Another Suite of Confusing Radiocarbon Dates for the Destruction of Akrotiri

There is evidence that at least some seeds contained an older contaminant. To further add to the confusion, new radiocarbon dates for the eruption of another Mediterranean volcano indicate that it could have occurred at the same time as the Theran eruption.

INTRODUCTION

At the 1986 Archaeometry Conference, Professor C. Doumas, Director of excavations at Akrotiri, provided a fascinating over-view of this remarkable site. He mentioned that, despite considerable radiocarbon testing, the date for the volcanic destruction of the town was still in dispute, and that there was a difference of a little more than 100 years between the so-called 'traditional' age of 1500 BC and the radiocarbon data. That was challenging, and representatives of the archaeometry groups at Oxford University and at Simon Fraser University proposed joint accelerator mass spectrometry (AMS) radiocarbon tests on small samples from the site. While attempts to obtain suitable samples were still under way, various re-evaluations seemed to strengthen the case for radiocarbon data (cf. Aitken et al. 1988). Further, signals at 1647 BC in Greenland ice-cores (Hammer et al. 1987) and 1627 BC in Californian and Irish tree rings (LaMarche and Hirschboeck 1984; Baillie and Munro 1988)were interpreted as confirming and refining the radiocarbon results. The remaining problem appeared to be to convince a few recalcitrant archaeologists to modify their chronologies (Manning et al. 1988).

At about this time, the Oxford group obtained a suite of seed samples from the site. While the new interpretations made the problem much less immediate, confirmatory studies seemed appropriate. Since none of us at the SFU laboratory had any archaeological experience in this area, we took the view-point that this exercise was to test our AMS system and to provide new, independent radiocarbon information. We made no attempt to enter into the details of the archaeological interpretations and restricted our reviews to the recent literature discussed above. However, it was disturbing to learn that the re-interpretations relied heavily on rejections of 'deviants' and 'outliers' in the original radiocarbon data. A closer review of these data showed that they were obtained primarily on materials from pots found in the ruins of the houses (seeds or unidentified organic material) or from pieces of charcoal. The 26 original radiocarbon dates are plotted in the upper portion of Fig. 1. The dispersion is disturbing, especially as the laboratories which provided these original measurements are known for the high quality of their work.

THE SAMPLES AND MEASUREMENTS

An original goal of this project was to obtain samples that would not be subject to the same problems that could have affected the materials originally dated. One possibility was to date bone from the site, as none of the previous measurements had been done on bone. Samples were obtained from the excavation and extensively tested at SFU for the presence of remnant collagenous protein. For the three samples tested, the acid-insoluble organic carbon concentration was at most ~0.3% total bone weight. This is much below that of normal bone and there was no measurable nitrogen in this material, indicating that it was probably not proteinaceous. Ion-exchange techniques were then used in attempts to extract amino-acids from the acid-soluble organic materials in the bone. The resultant product had carbon and nitrogen constants of ~12% and ~1% respectively, indicating that the majority of the carbon was from some source other than protein. This very poor collagen preservation seemed unusual, as we have obtained good protein yields and reliable dates from archaeological bones in nearby Crete. We speculate that this poor preservation is due to heating of the bone by the volcanic ash, which would make the collagen soluble in ground water. In any case, the samples were not considered to be worthy of dating.

All our subsequent measurements were then undertaken on samples of charred seeds of various food-stuffs found in pots in rooms 5 and 6 of the West House (R. Housley, pers. comm.). These foodstuffs are described as representing the three proposed final stages of occupation of the site: 1)the first earthquakes and abandonment; 2) re-occupation and second abandonment; and 3) the final eruption. A list and description is given in Table 1. While Table 1 groups the samples according to provenance, we considered each individual seed to be a separate sample. Our limited information on provenance strongly suggests that these seeds are from the same pots on which many of the original measurements were taken.

In three separate sets of determinations over the course of a year, we obtained about 70 measurements on 40 seeds. With the exception of sample preparation details, the general measurement methods are unchanged from those we have used in other studies, and are described in Nelson et al. (1986) and Vogel et al. (1987). This project required careful attention to measurement accuracy since it would be necessary to clearly resolve a time difference of about 100 radiocarbon years. Our normal practice is to include a number of known-age materials with each set of unknowns to provide a running test. From past experience, we have found that our measurement precision (at 1 SD) for each sample, as based on repeat determinations (not just counting statistics) is about 110 to 120 radiocarbon years, and the results obtained are randomly distributed about the true mean. Thus, it should have been possible to provide a sufficiently-accurate average age given measurements of many samples. Even so, since the accuracy requirements were more stringent than we normally encounter, we did measure the known-age materials more than we usually do. Three were used:

1. the ANU sucrose standard;
2. the ACT-1 test material (2400 ± 20 BP); and
3. the ACT-4 test material (15,170 ± 20 BP).

The latter two materials and ages were provided by M. Stuiver, University of Washington. The ACT-1 material clearly provides the more direct evidence of accuracy for this study.

The first set of measurements (Run a) led to an inconclusive result which seemed at the time to support the re-interpretations of the original data. We discussed these results at the 1988 Archaeometry Conference but did not publish then, as we wished to repeat the measurements. The second set (Run b) yielded an apparently very definitive but different result. We submitted these results for publication and sent copies to those colleagues directly concerned. This publication was rejected. Our third set of measurements (Run c) were taken to try to answer some of the questions posed by the first two runs. These results were again inconclusive. Even though we have not published any of our data, word has spread anyway and is adding to the confusion surrounding this problem. Here, we present and review the results of our measurement series.

DETAILS AND RESULTS. RUN a.

For the first determinations, we selected an individual seed from each sample and subjected it to the normal pre-treatment procedure for carbonized organic materials. The seeds were first soaked in 0.1 N HCl to remove inorganic carbonates, washed in distilled water, and then soaked in 0.1 N NaOH to extract contaminating 'humic acids'. A serious problem was immediately apparent. All the seeds were soluble in NaOH and even in neutralized sodium pyrophosphate, indicating that they were not carbonized, but at most only charred. This was disturbing, as it means that the seeds are themselves 'humic acids' and could easily have exchanged carbon with the ground-water. In retrospect, this finding is not surprising, as studies have shown that cellulosic materials (i.e. grass) covered by warm volcanic ash can rapidly be converted to humic materials (Shindo et al. 1986a, 1986b). While we should have rejected these samples as unsuitable, we decided to measure them anyway and then try to find methods for separating contaminants from such materials. The dissolved materials were re-precipitated with acid, an aliquot of each precipitate was taken from separate δ¹³C measurements, and the remainder was dated. A further complication for this run was that we were not satisfied with the performance of the AMS system, even though the results on the test samples indicated that our overall accuracy was excellent, as shown in the upper portion of Fig. 2.

The dates obtained for the Akrotiri materials in Run a are displayed in the second portion of Fig. 1. The large spread in the measurements is similar to that of the original studies. An attempt at a detailed comparison revealed some surprising facts. First, only a very few of the original measurements had had corrections made for isotopic fractionation. Our δ¹³C determinations indicated that these corrections should have been 30 to 50 ¹⁴C years, which is substantial in this context. Few, if any, of the recent detailed statistical re-evaluations of these data took isotopic fractionation into account. Next, the published date lists for the original measurements suggested that only some of the samples had been successfully pre-treated with NaOH. Apparently the Pennsylvania lab (B. Lawn, pers. comm., 1988) also found that many of the samples were base-soluble, and so a base wash could not be used.

Despite these problems, our new results seemed to confirm the older ones; on average, our younger ages supported an age for the eruption of about 3350 ± 50 BP. This conclusion was presented at the 1988 Archaeometry Conference, and we further suggested a means for resolving the small discrepancy between the tree-ring age and the ice-core age. As mentioned above, we chose not to publish these results, as we wished to repeat the measurements after resolving the problems of chemical pre-treatment and system performance.

DETAILS AND RESULTS. RUN b.

To resolve the purification problem, we decided to use a pre-treatment process with which we had been experimenting in other projects. Ultrasonic agitation is known to produce intense local concentrations of OH radicals capable of breaking complex molecules and solubilizing organic substances (Suslick 1988). Our experience had indicated that this method enhances extraction of humic fractions from other materials and that it mobilizes a qualitatively different and quantitatively larger humic fraction (Ramunni and Palmieri 1985). One of our Akrotiri samples (2055) still contained three large, well-defined seeds which we subjected to a detailed extraction test using non-basic solutions in an ultrasonic bath. Each seed was soaked overnight in a solution of either 6N HCl, 0.1 N HCl or in water, followed by successive 15 minute sonications in fresh solution. This extraction removed about 10-20 % of the carbon in each seed. The final residues consisted of shiny, black, flaky powders, with the exception that the 6N HCl residue had a brown amorphous material sticking to it. This strong hydrolysis was not considered further. Both the solutions and the residues from these tests were prepared for measurement, and a second suite of individual seeds from other samples was selected and prepared by soaking in 0.1 N HCl overnight and ultrasonically washing in water.

As before, a number of measurements of the ACT-1 material were made to provide an indication of accuracy. These results and those for unknowns are again given in Fig. 2 and 1 respectively, and the data from the extraction test of the three seeds are given in Fig. 3.

An immediate conclusion was that sample 2055 did contain an older material which was removed by the new pre-treatment process. We found that, in contrast to all earlier measurements, the results on the unknowns were as tightly grouped as one could expect for repeat measurements of a single sample. Further, the average age was found to be younger than those previously obtained, in entire accord with the discovery of the older extract. Since the average obtained for the ACT-1 material was slightly older than it should have been (see Fig. 2) the average age for the unknowns was likely to have been an older estimate. These apparently excellent results gave us considerable confidence that the problems of the previous run had been solved and that our results were reliable.

At the time, this data set was certainly the best in existence. Following earlier advice from our Oxford colleague, we submitted these results for publication. We also sent copies of the paper to other researchers working on the problem, with the stricture that it was not to be quoted or discussed until the paper was accepted. Our measured average age, when calibrated, indicated that the eruption would have occurred exactly at the traditional archaeological age, in direct contradiction to the recent conclusions, and the paper was rejected. We advised our colleagues that our data were withdrawn until further notice. To proceed, we needed to verify our results.

DETAILS AND RESULTS. RUN c.

Several tests were carried out in response to the logical critiques of our abortive paper. Unfortunately, these further tests were restricted by the small amounts of the samples remaining.

First, a suggestion was made that our measurements of the extracts of the three seeds from sample 2055 were inaccurate as they were done on very small amounts of extracted material (100-400 μgC). This criticism was unexpected as we have reported measurements on even smaller samples (e.g. Vogel et al. 1987). To provide a further demonstration, we interrupted another measurement series to date amounts of the ACT-1 material ranging from 50 to 6500 μgC. We were careful to treat these samples exactly as any other unknown. The results (Vogel et al. 1989) strongly indicate that our measured ages for the extracts are not suspect because of sample size.

A second valid criticism was that we had tested the extracts from the seeds of only one sample. For several other samples, the last remaining material was prepared using the same acid/water/ultrasound technique. Some of these did not yield enough extract for measurement since they were only bits of seed and amorphous powder.

These new extracts and residues and any remaining prepared graphite from earlier measurements were then dated in Run c. As before, the test samples were used to monitor accuracy. The ages determined for the ACT-1 material are given in the lower portion of Fig. 2. The average again indicates excellent accuracy.

The results of the extraction tests are given in Fig. 3. The repeat measurements on the Run b graphite for sample 2055 re-confirmed that these seeds contained an older contaminant. However, this was not the case for the other samples. For some, there was no significant difference in the ages for extract and residue; for one sample (2059) the extract was significantly younger than the residue. It is noteworthy that the prepared residues for this sample consistently yielded an age of about 3550 BP in four separate measurements over the three runs, while the one measurement of the extract was close to the postulated eruption age. This needs confirmation, but it adds further evidence that the carbon in these seeds represents more than one event.

A self-criticism of our Run b results was that, despite the information on accuracy provided by the test samples, there was an age-offset for all the data. For Run c, we wished to provide a completely independent test, preferably using known-age materials similar to those at Akrotiri. We were fortunate to obtain wood (from construction beams), a small piece of charcoal, and soil samples from a recent study of the Pompeii eruption at Vesuvius. This is the type-eruption used to describe volcanic activity of the sort that occurred at Thera, and its age (AD 79) is very well known since it was historically recorded by Pliny the Younger. We wished to determine whether we could accurately date this eruption, and whether there were any unusual contaminants in materials covered by that volcanic ashfall.

Our measured age for the sample best associated with the eruption (the charcoal) was 1940 ± 80 BP. The calibrated age (Stuiver and Reimer 1986) for this date is AD 66, with a possible age-range of 36 BC to AD 129. The measured ages for the two large beams were 85 ± 45 years and 145 ± 60 years older, which is very consistent with their size and use. While dates on soils are in general not very definitive, we dated chemical fractions that have been shown elsewhere to give good results (Campbell et al. 1967; Goh et al. 1984; Trumbore et al. 1989) and the results were consistent with the historical age. Details of this study are published elsewhere (Vogel et al. 1990).

These data again indicate that our measurement procedures should have been sufficient to resolve the Theran problem. However, the dates obtained on the remaining Theran seeds, as shown in the lower portion of Fig. 1, are not in good accord with those obtained in either of the first measurements. They are slightly older than those obtained in Run a, and shown the same wide variation as in that Run. They are in definite disagreement with the tight grouping of ages obtained in Run b.

INTERPRETATIONS

These data are highly confusing, and we have examined our procedures in great detail. There are three general possibilities that could cause these problems, all or some of which could be operative:

1. the seeds may not truly represent the eruptive event under consideration;
2. our AMS system is not accurate; and
3. there are variable amounts of contamination in the samples. We comment on these separately below.

1)      Sample association.

It is difficult for us to examine this possibility, as we have no knowledge of the excavators' procedures and interpretations. However, there are some general comments that can be made. Only a few of the samples that have been measured in any study appear to be related directly to the actual event of interest, the eruption itself. The samples are thought to be associated with the last human habitation of the site, but perhaps we should even re-examine that assumption. Do all these samples truly represent the same events? Could materials from the earlier Middle Bronze Age occupation have been mis-identified? Are the time differences between final abandonment and eruption truly small?

Since it may be very difficult to provide definitive answers to those questions, it may be as well to take a different approach. In the conclusion of our abortive paper, we argued that further tests should be made on new samples that could be directly associated with the eruptive event, and which would not be subject to the same problems that may have occurred at Akrotiri. We suggested that identifiable organic materials found underlying Theran ash at distant locations might be ideal. Apparently, such studies are now under way (Y. Maniatis, pers. comm., 1989).

2)      System accuracy.

We have examined the measurement uncertainty of our system for these runs in great detail. As noted above, the precision and accuracy of the AMS system were monitored by repeated measurements of three test materials. These samples were repeatedly prepared, not taken from a large batch-preparation, so that the variability would represent that of the unknowns.

The observed sample-to-sample variability of the measured ¹⁴C concentrations was about 1.5% at one standard deviation. (This corresponds to an uncertainty of ± 120 years). This estimate is very similar to the variability we found in a separate study (Wahlen et al. 1989)and we believe it to give a reliable estimate of our measurement precision for samples of age of a few thousand years.

The good correspondence between the known ages for the test samples and our measured average ages indicates that there were no systematic deviations, and that our estimates of measurement precision are also good estimates of measurement accuracy. Given these data, the average ages for the unknowns should have been of sufficient accuracy to clearly resolve the problem.

We also attempted to determine the relative contributions to variability from the preparation process and from the isotopic ratio measurement itself. During the runs, we made a number of duplicate measurements on two aliquots of the same graphite prepared from a single sample. (This is often possible as graphite production usually yields more than enough for one measurement.) These duplicates should then provide a conservative estimate of the variability of the isotopic ratio determination during these runs. We made a total of 104 such duplicate measurements during these series. Of these, 68 were on non-Akrotiri samples, mostly the test materials. The observed variability for these duplicates was about 1.8%. However, this includes the results obtained for the much older ACT-4 material, which we seldom determine to precisions better than ~2%. Exclusion of those data gives an observed variability of about 1.3%. Since that is very close to the value found above for the repeat measures, this suggests that our variability is essentially all due to the isotope-ratio measurement process. However, an examination of the data on 36 duplicate measurements of the graphites prepared from the Akrotiri samples showed almost twice the variation, about 2.1%. We can find no reason why graphite prepared from Akrotiri samples should show a greater variability on re-measurement than graphites prepared from the test samples.

3)      Contamination.

The problems with sample contaminants are complex, not only for this study but for all others. There is no doubt that three individually tested seeds from sample 2055 contained an older material. The observed age-difference for extracts and residues is too great to be due to measurement variability. Other samples tested either didn't contain an older material, or our pretreatment procedures did not separate one. One older sample produced an extract younger than the residue.

The possibility of contamination was perhaps the single issue that most concerned the referees of our abortive paper. There was a general concern that, as we had not chemically identified the contaminant, it was premature to say that it existed, and therefore the age data could not be considered reliable. The problem is even more complicated and wide-reaching than that. The most likely natural contaminating substances in these charred seed-remains are humic materials. Detection, identification and elimination of humic contaminants will be very difficult as the seed remnants themselves are humic acids. In this case, one can only state that if different fractions of a sample, however extracted, yield different radiocarbon concentrations, the sample must contain a contaminant.

Most of the original published radiocarbon ages for Akrotiri were made on samples that received less pre-treatment and testing than those given in Run b, and these are the very measurements on which the many interpretations and re-interpretations have been based. Application of our critics' objections to the original data would result in the rejection of most of it, and indeed the same can probably be said of almost all radiocarbon dates ever taken. We are left to wonder why this consideration is important for the samples we measured but not for those of others.

The situation is further confused by a comment we received from a colleague to whom we sent the paper. He had indirect information that the seeds may have been treated with a preservative in the field, perhaps without the excavation director being aware of it. More support for this rumour came from a researcher who had worked on some of the original measurements. It would seem that many of the samples were covered with some substance that had to be scraped away. [Note added in press: Following my presentation at the Congress, I was assured that the samples we measured had not been treated with preservative.]

We strongly suspect that at least part of the unusual variability seen in our studies is due to varying levels of contamination. Since it is likely that many of the seeds we measured are from the same pots as the samples originally dated, perhaps the unusual variability found in those studies also reflects this same problem.

In these circumstances, it is difficult to provide further interpretations for our data. Depending upon the assumptions that one makes at the outset, it is possible to support either of the two ages proposed for the eruption. For example, if one assumes that many of the samples do contain variable amounts of an older contaminant which is not always removed, then the younger ages represent the final occupation, and presumably, given the closest age for the eruption. The data from Run a and especially from Run b would then indicate that a ¹⁴C age of about 3200 BP would be appropriate. But one can legitimately argue that our measurements in Run b contain a non-random deviation despite the information provided by the test samples.

On the other hand, if one assumes that our overall accuracy is good despite non-random variations between runs, then an average taken for the whole data set should provide the best estimate. The weighted average for all our measurements is 3380 ± 10 BP (N=70 measures on 40 seeds). This corresponds amazingly well with the weighted average for all the original dates, 3367 ± 13 BP (N=26). The probability that two separate determinations on the same sample would yield these results is high (40%) and could be even higher if all the original dates were properly corrected for isotopic fractionation. Of course, judicious elimination of 'outliers' and 'deviants' would make the correspondence even better. While these extremely close results provide strong mutual corroboration for both data sets, they do not necessarily provide information on the event of interest. Large-scale averages of a variably-contaminated suite of samples will yield the same incorrect age, whether done a seed at a time by an AMS lab, or several grams of seeds at a time by a traditional lab.

One can continue these discussions ad infinitum without arriving at an interpretation which satisfactorily explains all the data.

CONCLUSIONS

Our research has. not provided a resolution to the problem of dating the Theran eruption. There are clearly underlying problems that we have not identified and eliminated. At one point we did sincerely believe that we had sorted out these problems, but we were mistaken, and we apologize to those whom we may have led astray.

However, our apologies are tempered by our concern that similar problems have beset many of the other attempts to date the Theran eruption. The original measurements can be criticized on many of the same grounds. Have any of us been measuring what we think we were measuring?

This criticism extends beyond the radiocarbon laboratories. Both the ice-core researchers and the tree-ring researchers have correlated signals in their respective records with the Theran eruption. The criticism has been levied that these correlations may be fortuitous (Pyle 1989) and we have unwittingly uncovered new evidence that this may indeed be the case. As a part of our test study of the AD 79 Pompeii eruption, we also obtained some samples from the earlier major eruption of Vesuvius, the so-called Avellino eruption. The age of this event was thought to be 3700-3800 BP, as determined by radiocarbon dating of soils (Delibrias et al. 1986) and humics (Alessio et al. 1971, 1974) without taking into account the finite age of these materials at the time of burial. We have dated new charcoal and soil samples obtained from a geological re-examination of this eruption, and we find that the date above is too old by about 400 years. Our best estimate for the age of this eruption is 3340 ± 30 BP (Vogel et al. 1990). This is almost exactly the age proposed for the Theran event, and although it was smaller, the Avellino eruption was of the type which could have produced the ice-core and tree-ring signals.

This startling observation prompted us to casually examine the record for other eruptions at this time period. There seem to have been a number of them. Mount Etna erupted at 3230 ± 110 BP (Gif-2776) and a smaller eruption took place in the Massif Central at 34990 ± 80 BP (Gif-6229). On the other side of the world, a massive eruption of Mount St. Helens in Washington State (The Yn eruption) spread ash over a large portion of western North America. The 3500 BP date for this eruption was based on soil and peat dates (Mullineaux 1986; Zoltai 1989) and if our experience at Vesuvius is any guide, these ages are likely to be a few hundred years too old.

Which of these eruptions, or any others of which we may be unaware, are represented in the ice-core and tree-ring records? Correlations with the Theran eruption may be correct, but the evidence is not compelling.

These observations lead to a final observation that is not related to the actual age of the Theran event, but may prove of interest in understanding the remarkable changes that occurred in human events in the Mediterranean area at about this time. The re-dating of the various Mediterranean volcanic eruptions appears to move them closer in time. This, combined with the great amount of archaeological evidence for severe earthquake activity, suggests that there may have been repeated and severe tectonic activity over a very wide area at the time. Can we directly link the great human social re-organizations of late Minoan times to a rapid movement of the complicated tectonic plate system that underlies the Mediterranean?

------------------------------------------------

-------------------------------------------