Trying to date Roman time
An attempt to absolutely date the parts of the Irish oak chronology with dendrochronology

PRELIMINARY
Contents

Abstract

We can demonstrate that long and well-replicated oak master chronologies from north-western continental Europe show a significant correlation towards supra-long Scandinavian pine chronologies. This feature has been used to try to absolutely date the floating parts of the Irish oak chronology. The result for the tree-ring chronology part connected to the European Roman complex is unexpected and controversial.

Introduction

In a recent paper (Larsson et. al. 2013) we described our interpretation of the oak measurement series made available by Queen's University Belfast (QUB) on the Internet.

We synchronized three Irish chronology parts which do not crossdate among themselves. Each of the three had to be separately dated against published data:

  • BelfastAD, spans the time AD 25 to 2006, absolutely dated with living trees.
  • LateBC, a chronology floating in time, spans the time 1155 to 69 BC.
  • BelfastLong, a floating bog-oak chronology covering 4615 years, spans the time 5452 to 837 BC.
With the above dating, BelfastAD and LateBC do not overlap: there are no tree-rings between 69 BC and AD 25. We were able to convincingly extend LateBC towards modern time with English Roman raw data kindly provided by Cathy Tyers, Sheffield University/English Heritage. The correlation of the English data series towards BelfastAD was much weaker. Though we could reconstruct the proposed linkage (Pilcher et. al. 1984), we do not regard this as a proof that the link is correct, as we never reached sufficient t-values.

LateBC and BelfastLong do indeed overlap 316 years, but with a very unsatisfactory correlation (corr. 0.17, t=3.0). A deeper analysis showed that the proposed linkage (Baillie et. al. 1983) most probably is wrong and that a large but unknown number of years has to be inserted between the two chronologies.

Full details of our work are available in the Belfast section, which includes our synchronized mean value curves and also our dating reports at single sample level.

To proceed towards an absolute dendrochronological dating of all the three parts of the Belfast chronology, we have to find new material to bridge the gaps or we have to find other absolutely dated continuous masters to compare our chronologies with.

In a recent QUB paper (Brown et. al. 2012), the authors find that 25 years of random sampling of oaks from Irish bogs yielded about 5000 samples, but did not yield sufficient material to close the remaining gaps in the Irish oak chronology. So they do not think that there is much hope that additional sampling will resolve the problems. For them the gaps increasingly appear to be real.

German dendro labs (Friedrich et. al. 2004), (Leuschner et. al. 1984) claim that they hold long absolute oak master curves which probably match right away towards the Belfast chronologies. But none of the labs has ever demonstrated this and all the raw measurement data is unpublished.

There are also supra-long absolute masters of other species. One of these is a pine curve from Northern Finland (Eronen et. al. 2002), (Helama et. al. 2008) which was kindly made available to us by Mauri Timonen for research at single sample level. Therefore we were able to check its synchronisation and found that this master curve indeed is continuous over more than 7600 years. The pine curve from Torneträsk in northern Sweden (Grudd et. al. 2002), given to us as a mean value curve by Håkan Grudd, matches the Finnish master clearly.

Our challenge now is to find a match between Irish oak and Scandinavian pine, i.e. both interregional and interspecies. Is this possible, and under which conditions? The first step will be to check this out with absolutely dated, recent material.

Methods

We use our program CDendro for dendrochronological synchronization and dating. CDendro is optimized for handling large data sets and for search of matches in large tree-ring data bases. A number of well-tried algorithms for crossdating and normalization are available, among others Baillie-Pilcher and Besançon index E. We usually look at several methods simultaneously and require a high score from all of them to accept a match. Crossdating can be done under interactive supervision with graphical tools and quality check functions.

When building our master references, we follow the advices given by QUB on their web page in connection with the publication of the raw data, and the methods described in Mike Baillie's book A Slice Through Time. When building site collections, samples shorter than 100 years, with an overlap less than 70 years, or with a t-value less than 6 are generally excluded. (When it comes to archaeological wood, this is of better known origin than a trunc from a bog, but it is generally shorter in length. For wood of archaeological origin we have therefore set the limit for exclusion at 75 years. This is valid for Willy Tegel's data.)

Results

Interspecies correlations on the British Isles

All British recent (absolutely dated collections sampled from living trees) pine collections we could get hold of are from Scotland. They were all placed on the ITRDB by Fritz Schweingruber. We excluded collections shorter than 200 years. The following diagram shows the included collections and their correlations towards each other and towards the sum of the rest.


(Olap=Overlap. The correlation coefficients are calculated as a mean from four different normalization methods.)

The British recent oak masters are our BelfastAD and an Irish, a Scottish and an English master from the East Anglia incident (see references).





As we believe that we need long and internally well replicated collections to get reliable interspecies correlation, we combined the pine collections to form "ScottishPine", and the oak collections to form "BritishOak". These two collections match each other with corr. 0.15, t=2.7 at 305 years overlap.

I.e. a corr coeff around 0.15 is what we can expect for an interspecies match on the British Isles, this means we need about 2000 years long series to see a t-value of 7!

Interregional correlations British pine towards Scandinavian pine

How strong is the correlation between pine collections from the British Isles and Scandinavia? We matched ScottishPine towards the Finnish supra-long pine curve and got corr. -0.01 at 304 years overlap. ScottishPine towards Torneträsk gave corr. -0.02. This means: No correlation what so ever between pine from Scandinavia and the British Isles! So we would have to find a geographically closer, absolute pine master. The Scottish Pine Project of St. Andrews Tree-ring lab is said to have the potential for a supra-long chronology for the Scottish Highlands, but it will probably take several years to reach that goal.

Interregional/interspecies correlations European oak towards Scandinavian pine

Is it possible to come closer to Scandinavia by using intermediate master collections from different European regions? We compared a number of recent oak masters with each other and with the combined pine masters from Finland and Sweden (ScandinavianPineAbs).





Again, bad correlation of the British Isles towards Scandinavia. The big surprise is the Danish oak master (Note 1.) which shows high correlations both towards European oak and Scandinavian pine, the properties of a good intermediate! But also the masters from Hamburg, Lübeck and France (from the East Anglia incident) show useful correlations, provided that the matching series are sufficiently long. Conclusion: there might be a possibility to match Scandinavian pine directly with north-western continental European oak, and such oak has a good chance to match Irish oak.

How significant is the "right" match of our own, trusted European oak masters towards Scandinavian pine? HollsteinValidatedExtended (see references), 1565 years long and dated to AD 1974, shows a corr. 0.05, t=2.2 for offset 0 (= "as dated") and appears as best match only if the end year interval is limited to ±50 years. A new master (Note 2.) compiled from French data included in the QUB material, and a lot of series from north-eastern France put on the DCCD by Willy Tegel, shows a better result: corr. 0.07, t=2.7 at 1431 years overlap. This match appears on fourth position if the end year interval is limited to ±500 years. First the combination of these two masters, plus BelfastAD, NorthEnglishMaster and EnglishMaster, gives sufficient significance to continue the experiment: third best match only if the end year interval is limited to ±500 years, alternatively on position 7 without limits (1984 years tested against any position of the 7638 years long reference curve).





Absolute dating of the European Roman complex and LateBC

For the first millennium BC and the first centuries AD we have about the same mixture of quite long validated masters. Our floating LateBC crossdates convincingly towards the European Roman time complex. This complex consists of a number of strong regional collections with good intercorrelation.



We combine a new master from north-eastern France (TegelRoman (Note 3.) ) with HollsteinValidatedRomanExtended (see references) without its KORB-member (which we suspect is corrupt in its younger end), NorthEnglishRoman, SouthEnglishRoman and LateBC to a 1441 years long mean curve with an assumed end year AD 286. As derived from the previous experiment with the masters of similar sites but with an absolute end date in recent times, we expect a match with about corr. 0.07 and a t-value of about 2.7 at least among the twenty best positions if the end year interval is limited to ±500 years.

Tested towards the Scandinavian pine master, offset 0 (= "as dated") does not appear in the list of scores (corr. 0.03). Instead there is a strong, distinct best match at offset 218 years which has a t-value of over 5 for the P2Yrs and Hollstein normalization methods. This match is also on top if no limitations are applied (1441 years tested against any position of the 7638 years long reference curve).




In order to extend the Roman collection with about 40 years, we touch up with a small collection of late Roman material from London (Note 4.) (assumed end year AD 328). This strengthens the match even further, P2Yrs and Hollstein now reach t-values of 5.8 and the discrimination is good for all normalization methods. We reach t-values as high as 6.5 with only the Finnish master.



To see if this high score is real or volatile, we apply the new offset to all involved subcollections and compare each of them with the Scandinavian pine master.



This reveals that all collections except NorthEnglishRoman, independent from each other, show unexpected high correlation towards Scandinavian pine at 218 years offset. The diagram also shows how the French and German Roman collections correlate to the Irish LateBC via the English Roman masters.

We also make a block analysis of this match between European oak and Scandinavian pine (block length 400 years, block distance 50 years), which demonstrates that the match is consistant over its whole length and that the assumed end year AD 328 does not appear at all.



(The Best around-column indicates if there is a better match within an offset of ± 2 years. In this case all blocks are at their best matching points.)

Absolute dating of BelfastLong

Beyond the first millennium BC, we have only Irish (BelfastLongVersion2) and matching English bog oaks (EnglishBog) to compare with Scandinavian pine. But the collection is well replicated and 4616 years long, so we try and see what's happening. Again we expect a match with maximum corr. 0.07 which would give a t-value of about 4.5 among the twenty best positions if the end year interval is limited to ±500 years. The assumed end date of the collection is -836 (837 BC).



The offset 0 (= as dated) does not show up (corr. 0.01). Instead there are twenty other matches with a very low correlation and poor discrimination.


Discussion

Dating of the Roman time complex

The match of the European Roman complex with LateBC against the absolute Scandinavian pine master looks like a "normal" interregional, almost acceptable match. Maybe the correlation is a bit low. But remember: this match is not only interregional, it is also between species! And it does not confirm a well-established truth (the 0-offset does not show up), on the contrary it turns our calendar upside down as it redates the Roman complex 218 years closer to our time. So, this match must be wrong. Or is it true?

Firstly, is it at all possible to see such high correlations between oak and pine over considerable distance? The answer is yes, as the Danish oak master shows corr. 0.15, t=6.6 at 1784 years overlap towards pine from high up in the Scandinavian north. With increasing distance, the correlation becomes weaker but seems to be significant for the north-western part of continental Europe, as the comparison with a number of different absolute masters demonstrates. However, the masters have to be sufficiently long and internally well-replicated.

Secondly, we would not have expected fully so high correlations for the regional mixture in our Roman complex. Though it is maybe just because we are dealing with Roman time material that we get better correlations. Do we know the real provenance of the samples? The Romans needed enormous amounts of timber to build their fortifications, therefore they certainly were in the timber trade. Mike Baillie (1995, p.35) remarked on this regarding the timbers of the Carlisle fort: "We began to wonder if the Romans created stores of wood from all over the Empire and then sent wagon loads, suitably mixed, to outposts like Carlisle!"
We would expect increasing correlations, if we had e.g. a certain percentage of timber from the Elbe valley in our collections. This can be illustrated with a 155 years long oak collection from Vehlow in Brandenburg, Germany (Note 5.), which shows a corr. coeff. 0.13 against Scandinavian pine. Or why not timber from the Southern Scandinavia costal region?

Thirdly, do we see a replication of the 218-years offset between our regional oak collections?
At least not too obviously, otherwise this match would have been observed long before.
Ireland: LateBC, redated to AD 150, shows corr. 0.10, t=1.1 against BelfastAD at 123 years overlap. Northern and Southern England: no sufficient overlap. Hollstein: HollsteinValidatedRomanExtended without KORB, redated to AD 504, shows corr. 0.18, t=1.7 against HollsteinValidatedExtended at 92 years overlap.
Tegel: TegelRoman, redated to AD 435, shows corr. 0.28, t=2.7 at 90 years overlap towards FranceAbsoluteAD, and corr. 0.27, T=2.9 at 113 years overlap towards an absolutely dated Rhine valley collection (R3) of Bernd Becker which we could get hold of. However, another of Becker's collections from the Danube valley (D4+D5), dated AD 679 to 338 BC and stretching over the Migration gap, confirms that there probably is an error in the South-German chronology and even its size: a 110 years long signature is repeated with 218 years interval. (Note 6.)

Dating of BelfastLong

The 0-offset does not show up either here, as it definitely does for the absolutely dated AD-masters, among the best matches of BelfastLong against the Scandinavian pine master, but in this case the correlations are really low and none of the matches is sufficiently discriminated from the others.

On the other hand we might get a hint if we could find an alternative match against LateBC. We redate LateBC according to its new, absolute date 218 years towards our time (new end date AD 150). Then we crossdate BelfastLongVersion2 against the re-dated LateBC and see if one of the found offsets against Scandinavian pine also shows up now. The only one that appears in both cases is offset -8 years. It has corr. 0.18, t=1.8 at 90 years overlap, which would mean that we have to insert 8 + 218 = 226 years in the gap at 950 BC.

Conclusions

Our proposed dating of the three parts of the Belfast oak chronology is based on dendrochronology only. The currently accepted QUB interpretation of the same raw data is based on radiocarbon measurement and subsequent fine-adjustment using dendrochronology (Baillie et. al. 1983). Though we in principle arrive at the same absolute date for the long prehistoric curve as the QUB researchers, and fully agree on the dating of the AD collection, our dating of the middle collection connecting to Roman times has an offset of 218 years compared to QUB's interpretation. Moreover, our interpretation provides a feasible solution for the error in the gap at 950 BC which we described in our previous paper (Larsson et. al. 2013).

Though our interpretation is the better alternative from a dendrochronological point of view, it is too early to regard it as a proof that it is real. Especially the exact offset of the long prehistoric collection has to be substantiated by replication, e.g. by comparison with a future absolute pine master from Scotland. As it is now, the -8 years offset is only a qualified guess.

Regarding the middle collection, we expect that all long, internally well replicated Roman oak master collections from north-western continental Europe show an offset of 218 years and not "as dated" among their top matches against absolute Scandinavian pine master curves. Suitable material to demonstrate this could be found e.g. in the still unpublished data sets from north eastern Germany used in a recent article by Buentgen et. al. (2011).

And of course, the demonstration of a trustable, supra-long and absolute German oak master would make the end of the discussion.

April 6 2013. Petra Ossowski Larsson, Lars-Åke Larsson

References cited

Baillie, M.G.L, Pilcher, J.R. and Pearson, G.W., 1983. Dendrochronology at Belfast as a background to high-precision calibration. Radiocarbon 25 (2), 171-178.

Baillie, M.G.L. 1995, A slice through time - dendrochronology and precision dating. ISBN 0713476540.

BelfastAD, LateBC, BelfastLongVersion2 and all English masters incl. dating reports are available at http://www.cybis.se/belfast

Brown, D.M. and Baillie, M.G.L., 2012. Confirming the existence of gaps and depletions in the Irish oak tree-ring record. Dendrochronologia 30 (2), 85-91.

Buentgen U, Tegel W, Nicolussi K, Mccormick M, Frank D, Trouet V, Kaplan J O, Herzig F, Heussner K U, Wanner H, Luterbacher J, Esper J, 2500 Years of European Climate Variability and Human Susceptibility, Science Express, 13 January 2011. doi:10.1126/science.1197175

CDendro and CooRecorder available at: http://www.cybis.se/forfun/dendro

East Anglia incident see: http://www.cybis.se/wiki/index.php?title=East_Anglia_Incident_chronology_files

Eronen, M., Zetterberg, P., Briffa, K.R., Lindholm, M., Merilainen, J., Timonen, M., 2002. The supra-long Scots pine tree-ring record for Finnish Lapland: Part 1, chronology construction and initial inferences. The Holocene 12(6): 673-680.

Friedrich, M., Remmele, S., Kromer, B., Hofmann, J., Spurk, M., Kaiser, K.F., Orcel, C. and Küppers, M., (2004). The 12,460-year Hohenheim oak and pine tree-ring chronology from central Europe. Radiocarbon 46 (3), 1111-1122.

Grudd, H., Briffa, K.R., Karlen, W., Bartholin, T.S., Jones, P.D., Kromer, B. 2002. A 7400-year tree-ring chronology in northern Swedish Lapland: natural climatic variability expressed on annual to millennial timescales. The Holocene 12(6): 657-665.

Helama S., Mielikäinen K., Timonen M. & Eronen M. (2008): Finnish supra-long tree-ring chronology extended to 5634 BC. Norsk Geografisk Tidsskrift-Norwegian Journal of Geography 62 (4): 271-277.

The collection can be downloaded here: http://lustiag.pp.fi/ClimateFromTreeRings_gb.htm (data available in March 2013)

An alternative download document is: http://lustiag.pp.fi/download.htm Look for the "Finnish over 7500-yr pine chronology."-link (a .pdf-document) and there look for a file named "adv7638.tuc" referred to as "Dataset2/Extended Ring-width dataset" or "The 5634BC chronology". You might have to edit it to make it readable in CDendro as the identity is missing for one of the recent collection members. (A simple solution might be to remove these lines with a text editor.) The original collection contains 1484 members. (Updated August 2014.)

Hollstein's validated masters are available in the Hollstein section at http://www.cybis.se/forfun/dendro/hollstein

Larsson, P.O. and Larsson, L-Å., (2013). Evaluation of the Irish oak chronology and its linkage. Tree-ring Research, in review. See also the Belfast section at http://www.cybis.se/forfun/dendro/hollstein

Leuschner von H. H. & Delorme A. (1984). Verlängerung der Göttingen Eichenjahrringchronologien für Nord- und Süddeutschland bis zum Jahr 4008 v. Chr. Forstarchiv 55, pp. 1-4.

Pilcher, J.R., Baillie, M.G.L., Schmidt, B. and Becker, B., 1984. A 7,272-year tree-ring chronology for Western Europe. Nature 312 (5990), 150-152.

QUB measurement series available at: http://chrono.qub.ac.uk/Resources/dendro_data/dendro.html

Schweingruber, F., ITRDB swed019, brit015, brit017, brit018, brit024 and brit026 on http://www.ncdc.noaa.gov/paleo/treering.html

Scottish pine project: http://www.st-andrews.ac.uk/~rjsw/ScottishPine

Tegel, Willy: 122 objects placed 2 june 2012 in the DCCD data repository at http://dendro.dans.knaw.nl

Notes

Note 1: WestDK, Chronology for Jylland and Fyn 200-1986, by Nationalmuseum in Copenhagen. Was available at their web site at least until 2009, but now seems to have disappeared. See also Cybis Wiki on WestDK

Note 2: FranceAbsoluteAD, a mean value curve of site collections incl. dating reports is available in the download section below. All sample identifiers with prefix "Q" or other letters denominate QUB measurement series; all other samples are retrieved from Willy Tegel's projects placed on the DCCD in 2012. As these samples are of archaeological origin, they are generally shorter and we therefore included samples down to 75 rings length in our site collections. The mean value curve is discontinuous but well replicated by HollsteinValidatedExtended.

Note 3: TegelRoman, mean value curve of site collections incl. dating reports is available in the download section below. All samples are retrieved from Willy Tegel's projects placed on the DCCD in 2012. As these samples are of archaeological origin, they are generally shorter and we therefore included samples down to 75 rings length in our site collections.

Note 4: BaynardChamb, Billingsgate1 and Tower from BitsPieces available in the Belfast section. Please note that these are the same collections making up the currently accepted bridge between Roman time and recent time (BelfastAD). The match towards BelfastAD has corr. 0.32, t=5.0 at 225 years overlap.

Note 5: Vehlow.fh, see the download section below.

Note 6:

Auwaldeichen Rhein (R3), AD 624 to 317. Described in: Becker, B. (1981), Fällungsdaten Römischer Bauhölzer anhand einer 2350-jährigen Süddeutschen Eichen-Jahrringchronologie, Fundberichte aus Baden Württemberg No.6, 369-386. The following diagram is taken from that article.


The youngest 110 years of TegelRoman show a double match against Auwaldeichen Donau (D4 + D5), both with offset 0 (= as dated) and offset 218:


Seems that Roman time has been "played" twice with 218 years interval, and that the error is located around AD 300 in the timber shortage period called the "Migration gap".



Dating reports for published mean value collections:

The purpose of these reports is to enable check of our work. Each link will open within a new tab.
Note: These tables and diagrams show how a large number of site collections are dated and how their individual members (samples) are crossdated towards each other. Some members of these collections do not match or overlap good enough towards the rest of their collection. In these cases their dating has been set by good matching towards the master or towards other site collections.
  1. FranceAbsoluteAD
  2. TegelRoman



Files to download:

Download file package for testing oak to oak and oak to pine matches

To get the Finnish pine data, see (Helama et. al. 2008)

You might check also with the ITRDB Swed019 curve (Schweingruber & Bartholin) from Torne Träsk, though this covers only AD 441-1990. It can be used to demonstrate the match towards the FranceAbsoluteAD curve which it matches with corrCoeff=0.08 (T=2.9) 1309 years. The 218 years offset match (RomanAll) overlaps with only 103 years but gives a correlation coefficient of 0.25.

The Danish WestDK oak curve, see (Note 1.) can be matched towards the Swed019 curve and then gives a corrCoeff=0.15/T=6.0.

Alf Bråthens oak curve from Western Sweden (see Cybis wiki on BrVQ AD 867-1975 gives a corrCoeff of 0.16 towards Swed019.



Last update: August 27 2014 (see Helama et. al. 2008)