cc: "Mitchell, John FB \(Chief Scientist\)" <john.f.mitchell@metoffice.gov.uk>
date: Wed, 16 Aug 2006 16:14:52 +0100
from: "Mitchell, John FB \(Chief Scientist\)" <john.f.mitchell@metoffice.gov.uk>
subject: Chpt 6 - last 1000 yrs
to: "Keith Briffa" <k.briffa@uea.ac.uk>

   Hi Keith
   I have tried to cindense what I think the main issues for the and what the response is
   below. The weakest area seems to be statistical significance and by implication the likely/
   very likely statements. I can't think of any easy solution - in the TAR for detection and
   attribution we used 95% limits on stats tests and them downrated them to allow for other
   uncertainties.

   I am interested in your comments
   John

   Issues
   1. Reliance on Bristlecone pine -
   Response - the issues are in calibration period- they agree with other indicators for the
   rest of the record
   2. Centring of principle components leads to "hockeysticks"-
   Response - this makes only a small difference when standardised data used.
   Comment - Would be useful to know which reconstructions do and donot make this assumption-
   this could strengthen the response

   3. The divergence issue-
   Response - it is only apparent in high latitudes, and only with some trees.
   Comment- Do we know what happens if we eliminate those records with a divergence problem.
   The wider issue is whether or not it is reasonable to extend the reconstructions outside
   the calibration range.

   4. There are different ways of verifying reconstructions and assigning significance levels(
   calibration period or seprate verifying period, different statistics)

   Response ?
   Comment- it is difficult in the text to gauge how well reconstructions are validated - eg
   using the calibration period to estimate errors as opposed to an independent period clearly
   makes a difference. This is important where "likely", "very likely"are used- based on what
   statistics? I think this is the area where I think the current response is weakest

   5. Robustness- Burger and Cubasch show a wide range of results using different assumptions-
   Response ?
   Mann makes a reasoned defence- there are other checks and tests which would rule out many
   of the arbitrary assumptions explored by Cubasch and Burger, but this is not clear in the
   response to M&M etc
