both GISS and HADCRUT for calculation of global temperatures, as well as NCDC themselves.
In May 2011, NCDC brought out Version 3, which “enhanced the overall quality of the dataset”, but made little difference in overall terms. However, only two months later in July, a Google Summer Student, a graduate called Daniel Rothenberg, was brought in to convert some of the GHCN software and make modifications to “correct software coding errors”. The result was Version 3.1, which went live in November 2011. (The full technical report is here).
It is this latest version that has thrown up the Arctic adjustments we are now seeing.
Until December, GISS used Version 2 unadjusted temperatures. Since then, they have changed to using Version 3.1 adjusted temperatures.
Basis of Homogeneity Adjustments
It is worth taking time to be clear why temperature adjustments are made (or should be). As far as GHCN are concerned, they explain their logic thus :-
Surface weather stations are frequently subject to minor relocations throughout their history of operation. Observing stations may also undergo changes in instrumentation as measurement technology evolves. Furthermore, observing practices may vary through time, and the land use/land cover in the vicinity of an observing site can be altered by either natural or man-made causes. Any such modifications to the circumstances behind temperature measurements have the potential to alter a thermometer’s microclimate exposure characteristics or otherwise change the bias of measurements relative to those taken under previous circumstances. The manifestation of such changes is often an abrupt shift in the mean level of temperature readings that is unrelated to true climate variations and trends. Ultimately, these artifacts (also known as inhomogeneities) confound attempts to quantify climate variability and change because the magnitude of the artifact can be as large as or larger than the true background climate signal. The process of removing the impact of non-climatic changes in climate series is called homogenization, an essential but sometimes overlooked component of climate analysis.
It is quite clear. Their algorithms should look for abrupt changes that are not reflected at nearby stations. It has nothing to do with “averaging out regional temperatures” as is sometimes claimed.
GISS also make homogeneity adjustments, but for totally different reasons. In their case, it is to make an allowance for the Urban Heat Island Effect (which is not spotted by GHCN because it is a slow change).
Effect of The Adjustments
Appendix A lists every current GHCN station with records back to 1940,that lie between Greenland, at a latitude of 56 W, around to a point about midway across Siberia at 86 E and which are situated close to the Arctic Circle. The table shows the adjustment made by GHCN for 1940 data. Out of 26 stations, the adjustment has reduced actual temperatures in 23 cases, many substantially. In contrast, 2 remain unchanged and only one has a positive adjustment (and this is insignificant). As a crude average, the adjustment works out at a reduction of 0.70 C.
These adjustments typically extend back to the beginning of the station records (though Reykjavik is an exception) and most continue at the same level till about 1970. ( Some of the Russian stations last longer – e.g. Ostrov Dikson’s disappears in 2009).
By 2011, however, the adjustments disappear at ALL of these sites. In other words, an artificial warming trend has been manufactured.
It is worth spelling out two points :-
1) Within this arc of longitude, there are no other stations within the Arctic Circle.
2) With the exception of Lerwick and Vestmanneyja, I can find no stations, in the region, below a latitude of 64 North with similar adjustments. Why is 64 North significant? GISS produce zonal temperature data, and their “Arctic” zone goes from 64 North to the Pole. Coincidence?
Is there any justification for adjusting?
Trausti Jonsson, a senior climatologist at the Iceland Met Office, has already confirmed that he sees no reason for the adjustments in Iceland and that they themselves have already made any adjustments necessary due to station moves etc before sending the data onto GHCN.
Clearly the fact that nearly every station in the region has been adjusted disproves the idea that these sites are outliers, which give biased results not supported by nearby stations.
GHCN were asked in January to investigate this issue and so far have failed to come up with any explanation. Unless they can do this, the assumption must be that the adjustments have been created by faulty software.
Discussion
In global terms, these few stations make no tangible difference to overall temperatures. However, they do make a significant difference to temperatures in the Arctic, which are derived from a small number of stations such as these and then projected over hundreds of miles.
Across much of the Arctic, temperatures were as high in the years around 1940 as they are now. History should not be revised at the whims of an algorithm.
What should happen next? In my view, GHCN should immediately revert to Version 3.0 until the matter is properly investigated and any issues resolved. They maybe just need to put Version 3.1 down as a bad experience and start from scratch again. I believe they also need to seriously review their Quality Control procedures and question how these anomalies were allowed to arise without being flagged up.
It should not be up to independent observers to have to do this.
References
1) GISS still archive the Version 2.0 data here. (Also GISS, following requests by me and others, have included a link to Version 2.0 on their main site).
http://data.giss.nasa.gov/gistemp/station_data_v2/
2) And can be compared with Version 3.1 here.
http://data.giss.nasa.gov/gistemp/station_data/
3) The adjustments can also be seen in graph format at GHCN here. (The station numbers can be obtained at GISS)
ftp://ftp.ncdc.noaa.gov/pub/data/ghcn/v3/products/stnplots/
Footnote
I originally set this table up yesterday, 9th March. Today I noticed a few had changed slightly, presumably at the monthly update, so have amended them. It appears GHCN are still fiddling with their algorithms as the same thing occurred last month.
steven mosher says:
The homogeniety adjustments have been tested rigorously in a blind test.
That is, temperature signals are held from the algorithm writers.
The temperature series are then “infected” with various forms of error.
The algorithms are then tested for their ability to restore the series to the truth.
I’ll suggest that everybody read the papers and the code.
The issue at hand is not the algorithms’ ability to find error where it exists, but their propensity to find and “correct” errors that aren’t there. That you so frequently provide condescending answers to the wrong question is telling. It is a strong indicator of bias.
Finally, GISS is old hat. The better methods do no ad hoc adjusting. And their answer comes out the same.
Their answer? As if there were only one “answer” that comes out of such a process. Some of the “answers” may be the same, but some of the “answers” change. That is the point of the adjustments – if they have any legitimacy at all, they must be changing something in a significant way. If they aren’t there is no reason to perform them. Again, answering the wrong question.
I’ve used unadjusted GHCN and adjusted GHCN. v2, v3, v3.1 The warming doesnt disappear.
Who said “dissapear”? Modifying the pattern and magnitude of the warming is all that is necessary to tell a tale. Again, answering the wrong question.
It CANNOT. if it disappeared, then UHA would be wrong as well.
UAH? That only goes back 30 years. The topic at hand is adjustment induced warming that conveniently tapers to zero over the UAH period of record. Back when there wasn’t an objective check, the adjustments are much greater in magnitude, and getting bigger with every iteration. That is the topic of this post. Again, answering the wrong question.
Understand. The world is warming. That warming has been going on since the LIA.
That is not the question. The question is whether or not there has been anything other than LIA recovery warming going on over the last 70-80 years. The answer to that question has been tied to the pattern of warming and the relative magnitude of warming between periods. Again, answering the wrong question.
Focusing on GISS is a waste of time. Old method. Smaller dataset. and confusing unnecessary adjustments.
What does it matter if the adjustments are confusing and unnecessary, if (as you allege) they have no effect? If all datasets provide the same “answer”, by what justification do you disparage any of them? That the “new and improved” methods’ unnecessary adjustments are easier to make?
Personal investment: Achieves buy in, returns sell out. Something we must all be vigilant against.
The Canadian band The Guess Who said it best:
F-I-D-D-L-I-N spells fiddlin’
D-R-I-N-K-I-N spells drinkin’
I’ll tell you something you might not know
You can’t go fiddlin’ without your bow
F-I-D-D-L-I-N spells fiddlin’
(from the song “Fiddlin'”, on the LP “So Long Bannatyne”)
@ur momisugly Nick Stokes
There may be other stations with cooling trend adjustments. Has it occurred to you though that they might be correct?
Alternatively, they may be in error too, which does not give anybody much confidence.
@ur momisugly Roman M
The plots in reference 3 above do not seem to exist. All of the subdirectories that I checked were empty.
The “Last Modified” date on all of them reads 12/03/2012 5:14:00 AM.
I have had the same problem since Friday. I assumed GHCN were doing the monthly update, but that usually only takes a day.
The link should give a graph like this.
http://notalotofpeopleknowthat.wordpress.com/2012/01/20/ghcn-say-it-cant-be-that-cold-in-greenland/
Paul,
I noticed them disappear before (a while back) and assumed they were only a temporary feature while NCDC was working on the adjustments, so there is some sort of update/cycle happening. I learned download any I needed (e.g. http://diggingintheclay.wordpress.com/2011/04/17/cherrypicking-in-bolivia/). Glad you spotted the V3.1 thing. It is hard to keep up with the various versions.
Nick Stokes says:
March 12, 2012 at 3:21 am
You’re missing what Steven is saying. Posts like this make people think that someone has replaced the old data with new. But they haven’t. GHCN publishes its unadjusted data set, and that is what it says, unadjusted. The original readings.
And you can use these to construct a temperature index. Steven has done it. So have I, and many others. And we get essentially the same result as GISS, without using any of their adjustments.
=============================================================================
The keyword here Nick, is “essentially”.
Nobody with any credibility can argue that the globe has not warmed somewhat since the start of the GISS record in 1880 : It is the precise extent and, equally crucially, timing, of the warming, which is exercising everybody’s minds.
The raw temperature constructions by both yourself and Zeke (kudos to you both for your work on these) show the same thing – that the GISS adjustments have 2 effects :
The first is to slightly increase the extent of the warming.
The second (more important in my view) is to adjust the timing of the warming, so that more of it occurs later in the record. This is crucial because it has the effect of “improving” the correlation between rising temperatures and rising levels of atmospheric CO2, and improving it at the expense of that inconvenient “natural variability” which the CAGW brigade would so love to sweep under the carpet.
That’s the real issue here……….
A few questions for Paul Homewood…
In “THE U.S. HISTORICAL CLIMATOLOGY NETWORK MONTHLY TEMPERATURE
DATA, VERSION 2” by Matthew J. Menne, Claude N. Williams Jr., and Russell S. Vose, it states:
* Do you think that changing the collection times would create a bias that SHOULD be corrected?
The paper also states:
* Do you think that changing the thermometers would create a bias that SHOULD be corrected? Do you disagree with the Quayle’s conclusions for the direction or magnitude of the shifts?
* Finally, this particular paper specifically covers US stations in the HCN – should similar adjustments be made internationally? Do you know if such changes (thermometers or collections times) were made in any of the stations you list, and if so, when was this done for the stations? Should these changes in conditions be “corrected” or should the data be left in its raw, inconsistent form?
“and to an average rise in minimum temperatures of 0.3°C ”
Or was the bias created by short MMTS cables that brought thermometers closer to heated homes?
@ur momisugly Tim Folkerts
* Do you think that changing the collection times would create a bias that SHOULD be corrected?
* Do you think that changing the thermometers would create a bias that SHOULD be corrected? Do you disagree with the Quayle’s conclusions for the direction or magnitude of the shifts?
Thanks Tim for pointing these out. My understanding from the Iceland Met Office is that their “actual” temperature data is already adjusted for these factors.
Secondly, the GHCN algorithm is not designed to pick up gradual changes such as these, which still raises the question why they have suddenly appeared now.
Reblogged this on Climate Ponderings and commented:
Following the Arctic GISS pea…………
I am not keen on Steve Mosher using the following terminology:
..The algorithms are then tested for their ability to restore the series to the truth..
Firstly, the truth may well be ‘out there’ but without time travel – it will never be known in such a factual or accurate sense. Any algorithms output must logically be an ‘estimate’ per se. (which brings an interesting point on error bars – if we have a series with a +/- 1 deg F error bar, and we use an algorithm to infill or correct a series – is the error range significantly increased for ‘that’ corrected data? just curious!)
Secondly, if we don’t know the truth, how do we know that an algorithim is restoring it! LOL – Yeah, I’m being a bit facetious – but even testing an adjacent stations ‘good’ data with an algorithm won’t necessarily mean that algorithm holds true for the ‘bad’ station data? Again, just sayin….
Paul replies:
My understanding from the Iceland Met Office is that their “actual” temperature data is already adjusted for these factors.
This seems like a key factor here, that gets to the core of your post. I think you ought to be more more certain about a central tenet of your claim.
Secondly, the GHCN algorithm is not designed to pick up gradual changes such as these,
1) These are NOT gradual changes. Up until one day a station has an old thermometer; the next day it has a new thermometer. Up until one day the temperatures are recorded in the afternoon; the next day it is recorded in the morning. These are clear step-functions in the conditions.
2) I am not an expert on the algorithms they use, but they claim in the paper:
and
This makes it seem that they are indeed trying to deal with gradual changes (eg urbanization or plant growth). Do you have support for your claim that the “algorithm is not designed to pick up gradual changes” ?
which still raises the question why they have suddenly appeared now.
So what is the answer? What changes “suddenly appeared” in V3 compared to V2? Are these changes indeed justified or not? Simply noting that the adjustments make it “worse than we thought” says NOTHING about whether the adjustments were legitimate. Some adjustment do indeed make it “worst than we thought”; they do indeed make it seem like people are “cooling the past”. Either show that 1) the adjustment are wrong, or 2) they left out (inadvertently or intentionally) adjustments that would be in the other direction or 3) accept that the adjustments are proper.
Is there a specific adjustment in the new Version 3 that you think is incorrect?
The possibility of a “double adjustment” to the Iceland data is definitely worth a look. On the other hand, if the adjustments were made by the Iceland Met, then there should be no signal left to detect and the algorithm should not have found anything to correct. I would enjoying hearing a more specific discussion on the issue of how (if at all) Iceland Met adjusted the data, and how GHCH adjusted the potentially-preadjusted data.
Tim, I suggest you look at the blog of the icelandic met office at
http://icelandweather.blog.is/blog/icelandweather/
Please also look at the sequence of posts at Paul Homewood’s blog notalotofpeopleknowthat.
Many of your questions have already been answered. The adjustments are not justified. For example, there was an abrupt and definitely genuine cooling in Iceland in 1965, clearly apparent in the raw temperature data and seen in the SST data and well established in the literature. This is largely removed by the adjustments.
I emailed GHCN on Jan 17th asking about these adjustments. I have had no answer except an acknowledgement saying “Please stay tuned for further updates”.
@ur momisugly Tim Folkerts
Paul replies:
My understanding from the Iceland Met Office is that their “actual” temperature data is already adjusted for these factors.
This seems like a key factor here, that gets to the core of your post. I think you ought to be more more certain about a central tenet of your claim.
Trausti Jonsson of the Iceland Met confirms that their data is already adjusted and both datasets can be accessed on his site here.
http://icelandweather.blog.is/blog/icelandweather/
Secondly, the GHCN algorithm is not designed to pick up gradual changes such as these,
1) These are NOT gradual changes.
All the changes you talk about, of course, did not happen on one day, but were gradually introduced. If you are right, though, we should be seeing similar adjustments across the world, not just Arctic regions.
Either show that 1) the adjustment are wrong, On the contrary it is up to GHCN to show they are right, which they have failed to do. Meanwhile the Iceland Met say as far as they concerned
“The adustments for 1940 in Icleand are way out of line. “
Zeke Hausfather (21:44:01) : To me at least the results aapper indistinguishable: That’s because you didn’t do the homogenize and Grid / Box steps as GIStemp does. So first take your Eureka temps and spread them 1000 km in all directions as fill in’ and homogenizing to any stations missing data or that were discontinued after the baseline. THEN take those and spread them another 1200 km into empty grid boxes. I make that about 2200 km RADIUS of influence. That’s how GISS does it. And that’s why the GISS graph has a small box for Eureka (the first image up top with a mostly grey arctic Canada) but then the whole thing turns blood red when you smear the data around ala GIStemp.Look, if you’re going to play climate scientist’ you really must learn all the tricks of the trade. Try reading Hansen’s papers for starters. The Reference Station Method and Optimal Interpolation would be good search terms to start with. For advanced study, read the GIStemp source code. I know where you can read it on line So, go back to your map, and draw a 1200 km radius circle around Eureka. That is the MINIUM area it will directly be used to fabricate the Grid Box anomaly. Now draw a 1000 km radius circle. Any OLD stations in that radius will be homogenized with Eureka (now, we don’t know how many or how much). Then put a 1200 km radius around each of THEM. That’s the ultimate reach of the data. Well, maybe not the ultimate ultimate I did leave out one additional reach’ step After the 1000 km homogenize’ there is an added 1000 km UHI’ adjust. To the extent it’s backwards’ you could get bogus warming from it. Not that that ever happens. Well, not more than 1/2 the time . then after the UHI correction’ it goes to the Grid/ Box step. So in theory you could chain this out to 3600 km radius. But I’m sure that rarely happens. After all, you would need to have nearly no other stations nearby, since the code uses the closest stations first. And I’m sure there must be dozens of stations up there’ What, only one you say? Who knew? So just remember, this is climate science’. You can’t expect to apply simple mathematics to it and find the warming influence. So your toy world’ experiment was doomed from the beginning.