• Embed Doc
  • Copy Link
  • Readcast
  • Collections
  • CommentGo Back
Download
 
From: "Michael E. Mann" <mann@xxxxxxxxx.xxx>To: Tim Osborn <t.osborn@xxxxxxxxx.xxx>,Tom Crowley <tcrowley@xxxxxxxxx.xxx>, PhilJones <p.jones@xxxxxxxxx.xxx>Subject: Re: Fwd: Soon & BaliunasDate: Wed, 12 Mar 2003 12:12:xxx xxxx xxxxCc: rbradley@xxxxxxxxx.xxx,mhughes@xxxxxxxxx.xxx,srutherford@xxxxxxxxx.xxx,k.briffa@xxxxxxxxx.xxx,mann@xxxxxxxxx.xxxDear Tim,Thanks for your rapid replies and your help. This is all very useful.Well, lets see what this gives...There are some notable differences just between our relative comparisons of thedifferentseries which must have something to do with the relative scaling and aligning oftheseries. The position of Crowley and Lowery, in particular, is quite inconsistentbetweenour respective comparisons. When we scale the various series to the full N. Heminstrumental annual mean CRU record 1xxx xxxx xxxx, we get a a very differentrelative orderingof the different series, as shown in the attached figure from my Scienceperspective piecefrom last yearThis should not, however, influence the EOF decomposition if all series are zero-mean andstandardized prior to the EOF analysis, but the scaling and alignment of theresult, in theend, will be sensitive to all of these various issues.So, in short, lets see what we get, and then discuss any similarities/differencesw/ yourresult, then make a decision as to what to show in the Eos piece. I'm sure we cancome upw/ something we're all happy with...Please do send us your & Keith's preferred version of the MXD reconstruction--we'll collectthe others from the individual sources (most we already have, I think)...,mikeAt 04:53 PM 3/12/2003 +0000, Tim Osborn wrote:At 16:29 12/03/03, Michael E. Mann wrote:but there are many variables here [not the least of which is the choice of scalingtheseries to an extratropical summer mean, which as we have argued before, we don'tthinkis appropriate for a full N. Hem mean because of changes in meridional temperaturegradient over time, and the choice of calibration period--I wonder if 1xxx xxxxxxxxor1xxx xxxx xxxxgives a more stable result).True, but as I indicated I have tried alternatives. The attached is what I getwithannual mean temperature as the target series - still taken only from land >20Nthough[but I have extracted that domain from your spatial reconstructions to produce thetimeseries that I used for "Mann et al." - which should make it reasonably appropriate
 
backto 1400 at least]. I have also tried different calibration periods (including notcalibrating against instrumental data at all!). All give qualitatively similarresults- see attached .pdf and compare with the first one I sent.The point is, that (I believe) the approach will introduce a *new* result andwhile thatis interesting it wouldn't be appropriate for a short EOS piece - and having foundthisout, I was trying to save you the effort.But, on reflection, it would be good if you went ahead and did this anyway,because theresults might well be useful to publish in another paper, even if they weren'tdeemedsuitable for the EOS piece.I could provide the 7 series that I have used, but would prefer that you got themfromthe original sources to ensure that you have the most up-to-date/correct versions.CheersTimDr Timothy J Osborn | phone: xxx xxxx xxxxSenior Research Associate | fax: xxx xxxx xxxxClimatic Research Unit | e-mail: t.osborn@xxxxxxxxx.xxxSchool of Environmental Sciences | web-site:University of East Anglia __________| [1]http://www.cru.uea.ac.uk/~timo/Norwich NR4 7TJ | sunclock:UK | [2]http://www.cru.uea.ac.uk/~timo/sunclock.htm______________________________________________________________Professor Michael E. MannDepartment of Environmental Sciences, Clark HallUniversity of VirginiaCharlottesville, VA 22903_______________________________________________________________________e-mail: mann@xxxxxxxxx.xxx Phone: (4xxx xxxx xxxxFAX: (4xxx xxxx xxxx[3]http://www.evsc.virginia.edu/faculty/people/mann.shtmlAttachment Converted: "c:eudoraattachmannpersp2002.gif"References1. http://www.cru.uea.ac.uk/~timo/2. http://www.cru.uea.ac.uk/~timo/sunclock.htm3. http://www.evsc.virginia.edu/faculty/people/mann.shtmlOriginal Filename: 1047503776.txt | Return to the index page | Permalink | EarlierEmails | Later EmailsFrom: Tim Osborn <t.osborn@xxxxxxxxx.xxx>To: "Michael E. Mann" <mann@xxxxxxxxx.xxx>,Tom Crowley <tcrowley@xxxxxxxxx.xxx>,Phil Jones <p.jones@xxxxxxxxx.xxx>Subject: Re: Fwd: Soon & BaliunasDate: Wed, 12 Mar 2003 16:16:16 +0000Cc: Malcolm Hughes <mhughes@xxxxxxxxx.xxx>,rbradley@xxxxxxxxx.xxx,mhughes@xxxxxxxxx.xxx,srutherford@xxxxxxxxx.xxx,k.briffa@xxxxxxxxx.xxx,mann@xxxxxxxxx.xxx<x-flowed>This is an excellent idea, Mike, IN PRINCIPLE at least. In practise,
 
however, it raises some interesting results (as I have found whenattempting this myself) that may be difficult to avoid getting bogged downwith discussing.The attached .pdf figure shows an example of what I have produced (NB.please don't circulate this further, as it is from work that is currentlybeing finished off - however, I'm happy to use it here to illustrate my point).I took 7 reconstructions and re-calibrated them over a common period andagainst an observed target series (in this case, land-only, Apr-Sep, >20N -BUT I GET SIMILAR RESULTS WITH OTHER CHOICES, and this re-calibration stageis not critical). You will have seen figures similar to this in stuffKeith and I have published. See the coloured lines in the attached figure.In this example I then simply took an unweighted average of the calibratedseries, but the weighted average obtained via an EOF approach can givesimilar results. The average is shown by the thin black line (I've ignoredthe potential problems of series covering different periods). This was alldone with raw, unsmoothed data, even though 30-yr smoothed curves areplotted in the figure.The thick black line is what I get when I re-calibrate the average recordagainst my target observed series. THIS IS THE IMPORTANT BIT. The*re-calibrated* mean of the reconstructions is nowhere near the mean of thereconstructions. It has enhanced variability, because averaging thereconstructions results in a redder time series (there is less commonvariance between the reconstructions at the higher frequencies comparedwith the lower frequencies, so the former averages out to leave a smoothercurve) and the re-calibration is then more of a case of fitting a trend(over my calibration period 1xxx xxxx xxxx) to the observed trend. This resultsin enhanced variability, but also enhanced uncertainty (not shown here) dueto fewer effective degrees of freedom during calibration.Obviously there are questions about observed target series, which series toinclude/exclude etc., but the same issue will arise regardless: theanalysis will not likely lie near to the middle of the cloud of publishedseries and explaining the reasons behind this etc. will obscure the messageof a short EOS piece.It is, of course, interesting - not least for the comparison withborehole-based estimates - but that is for a separate paper, I think.My suggestion would be to stick with one of these options:(i) a single example reconstruction;(ii) a plot of a cloud of reconstructions;(iii) a plot of the "envelope" containing the cloud of reconstructions(perhaps also the envelope would encompass their uncertainty estimates),but without showing the individual reconstruction best guesses.How many votes for each?CheersTimAt 15:32 12/03/03, Michael E. Mann wrote:>p.s. The idea of both a representative time-slice spatial plot emphasizing>the spatial variability of e.g. the MWP or LIA, and an EOF analysis of all
of 00

Leave a Comment

You must be to leave a comment.
Submit
Characters: ...
You must be to leave a comment.
Submit
Characters: ...