How is the Urban Heat Island Effect accounted for in Temp records?

[DnD Addict]

How is the Urban Heat Island Effect accounted for in Temp records?

I'm asking this, cause I recently was looking at some weather stations records at the NCDC and noticed that many cities do not have more then one recording station and got to wondering how the effect could be removed without parallel measurements.

Are they somehow able to guess at what the temperature distortion will be given for a city of A type and B size? If so, then how accurate are there estimations suppose to be?

 

[mgb_phys]

The temperature record for a city isn't measured in the downtown core - normally it's at the airport because they have had an interest in and need for meterology for longest.

 

[Xnn]

They compare temperature readings between pairs of stations within a particular region. The difference series is compared against a number (up to 40) of other highly correlated stations pairs within the region. If atmospheric conditions are such that the urban heat island effect is significant, then the offending data is rejected and not used.

So, when the urban heat island effect is detected, the data is rejected.

 

[Andre]

Maybe demonstrate where we can check that.

It seems that this could be a rather hot issue in the climate gate saga.

An interesting case is http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=210476620003&data_set=1&num_neighbors=1 for instance. Firstly would the appearance in the GIStemp list mean that the this station is used to generate a global temperature graph?

Anyway, compare that rise with it's closest rural neighbours, http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=210476400010&data_set=1&num_neighbors=1, http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=210476680010&data_set=1&num_neighbors=1, http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=210476750000&data_set=1&num_neighbors=1, http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=210476900010&data_set=1&num_neighbors=1, etc and gone is the global warming in local Tokyo. All urban heat island effect.

As soon as this is all over, a lot of scientists with the current status of crackpot are going to have a lot to say about this.

 

[sylas]

So, when the urban heat island effect is detected, the data is rejected.

Actually, that isn't true. The Urban Heat island effect is managed by estimating the magnitude of the effect and producing a modified record for the station. Data is rejected only if there are indications of errors in the data that cannot be identified and compensated.

There's a fair bit of literature on this, and a number of different methods that have been applied. The urban heat island effect has a comparatively small impact; but it is sufficient to require this kind of additional processing; and it is not limited to stations in downtown areas.

I'm proposing to add a post on this, with a selection of references, but I'm posting this quickly now just because Xnn is very much on my side of this whole issue, and I generally prefer to correct mistakes from my own side. Sorry, Xnn!

Cheers -- sylas

 

[Xnn]

Sylas;

If you have more information on this, that'd be good.
I got my information from the NCDC referring to Version 2
of the Historical climate serialized temperature data set.

Urbanization Effects
In the original HCN, the regression-based approach of Karl et al. (1988) was employed to account for urban heat islands. In contrast, no specific urban correction is applied in HCN version 2 because the change-point detection algorithm effectively accounts for any "local" trend at any individual station.

http://www.ncdc.noaa.gov/oa/climate/research/ushcn/

 

[sylas]

Sylas;

If you have more information on this, that'd be good.
I got my information from the NCDC referring to Version 2
of the Historical climate serialized temperature data set.
http://www.ncdc.noaa.gov/oa/climate/research/ushcn/

That is saying the same thing I am saying. There is an urban island effect, and it is handled by corrections applied to the data. A description of the steps in processing for version 2 can be found here: Version 2 Processing Steps (at NCDC).

In summary, the steps are:

1. Initial database construction: monthly data is obtained by averaging daily data, for each station. Months with insufficient days are omitted. A quality check removes a small number of extreme outliers that are likely subject to an unknown error. End result is monthly averages from raw daily data, with no corrections, but a small number of months omitted.
2. Based on metadata for each station, a time of observation bias is applied. This corrects for small changes that arise if there are changes to the time of day at which readings were obtained. This mostly applies to historical data, when observation meant reading a thermometer to a defined schedule.
3. A homogeneity process is applied, which is designed to pick up historical changes at a station; including changes that do and do not appear in the meta-data. (Changes can be things like relocation of the station, introduction of new instruments, maintenance of the housing, and so on; all of which may introduce a small step shift in the record.)
4. Missing values are inferred based on data from surrounding stations.

The documentation explains how urbanization effects are handled as a natural outcome of the homogeneity process of step 3.

Version 1 of the USHCN is what I am personally most familiar with, since I did a number of experiments on it myself, and have written various programs to process and use that data. To assist users like me, the version 1 data is still available. A description of version 1 processing is available here: United States Historical Climatology Network (USHCN) Version 1 (at NCDC).

There are some differences in the processing steps of version 1. It uses the same basic steps, but the algorithms are a bit different. In this case, an additional final step was applied that explicitly corrects for urbanization; in version 2 this is now addressed in step 3.

The version 1 data description includes a graph in which the effect of the urban adjustment step is made clearly visible; it is the purple line at the bottom of the graph. This figure is linked from the NCDC page for version 1.

The effect is to alter the trend over the century by about 0.1 degrees F. Note that all these steps are applied to each individual station. In particular, for version 1 you can see the urbanization step individually at each station as well. The effects are not uniform, of course; the purple line in the graph above is merely a measure of the impact of all adjustments on the USA combined temperature.

All the pages also come with bibliographies and references. There are published accounts of the urbanization correction applied, which include tests on its reliability and further references to associated works. In particular note Karl, T.R., H.F. Diaz, and G. Kukla, (1988) for the version 1 page.

Also, in [post=2464019]msg #17[/post] of thread "Climate Science Update" I provide links to the three main datasets for global anomalies, along with a link to a freely available preprint of the formal publication of the dataset and its construction. Each of these three publications includes a description of how the urbanization effect is managed for that set, along with discussions and references to relevant literature.

One simple method to check for the validity of urbanization correction is to perform an independent replication using only rural stations. I have done such a replication as a private investigation, using the version 1 of USHCN, and the results are sufficiently good to confirm that the urbanization corrections applied are appropriate and not being distorted by any failure to deal with this properly. I describe this project briefly in [post=2469543]msg 357[/post] of thread "CRU hack" in the politics subforum; that thread is now locked (and I am also deliberately not making any comment on the hack affair until the staff have decided how to proceed.) I am thinking of writing that project up more thoroughly as a submission to the independent research subforum. The results I obtain are about the same as a similar project that the NCDC did a year or so after my effort, though of course they would tend to be more reliable than my effort.

Cheers -- sylas