To my readers, There are three items of interest on this subject in this post.
1. There are some new postings from Steve McIntyre (and here) that show us the hidden decline in Briffa’s tree ring data – basically it appears that Briffa only published the data up to 1960, and that ignoring the tree ring proxies past 1960 because they don’t match the instrument data! In McIntyre’s re-plot there is a distinct decline in the tree ring proxy data. Now to me this doesn’t show us that the world was actually cooling – the instrument data appears to indicate warming. But what is does show is that the tree ring proxies are not good proxies to global temperature. So we should be loathe to trust them as a historical record of temperature.
This of course makes sense to me. How can the density of tree rings be correlated to temperature without knowing the levels of precipitation, sunlight intensity. Also, tree ring data give annual average tree growth so information about whether a year started dry, wet, cold or warm and CHANGED during the year cannot be easily detected in the tree rings. I’m no dendro-chronologist, but I see unanswered questions.
2. Phil Green published a good article in the National Post today, mostly comparing the daily average minimum temperatures for Europe and North America to the monthly average anomaly temperatures that CRU published for the IPCC. He makes some very good points about data sharing – if Courtillot has data that doesn’t match the CRU data, why won’t CRU share the data so he can compare more closely?
Green also makes some very good points about the averaging/smoothing of data that CRU has done. If you have a bunch of data, averaging it out will remove a lot of the information. If Courtillot has daily data from 44 European stations and 153 US stations for most of the 20th century, then CRU must have at least that much data. But why then do they average the data out on monthly and annual bases? I recognize the need to grid the planet for the computer models, and that there has to be some relationship between the measured stations and the grid locations (which is an interesting mathematical challenge) – but the gross averaging of data seems questionable to me.
Further, the use of anomalies instead of actual temperatures is misleading because the selection of the normal period changes the appearance and interpretation of the results.
3. Along the lines of these posts, Chemical Engineering Progress, published by AIChE, has a great article (membership required) this month about numerical errors and gives a great example about how different plotting methods can provide very different assessments of how well two datasets are correlated. This has a lot to do with data fitting and explains why using the wrong method can obscure the important information. Here are two graphs from that article:
The captions of the two figures are self explanatory. In this case, the first method appears to show good correlation, whereas in the second it is clear that there is a distinct curvature, indicating the correlation is not very good. This is exactly the same problem as comparing Courtillot’s daily data with the heavily smoothed CRU data. The clearly significant averaging and smoothing of the data (and the use of “anomalies”) has obscured the useful information.