“This a complicating factor for the homogenization of daily temperature series.”
WUWT first reported on the issues with the KNMI De Bilt weather station in October 2009 here. About that time, Dr. Pielke Sr. and I were given access to a KNMI preliminary report on the siting and subsequent bias problems at De Bilt, but we decided to wait until the final report was available before saying anything about it. Those waiting on GMU/Wegman take note – universities move like molasses, chill. Why is this station important? It just so happens that KNMI De Bilt is the only station in the Netherlands used for NASA GISTEMP, and now it has been shown to have problems related to siting.
The issue became front page news in the Netherlands, partly because the KNMI climate data was challenged by a private meteorological firm because it always read too warm:
Weather specialists from the Wageningen-based Meteo Consult have been expressing their distrust for years, because the KNMI figures in De Bilt were always a bit warmer than in Cabau, 16 km away, where there is also a KNMI thermometer. The position of both places could, according to Meteo Consult, not explain the temperature difference of on average half a degree (Celsius). It was also not taken into account that De Bilt is located in a more built-up, and probably therefore warmer, surroundings than Cabau, near IJsselstein.
Above: GISS Temperature plot for De Bilt KNMI – notice the step function. Click for source data.
To KNMI’s credit, they have conducted an exhaustive parallel study to determine the magnitude of effects from siting changes. The results and their conclusion show clearly that siting does matter.
Dr. Roger Pielke Sr. writes:
There is an important, much-needed, addition to the scientific literature which adds to our conclusions in
Fall, S., A. Watts, J. Nielsen-Gammon, E. Jones, D. Niyogi, J. Christy, and R.A. Pielke Sr., 2011: Analysis of the impacts of station exposure on the U.S. Historical Climatology Network temperatures and temperature trends. J. Geophys. Res., in press. Copyright (2011) American Geophysical Union.
that siting of climate reference stations do matter in terms of long term temperature trends and anomalies. This new report is
(16 MB PDF from KNMI – alternate, faster download here KNMI_DeBilt_WR2011-01)
The summary reads (emphasis added)
Air temperature measurements at the KNMI-observatory in De Bilt are important mainly because the observatory has a long and relatively homogeneous record and because its observations often serve as an indicator of changes in climate for the Netherlands as a whole. Among others, relocations of the temperature measurement sites and (gradual) changes in surroundings influence the measurements. To improve the homogeneity of the long-term temperature record and to study the representativeness of the current measurements, a parallel experiment was carried out at the observatory of KNMI in De Bilt from May 2003 through June 2005.
Five sites at the KNMI-observatory, including the (at that time) operational site WMO 06 260 (further denoted as DB260), were equipped with identical (operational) instruments for measuring temperature and wind speed at a height of 1.5 m (see for an overview of the sites Figure 1.1). The instruments were calibrated each half-year and the calibrations curves were used to correct the data to minimize instrumental errors. With the measurements at the Test4 site (operational site since 25 September 2008) as a reference, the temperature differences between the sites were studied in connection with the local wind speed and its differences and operationally measured weather variables at the KNMI-observatory. In September/October 2004 the area west of the operational site DB260 was renovated and made into a landscaped park. From 1999 onwards that area slowly transformed from grassland into a neglected area with bushes (wasteland). The parallel measurements provided the opportunity to study the impact of this new inhomogeneity in detail.
The results show that changes in surroundings complicate or impede the use of present-day parallel measurements for correcting for site changes in the past. For instance, the (vertical) growth of the bushes in the wasteland area west of DB260, caused increasing temperature differences between the operational site DB260 and four neighboring stations. The effects were most clearly visible in the dry summer of 2003, when the mean monthly maximum temperatures at DB260 were up to about 0.4C larger than those at the reference Test4. This increase was more than counteracted by a decrease in the mean monthly minimum temperature of up to 0.6C. After the renovation of the wasteland area, the temperature differences between DB260 and Test4 became close to zero (< 0.1C). The comparison of DB260 with four neighboring stations showed that the renovation restored to some extent the temperatures of the old situation of before the year 1999. However, the land use west of the DB260 has been changed permanently (no longer grassland as in the period 1951-1999, but landscaped park land with ponds). Therefore, operational measurements at DB260 became problematic and KNMI decided to move the operational site to the Test4 site in September 2008. The Test4 site is the most open of five sites studied in the report.
The results increase our understanding of inter-site temperature differences. One of the most important causes of these differences is the difference in sheltering between sites. Sheltering stimulates the build up of a night-time stable boundary layer, decreases the outgoing long-wave radiation, causes a screen to be in the shade in the hours just after sunrise and before sunset, and increases the radiation error of screens due to decreased natural ventilation. Depending on the degree and nature of the sheltering, the net effect of sheltering on temperatures may be a temperature increase or decrease. DB260 is a sheltered site where the net effect is a decrease of the mean temperature (before the renovation). The former historical site Test1 is an example of a site where the net effect is a temperature increase. The monthly mean minimum temperature at Test1 is up to 1.2C higher than the reference and the maximum temperature is up to 0.5C higher than that at Test4. The mean temperature at Test1 is, however, only slightly higher than the mean at Test4. This is caused by the relatively low temperatures in the hours after sunrise and before sunset, when the screen at Test1 is in the shade. Both the Test1 and Test4 location are probably not affected by the renovation.
The renovation of the wasteland area causes not only a shift of the location of the pdf of the daily temperature differences but also a change in the shape. This means that for the homogenization of daily temperature series it is not sufficient to correct only the mean.
We showed that the magnitude of the inter-site temperature differences strongly depends on wind speed and cloudiness. In general the temperature differences increase with decreasing wind speed and decreasing cloudiness. Site changes directly affect wind speed because they are usually accompanied by changes in sheltering. Some effects, like the built up and (partly) breaking down of the stable boundary layer near the surface, are highly non-linear processes and therefore difficult to model. The fact that these processes are mostly active at low wind speeds (< 1.0 m/s at 1.5 m) further complicates the modeling. Regular cup anemometers are not really suited to measure low wind speeds. Operationally these anemometers have a threshold wind speed of about 0.5 m/s and this threshold wind speed often increases with the time during which the anemometer is in the field. In addition, anemometers are mostly situated at a height of 10 m. During night-time stable conditions the correlation between wind speed at 10 m and wind speed at screen height is weak. This complicates the homogenization of daily temperature series.
This is a timely posting as just yesterday I uploaded a page to my site detailing the history of the BoM’s official recording location since 1897 for the Western Australia capital city of Perth.
http://www.waclimate.net/perth-temperature-history.html
The top of the page deals with claims by the Bureau of Meteorology that Perth has just endured record maximum summer temperatures. It demonstrates a .1C difference between the BoM’s method of averaging rounded monthly temperatures and a more sensible method of averaging the actual temperatures recorded on each day.
The rest of the page shows what happens when Perth’s temperatures were recorded at one location (61 metre elevation on the outskirts of the city) from 1897 to 1967, to another location (19 metre elevation two kilometres away in the middle of the city) from 1968 to 1992. Both locations contributed to the records of the one station, Perth Regional Office (9034), which is the benchmark site against which all modern temperature readings are compared for this city.
Perth Regional Office saw a sudden mean temperature increase around .8C as of 1967.
Location 9034 was closed in 1992 when a new official Perth temperature site (Perth Metro – 9225) was set up four kilometres to the north (elevation 25 metres). Average maxima immediately increased in “Perth” by about .4C – thus the claims of record maximum temperatures by the BoM. However, average minima immediately fell by about 1.6C.
If you have a look at Perth’s historic temperature records and locations, you’ll be left wondering why the BoM wants to make any claims about comparative historic records in this capital city beyond those at its current (stationary) location established in 1992.
“We showed that the magnitude of the inter-site temperature differences strongly depends on wind speed and cloudiness. In general the temperature differences increase with decreasing wind speed and decreasing cloudiness. Site changes directly affect wind speed because they are usually accompanied by changes in sheltering.”
I can imagine, due to the particular arrangement of shelter, that the direction of wind may also be important. Wind from one direction having a wholly different effect than that from another.
Sensor readings are suspect when sensors are placed in the boundary layer. Why should this engineering principle not apply to climate science?
Geoff Sherrington says:
May 18, 2011 at 6:24 pm
The “acceptable solution” is to work exclusively with long, intact, raw records from vetted small-town stations. Unfortunately this narrows the geographic coverage drastically and is vulnerable to charges of “cherry-picking” from bureaucratic minds intent on using ALL the data, no matter how corrupt. Working this way obtains wider uncertainty bands, but it circumvents the most egregious systematic bias.
Kev-in-Uk says:
May 18, 2011 at 4:18 pm
In the USA , Matthew Fontaine Maury’s 19-th century visionary plan of establishing a network of small-town stations for reliable monitoring of temperatures produced a national treasure that is being destroyed version by version in USHCN. . Analysts might permitted to do what they might to data compilations, but archivists have no right to manufacture/invent data by altering what was recorded by instruments. That should be made a federal crime.
Tonight NASA is hard at work:
http://oi55.tinypic.com/or822a.jpg
Creating serious ideas:
http://oi51.tinypic.com/20ho86e.jpg
[maybe ask Josh to join in? ~ac]
The “De Bilt” site is next to/ part of the Dutch Royal Meteorological Institute KNMI of which Buys Ballot was the first governor and which has the longest record in the Netherlands. It has great historical and cultural value. Temperature readings are not biased they just are what they are, it is the conclusions drawn that are biased. Of course siting issues play a major role in an over 100 year record: in this time span the nearby Ijsselmeer (lake) resulted from damming off the Zuiderzee (sea), and one third of the area of the former Zuiderzee was turned into new land (Flevoland-polder). In this centre of one of the most densely populated countries in the world large surfaces of roads have been covered with tarmac and the nearby city of Utrecht grew several times ist original size in the timespan of the readings. It is not the temparature or temperature readings that is inaccurate or biased or wrong , it is the naïvety that the local data series is a usefull data set for “global” warming analysis that is biased and naïve. It is refreshing that despite the Dutch gouvernment’s CO2 doctrine and related restricted freedom of speech for the institute’s staff, such a report has been published.
De Bilt goes back much further than NASA cuts it off at. It’s one of only two very long running thermometer records that actually forms a little local hockey stick:
http://oi47.tinypic.com/2zgt4ly.jpg
Luckily, nobody’s ever heard of De Bilt.
The rest of the old records don’t show a recent spike:
http://oi52.tinypic.com/2agnous.jpg
“It is good to be learned in the things that are hidden from the wise and the intellectual ones of the world but are revealed, as if by nature, to the poor and simple, to women and little children.” – Vincent van Gogh (letter to Theo van Gogh, 1878)
This particular theme is probably repeated all over the world. Takes me back to the incident that got my suspicions aroused about the CAGW scam when a Greenpeace activist and avid supporter of one G Monbiot attacked me (in the Guardian of London’s CIF, which I rarely read now) for asking if standards existed for siting and reading instruments.
It’s quite timely that Chris Gillham posted yesterday about similar problems with the Perth (Aus) site(s).
That faint roaring noise we can hear in the background is the climate shibboleths falling, one after the other and in rapid succession.
JFD says: May 18, 2011 at 5:55 pm
I did a study of the De Bilt data set a couple of months ago. Using a linear curve fit to De Bilt (or any data set with breaks or steep step functions) is an incorrect procedure. The data can be broken down into segments and analysed with a linear fit or some other curve fitting method but a linear fit across steep step functions is not proper.
I disagree. A linear regression calculation is not a ‘curve fit’ like a polynomial or a running mean would be – it is simply a statistical method of determining the long term average trend of data over a given period. While I admire your detailed analysis and your general conclusions, I question both the usefulness of measuring a trend over a short time interval and also whether it is possible to draw meaningful conclusions on causes, a point you make yourself when you draw up a list only to express some doubt that it is possible to decide between them (but my bet is also on the AMO).
Eyeballing the temperature chart in the original article, there does indeed appear to have been a sudden anomalous drop in temperature that lasted (with the usual annual ups-and-downs) from the mid 1950s to the late 1980s. This then reversed around 1990, and carried on very much as it had before the mid-1950s. But, instead of seeing that as simply an extended low temperature period that subsequently corrected itself around 1990, the eye is instead easily tricked into seeing an ‘alarming’ climb all the way from the mid-1950s up to the present day.
To illustrate this psychological effect, I downloaded exactly the same data as you did and generated the following chart, containing three different linear trends for three different time spans as follows:
http://www.thetruthaboutclimatechange.org/KNMIDeBilt.html
1. The blue linear trend line over the full 130 year period shows a statistically insignificant fall of 0.06degC per century.
2. The grey linear trend line over the first 75 years period to 1955 shows a statistically insignificant rise of 0.1degC per century.
3. The red linear trend line for the 55 years from 1956 to 2010 shows a very alarming rise of 3.5degC per century.
The shorter red linear trend is clearly the anomaly, an artifact of the chosen method of analysis.
This demonstrates in a dramatic way how easy it is to get fooled into false alarmism by selecting a short term trend while ignoring the full long term linear trend of the data series. And, more importantly, an exactly similar thing has happened to the world temperature series, as is shown in the following chart:
http://www.thetruthaboutclimatechange.org/temps.html
The official world temperature data shows a decidedly un-alarming long term average trend of 0.41degC per century. But superimposed on this is a roughly sinusoidal cycle of around 67 years duration with an amplitude of plus and minus 0.25degC. Climate alarmism gathered pace during the last 35 years of the 20th century precisely because from 1964 to 1998 the world was on the upward swing of this cycle. The temperature rise over that period was 0.83degC. That’s equivalent to an alarming rise of about 2.4degC per century, quite close to the 3degC rise that some CAGW enthusiasts had been predicting from CO2 theory by 2100, but quite atypical when compared with the 160 year temperature trend of only 0.41degC.
During the 1980s and 1990s, climate alarmists began increasingly to believe that this late 20th century high rate of temperature rise was the clear ‘fingerprint’ of global warming they had been expecting. They assumed that it was set to continue at that rate until 2100. Disappointingly for them, the temperature rise tailed off subsequently and it now looks like it will be heading back towards maintaining a long term un-alarming average of around 0.41degC.
If this does happen, as most skeptics expect, the next decade will doubtless see the end of the alarmist bandwagon. If not, the skeptics will be the ones having to eat humble pie. It’s as simple as that.
Alexander Vissers says:
May 19, 2011 at 4:25 am
“Temperature readings are not biased they just are what they are, it is the conclusions drawn that are biased. ”
You’re entirely correct. The assumption that a particular station record is representative of temperature variations over a much larger area than just the instrument site proves unwarranted in many cases. The systematic bias that I refer to is relative to temperatures in that larger area.
KNMI immediately corrected the temperature data of De Bilt as soon as Meteo Consult raised the alarm over the De Bilt data. Although the whole matter put KNMI red with shame, they generally deliver reliable data, particularly compared with the temperature data from GISS showed in this article. The suggestion as if KNMI played false to increase De Bilt temperature can be ignored: the temperature range of the mean monthly temperature in De Bilt did not show any increase during the last 14 years, lineair trend was zero. See:
http://www.klimaatgek.nl/cms/?De_feiten:Temperatuur_De_Bilt
Chris Gillham, I liked your article about the Perth temperature records. It must be very difficult to manage the transition from one weather station to another. I’m struck by a few things in Perth’s recent weather history. Firstly, we don’t seem to get the same really hot days we used to. Days over 42C are few and far between. Secondly, we do get a lot of very warm nights in summer. The winters should be colder, since it gets freezing when it doesn’t rain, and it doesn’t rain much these days. (Freezing in Perth is any time the minimum temperature drops below 5C (~40F) – we don’t really have cold weather).
I guess the important thing is to get as broad a range of weather stations as possible, so that errors here cancel out errors there, giving a reliable average.