Airport Warming
Posted by drj | Filed under Uncategorized
More or less on a whim I split the GHCN data into two sets: Those stations marked as being at an airport; those stations not marked as being at an airport. This is easy to do because the v2.inv file puts an ‘A’ in column 81 (counting from 0) for airport stations.
Here’s the airport versus non-airport comparison for ccc-gistemp:
Certainly for the most recent 50 years it doesn’t seem to matter much whether you use exclusively airport based measurements or exclude airport based measurements (considering the global anomaly).
My earlier post about the 1990s station “dropout” used a similar technique of splitting the input data into two sets.
May 16th, 2010 at 1:37 am
So you run step 0 for the whole thing, then split the input, and run steps 1,2,3 and 5 for each subset separately?
Can you give sample sizes?
I’m curious about the airport subset, maybe in step 2. I thought urban stations get whacked if they don’t have enough rural neighbors to do the UHI adjustment. If you have an airport-only subset, then you need neighboring rural airports, which seems a little restricting. I’m curious how many stations survive that step, and how many get whacked.
May 18th, 2010 at 11:42 am
Why do you show the R²-value?
It’s nearly irrelevant – the standard error contains more information.
May 18th, 2010 at 2:20 pm
@physicist: Mostly because it just happens to be computed.
Showing something more intelligent would have to take into account the autoregression in the data (which requires a bit more thought). [drj: 2010-05-19: oops, I mean autocorrelation here, not autoregression]
Actual answer: I didn’t show R² initially, and then Nick added the code.
May 18th, 2010 at 3:47 pm
Carrot Eater: What I’ve done in my own studies was to actually insert various boolean checks in Step2 where the code drops “short” records, and have the code also drop records with whatever characteristic I care about (eg records ending after 1992/records with brightness less than X/etc.). It seems to me that this is less labor intensive than splitting the GHCN dataset.
May 18th, 2010 at 8:24 pm
I think you might be introducing some bias due to differing spatial coverage. I’ve been working on a similar project for the last few days, and get the same results as you for all stations of each type, but the difference mostly disappears when looking at only grid cells with both airport and non-airport stations.
May 19th, 2010 at 8:03 am
@Zeke: That seems entirely likely. Visualising the spatial coverage is on the infinitely long todo list.
May 21st, 2010 at 10:17 pm
drj,
I’m curious if you or Nick might be able to lend us a hand unraveling a little mystery: http://rankexploits.com/musings/2010/the-great-gistemp-mystery/
May 25th, 2010 at 8:24 am
How do you have ‘airport’ annotated data prior to 1928? Is is data from sites that have since become airports?
May 25th, 2010 at 2:13 pm
@MikeMcc: Yes. The airport tag is in static metadata for each station: latitude, longitude, altitude, vegetation type, satellite brightness, and so on.
August 19th, 2010 at 3:16 pm
[...] and UHI will be addressed in forthcoming pieces. Given the preliminary work done on airports. (and here) and latitude to date, we can confidently say that the entire debate will come down to two basic [...]