As an Old Data Hand –
I really got interested in this when I saw the data from New Zealand – showing that the raw and adjusted temperatures were showing different trends:
Specifically the raw data was showing no trend and the adjusted data was showing an upward trend. We ODHs know that if you can’t see it in the raw data, it probably isn’t there. Now that’s not always true and sometimes the data needs to be transformed in an amazingly elegant way and suddenly it fits a new pattern. Then you might get a Nobel Prize, or maybe a word of thanks – but it’s mostly it doesn’t happen.
So I was even more interested in ‘Climategate USA’, the revelation that the number of observation stations used in producing the GISS Land-surface air temperature anomalies (Global Temperature data for the newbies) had fallen from over 6000 to about 1500 during the 1990s and onwards. If there’s one thing we ODHs know, it’s that the easiest way to find what you expect in the data is to select the observations that you include in the result.
(The second easiest way is to find something else to normalise it against!)
There are a lot of caveats and speculation after the exciting graph.
So I did a bit of work with the raw data that was made available through realclimate.org (Climate Science from climate scientists) which is the GHCN v.2 (Global Historical Climate Network: weather station records from around the world, temperature and precipitation) . I had previously checked that I got the same pattern from this data regarding the New Zealand divergence and also the Darwin station fuckup, so I am reasonable sure that I am dealing with the right sort of data.
I certainly get the right sort of numbers against the station count.
I thought that if the stations observed vary from year to year, it might be interesting to look each year in terms of ONLY stations that reported in that year and the previous year AS WELL (we sometimes call this a constant sample).
Obviously the absolute temperatures can vary very considerable, if a lot of cold stations were included in Years 1 and 2, but dropped in 3, then the Year 2 absolute would be much lower than the Year 3 absolute. However I can’t really see why the change between the two years shouldn’t be a fairly valid statistic.
I can then apply the change to a base year and compare the trend between GISS, and My Constant Station trend. My data is of course unweighted by geography and has no interpolation. So I include my All Station data (also re-aligned to starting point GIFF base temperature).
I can then take the difference between my result for the constant sample stations and my result for all stations and apply that difference to the GISS temperature data (I’ve used absolute difference rather than an index ) to show what the GISS data might have looked like on a constant sample basis. (GISS+Constant sample)
Notes and queries: Am I comparing Apples and Pears? I don’t really know – I’m using the GISS Global Land Station data so I hope not.
The series starts in 1946 as this is a bit of ground zero in the GISS data I’m using with a zero degree anomaly.
Things I haven’t take account of are probably too many to mention. The number of observations in a year from a station are taken into account in the averaging but if one year is January only and the next July only I haven’t bothered. I reckon these things tend to average out.
One interesting thing is how much more variable year to year the GISS data is than my All Station data – that’s the result of me using the data straight and not weighting the stations in any way (for instance by interpolation). I know this is land temperatures only, but logically I expect global temperatures to be pretty consistant year to year – of course possibly with a trend – because it’s the same whole Earth – is that a wrong assumption?
Of course the same analysis is possible for any country with a reasonable number of sample stations – although I haven’t done it
And by the way Jon Sadleir is a cunt.
February 7, 2010 at 5:54 pm
It looks like you have De-Hansenified GISTEMP. Congratulations! A very elegant method.
February 8, 2010 at 12:14 am
I can then take the difference between my result for the constant sample stations and my result for all stations and apply that difference to the GIFF temperature data (I’ve used absolute difference rather than an index ) to show what the GIFF data might have looked like on a constant sample basis. (GISS+Constant sample)
Ought those GIFF tokens be GISS?
Very nicely done analysis, BTW. Could you, perhaps, add a bit of interpretation? What does “all stations” tell you vs. the “constant sample”, for example. What does the downshift in GIStemp mean? as another.
I think the “constant sample” says that it’s the dropout of cold southern Campbell Island that lets N.Z. warm (that knee in the ‘all data’ ) and with only a normalized constant ‘instrument’ you measure no warming trend. Yes?
Your graph then has GISS + Constant Sample. Ought that to be GISS minus Constant sample? With the black line then showing the result of the GISS processing as compared to the actual Constant Sample from the input data? If so, that GISS/ConstantSample line could be interpreted as “The GISS Transform” function applied to the input data?
Or am I projecting too much on to the graph?
February 8, 2010 at 2:38 pm
Thanks fot the correction – for GIFF read GISS (too much thinking about uploading graphics).
I’ll start at the end: GISS + Constant Sample. This is actually the GISS annual series http://data.giss.nasa.gov/gistemp/tabledata/GLB.Ts.txt
+ the difference between my All Station series and My Constant Station Series. (which I treated as a nagative).
What I am trying to do is to see what the effect of changing the stations in the data is across the whole data (where you can’t just use the stations reported continuously because they are almost all US stations)
My All Station data does not show the same trend as GISS and I assume that this is because it is not homogenized or gridded, but the difference between my All Station and Constant station data tries to show the effect of not having constant stations.
So I’ve then applied that difference to the GISS data series in order to speculate what that data would look like if it had always had the same sample of stations.
But I must be clear that it is speculation.
February 10, 2010 at 10:48 am
Thank you for the clarification. OK, adding a negative number. Got it.
“But I must be clear that it is speculation.”
Oh! So only modestly better than the IPCC citations of “peer reviewed” articles!? 😉
(Hey, I’m sure your “speculation” is better founded that mountain climbing magazines on glacier melting…)
February 11, 2010 at 6:27 am
EM Smith said that this might be a good read.
I agree. P.G.