Comments on: Overview et al

By: David H

David H — Thu, 24 Mar 2005 15:55:25 +0000

In #26 Hans mentioned UHI and if you look at the CET data shown at http://www.environment-agency.gov.uk/commondata/103196/304681 , it looks as if the CET is now increasing much faster than the global average. If you look at http://minerals.usgs.gov/minerals/pubs/of01-006/cement.pdf and plot the world cement production figures (asphalt is similar) you get another hockey stick. Most of what was ever made was made after WWII and most of that is still about, causing UHI. Those of us that can remember the 40′s and 50′s know that not only are there more roads but they are all much wider and that buildings have increased in size as well as greatly in number.

One can argue how big UHI is but I have not seen anyone questioning its existence. This means that we are making corrections to some readings which are of the same order as the perceived trend while ignoring UHI in others. The accuracy of land based sensors has to be questionable and in any case the amount of thermal energy they represent is tiny compared with the oceans.

The IPCC used various scenarios to model the likely effect of increasing CO2. Has anyone tried modelling these same scenarios assuming the observed temperature rise is that shown by the satellites or sea surface?

The other thing to note from the CET data is that it was as about as warm in Central England in the 1730′s as the present global average. That may answer the question as to why Mann used the 40 or so exceptionally warm years but not earlier very cold ones.

By: Hans Erren

Hans Erren — Thu, 24 Mar 2005 07:45:25 +0000

re # 29
I give up

By: Michael Mayson

Michael Mayson — Thu, 24 Mar 2005 03:22:54 +0000

Re #26 and #28 – Points taken. Here is a short history of thermometers http://www.chymist.com/Temperature.pdf with pictures of some of Fahrenheit’s early instruments – around 1700AD. Note the graduations ( between 0 [freezing of salt and water mixture] and 96 [body temperature] are at 4 degree intervals – a resolution of around 2 degrees C.

By: Michael Ballantine

Michael Ballantine — Wed, 23 Mar 2005 16:35:00 +0000

OK. So, where is the calibration and correlation data for the instruments in question? Correlation, at the time, would have been relatively easy. Stick all the thermometers in a big pot of water and read what they all say over their full range of intended measurements. Use whiskey if you want to check below zero. This will give you correlation data between the devices but only to the limit you can read them. With really good eyes you might read them to 1/3 degree and that would be the practical limit of resolution.
It is all well and fine to speculate that the errors equalize out to something small but the effective resolution is still no better than 1/3 degree. That is 30 times less than the implied resolution of the listed data. And we still don’t have the calibration data. For all we know, a third of the thermometers could be out by up to 2 degrees in the middle of their range. That creates a large offset that varies with temperature. As new thermometers are added, if the correlation isn’t done then the offset of the group could change with time and could easily be larger than the actual temperature change.
Lots of uncertainty and sources of error and no way to pin it down without the calibration data. Meanwhile, charting to 0.1 resolution maybe. Charting to 0.01 resolution, no way.

By: John Davis

John Davis — Wed, 23 Mar 2005 15:49:10 +0000

Re 25
It’s important to distinguish between resolution and accuracy. Resolution the degree of apparent precision with which a reading can be taken, accuracy is the difference between the observed reading and the actual thing being measured.
If a single instrument is used, and it has a resolution of, say, 1 degree C, the result of averaging N readings will have a usable resolution of 1/rootN degrees C. However this won’t make any difference at all to any accuracy errors in the offset or linearity of the instrument which will remain unchanged.
If you have a large array of instruments – as long as you can justifiably assume that the non-linearities and offsets are normally distributed – then you can reduce the offset and non-linearity errors in the same way. This isn’t magic, it’s just saying that for every negative error in the population there is probably a roughly equal positive error somewhere. So the absolute accuracy of a whole bunch of thermometers IS better.

By: Hans Erren

Hans Erren — Wed, 23 Mar 2005 15:32:07 +0000

Your discussion of error propogation is mathematical magic. Actually it's undergraduate physics, I thank much to my tutor Jan Kuperus, who also translated the classic by into Dutch. John replies: The "classic" book costs £110 on Amazon!

By: Hans Erren

Hans Erren — Wed, 23 Mar 2005 14:53:31 +0000

re #21

Michael,

You are changing the subject from a statistical error in a well defined metric (i.e. the average of 365 thermometer readings with uncertainty 0.5 C), to the uncertainty of systematic errors, which is a complete different subject. This touches upon the problem of station homogenisations and UHI corrections. which are shown to be in the order of 1K, and can be adjusted by comparing adjacent stations.

By: Michael Ballantine

Michael Ballantine — Wed, 23 Mar 2005 13:25:48 +0000

Re #21
Your discussion of error propogation is mathematical magic. The error being reduced is the errors introduced by the various mathematical calculations. This sort of hand waving does nothing to change the inherent instrumental error which must be applied for each instrument at the end of the calculations. The various climate scientists are so caught up in their mathematics, theories and models that they have totally lost sight of the limits of the raw measurements.

By: Michael Mayson

Michael Mayson — Wed, 23 Mar 2005 11:51:47 +0000

Re #17: Hans, isn’t it the case that “the error of a summation of values is proportional to the root of the squared summed errors” is only true if measurement error is normally distributed about the reported measurement. If the measurements in your example had a bias error of .5 then the total summation error would be 15 with an average of 0.5.

By: Michael Mayson

Michael Mayson — Wed, 23 Mar 2005 11:28:43 +0000

Re 22: Well I got that wrong! The new Scientist article as May 2004. My question now is why haven’t we heard more?