Temperature Analysis: It’s Climate Science’s Own Pepsi Challenge
February 25, 2010Posted by on
One thing Phil Jones, the CRU and others like to point out is how well X dataset matches to Y’s “independent” dataset. Well that is at best only a half truth, half lie statement.
They are not independent since they all work from the same basic “raw” datasets. The CRU, GISS and NCDC did not set up their own thermometers all over the globe and take readings, they all rely on the same readings, from the same thermometers for their analysis.
Where they are “independent” is how they do their analysis. CRU claims they use the methods spelled out in the multiple Jones et al papers and the Brohan et al 2006 paper. NCDC as of right now does their global analysis on the papers published by Dr. Petersen. GISS does their work based on the the Hansen et al 1987 and Hansen et al 2001 papers.
The one I’m most familiar with is the Hansen 2001 paper and in it Dr. Hansen spells out clearly that each of these analysis methods are different each with their own inherent strengths and weaknesses.
Since they treat the data differently they can and do reach different results. Then to top it off each country that supplies the raw data via CLIMAT reports do their own analysis of the data as well using their own methods, which leads to an important question?
Is what these various countries sending in true “raw” data or data that they themselves have done work on such as preliminary homogenization adjustments?
Who knows. NCDC can’t be 100% sure because they are not the ones doing the readings they just get what is sent to them and if they are not sent the correct metadata they wouldn’t necessarily know if they are getting “raw” or pre-processed data.
So in the end when you look at for example the outputs from these various groups most people are “picking” one over the others as being the “best” or “most accurate”. As an example of the differences between datasets just head over to the Blackboard and follow Lucia’s posts on the monthly temperature anomalies. Example HadCrut has the Jan. Anomaly at .47° C, GISS had it at .71° C and NCDC had it at .6° C.
So people seem to pick and choose which analysis they believe on what they usually see in a gridded map or in a line graph, not on the method of how the numbers were derived. They are placing faith that one method is “better and more accurate” then the others.
To hopefully show this I’m taking a part of the work I’m doing going through the Canadian records and I will and put up two comparative charts that will contain anomalies computed on the 1961-90 baseline. These two charts will have three time series on them, one will be the GISS analysis, another the NCDC analysis and the third will be the Canadian numbers. The catch is I won’t tell you on the graph which is which. Then I will show you a graph of all the different thermometer series used for each station in the GHCN “raw” data file in a seperate graph to see what they all started with.
Finally at the end I will reveal which is which and let people see how their own internal bias is working. Will they pick the same analysis for both graphs or will they switch analysis based on other factors (visual factors)? Basically this is the old 1980’s Pepsi Challenge where the person tried both Coke and Pepsi without knowing which was which and at the end was shown which “taste” they truly preferred.
First lets start with Alert NWT Canada WMO # 71802:
I’m not giving you trend lines either, you pick the series from what you see there. Now I will give you this from the GHCN “raw” file:
Now I bet you noticed one feature from the “raw” data that corresponds very nicely with one of the adjusted anomaly series: That great big dip in 1973 but did you notice that all the analysis dismiss that huge increase in 1963 on the same thermometer as the 1973 dip? Also notice that the increase in 1990 seems to have vanished as well from the adjustments. I’ll let you think on how those two omissions might have effected the trend lines.
Now we move onto Eureka NWT Canada WMO # 71917:
Oh I may or my not have switched the analysis around between figure 1 and figure 3 so that what was Choice A in figure 1 may or may not be Choice A in figure 3.
Now lets look at the GCHN raw from this WMO ID:
Now I bet you noticed that my comparison graph didn’t go include the later years of the 2000’s but the raw shows that there is data in those years. The problem is that one of the analysis methods drops those years so I had to cut off the others so they matched.
So make your picks which one you think is the closet to the real temperature and see if you picked the same analysis for both comparison with out knowing before hand who did the analysis.
Here is the answers:
Figure 1 A = Canadian, B = GISS, C = NCDC
Figure 3 A = NCDC, B = Canada, C = GISS