I observed the other day that Eli Rabett had done a bait-and-switch, in which Eli showed a picture of a well-located new station at Niwot Ridge (started only in 2003) as a supposed refutation of criticisms of incinerators, barbecues and tennis courts at USHCN stations. In response, Eli observed his right to “carrot pick”. Rosanne D’Arrigo once told an astonished NAS panel that you have to pick cherries if you want to make cherry pie and I guess the same applies to carrots.
However little Eli Rabbitt seems to have quickly retreated from the USHCN garden and decamped overseas, where his latest carrot pick is the De Bilt weather station in Holland, a site that is incinerator and barbecue free. The De Bilt station has moved a couple of times in the past century, but each move is relatively well-documented; there are photos not just of the current location but of the location in the 1930s and even the 19th century.
When I re-examined the De Bilt data, I was intrigued both by the size of the adjustment for a seemingly innocuous move of a couple hundred meters from a pagoda hut to a Stevenson hut and by the large discrepancy between the adjustments applied in 3 standard versions of the data. If the KNMi adjustment is “right”, then errors in the GISS and GHCN adjustments are up to 0.5 deg C in the 20th century and up to 1 deg C in the 19th century – extremely large errors in a well-studied site.
Anyone thinking about De Bilt needs to visit Hans Erren’s webpage , especially here where he has examined the De Bilt data from a variety of angles and I’m sure that Hans will correct any mis-steps here.
Let me re-state my view on these temperature series again: personally I have no doubt that temperatures are warmer today than in the 19th century. Are they warmer than the 11th century? I don’t know. Is it warmer now than the 1930s? Probably, but the size of the difference seems to depend to an alarming extent on adjustments. That’s not to say that the adjustments are wrong, but, if one is going to rely on specific estimates of the difference, one needs to understand the nitty-gritty of the adjustments that underpin the conclusion and why the adjustments are so different between KNMI, GISS and GHCN.
The “De Bilt” series begins in 1706, but is a splice of measurements taken in 5 different locations: Delft/Rijnsburg (1706-1734), Zwanenburg (1735-1800 & 1811-1848), Haarlem (1801-1810) and Utrecht (1849-1897) reduced to De Bilt and De Bilt (1898-present).
So far I have identified six different versions of the “De Bilt” series: GISS raw, GISS adjusted, GHCN raw, GHCN adjusted, KNMI and Labrijn (the latter emailed to me by Hans, who obtained it from the authors.) GHCN raw and Labrijn are the only versions that go back to 1706; the two GISS versions begin in 1881; the GHCN adjusted version in 1851 and the KNMI version in 1901. Four versions extend up-to-date, but the GHCN adjusted ends in 1991 for some inexplicable reason and the Labrijn version in 2000. The Labrijn and KNMI versions are pretty much identical for their period of overlap (1901-2000) and the GHCN and GISS raw versions are essentially identical for their period of overlap (1881-2006).
The 1950 Station Move
De Bilt metadata is discussed here which includes some interesting pictures (see also pictures at other Dutch sites). In the late 19th century, the station was at Utrecht, not De Bilt, moving in 1898 from Utrecht to the De Bilt “pagoda hut”, shown in the picture (top left below). In 1950, the station was moved again, this time several hundred meters away (see bottom left map from Hans Erren) and re-located in a Stevenson hut (see top right picture – this is the figure from Eli Rabett.) In 1993, an electric sensor replaced the thermometer. The adjustments are shown in bottom right and are discussed below the pictures.
Bottom right: Adjustments from GISS-GHCN raw data. GISS-GHCN version minus 4 versions.
This seemingly innocent move of a few hundred meters at the same elevation from one responsible scientific observation location to another resulted in a measurement discontinuity estimated by KNMI as approximately 1 deg C. Hans Erren has a similar estimate. Both estimates are obtained by comparing the De Bilt series over this period with nearby stations e.g. Uccle which did not move at this time – a strategy similar to that of the USHCN Station Adjustment Program.
To keep this in perspective, this “inhomogeneity” incurred by this move of a few hundred meters is substantially more than the entire amount of global warming experienced to date. So when people are thinking in terms of signal and noise, here’s an example of the type of “noise” that is introduced by a seemingly innocuous station move. So a move from the roof top of the Boulder fire station to the NIST building may well introduce a discontinuity as well. If all your stations have discontinuities, the estimation process would appear to become very formidable very quickly. There is some recent literature on these sort of adjustments, but the entire topic bears examination by third-party statisticians of Wegman calibre.
Now let’s look at the differences that arise from different adjustment schemes. See bottom right above. The KNMI and Labrijn already appear to contain an adjustment for the move from the pagoda hut. I presume that the online KNMI version derives in some way from the Labrijn version as these are essentially the same. Hans Erren has his own proposed adjustments shown here , which is in the same “family” as the other Dutch adjustments. A caveat here : my calculation of the GISS adjustment differs somewhat from Hans’ (we’ll probably try to reconcile calculations at some point) although the shapes are similar; and Hans’ figure does not include the GHCN adjustment version.
As noted above, the station moved from Zwanenburg to Utrecht in 1849 and from Utrecht to the De Bilt pagoda hut in 1898. In the 19th century when the station was located in Utrecht, the GHCN adjustment is very different than the KNMI adjustment, with the GHCN adjustment up to 1 deg C larger. It’s hard to understand why the GHCN adjustment should be so much larger. Also the GHCN adjustment becomes effective in little steps, while KNMI has a 1 deg C adjustment in 1950. GHCN and GISS both implement changes in a series of small steps, while KNMI has one large step in 1950. It’s curious that the move from the pagoda hut to the Stevenson hut occasioned a big adjustment, but the move from Utrecht to De Bilt didn’t trigger any change in the KNMI-Labrijn version. Also that the Huis Zwanenburg site, pictures by Hans Erren here had no material adjustment relative to the Stevenson hut.
So even in a relatively well-controlled series we’re talking substantial adjustments. If one views the adjustment as a statistical estimate, then how does one estimate the error distribution? In a trivial sense, the GHCN, GISS and KNMI adjustments cannot all be correct and, if either of the GHCN or KNMI adjustments is “correct”, then the error in the other one is about 1 deg C, an amount that seems very high in a well-measured instrumental site for a period that isn’t long ago. Even in the 20th century, adjustment errors of about 0.5 deg C are necessarily occurring in at least one of the series.
I haven’t waded through GHCN and GISS adjustment methodologies to see exactly what they are doing and am nowhere near to replicating their methodologies. (Wouldn’t it be nice if they just posted their codes so one could see exactly what they were doing, instead of having to decode obscure and poorly written methodological descriptions?) But let’s say that the KNMI adjustments are “right” and that the GISS and GHCN adjustments are “wrong” in this spot check. Then that would appear to indicate that the GHCN and GISS methodologies were wrong in some way and there would be no reason to assume that they were randomly wrong.
Corrections have been made for:
- relocation combined with a transition of large open hut to a wooden Stevenson screen (September 1950).
- relocation of the Stevenson screen (August 1951).
- lowering of Stevenson screen from 2.2 m to 1.5 m (June 1961).
- transition of artificial ventilated Stevenson screen to the current KNMI round-plated screen (June 1993).
- warming trend of 0.11°C per century caused by urban warming.
Please note the following:
- the transition of the wooden Stevenson to a PVC Stevenson screen (August 1980) did not have a noticeable effect on the daily mean temperatures; consequently no corrections had to be applied.
- from 1951 onwards, KNMI shifted from so-called climatological temperature measurements to synoptical measurements. However, the climatological measurements continued to 1970. For the 1951-1970 period these measurements may be considered as more reliable than the synoptical measurements. Therefore, in the homogenized series the climatological measurements extend to 1970.
the present version of the series may change as soon as new results come available from the efforts of KNMI to homogenize the temperature series of De Bilt on a daily base, taking into account weather-dependent corrections.
There is nothing in this that is inconsistent with the above commentary. The 1950 relocation from the “large open hut” is the relocation from the “pagoda hut” described above, which yielded an adjustment of 1 deg C. I’m not objecting or arguing about this adjustment, merely observing that the changes as documented are not large and still yield an effect larger than observed global warming in the past century – something that has to be considered in signal-noise statistics when one is dealing with estimation of undocumented station changes.
Second, Eli has not commented on the discrepancy between GISS and GHCN adjustments and the KNMI results. So while the additional reference is interesting and further documents the points that I made above, it does not contradict them. Eli has simply thrown up a diversion as though it were an academic comment.