Monday, 21 March 2016

Cooling moves of urban stations

It has been studied over and over again, in very many ways: in global temperature datasets urban stations have about the same temperature trend as surrounding rural stations.

There is also massive evidence that urban areas are typically warmer than their surroundings. For large urban areas the Urban Heat Island (UHI) effect can increase the temperature by several degrees Celsius.

A constant higher temperature due to the UHI does not influence temperature changes. However, when cities grow around a weather station, this produces an artificial warming trend.

Why don’t we see this in the urban stations of the global temperature collections? There are several reasons; the one I want to focus on in this post is that stations do not stay at the same place.

Urban stations are often relocated to better locations, more outside of town. It is common for urban stations to be moved to airports, especially when meteorological offices are moved to the airport to assist in airport safety. Also when meteorological offices can no longer pay the rent in the city center, they are forced to move out and take the station with them. When urban development makes the surrounding unsuited or when a volunteer observer retires, the station has to move, it makes sense to then search for a better location, which will likely be in a less urban area.

Relocations are nearly always the most frequent reason for inhomogeneities. For example, Manola Brunet and colleagues (2006) write about Spain:
“Changes in location and setting are the main cause of inhomogeneities (about 56% of stations). Station relocations have been common during the longest Spanish temperature records. Stations were moved from one place to another within the same city/town (i.e. from the city centre to outskirts in the distant past and, more recently, from outskirts to airfields and airports far away from urban influence) and from one setting (roofs) to another (courtyards).”
Since relocations of that kind are likely to result in a cooling, the Parallel Observations Science Team (ISTI-POST) wants to have a look at how large this effect is. As far as we know there is no overview study yet, but papers on the homogenization of a station network often report on adjustments made for specific inhomogeneities.

We, that is mainly Jenny Linden of Mainz University, had a look in the scientific literature. Let’s start in China were urbanization is strong and can be clearly seen in the raw data of many stations. They also have strong cooling relocations. The graph below from Wenhui Xu and colleagues (2013) shows the distribution of breaks that were detected (and corrected) with statistical homogenization for which the station history indicated that they were caused by relocations. Both the minimum and the maximum temperature cool by a few tenth of a degree Celsius due to the relocations.

The distribution of the breaks that were due to relocations for the maximum temperature (left) and minimum temperature (right). The red line is a Gaussian distribution for comparison.

Going more in detail, Zhongwei Yan (2010) and colleagues studied two relocations in Beijing. They found that the relocations cooled the observations by −0.81°C and −0.69°C. Yuan-Jian Yang and colleagues (2013) find a cooling relocation of 0.7°C in the data of Hefei. Clearly for single urban stations, relocations can have a large influence.

Fatemeh Rahimzadeh and Mojtaba Nassaji Zavareh (2014) homogenized the Iranian temperature observations and observed that relocations were frequent:
“The main non-climatic reasons for non-homogeneity of temperature series measured in Iran are relocation and changes in the measuring site, especially a move from town to higher elevations, due to urbanization and expansion of the city, construction of buildings beside the stations, and changes in vegetation.”
They show an example with 5 stations where one station (Khoramabad) has a relocation in 1980 and another station (Shahrekord) has two relocation in 1980 and 2002. These relocations have a strong cooling effect of 1 to 3 degrees Celsius.

Temperature in 5 stations in Iran, including their adjusted series.

The relocations do not always have a strong effect. Margarita Syrakova and Milena Stefanova (2009) do not find any influence of the inhomogeneities on the annual mean temperature averaged over Bulgaria. This while “Most of the inhomogeneities were caused by station relocations… As there were no changes of the type of thermometers, shelters and the calculation of the daily mean temperatures, the main reasons of inhomogeneities could be station relocations, changes of the environment or changes of the station type (class).

In Finland, Norway, Sweden and the UK the relocations produced a cooling bias of -0.11°C and relocations appear to be the most common cause of inhomogeneities (Tuomenvirta, 2001). The table below summarises the breaks that were found and what the reasons for them were if this was known from the station histories. They write:
“[Station histories suggest] that during the 1930s, 1940s and 1950s, there has been a tendency to move stations from closed areas in growing towns to more open sites, for example, to airports. This can be seen as a counter-action to increasing urbanization.”

Table with the average bias of inhomogeneities found in Finland, Sweden, Norway and the UK in winter (DJF), spring (MAM), summer (JJA) and autumn (SON) and in the yearly average. Changes in the surrounding, such as urbanization or micro-siting changes, made the temperatures higher. This was counteracted by more frequent cooling biases from changes in the thermometers and the screens used to protect the thermometers, by relocations and by changes in the formula used to compute the daily mean temperature.

Concluding, relocations are a frequent type of inhomogeneity. They produce a cooling bias. For urban stations the cooling can be very large. For the average over a region, the values are smaller, but especially because they are so common, they will have most likely a clear influence on global warming in raw temperature observations.

Future research

One problem with studying relocations is that they are frequently accompanied by other changes. Thus you can study them in two ways: study only relocations where you know that no other changes were made or study all historical relocations whether there was another change or not.

The first set-up allows us to characterize the relocations directly, to understand the physical consequences to move for example a station from the center of a city / village to the airport. In this way the differences are not subject to other changes specific to a network. So, the results can be easily compared between regions. The problem is that only a part of the parallel measurements available satisfy these strict conditions.

Conversely, for the second design (taking all historical relocations, also when they have another change) the characterization of the bias will be limited to the datasets studied and we will need a large sample to say something about the global climate record. But on the other hand, we can also analyze more data this way.

There are also two possible sources of information. The above studies relied on statistical homogenization comparing a candidate station to its neighbors. All you need to know for this is which inhomogeneities belong to a relocation. A more direct way to study these relocations is by using parallel measurements at both locations. This is especially helpful to study changes in the variability around the mean and in weather extremes. That is where the Parallel Observation Science Team (ISTI-POST) comes into play.

It is also possible to study specific relocations. The relocation of stations to airports was an important transition, especially around the 1940s. This temperature change is likely large and this transition quite frequent and well documented. One could focus on urban stations or on village stations, rather than studying all stations.

One could make a classification of the micro and macro siting before and after the relocation. For micro-siting the Michel Leroy (2010) classification could be interesting; as far as I know this classification has not been validated yet, we do not know how large the biases of the 5 categories are and how well-defined these biases are. Ian Stewart and Tim Oke (2012) have made a beautiful classification of the local climate zones of (urban) areas, which can also be used to classify the surrounding of stations.

Example of various combinations of building and land use of the local climate zones of Stewart and Oke.

There are many options and what we choose will also depend on what kind of data we can get. Currently our preference is to study parallel data with identical instrumentation at two locations, to understand the influence of the relocation itself as well as possible. In addition to study the influence on the mean, we are gathering data on the break sizes found by statistical homogenization for breaks due to relocations. The station histories (metadata) are crucial for this in order to clearly assign breakpoints to relocation activities. It will also be interesting to compare those two information sources where possible. This may become one study or two depending on how involved the analysis will become.

This POST study is coordinated by Alba Guilabert and Jenny Linden and Manuel Dienst are very active. Please contact one of us if you would like to be involved in a global study like this and tell us what kind of data you would have. Also if anyone knows of more studies reporting the size of inhomogeneities due to relocations, please let us know. I certainly have seen more such tables at conferences, but they may not have been published.

Related reading

Parallel Observations Science Team (POST) of the International Surface Temperature Initiative (ISTI).

The transition to automatic weather stations. We’d better study it now.

Changes in screen design leading to temperature trend biases.

Early global warming.

Why raw temperatures show too little global warming.


Brunet M., O. Saladie, P. Jones, J. Sigró, E. Aguilar, et al., 2006: The development of a new daily adjusted temperature dataset for Spain (SDATS) (1850–2003). International Journal of Climatology, 26, pp. 1777–1802, doi: 10.1002/joc.1338.
See also: a case-study/guidance on the development of long-term daily adjusted temperature datasets.

Leroy, M., 2010: Siting classifications for surface observing stations on land. In WMO Guide to Meteorological Instruments and Methods of Observation. "CIMO Guide", WMO-No. 8, Part I, Chapter 1, Annex 1B.

Rahimzadeh, F. and M.N. Zavareh, 2014: Effects of adjustment for non‐climatic discontinuities on determination of temperature trends and variability over Iran. International Journal of Climatology, 34, pp. 2079-2096, doi: 10.1002/joc.3823.

Stewart, I.D. and T.R. Oke, 2012: Local climate zones for urban temperature studies. Bulletin American Meteorological Society, 93, pp. 1879–1900, doi: 10.1175/BAMS-D-11-00019.1.
See also the World Urban Database.

Tuomenvirta, H., 2001: Homogeneity adjustments of temperature and precipitation series - Finnish and Nordic data. International Journal of Climatology, 21, pp. 495-506, doi: 10.1002/joc.616.

Xu, W., Q. Li, X.L. Wang, S. Yang, L. Cao, and Y. Feng, 2013: Homogenization of Chinese daily surface air temperatures and analysis of trends in the extreme temperature indices. Journal Geophysical Research Atmospheres, 118, doi: 10.1002/jgrd.50791.

Syrakova M. and Stefanova M., 2009: Homogenization of Bulgarian temperature series. International. Journal Climatology, 29, pp. 1835-1849, doi: 10.1002/jov.1829.

Yan ZW; Li Z; Xia JJ. 2014. Homogenisation of climate series: The basis for assessing climate changes. Science China: Earth Sciences, 57, pp 2891-2900, doi: 10.1007/s11430-014-4945-x.

* Photo at the top "High Above Sydney" by Taro Taylor used with a Attribution-NonCommercial 2.0 Generic (CC BY-NC 2.0) license.


  1. Victor

    You ask; 'if anyone knows of more studies reporting the size of inhomogeneities due to relocations'

    I assume you are already well aware of the EU 'Improve' project carried out by Jones and Camuffo a few years ago

    I think it covers the sort of things you may be interested in. The data consists of a thick book and a cd that contains the original and corrected data. It concerns Historic European temperature sets so may be before the time scales you may be interested in.


  2. Hi Tony, thanks for the tip. I was aware of IMPROVE, but it was before I started working on this topic and I did not know that they also worked relocations. Makes sense given that many of the long series have started in cities.

  3. Victor

    The past masters at station relocation are of course CET. David Parker has sent me a number of articles he has written on station relocation as stations are changed as they no longer represent the type of terrain they were originally selected for.

    The history of Oxford is especially interesting but Ringwood is also notable, as over the years it became too warm due to the expansion of Manchester airport and they had to change the location. There must have been at least a dozen changes with CET stations over the last 30 years or so, all well documented.

    I know you have met David so it may be you ate already pursuing this with him.

    due to britains small size and likely uhi, the regional stations used for global and national calculations also frequently change and are well documented so again this may be of some interest to you.


  4. Noted this on twitter, but many center cities are less dense than they were in the early 1900s (Paris is a good example).

    Another interesting case is the Central Park Station in NYC, where the park trees grew in over the last century

    Eli would not be surprised if this is the case in many other cities.

  5. Tony, thanks. Talked to David a lot about the transition to Stevenson screens for which he wrote the main article up to now. Will also ask him about relocations.

    Eli, interesting idea. Had not expected that. I was thinking of Eastern Germany for a similar example, which was depopulated after the collapse of the Berlin wall.

  6. Victor

    Eli makes an interesting reference to New York. It was one of three station in the US where I looked at the station siting

    “Central Park appears to be greatly influenced by a number of factors. The reservoir in front of Belvedere-built 1842- was emptied in 1917 after the park became semi derelict. It then became ‘Hooverville’ –a shanty town-during the Great Depression.

    Then in the 1940’s the area was grassed over and became The ‘Great Lawn,’ but then became ‘packed as hard as ashpalt’ before being renovated in the 1970’s. So lots of changes to the weather station environs. The current location is shown in the photo below. (reference is made to the shading)

    population soared

    1820 123,706

    1850 515,000

    1880 1.5 million people (note inclusion of Bronx)

    1990 7.5 milion people

    My estimation would be that virtually every station is affected by some significant factor or other over its life, from physical moves, to Uhi, to changes in instrumentation or observer, to shading to depopulation to altitude.

    That is why I follow Hubert Lambs maxim that as regards historic temperatures ‘we can understand the tendency but not the precision’ that is to say that the general trend is probably somewhere about right but we have no idea of the real temperatures to within tenths of a degree of accuracy.


  7. I would also expect that nearly any longer station has been affected by inhomogeneities. The world was so different a century ago. The US climate reference network makes a brave attempt to avoid this going forward. I expect that they will be able to get less inhomogeneities, but also the reference network will have some breaks in a century. Homogenization will always be important. Rather than wishing it away, I wish people would study how to make it better.

    The uncertainty due to inhomogeneities is certainly larger than a tenth of a degree Celsius. That is also why I have never understood the people who confidently claimed that we had a hiatus in the last years. Even if it would have been a statistically robust feature, I would never have trusted the data so much for such a minute deviation.


Comments are welcome, but comments without arguments may be deleted. Please try to remain on topic. (See also moderation page.)

I read every comment before publishing it. Spam comments are useless.

This comment box can be stretched for more space.