cc: David Easterling <David.Easterling@noaa.gov>, trenbert@ucar.edu,  Aiguo Dai <adai@cgd.ucar.edu>, l.alexander@bom.gov.au, Byron Gleason <Byron.Gleason@noaa.gov>
date: Tue Jul 26 16:55:44 2005
from: Phil Jones <p.jones@uea.ac.uk>
subject: Re: DTR follow-up
to: "Russell Vose" <Russell.Vose@noaa.gov>

    Russ,
         The work with Adrian will help in the long-term, but not in the short term here (i.e.
    Aug 12). Let's discuss that possibility later.
         Also can you arrange for WWR91-00 to get sent at some time. Is it all on a
    CD, for example. That would be fine. If books have been produced, they are always
    useful - and do get used in our library. We would like them, not asap, so if books
    are available, they can come on a slow boat.
         The HC said they had found some errors with station level and SLP. It was
    Rob Allan, so only looking at pressure.
        As for the matter in hand. Can you send this group, time series plots for
    1950-2004 for annual max/min/dtr, so we can compare with the figure we
    currently have in the draft ? Dave a copy of this. Maybe plot with/without the
    suspect Argentinian stations.
        An issue with these plots might be the last 2-3 years. Do they significantly
    impact the trends, for example? Aiguo's data must use synops for 2001-4. You
    note problems using this source via your pt 7 earlier.
        We do need resolution, but I'm not sure we will get it.
    Cheers
    Phil
   At 15:26 26/07/2005, Russell Vose wrote:

     Phil:
     Thanks for the feedback.
     How much longer would you recommend I continue to look at regional-scale discrepancies?
     I'm wondering because my data "sensitivity" tests aren't having a huge impact.  For
     instance, I removed all Argentinian stations derived from GHCN-Daily (i.e., the those
     with data disagreements with other sources from 1984-87).  The DTR trends dropped as a
     result, which was expected, but most trends are still slightly positive in that area.
     As far as the work with Adrian Simmons goes -- might be a really useful Q/A tool.  We
     should discuss further because it could impact how I merge stations in the future.
     The global/annual DTR trend from 1979-2004 is 0.0036 C/dec if you include the suspect
     Argentinian stations and -0.0033 C/dec if you exclude them.  Neither is anywhere close
     to statistically significant.
     Both are based on a 1979-2004 base period, 21 years of data, etc.
     We might have sent the WWR 1991-2000 data to the HC.  Hope they got the clean data!
     Just let me know when you want it.
     Regarding the impending deadline -- I'm on it, or it's on me!
     Phil Jones wrote:

      Russ,
         Thanks for the summary.  The results you're getting in 4) and 5) seem to confirm
      that you've got the overall numbers right, but the detail for South America and
      other regions will be where the goodness of the results stands or falls. So,
      continue the work on sorting this out.
          The work we planned with Adrian Simmons with the ERA-40 data would
      seem more important. This highlighted a few problems with the CRU data for
      mean temperature. My guess is that a similar comparison might highlight
      many more problems, most of which would likely be resolved through looking
      at the issues of dataset construction that rest of the decisions you allude to
      in 7).
         The points you make in 7) just highlight how difficult it is to put together
      datasets of this kind, particularly when trying to use daily and hourly data.
         We are working to a tight deadline, as you know. What is the DTR trend
      over all for 1979-2004?
         In the longer term I'd like to get a copy of the WWR dataset for the 1990s.
      The HC seem to have a copy of this by the way !  No rush, as I've far too
      much on at the moment.
      Cheers
      Phil
     At 22:37 25/07/2005, Russell Vose wrote:

     Hi guys...
     About two weeks ago I received several e-mails expressing surprise (and concern) over
     Fig. 3.2.11, the global map of gridded DTR trends for 1979-2004.  Over the past few days
     I've conducted several sanity checks to address these concerns.  In general, these
     checks indicate that the analysis is accurate over large spatial scales, although some
     areas may be suspect due to limitations in the underlying station network.  In the
     following I discuss the various checks performed as well as the results obtained
     therein.
     1. As a first step I double-checked the calculations by hand and didn't catch any
     problems.
     2. I computed trends for 1950-93, the Easterling et al. analysis period, using exactly
     the same base period as in the Science paper.
     As global/annual trends derived from the "NCDC" dataset were within 0.02 C/dec of Dave's
     results (slight differences are to be expected given differences in spatial coverage).
     Furthermore, trend maps prepared from the NCDC dataset are VERY similar to Easterling et
     al.
     3. I evaluated trends for 1979-2004 using two record length thresholds: 18 years (67%)
     and 21 years (80%).  Note that the former was used in the map you received a couple of
     weeks ago.  The latter contains slightly less noise, but the patterns are essentially
     the same.  Specifically, both maximum and minimum temperature increased over virtually
     all areas except northwestern Australia and parts of southern South America.  As already
     discussed in various e-mails, the DTR pattern was much more variable, with somewhat
     surprising increases in the same areas (as well as the United States and Europe).
     4. I took Phil's suggestion and compared the max/min trends in the NCDC dataset to the
     mean temperature trend in GHCN version 2.0, which is used here operationally.  For
     1979-2004, the maximum temperature trend is 0.284 C/dec, the minimum temperature trend
     is 0.280 C/dec, and the mean temperature trend is 0.269 C/dec (again, slight difference
     likely results from differences in spatial coverage).  Few areas have had decreasing
     mean temperature trends during that period, but those that did almost always had a
     decrease in the maximum and/or minimum (e.g., southern South America, northwestern
     Australia).
     5. I followed up on another of Phil's suggestions and evaluated trends from 1950-2004
     (using a base period of 1979-2004, requiring 18 years of data during the base period and
     36 during the trend period -- not that the results are very sensitive to these
     thresholds).  As expected, the DTR map is much smoother than for 1979-2004.  On the
     global scale, the maximum temperature trend was 0.144 C/dec, the minimum was 0.204
     C/dec, and the DTR was -0.058 C/dec.  The maximum and minimum trends are considerably
     larger than Easterling et al. (1997) whereas the DTR is smaller.
     6. I have made numerous time series plots for southern South America in an effort to
     explain the somewhat surprising trends there.  At first I was highly confident in my
     data for that area because it is a compendium of records from the Easterling et al.
     (1997) paper, the new addition of World Weather Record (which no one has yet), CLIMAT
     reports, and GHCN-Daily.  In many cases I have data for the same station from more than
     one source, and the time series plots suggest that the data during the overlap periods
     are very similar.  However, by the end of today I had identified 13 stations that may
     have suspect data from 1984-87.
     Specifically, the GHCN-Daily data for this period disagrees with the other sources in a
     manner which may result in an increasing DTR trend.
     Unfortunately, I have no way of knowing which source is correct (majority doesn't rule
     in this business).  I'll try to sort that out in the next day or so.
     7. For the record, we actually computed monthly mean maximum and minimum temperature for
     all stations in the Integrated Surface Hourly (i.e., synoptic) database for the entire
     period of record.  There are many decisions one must make along the way (e.g., what
     observation time should one use, should one make assumptions about missing "period
     quantities", how many days must have data when computing a monthly mean).  We basically
     found those decisions had a huge impact on the results.  Consequently, I made only very
     limited use of these data in the analysis -- basically applied them to fill a few holes
     or extend the record when the agreement seemed reasonable for certain quantitative
     checks and on time series plots.
     --
     Russell S. Vose, Chief
     Climate Analysis Branch
     National Climatic Data Center
     151 Patton Avenue
     Asheville, North Carolina 28801
     Phone: (828) 271-4311
     Fax: (828) 271-4328
     E-mail: Russell.Vose@noaa.gov

     Prof. Phil Jones
     Climatic Research Unit        Telephone +44 (0) 1603 592090
     School of Environmental Sciences    Fax +44 (0) 1603 507784
     University of East Anglia
     Norwich                          Email    p.jones@uea.ac.uk
     NR4 7TJ
     UK
     ----------------------------------------------------------------------------

     --
     Russell S. Vose, Chief
     Climate Analysis Branch
     National Climatic Data Center
     151 Patton Avenue
     Asheville, North Carolina 28801
     Phone: (828) 271-4311
     Fax: (828) 271-4328
     E-mail: Russell.Vose@noaa.gov

   Prof. Phil Jones
   Climatic Research Unit        Telephone +44 (0) 1603 592090
   School of Environmental Sciences    Fax +44 (0) 1603 507784
   University of East Anglia
   Norwich                          Email    p.jones@uea.ac.uk
   NR4 7TJ
   UK
   ----------------------------------------------------------------------------
