Validating the dendrochronological dating of Roman oak wood in western Europe

by Torbjörn Axelson , January 2011 (updated Jan.16. and Jan. 29.)

There are huge amounts of dendrochronological oak samples from all around Europe from recent and down to early medieval time. There are also great amounts of samples found in Roman archaeological contexts. But the connection between those complexes is weak. It is a severe lack of useful samples during the late Roman time. In most cases it seems hard to extend the chronologies from present time down to more than 4'rd or 5'th century AD, and the Roman chronologies up to more than about AD 150 – 300. The only publicly available chronology covering the gap is the South German curve, by Berndt Becker, and after the “East Anglia Incident” 2009, also available in an updated version, there named “A100”. But is this chronology and its bridging of the late roman gap correct? The answer of that question is not obvious. The first published chronology with site-wise chronologies possible to examine is those by Ernst Hollstein, 1980. They were retrieved from diagrams and made available online by Lars-Åke Larsson some years ago. Those data appeared generally to be of high quality and two useful collections were built by him: One from present time and down to AD 401, “HollsteinBestLateToYear401” and another 340 BC – AD 336, “HollsteinBestRoman”. The bridging of the gap AD 337 – 400, presented by Hollstein, turned out to be extremely weak if at all existent. Instead an unexpected and very strong match for the “HollsteinBestRoman” towards “HollsteinBestLateToYear401” 207 years too late appeared! The correlation between them, T=7.4 (Proportion of last two years normalization – if other not is stated this method is the one referred to in this article), T[BaPi]=6.1, overlap=142, would normally be regarded as a safe level. Despite the incompatibilities between this “-207 years offset” and the historical records, I have been looking for corresponding matches in other chronologies but not been able to find anything at all at this offset, yet. In my opinion the -207 offset is due to some mistake by Hollstein, although it is unclear to me what it is.

The bridge proposed by Becker was confirmed 1983 by Baillie et al., Queens University, Belfast, reporting a strong match (T[Baillie/Pilcher]=6.5) between the old end of their chronology for Northern Ireland from present down to 13 BC and a Roman chronology for Southwark, southern England, by Ian Tyers 252 BC – AD 255. The same bridge was also confirmed by Leuschner and Delorme in Göttingen 1984/85 (See for instance Baillie: A Slice Through Time, 1995, pp. 36 ff.) None of those chronologies are available for “outsiders”.

From a very suspicious point of view, the objection could be risen, that it all is a conspiracy. Maybe driven by piety – or fear – not to destroy the accepted historical chronology. To me this seems rather unrealistic. For most scientists it would have been a very exciting thing to be able to do such a fantastic discovery as to prove the Western European AD-chronology being some hundreds year shorter than expected. And in 1980th certainly no one doing such a discovery, would have risked a fate like that of Galilei in either West Germany or United Kingdom! But what would certainly have been worth to fear, was the situation when other dendrochronologists, sooner or later, would have found such severe errors in his or her published chronology – and in this case it would not even have helped locking in the measurements and chronology itself. Not even that the honor for the discovery would be given to someone else, but far more the shame for doing such mistakes (intentionally!), would surely have been a real nightmare. As I also have seen some of the hard words written against a Swedish dendrochronologist making mistakes when compiling a chronology for Western Sweden at the same time, I can not really imagine anyone intentionally make himself the target for such potential criticism – especially without any obvious benefit. If the time span covered by the AD-chronology would be significantly shorter than expected, there would of course have been a lot more overlapping samples. So the risk not finding such a bridge, if the true one, seems also very small in my opinion.

Anyway, we are now in the fortunate situation that we have not only to relay upon texts and psychological considerations if the generally accepted bridging is correct or not. We also have the opportunity to (to some extend) independently investigate the link ourselves, using openly published measurements and chronologies – and also see if there may still be any alternatives, resulting in a shorter or a longer “AD-chronology” than the accepted one.

A Danish – Dutch – German bridge

There is a chronology for western Denmark (Jylland and Fyn), WestDK AD 200 – 1986, by the Danish National Museum. According to its description, this chronology is built solely on Danish material. It is also mentioned that the Museum has a longer version of this curve down to 109 BC, but not published as it was in need of external [secret?!] references to be verified. This chronology do overlap the gap between the two Hollstein chronologies, although the distance is to long to suggest any great expectations. Fortunately there are two collections available in the ITRDB from the Netherlands by Ester Jansma et al.: neth002.rwl and neth006.rwl, which covers AD 248-395 (as dated). They dates towards WestDK, although with not too impressing values: corr=0.31, overlap=147, T=4.0. I.e values not strong enough to certify a dating, although the block AD 280-362, gives corr=0.56, overlap=82, T=6.0 – and as it is not uncommon that distant chronologies, as those, sometimes are varying according to corr values from period to period in a way like this.

The relation between the Netherlands AD 248 – 395 and HollsteinBestRoman gives: overlap=59, corr=0.40, T=4.1. That is not a strong match – but a match. Towards “A100”, Southern Germany, the overlap is the all available 147 years, but only with corr=0.34, which gives T=4.4. This means that there is a weak match at the point proposed by Becker. (At offset -207 is nothing of interest).

Northern Ireland – England bridges

In April 2010 Queens University Belfast (QUB) decided to make more than 9500 pieces of raw data measurements available to the public. This data contains material of which a lot of complexes can be built up, among them a North Irland/Irland curve extending back to at least first century AD. The oldest end, which at this stage were identified, with guidance from available meta data, consists of chronologies from five sites: Teeshan, Mill Lough, Balloo, Alistragh and Ballinderry. All are Northern Ireland sites.

This chronology do not match any of the continental Roman chronologies available, but there are now also two collections of site-wise English chronologies available, one for Southern England (London area) and one for Northern England. Most of them are chronologies assembled and published by Petra Ossowski Larsson, using data from English Heritage / Sheffield University, not yet free. From the QUB material it is possible to build at least two English Roman collections: Carlisle-B, 264 BC – AD 103 (included in the NorthEnglishRoman-collection) and Alcester-A, 139 BC – AD 99 (not included). The mean of the SouthEnglishRoman collection dates towards HollsteinBestRomans with T=7.8 and towards ITRDB:neth5+9+10+11+12+13 with T=4.1 all at AD 207. NorthEnglishRoman dates with T=6.0, 358 years towards SouthEnglishRoman at AD 103. The EnglishRoman chronologies are safely anchored in the Western European Roman complex.

If we try to test the five Northern Ireland chronologies as they are, towards the SouthEnglishRoman the result is not very convincing:


Cybis CDendro, Algorithm: Proportion of last two years growth (2,0,T
Correlations between chronologies and SouthEnglishRoman
dated to 207 with corr >= 0.12 and with overlap >= 125
Results sorted according to decreasing correlation coefficient values.

                         T-  Over
                Corr   Test   lap
all...          0.26   4.00   219   based on 5 members
Teeshan         0.31   3.57   125     581  (Crannog)
MillLough-AD    0.29   4.48   219     611  
Balloo_312AD    0.18   2.55   190     312  
Ballinderry-A   0.15   2.00   182     618  
Allistragh      0.12   1.51   168     337

When looking closer to the five Northern Ireland chronologies covering parts of the first century AD, we can see they have rather long tails in their old ends consisting of only one sample or sometimes two with low agreement. It means that the quality of the old end is not verified: there may be missed or extra rings, there may be untypical growth patterns due to non-climatic factors affecting the young trees or plants which produced those oldest rings. So a more fair match towards the English Roman chronologies, may be found if we cut away the weak and in a way unconfirmed old tails of each of our five chronologies. In the kept parts there are usually two or three samples depth depending on the visual agreement between them.

A mean of those five truncated site-wise chronologies, forms a collection, “BelfastAD-OldEnd-Truncated”, going back to AD 35. It gives a T=4.8 over 173 years (corr=0.35) towards the SouthEnglishRoman. 

                    T-  Over
            Corr   Test   lap
all...        0.34   4.66   172   based on 5 members
MillLough     0.36   5.10   172     611  Block 0-575 of MillLough-AD
Teeshan       0.30   2.51    66     581  Block 37-476 of Teeshan
Ballinderry   0.23   2.49   111     618  Block of Ballinderry_AD
Allistragh    0.17   1.84   112     337  Block 0-241 of Allistragh
Balloo        0.17   2.11   157     312  Block 0-261 of Balloo_312AD

Towards the NorthEnglishRoman, however, the correlation value is unexpected low, only 0.16 over 68 years. Despite that, here is actually another weak bridge over the Late Roman gap, at the point predicted by Becker.

Lisbarnet Cottage

The next step was to try finding anything which may extend the “BelfastAD-OldEnd-Truncated” backwards. To do this, I decided to scan through the entire database of about 9500 samples towards a block out of the older part of the mean sample of them. This method gives a number of samples with T>5, of which all with overlap less than 80 are to be neglected – to check those short sample would just be a waste of time as most of them are false – as well as those known to be from other geographical contexts than around Northern Ireland, and those from other time spans as well as those not being oak (i.e pine-samples). The rest needs confirmation, because some of them may still be false high. Anyway, in this run two samples from Lisbarnet cottage come up as the two best ones(!): Q2672 (T=6.0 over 116 years) and Q2677 (T=5.5 over 103 years). They cross-date each other weakly at the implied position, which means both are there “independently” of each other! The total number of samples marked “Lisbarnet” or “Lisbarnet Cottage” are just 18, so the chance to have two of them in the very top of list, out of those several thousands samples is small indeed! From Q2677 as a base, a collection consisting of Q2666, Q2667, Q2677 and Q2680 could be constructed. Q2672 which gives T=6.0 at AD 161 towards the “BelfastAD-OldEnd-Truncated”, matches fine towards the full length Balloo (T=5.1, 111 years), but just T=4.4 towards the rest of the Lisbarnet-A colletion, although a bit higher with Ba/Pi or BesIE-normalization. A possibly problematic block is the youngest about 20 years of Q2667, which forms the younger end of the Lisbarnet collection. If that block is removed, the T-value for the whole Lisbarnet collection towards Balloo will increase to 6.8 (T[BaPi]=8.8).

It may be noted that also another Lisbarnet-collection (Lisbarnet-B) is possible to build, consisting of Q2665, Q2668 and Q2674, with a strong match towards the very long Northern Ireland BC-chronology. The samples are taken from bog oaks used in the roof of a cottage from 18'th or 19'th century, in Lisbarnet about 15 km south east from Belfast and just a few km north of Balloo.

I did an other search through all the QUB-samples now using the Lisbarnet-A as reference, finding Ballinderry Q9848 (AD 171, T=5.35, 119 years) not previously included in the Ballinderry collection. This sample turned out to be a key-sample. Another interesting match is Carlisle Q4965 (AD 42, T=5.11, 91 years, which is how it is dated also within the Carlisle collection, i.e in the Roman context).

Ballinderry

Starting from the Ballinderry Q9848 sample, a collection called Ballinderry-B, consisting of five samples: Q9848, Q9878, Q9893, Q9894 and Q9895, could be constructed covering 183 BC – AD 171, although the old end is thin – only one sample until 148 BC and two until 41 BC. The two long samples diverge rather much during the period before about 100 BC, so that part seems weak. This new Ballinderry-B correlates with T=5.0 over 136 years towards the “BelfastAD-OldEnd-Truncated”. Comparing with the full length Ballinderry-A chronology it gives T=5.8 over 146 years. It is also worth noting that the very dramatic growth release occurring after AD 95, found in the oldest sample, Q9883, of Ballinderry-A also appears in the samples of Ballinderry-B. Such a dramatic increase of growth rate (more than three times in a few years!) usually appears due to changes in the local environment – not due to general climatic factors. Those samples are from Ballinderry just south of Portmore Lough to the East of Lough Neagh, and in the Lough Neagh plain. So the dramatic growth release may very well be caused by for instance a temporary decreased water level in the lakes or a temporary fertilizing caused by a flooding. In any case, the fact that the growth release after AD 95 is clearly visible on all the Ballinderry samples identified to cover the event, gives a strong additional support for the dating of the Ballinderry-B collection. Cross correlations between the Ballinderry, Balloo and Lisbarnet chronologies gives the following pattern:

 

 

Rest

Ballinderry-A

Ballinderry-B

Balloo_312AD

LISBARNET-A

 

Years

  CorrTTest Olap

  CorrTTest Olap

  CorrTTest Olap

  CorrTTest Olap

  CorrTTest Olap

Rest

 

 

  0.38  6.9  287

  0.44  7.2  220

  0.37  6.8  295

  0.40  6.4  222

Ballinderry-A

594

  0.38  6.9  287

 

  0.42  5.5  146

  0.26  4.5  287

  0.21  2.5  148

Ballinderry-B

354

  0.44  7.2  220

  0.42  5.5  146

 

  0.28  3.6  154

  0.37  5.8  220

Balloo_312AD

296

  0.37  6.8  295

  0.26  4.5  287

  0.28  3.6  154

 

  0.42  5.7  156

LISBARNET-A

223

  0.40  6.4  222

  0.21  2.5  148

  0.37  5.8  220

  0.42  5.7  156

 

Other possible samples from the critical time span:

Further searching did not result in any more site wise collections, but one single samples, which seems to be from the actual time span was found:

English Roman towards the extended North Ireland

The Ballinderry-B collection gives T=4.3 towards NorthEnglishRoman over 285 years (Q9848, The oldest Ballinderry sample, alone gives T=4.7 (T=5.1) over 285 years depending on a decreased correlation for the very oldest part of the sample). The Lisbarnet-A chronology gives only T=3.4 over 152 years towards the NorthEnglishRoman and adding Lisbarnet-A to a collection containing the Ballinderry-B chronology will give about the same T-value. Adding the Toome Q4185 will give T=4.5. Looking at the curve diagram we will find that the match is not “homogenous” but varies a lot over time. The correlation is very good in a segment of about 25 years around AD 25 but even negative the two decades centered around 50 BC and AD 70 respectively.

Ballinderry-B towards Annaholty bog and Navan/Dorsey

The Q9848 sample of Ballinderry-B covers the youngest part of the time span for samples from Annaholty bog (ending at 69 BC, “Roman dating context” ) in south western part of Ireland and for some samples from Navan (fort) and Dorsey (also dated in the “Roman context”) only about 35 and 50 km respectively, south of the Ballinderry site. All are from archeological sites and, I assume, mainly grown on firm ground. The Annaholty samples (from 5 stems – Q11040 – 43, 45) do all show about zero correlation at the “as dated” position. The Navan and Dorsey samples with an overlap of at least 40 years gives all at least positive correlation values, but far from the level confirming the date:

                  T-  Over
         Corr   Test   lap
all...   0.25   2.40    88   based on 12 members
Q4902    0.42   3.14    49    -133  NAVAN
Q4633    0.40   3.21    55    -127  DORSEY
Q2977    0.39   2.65    41    -141  DORSEY
Q2989    0.37   2.68    48    -134  DORSEY
Q2887    0.35   2.31    41    -141  DORSEY
Q2888    0.25   1.89    54    -128  DORSEY
Q9735    0.21   1.95    83     -99  NAVAN
Q2979    0.21   1.67    65    -117  DORSEY
Q2120M   0.20   1.93    88     -94  NAVAN
Q4629    0.15   1.18    62    -120  DORSEY
Q2993    0.13   1.03    66    -115  DORSEY
Q4657    0.10   0.75    57    -125  NAVAN



So even if there is a “formal continuity” of ring width data from Ireland covering the BC/AD gap (according to the “zero-hypothesis”), the “bridge” here is so weak that it is practically useless for confirmation of the dating of the “Roman block”.

Conclusion

The bridging of the late Roman gap found by Becker and others in early 1980th (“the zero-offset hypothesis”) seems to be slightly confirmed by the Danish-Dutch-German bridge and by bridges between Northern Ireland and the English Roman complex, although none of them are strong enough to fully ensure the date. On the other hand, no iteratively upcoming offsets of notable strength, other than “zero” have yet been found. Extra attention has been offered the -207 years offset proposed by Larsson's analysis of the Hollstein material, but nothing at all supporting that date have been found anywhere else.

So, at this moment, no alternative bridging proposals for the Late Roman gap are found and the bridges supporting the “zero-hypothesis” found so far are:

It seems very unlikely that those bridges, which independently from two directions supports the South German bridge of the Late Roman gap, would be randomly generated. It seems, so far, that the bridging of the Late Roman gap, proposed by Becker, most likely is correct. Despite that, it seems to me that other dendrochronologists and laboratories in possession of dendrochronological measurements useful to bring more light upon this issue, should publish them in order to bring the possibly still remaining uncertainty to an end. In my opinion this is an obligation for the European dendrochronological community. Queens University Belfast, National Museum in Copenhagen, Ester Jansma and English Herritage / Sheffield University have all contributed in a commendable way. Hopefully others will follow their example.

Acknowledgments

Thanks to Mike Baillie and David Brown at QUB, for kindly answering questions about several samples and sites, and for critical questions on earlier versions of this text, helping me to (hopefully) clarify and improve it. Thanks to Petra Ossowski Larsson and Lars-Åke Larsson, who have compiled, retrieved and made available on Cybis.se, a huge amount of chronologies, necessary for this study.

References

This and other dendrochronological studies by Torbjörn Axelson are found at taxelson.se/dendro/obj/nytt.htm (Swedish mainly)