cc: David Viner <d.viner@uea.ac.uk>
date: Thu Jul 15 18:29:12 2004
from: Tim Osborn <t.osborn@uea.ac.uk>
subject: RE: UKWIR CL04\C
to: "Nick Reynard" <nsr@ceh.ac.uk>, "Christel Prudhomme" <CHRP@wpo.nerc.ac.uk>

   Dear Nick and Christel,
   here's some initial comments, more will follow (David and I are away on various odd days in
   the next week or so, so we haven't yet sat down together and talked this through - hence
   the cc to David).
   Later comments/changes will be put into the document itself.  These are about the project
   as a whole, and our overall contribution.
   (1) Uncertainty.
   The model I like best is the separation of uncertainty in global-mean temperature change
   (GM_dT) from the uncertainty in relating GM_dT to local change in P, T, PE.
   PDFs of GM_dT already exist that take into account uncertainties in emissions scenarios,
   carbon cycle, ocean heat uptake and climate sensitivity (e.g. Wigley and Raper) and/or that
   have been determined by reference to the rate of observed warming (e.g. Knutti et al.) -
   and we could say that would make use of later updates on this (e.g. output from the large
   Met Office ensembles or climateprediction.net studies).
   Relating the GM_dt to local change can then be done via "pattern scaling" methods, with
   patterns of change derived from GCMs, RCMs, SDSMs - whatever we choose to use.
   This separation of pattern from global-mean response implies the pattern of response is
   independent of the magnitude of response.  I think this is justified, given that there is
   limited evidence of a systematic difference in patterns between GCMs with small or large
   climate sensitivities.  And even if there is, the separation suggested here just slightly
   widens the range of uncertainty - which, given that the available GCMs do not span the full
   range anyway, is not a bad thing.
   As we discussed at the end of the London meeting, pattern scaling is not appropriate if we
   need to scale series of actual data, rather than *changes*.  Which leads me on to...
   (2) Variability.
   It is not clear to me from the first draft whether the focus is only on changes in mean P,
   T, PE, or whether changes in variability (at monthly, at daily time scales) will be
   considered throughout (as opposed to only considering them in task 3.3).  If the focus is
   on changes in mean climate, then that simplifies things but at the risk of omitting
   potentially important changes.  We have already developed a methodology (and ClimGen
   software, which could be adapted for this project) to generate series using scaling *and*
   including changes in variability of monthly means - thus including variability changes does
   not preclude a pattern-scaling approach to generating scenario uncertainty/probability.
   Going down to changes in daily variability is possible in a more limited way, though
   probably of less importance in a groundwater-focussed project like this.
   (3) Detection.
   Detecting expected changes in the observed record needs a careful consideration of the
   signal-to-noise ratio and whether we should *expect* to detect a signal, as I'm sure you're
   aware (I see, for example, the statement about taking variability into account).  To be
   worthwhile, this analysis must attempt to strengthen the signal-to-noise ratio as much as
   possible, and hence enhance the detectability of the climate change.  This can be done by
   careful combination of individual records (e.g. Osborn work on precip "intensity"),
   probably guided by the expected pattern of change from the scenario work (another reason
   for leaving task 3.1 till the end of the project!).  But it is also possible to enhance the
   signal-to-noise ratio by separating the climate into two components - that caused by
   atmospheric circulation variability and that caused by other factors.  Osborn and Jones
   (2000, Atmospheric Science Letters) showed that the signal-to-noise ratio of the latter
   component is much stronger and easier to detect (especially in relatively short records)
   for UK temperature and precipitation.  It would be nice to take a similar approach to the
   detection of streamflow trends.
   As you may have guessed from these comments, I'd like to see CRU having a role in tasks
   1.4, 3.1 and 3.3, in addition to the 1.2 and 1.3 that you have put us down for.  What do
   you think?
   Cheers
   Tim
   At 12:29 14/07/2004, Nick Reynard wrote:

     Hi everyone,
     I attach a daft of the UKWIR bid for you to look at.
     There are obvious places for input from each of you, but also comments
     (general on structure etc and specific) and additional /change to text
     is welcome.
     Once we have been through one iteration on this we all start thinking
     about our costs, so could I have first comments back asap please.
     Our ultimate deadline if July 26th, although we need it to go by next
     Wednesday (21st) if possible.
     Cheers,
     Nick
     >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
     Nick Reynard
     Head, Risk Analysis and Trends Group
     Hydrological Risks and Resources Section
     Centre for Ecology and Hydrology
     Maclean Building
     Crowmarsh Gifford
     Wallingford
     Oxon, OX10 8BB
     Tel: 01491 692402
     Fax: 01491 692430
     >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
