This issue was also recently covered on the Climategate blog here
Despite its assurances, GISS has adjusted the temperature records of two sites at Mackay to reverse a cooling trend in one and increase a warming trend in another. This study presents evidence that this is not supportable and is in fact an instance of manipulation of data.
I decided to have a look at the temperature records of the weather stations closest to where I live, near Mackay in North Queensland. The Bureau of Meteorology lists 3 current stations: Mackay MO, Mackay Aero, and Te Kowai Exp Station, plus the closed station Mackay Post Office. GISS has a list of nearby stations. One is “Mackay Sugar Mill Station”. I had never heard of it. Te Kowai Exp Station, only a few kilometres from Mackay, is in fact at the same co-ordinates as Mackay Sugar Mill. I checked on AIS for the GHCN site, and there is Mackay Sugar Mill on the map. The co-ordinates given by GHCN put it in the middle of a cane paddock 600m to the south of Te Kowai Sugar Experiment Station, so that’s definitely it! (If not, it’s identical in every other way!) And that is the closest weather station to my home, so I became even more interested.
Te Kowai is an experimental farm for developing new varieties of sugar cane, run by scientists and technicians since 1889. It has a temperature record of over 100 years with only a couple of gaps. So in fact it’s an ideal rural station for referencing a nearby urban station, as it should have a similar climate.
I plotted data from BOM for maxima and minima and obtained the means for Te Kowai, all Mackay city stations, all GHCN stations in our 5 x 5 grid, and several other towns and cities with long records (Te Kowai’s starts at 1908). This is because “ In our analysis, we can only use stations with reasonably long, consistently measured time records.”
GISS combines GHCN data from all urban stations at the same location, and then homogenises this with data from neighbouring rural stations. So I then plotted the same-location data and the post-homogenisation data.
A problem that appeared immediately is that the GISS annual mean runs from December to November, while BOM’s raw data is for calendar years. Most of the time it matches pretty well, but there are several examples of poor quality data. Another problem is that BOM does not compute a mean for any year with even one month of data missing, while GISS tolerates several missing months.
Here are graphs of the results.
Read his entire post here