KEY URLs

Computer Package

Source Data

The matrix of 112 proxies as received by the authors from Scott Rutherford, as requested by Prof. Mann, is in the file pcproxy.txt, which is about 1 MB. You can download a dressed-up Excel version (3 MB) for easy viewing of the problem areas from here. This is identical to the file pcproxy.txt was located at Mann's FTP site at the URL ftp://holocene.evsc.virginia.edu/pub/sdr/pcproxy.txt on October 29, 2003. Rutherford also sent the following file showing the roster of proxies.  The data in this file is identical to the data in the Mann FTP file ftp://holocene.evsc.virginia.edu/pub/MBH98/PCS/multiproxy.inf. 

The Computations

A step-by-step guide to the R scripts that generate Figures 6 and 7 is here at the computations page.

Response by Prof. Mann to File Issues  

Audit Issues - Jan. 14, 2004

The audit issues listed below were stated at the time of our E&E article. A considerable amount of new information has surfaced in respect to the data and methodology of MBH98, especially with respect to the MBH98 data at Mann's FTP site.  Here is the status of the following issues (italics are Jan. 14, 2004 comments).  See a step-by-step explanation of the errors in the data base is here at the audit page.  The comments compare the data in the file pcproxy.txt formerly located at Mann's FTP site and the data in FTP/MBH98.

1.       Does the database contain truncations of series 10, 11 and 100? (and of the version of series 65 used by MBH98)? 

        FTP/MBH98 contains two parallel versions of series 10, 11 and 100. The versions of series 10 and 11 collated into pcproxy.txt occurs as FTP/HIST/MANNETAL97/england.dat and FTP/HIST/MANNETAL97/europe.dat. Both are truncated from the source data and are summer data (see issue #10). Non-truncated versions of annual data also occur at the FTP site as FTP/HIST/cengtemp.ann and FTP/HIST/pfister.ann.  The Central Europe data at FTP/MBH98 goes only to 1979, but a more up-to-date version is at WDCP.  For series 100, both truncated and non-truncated versions are at FTP/TREE/ITRDB/MISC as series chin04.txt and chin04-orig.txt. The version of series 65 is identical to the version of this series at Briffa, but is truncated; there is no duplicate version of this series. It seems clear that the truncated versions of these series has been used.  

2.       Are the 1980 values of series #73 through #80 identical to 7 decimal places? Similarly for the 1980 values of series #81-83?  And for the 1980 values of series #84 and #90-92? What is the reason for this?   

        The file pcproxy.txt formerly located at Mann's FTP site, to which we were directed as being the data used in MBH98, definitely contains the problems described, but MBH have stated that this file was not actually used in MBH98 calculations. Series #72-80 and #84-92, as well as series #81-83, in pcproxy.txt are spliced from series contained in subdirectories in FTP/MBH98. For example, series #72 is a splice of the pc01.out series in subdirectories BACKTO_1400, BACKTO_1450, BACKTO_1500, BACKTO_1600 and BACKTO_1700 contained within the directory FTP/TREE/STAHLE/SWM. We identified this splicing and discussed it in our first comment on November 12, 2003.  A collation error was introduced in this splicing process, resulting in identical 1980 values in many series. The collation error itself only created an offset of one year and would not lead to major consequences in the MBH98 reconstruction; however, as a quality control check, it led to the identification of the splicing problems in pcproxy.txt which do have effects on the MBH98 reconstruction. In our own collation of an "MBH98" database, we were obviously very attentive to the defects in these principal component series. We accordingly re-collated all of the underlying tree ring series available at WDCP and calculated principal components using a standard algorithm to avoid this problem. Professor Mann, through David Appell, has disseminated the false information that we failed to observe the defects in these series and that our own calculations were thereby compromised ("garbage in, garbage out".) However, we studiously avoided use of these compromised series in pcproxy.txt. Professor Mann has provided no evidence beyond his assertion that the error made in collation of pcproxy.txt was not also made in the MBH98 calculations. Despite this lack of concrete evidence, we believe that it is quite possible that the PC series in the various FTP/MBH98 were read in correctly to the MBH98 calculations and that this issue pertains only to the pcproxy.txt collation. Professor Mann has also claimed that this collation was made at our request, because we did not want to use the FTP site. This is false. The files pcproxy.txt and pcproxy.mat on Mann's FTP site were dated August 8, 2002, long before our request, and shortly after the apparent commencement of Mannb's FTP site on July 30, 2002. The files have now been deleted. There was also a reference to the file pcxproxy.txt in a graph on Scott Rutherford's webpage in 2001 (according to the Wayback archive).  We made no request for a collation and wanted to see the data as it was used in MBH98; indeed, not only did we not request a collation, we did not want one. Rutherford's email of April 2003 refers to subdirectories, but notes that these references would be of no use to us.  The exact date at which the directory FTP/MBH98 was located on the public part of the server is very much an open question. Professor Mann has refused to provide this date and the webmaster at UVA says that he has no information as the site is not under his control. 

3.       Where are the calculations of principal components for series in the range #73-92 that would show that these have been collated into the correct year? Do you have any working papers that show these, and if so, would you make them FTP or otherwise publicly available?  

        Original principal component series are located in the BACKTO subdirectories (referred to in Issue #2) and are currently publicly available at FTP/MBH98 (see Issue #2). As discussed in our Response-Part 1, there are many more PC series in these subdirectories than actually used in MBH98. MBH stated in November that the principal components for each region were calculated separately for each reconstruction interval.  This is not possible, as there are insufficient subdirectories to correspond to each subinterval and there are inconsistencies between the subdirectories available for (say) the Stahle/SWM, NOAMER and Vaganov networks. MBH have also stated that they used 159 series, although there is no such information in Nature. After subtracting 81 non-PC series which are adequately identified, this leaves 78 PC series to be identified from the subdirectories.  A complete specification of the stepwise procedure therefore requires a table setting out the number of PCs used from each region in each reconstruction interval and a second table showing the subdirectory used. We have requested this information from MBH without success. It does not appear possible to fully reconcile the various pieces of information regarding the stepwise reconstruction. For example, information is provided at Nature SI regarding the number of indicators used in each interval. For the non-PC indicators, the number available in each interval is known, permitting the number of PC indicators to be calculated by difference. This indictes 3 PC series in 1400 and also in 1450, but the Nature SI shows the Vaganov PC1 initiating in 1450. We have been unable to reconcile the alleged use of 78 different PC series with the number of series shown in Nature SI for each interval. We surmise that either there are subdirectories used in MBH98 which are missing from FTP/MBH98 or the 159 series is incorrect, but cannot determine which on present information. For our own further analyses, we have made an estimate of the required tables, reconciling to the Nature disclosure as much as possible.   

        The FTP/MBH98 actually contains Fortran programs used to calculate the principal components. We have looked closely at these programs, which contain highly unusual procedures, and plan to report on these in detail. We believe that this will prove to be very important in understanding MBH98 results. 

4.       Do the following series contain "fills": #3, #6, #45, #46, #50-#52, #54-#56, #58, #93-#99? 

        All of these series contain fills in the FTP/MBH98 as well as in pcproxy.txt. There is a notice in the SI that certain series are extended through "persistence" at the end of their range  i.e. continuing the last value to the end. There is obviously no need for this extrapolation in making calibrations and it introduces an unnecessary risk of introducing a bias, but the effect is probably not a large one. Series #3 is different in that there is a significant gap of 13 years in the series between 1952 and 1965, which is interpolated. This gap points to a potentially more serious problem in that two completely different corals are used before and after this gap and the two series are inhomogeneous, with one series substantially adjusted. The calibration of this series using these different corals is highly vulnerable to the adjustment factor.  Additionally, the MBH98 data contains a unique unannotated "front-end" splice of series #53 for the years 1400-1403, which appears to have been done for the sole purpose of influencing the AD1400-1450 results, upon which this plug has a material effect.

    5.       How did you deal with missing closing data in the following series: #11, #102, #103, #104, #106 and #112?  

        These appear to have been filled in the same way as above (Issue #4), but at a later stage in the calculations (presumably, during the temperature index calculations themselves). Again, we do not expect that this will make a big difference, but, again, there seems little point in introducing unnecessary extrapolations into either calibrations or the estimation of he MBH98 temperature index in the 1970s. 

6.       What is the source for your data for series #37 (precipitation in grid-box 42.5N, 72.5W)?  Did you use the data from Jones-Bradley Paris, France and if so, in which series?  More generally, please provide, identifications of the exact Jones-Bradley locations for each of the series #21-42. Where are the original source data? 

        This issue has been discussed at some length at sci.environment (see Auditing the Auditors #6). Series #37 (from a grid-box in New England) appears almost certainly to be a transformed version of the Paris precipitation data at GISS. It matches exactly to the GISS2 data ( and substantially but not exactly to the Climate since 1500 data), but it probably from the GISS1 dataset (which remains unchanged) and the existence of this error in MBH98 is confirmed. The attribution to Jones-Bradley does not seem to be correct. A virtually identical misplacement of the Toulouse precipitation data to North America as Series #34 is certain. These misplacements can be summarized as: "The rain in Maine falls mainly on the Seine."  The most likely identification of these sites, in my view, is the following, with geographical errors or identification issues bolded. The exact edition of some of the underlying series has not been determined in all cases. At the FTP site, one precipitation series FTP/INSTR/PREC/prec-1820-02.dat is excluded without any explanation.

        In M&M, we verified connections, but not identity, between Climate since 1500 data and several proxy series. The actual MBH98 data seems to be derived from data at GISS, but not necessarily to the version currently at their FTP site. However, all sites are reasonably firmly located. Two series shown at the FTP site are not included in the MBH98 calculations for reasons not stated, including the well known De Bilt series.. 

7.       Did you use summer (JJA) data for series #10 and #11 rather than annual data. If so, why?    

        See issue #1 above.

8.       Does your dataset contain obsolete data for the following series: #1, #2, #3, #6, #7, #8, #9, #21, #23, #27, #28, #30, #35, #37, #43, #51, #52, #54, #55, #56, #58, #65, #105 and #112?    

        The existence of these obsolete versions is not just in pcproxy.txt, but is also in FTP/MBH98. The use of the obsolete version of series #56 has been stated by MBH to be important in their 15th century results; however, the version at WDCP is more recent and includes much more recent data.

9.       Do you use the following listed proxies: fran003, ital015, ital015x, spai026 and spai047?  If so, where?  

        These series are not used in the FTP site. In addition, we have identified many series, stated in SI to have been used in the tree ring networks for PC calculations, which were not used, 

10.   Did you commence your calculation of principal components after the period in which all dataset members were available for the following series: #69-71, #91-92, #93-95, #96-99?

11.   What is the basis for inclusion of some tree ring sites within a region in regional principal component calculations and others as individual dataset components?    

        No satisfactory answer to this appears on the horizon.

12.   Did you commence your calculation of principal components before the period in which all dataset members were available for the following series: #72-80, #84-90? If so, please describe your methodology for carrying out these calculations in the presence of missing data and your justification for doing so?  

        MBH have now stated that they reduced the roster of sites in earlier periods in order to calculate principal components. This deals with this issue, but see Issue # for residual issues.

13.   What is the explained variance under your principal component calculation for the period of availability of all members of your selected dataset?  Would you please make  your working papers that show this FTP or otherwise publicly available?

        See issues # above. 

Further Issues -January 14, 2004

1.    Please provide a table showing the exact number of PCs used in each region for each reconstruction interval, together with a table showing the subdirectory used for each region in each step.

2.    Please explain why subdirectories are not available for every reconstruction interval, if PCs were calculated separately for each reconstruction interval, together with the reasons for the specific choice of periods.

3.    Please explain the inconsistencies between the series stated to be used at SI and the series actually used (according to the FTP site).