Guest post by John Goetz
The GISS temperature record, with its various adjustments, estimations, and re-estimations, has drawn my attention since I first became interested in the methods used to measure a global temperature. In particular, I have wondered how the current global average can even be compared with that of 1987, which was produced using between six and seven times more stations than today. Commenter George E. Smith noted accurately that it is a “simple failure to observe the standard laws of sampled data systems.” GISS presents so many puzzles in this area, it is difficult to know where to begin.
My recent post on the June, 2009 temperature found that the vast majority of temperatures were taken from airports and urban stations. This would cause some concern if the urban heat island (UHI) effect were not accounted for in those stations. GISS does attempt to filter out UHI from urban stations by using “nearby” rural stations – “nearby” meaning anything within 1000 KM. No attempt is made to filter UHI from airports not strictly listed as urban.
If stations from far, far away can be used to filter UHI, then it stands to reason some stations may be used multiple times as filters for multiple urban stations. I thought it would be amusing to list which stations were used the most to adjust for UHI. Fortunately, NASA prints that data in the PApars.statn.use.GHCN.CL.1000.20 log file.
The results were as I expected – amusing. Here are the top ten, ranked in order of the number of urban stations they help adjust:
| Usage | Station Name | Location | From | To | Note |
| 251 | BRADFORD/FAA AIRPORT | PA / USA | 1957 | 2004 | Airport |
| 249 | DUBOIS/FAA AIRPORT | PA / USA | 1962 | 1994 | Airport |
| 249 | ALLEGANY STATE PARK | PA / USA | 1924 | 2007 | Admin Building |
| 246 | PHILIPSBURG/MID-STATE AP | PA / USA | 1948 | 1986 | Airport |
| 243 | WELLSBORO 4SSE | PA / USA | 1880 | 2007 | Various Farms |
| 243 | WALES | NY / USA | 1931 | 2007 | Various Homes |
| 241 | MANNINGTON 7WNW | WVa / USA | 1901 | 2007 | Various Homes |
| 241 | PENN YAN 8W | NY / USA | 1888 | 1994 | Various Homes |
| 237 | MILLPORT 2NW | OH / USA | 1893 | 2007 | Various Farms |
| 235 | HEMLOCK | NY / USA | 1898 | 2007 | Filtration Plant |
Unfortunately, having three of the top four stations located at airports was the the sort of thing I expected.
Looking a little further, it turns out all of the top 100 stations are in either the US or Canada, and none of those 100 stations have reported data since 2007. (By the way, #100 is itself used 147 times.) Several of the top-100 stations have been surveyed by surfacestations.org volunteers who have documented siting issues, such as the following:
- Mohonk Lake, N.Y. (197 times) – much too close to ground, shading issues, nearby building
- Falls Village, Conn. (193 times) – near building and parking lot
- Cornwall, Vt. (187 times) – near building
- Northfield, Vt. (187 times) – near driveway, building
- Enosburg Falls, Vt. (180 times) – adjacent to driveway, nearby building.
- Greenwood, Del. (171 times) – sited on concrete platform
- Logan, Iowa (164 times) – near building, concrete slabs
- Block Island, R.I. (150 times) – adjacent to parking lot and aircraft parking area.
The current state of a rural station, however, is an insufficient criterion for deciding to use it to adjust the history of one or more other urban stations. The rural station’s history must be considered as well, with equipment record and location changes being two of the most important considerations.
Take for example good ‘ole Crawfordsville, which came in at #23, having been used 219 times. As discussed here, Crawfordsville’s station lives happily on a farm, and does seem to enjoy life in the country. However, up until 16 years ago the station lived in the middle of Crawfordsville, spending over 100 years at Wabash College and at the town’s power plant.
| Mohonk Lake, N.Y. (197 times) – much too close to ground, shading issues, nearby building |
| Falls Village, Conn. (193 times) – near building and parking lot |
| Cornwall, Vt. (187 times) – near building |
| Northfield, Vt. (187 times) – near driveway, building |
| Enosburg Falls, Vt. (180 times) – adjacent to driveway, nearby building. |
| Greenwood, Del. (171 times) – sited on concrete platform |
| Logan, Iowa (164 times) – near building, concrete slabs |
| Block Island, R.I. (150 times) – adjacent to parking lot and aircraft parking area. |
steven mosher (22:55:15) :
for those interested in getting gisstemp working.. along with some corrections, compiler issues etcs…
http://clearclimatecode.org/
They are doing a re-write into Python. While that is a worthy goal (and they have lots of good info about what they ran into), I have a slightly different goal. I want to have the original code running with minimal changes (and all of them clearly causing no change of function) so that any “issues” found are clearly not the fault of the “port” to another language.
Basically, I’m taking more of a “forensics” approach with minimal change to the “evidence” in the lab…
DaveE (15:32:33) : You do need a specific compiler to get that code to run unaltered.
I’ll see what I can find & let you know.
In the STEP4_5 parts is imports a file written in FORTRAN “unformatted” which is actually highly formatted… and dependent on your computer “endian” type. (are numbers stored with the most significant digit first or the least significant…) So you must either run the code on a “bigendian” box like a Sun SPARC (or most other large workstations, but not all) or if you run on an x86 type chip (I.E. on a PC or anything using a gaggle of them like some of the Intel Pentium thousands of processor “supercomputers”) you need a complier with a “convert=swap” flag in the file open command.
The newer (last time I looked, it was beta) release of the g95 compiler claims to support that feature.
Mary Hinge (06:31:39) : Like every other science there is never a straight correlation,
Where do you get this stuff? In MOST of science there very much is a straight correlation. Drop a rock, it falls. Drop it again, if falls. And with mathematically repeatable precision. Good enough that pendulum clocks were the standard for precision for centuries.
It is the odd and unusual cases that have non-intuitive correlations. And while there are enough of those to “keep it interesting”, to assert they are the dominate feature of science is bogus, to say “never a straight correlation” is just so wrong as to stop a reader in their tracks. Which I did. (No idea what else you said, it all became “uninteresting” at that point.)