cc: k.briffa@uea date: Wed Apr 14 11:35:55 1999 from: Tim Osborn subject: Re: stuff to: mann@snow.geo.umass.edu At 11:50 AM 4/13/99 -0400, you wrote: >p.s. I had a good conversation w/ Phil in Asheville about our mutual >climate reconstruction projects, and we're hoping to work more closely >in the future. Phil indicated that you are an integral part of that, >and I would certainly welcome working w/ you on sharing data, codes, etc. >I start a permanent position at UVa this fall and will likely have >a postdoc on this project. I'm hoping he and you can communicate >frequently and potentially work together to resolve some of the key >questions Phil, Keith, and I have been discussing. I hope this sounds >like something you're interested in? Mike Such collaboration sounds good and hopefully beneficial to all. I've just started the 2-year project on April 1st, so I'm getting back into the paleodata frame of mind! I think I have travel money for one trip to the US over the two years; I've no definite plans yet, but will consider going to an AGU conference and combining it with meeting several US colleagues. Thanks for your comments. I've also received a subsequent e-mail from Julia Uppenbrink: she says she will go ahead and edit our original draft, and put queries in the places where your comments apply. Keith and I will then make appropriate adjustments when we get that revision - probably next Tuesday. Some of the comments deal with issues that are getting a bit beyond my expertise, and I haven't discussed them with Keith yet. But I thought you might like to hear my initial reaction to them - mainly so that you can, if you wish, clarify anything that I seem to be mistaken about. (1) Removing the word 'actually' might make this sound like a statement of interest rather than a criticism. (2) We will look again at our points concerning the uncertainty ranges. Were they computed from the goodness of fit over both the verification and calibration periods, or are you saying that they were computed from only the latter period but the former period would have given very similar results? In terms of the time-scale issues, I thought that the figure in your GRL paper showing the spectra of the calibration residuals implied the AD 1000-1980 network did (relatively) less well at capturing the lower frequency variations. (3) Keith knows more about this than I do, but the point we were making related less to the relative decline in tree-ring-density over the very recent decades, and more to the enhanced ring-width growth from 1850 onwards. The suggestion is that some non-climatic signal is in ring-width data from a range of locations (not just ITRDB PC#1), so is it correct to assume a priori that such a signal does not exist in the northern treeline series that you are using to make the correction? I'll talk to Keith about this further. (4) Adding the NINO3 reference should be fine. (5, and also part of point 3) The three tree-ring-width series used in the 2000-yr series have been standardised in such a way that none of the variance is lost at any time scale. Assessing whether it has been successful in this respect is, of course, difficult. We will, I'm sure, study this further (e.g., by comparing the zero-frequency end of the spectra and with a red-noise fit, as you have done), but I think that the 2000-yr series has merit for the Science paper and I would hope that it would be relevant to the IPCC review too. In terms of the baseline level and region/season, I think we have been extremely consistent, better perhaps than previous intercomparisons. We will explain further in a proper paper that we are preparing, where space will permit fuller details. Our premise is that if each series represents - and is calibrated to - a different region and/or season, then comparison of them will show up a combination of (i) differences that should be there because of region/season and (ii) differences due to the errors in each series. So, we have taken each series and re-calibrated it against a common season and regional average, and over the same period (1881-1960). We are then saying: what can each record tell us about past Apr-Sep temperature variations over land north of 20N? Justification for this is based on the seasonal responses and location of proxies in the Briffa (*2) and Overpeck curves and the fact that the Jones curve does better for that season/region than for the Jun-Aug NH land+marine series. For you curve, we could have asked you to compute the land-only area average, but we just used the Annual NH land+marine estimate because that still does better than the other records. Having re-calibrated, the comparison shows much smaller differences between the curves than would have otherwise been obtained. All are also referenced to the 1961-1990 average, BUT only because they have been calibrated against temperature data that are anomalies wrt 1961-90. The calibration is over 1881-1960, so the mean of all curves over that period is the same as the mean of the instrumental data over that period. They are not then forced to have the same mean over any other period. The fact that they don't is a combination of the different curve shapes, the loss of variance in the regression-based calibration (the higher r, the more variance captured), and the different timescale responses (some series respond more strongly to high-frequency variations than to interdecadal variations, and the calibration can, therefore, suppress some of the latter to avoid overdoing the former). I don't think, therefore, that you can say the entire series are too high or too low, nor to exclude them from IPCC on that basis. Finally, even though the curves are not annual & hemispheric, all are still relevant to climate change and IPCC because any warming signal is likely to be large over the land areas north of 20N. That's a longer response than I'd planned. I'll talk things over with Keith when he returns on Monday. Best regards Tim