Errors In GISS Classification Of Urban & Rural Stations
By Paul Homewood
Peter O’Neill left a comment on my post about GISS UHI adjustments a couple of days ago. I thought it worth posting it up with the table he had included.
It appears from the work Peter has done that many stations that GISS classify as rural are, in fact, urban. The errors are arising because of incorrect latitude/longitude data held within the GHCN metadata.
Peter has taken a sample, which suggests as many as 20% of stations could be wrongly classified.
If true, this would present two problems:-
1) These stations would not be being adjusted for UHI.
2) Other stations, correctly classified as urban, are adjusted for UHI by comparing with nearby rural stations. If some of these “rural” stations are in fact “urban” after all, then the first station may not be being adjusted correctly (or indeed might be adjusted the wrong way,which, as we have already seen , has been happening a lot).
This was Peter’s comment in full :-
There is a further factor at play in relation to GISTEMP homogenisation adjustment. Urban and rural stations are identified using GHCN inventory latitude and longitude values to look up night light radiance values. Hansen et al 2010 claims:
Station location in the meteorological data records is provided with a resolution of 0.01 degrees of latitude and longitude, corresponding to a distance of about 1 km. This resolution is useful for investigating urban effects on regional atmospheric temperature”
but a substantial number of GHCN station locations show errors far in excess of that claim (small differences may simply reflect station relocations, but GHCN locations 50 km or more distant from the named town, port or airport are almost certainly in error, as also are station locations out at sea but named as onshore).
The result of looking up the night light radiance at a point distant from the actual station location is often misclassification of urban stations as rural, or vice versa.. A possible consequence of urban stations wrongly classified as rural is that such a station, if UHI affected, will be allowed by the GIStemp homogeneity adjustment algorithm to contribute this warming UHI effect to the mean trend of “nearby rural” stations with which the long-term temperature trend of an urban station will be made agree. This possibility does not appear to have been considered by Hansen et al., although it would, if the contribution is sufficient, lead to a false adjustment which would appear to be for “local anthropogenic cooling”. Such false adjustments may or may not in fact make a significant contribution (accurate location metadata is needed to determine whether or not this is the case).
The scale of the problem with GHCN station locations may be illustrated by considering the sample of GISTEMP rural stations with records extending back at least 90 years and still reporting in 2010, which has been used for confirmation by comparison with GISTEMP. Almost all the non-North American stations in this sample are WMO stations, and using the WMO metadata instead of the GHCN metadata would lead to reclassification of nearly 20% of these stations as urban rather than rural.
GISTEMP have been made aware of the quality problems with GHCN metadata. As early as December 2009, before the change to global use of night light radiance was implemented, I spotted a change in the code added to test global use, and notified Reto Ruedy that this might be unwise, citing as an example the two Cherbourg stations which I had spotted were located out at sea by GHCN (and while the airport would still remain rural when relocated back on land, Cherbourg Chantereyn would become bright and urban if relocated back to land). At this stage I was not aware of the scale of the problem, but once the change to global night lights was implemented in early 2010 (and the drafts of Hansen et al 2010 for comment appeared on the GISTEMP site) I spent a few minutes checking for other European stations located out at sea, and quickly became aware that the Cherbourg stations were not unique in this respect, so I also looked for “airport” stations located well away from the named airport, and notified both GISTEMP (Reto Ruedy, and Dr Hansen as a comment on his draft) and GHCN that there were some substantial errors in the metadata. The initial list I sent to GISTEMP and GHCN was: [see below]
(and this early list is far from complete, and misses a substantial number of errors well in excess of the smaller distances at the bottom of this list found later)
* Errors related to Tenkodogo (=Zaragoza Airport), Albacete (=Ibiza Airport), Soria (=a Lerida station) have been addressed in the September 2010 GISS analysis update. Reto Ruedy indicated on August 23rd 2010 (coincidentally, the date on which Hansen et al 2010 was accepted for publication) that he was “not surprised at all that there are serious mistakes in this inventory file”, and that this list would be looked into, confirming that “Tenkodogo” was being dealt with. Since then I have not seen any discussion by GISTEMP of the possible problem arising from station location errors, although a substantial part of Hansen et al 2010 is based on the assumption that homogeneity adjustments can be made based on urban/rural classification by night lights using the GHCN metadata. Recent GISTEMP code however shows provision has been made to use BEST urban/rural classification as an alternative, so something may be in the pipeline.
I did try to submit a comment to Reviews of Geophysics, but this was not accepted “because Reviews of Geophysics does not accept Comments/Replies since they are not a review.”. The rejection continued “However, the topic of your Comment is potentially of great interest to the Climate Community and therefore I suggest that you submit it as a stand-alone paper to GRL or JGR. A paper that can quantify the size of the impact of the corrections would be an important contribution to the field.”, and I have been continuing to pursue sufficient revised station locations to achieve this quantification, a rather massive task. This does however leave the rather unsatisfactory situation for now that urban/rural classification based on GHCN metadata in which unwarranted confidence is expressed is allowed at appear as state of the art, without comment.
To finish, it would be unfair to withhold a guess at the impact of metadata correction based on the revised metadata collected so far, even though I am not yet satisfied that I have collected enough. Use of revised station locations does seem to reduce the incidence of “urban cooling” corrections, so cancelling out less of the “urban warming” corrections. A rough upper bound on the effect would be a reduction of the warming trend over the past 100 years for global temperature anomaly (meteorological stations only) by approximately 0.085°C. Not huge, but rather more than the value of the order of 0.01°C suggested as the “effect of urban adjustment on global temperature change”
More information is available at “Peter O’Neill’s Blog” – http://oneillp.wordpress.com/