cc: "Keith Briffa" date: Fri, 20 Aug 2004 09:55:23 +0100 from: Tim Osborn subject: Re: Question on Paleo Data to: Mike MacCracken , Jonathan Overpeck Dear Mike, answers to all your questions together: At 18:26 18/08/2004, Mike MacCracken wrote: >a. As I understand it, the instrumental data were for land north of 20N Yes. >b. As i understand it, the Overpeck data were in standard deviations, and >were converted to degrees by normalizing with respect to the isntrumental >data above, even though areal coverage is somewhat different? Or were the >calibrations done for the land areas covered by Overpeck? The latter would be a good idea, but is not what we did. We converted them to degrees by calibration against the instrumental data for land north of 20N, even though (as you point out) the proxy data to do cover all this area (only the more northern parts are covered). The calibration was done by linear regression rather than "normalisation", so the mean matches the instrumental data mean (and hence the calibrated reconstruction represents degrees anomalies from the 1961-1990 mean, because the instrumental data to which it is calibrated were expressed as anomalies from that period's mean) but you will find that the standard deviation of the calibrated reconstruction is lower than the standard deviation of the instrumental data. This is in common with all regression-based predictions, where the predictors are not perfect representations of the real temperatures, and thus some of the variance is not captured by the reconstruction. >c. Is there a reason the instrumental data stop in 1994 (we would want to >continue the curve to the present)? Is this just because the paleo data do? >Do you have the values for later for this land area (or we could perhaps >derive)? The last instrumental data value is in 1993, and this is because that is the last year that any of the reconstructions have data (I think the file header says this). The reason why I stopped then, was that when I filtered each curve to draw the graph, I was concerned that very warm post-1993 instrumental values might pull the filtered curve up and away from the reconstructions - and they would appear to be a worse fit to the real temperatures than they actually are (because, without reconstructed values for the warm post-1993 period, they don't have the same opportunity to have their filtered value pulled up buy high later values). This is only of concern if the smoothed series are extended right to the end of each record by padding them past the end of the series. I could compute post-1994 values for the instrumental data, though only to 2000 because that version (CRUTEM1) of the Jones et al. data set was replaced by a newer version (CRUTEM2) a couple of years ago. I could of course replace the entire series with the CRUTEM2 data, but then it wouldn't be the series against which I originally calibrated the Overpeck data. I don't know how different the CRUTEM2 and CRUTEM1 data are for this regional average, so visually it might not make much difference. Perhaps its best if I send you both series updated to their ends (2000 for CRUTEM1 and 2003 for CRUTEM2) and you can decide what to do with them. >d. Is there a reason for using a 50-year filter when the standard climate >period is 30 years? Would it be a problem if 30 year running avg were used >(or also plotted) in that the normalization period is 1961-90? The choice of filter length is arbitrary, so a 30-year one is fine. At 03:13 20/08/2004, Mike MacCracken wrote: >Tim and Jonathan--In plotting up the "calibrated" Overpeck data provided by >Tim, he indicated that it was year by year data. So, we were going to do >some time-averaging (say running 30-year average centered on years 1585 to >1970 as data seem to go from well before 1585 to about 1985). We are >rechecking our technique, but doing this seemed to give the same curve as >Jonathan's plot (though with different scale) which we recall he had >time-averaged. I'm slightly confused here, as I think the data I sent start in 1600. Have you picked out the right column - it's the one labelled '5', which isn't the 5th column, because the first column is the year. And I think the last value is in 1990. I've just taken a look at the MS Excel file that you sent, and I think the problem might be that while the columns of data were imported into individual spreadsheet columns, the column headings were put into a single cell. Thus the label '5' isn't lined up with the column it should be above (in fact it appears to be lined up with the column that should be labelled '4'). >Can you confirm if that the paleo data are indeed year by year data, or are >they time-averaged? The file I sent has one value per year. But, Overpeck's original data had 1 value every 5 years, with the first labelled 1600 and the last labelled 1990. To facilitate plotting etc., I linearly interpolated between each 5-yearly value to provide values for intermediate years. Thus, although the file that I sent does NOT have any subsequent filtering applied to it, the data never included any information on time scales below 5 years. At 03:15 20/08/2004, Mike MacCracken wrote: >One additional reason for asking is that the paleo data only go to about >1984--which with 30-year averaging would mean real data could go to about >1998 but with centered averaging only go to 1984. How close to present do >the individual year data go? Just to re-iterate, the data I originally got had 1 value every 5 years. I guess that Jonathan can say more about what went into each of these values. Best regards Tim Dr Timothy J Osborn Climatic Research Unit School of Environmental Sciences, University of East Anglia Norwich NR4 7TJ, UK e-mail: t.osborn@uea.ac.uk phone: +44 1603 592089 fax: +44 1603 507784 web: http://www.cru.uea.ac.uk/~timo/ sunclock: http://www.cru.uea.ac.uk/~timo/sunclock.htm