Radiocarbon Dating by AMS of the Destruction of Akrotiri

The existing radiocarbon dates have been used to support the argument for an earlier time range but this interpretation has not gone undisputed. Following the 1986 Athens Archaeometry Conference, the AMS groups at Oxford and Simon Fraser Universities (SFU) embarked on a programme of accelerator dates on short-lived carbonized seed material excavated from the West House at Akrotiri. In this paper we report on two new series of accelerator radiocarbon dates measured on the Oxford AMS facility. Our first radiocarbon analyses support the view that the archaeological date of c. 1500 BC is too young, a conclusion which is in general agreement with the tree-ring and ice-core data. A second series of measurements on the same samples using a different chemical pretreatment method, pioneered by the SFU group, suggests that there might be an older contaminant in the seeds which is biasing all the published radiocarbon dates. These conclusions are supported by SFU's findings. However, the resulting dates presented here, following the removal of the contaminant, cast doubt on SFU's assertion that the contaminant is of sufficient significance to favour the traditional date of c. 1500 BC. Our measurements indicate that a date in the latter half of the 17th century BC is the more likely, although the form of the calibration curve at this period precludes firm conclusions on the matter.

INTRODUCTION

The dating of the cataclysmic eruption of Thera, whose remains form the caldera of Santorini in the Aegean Sea, has remained contentious because not all the lines of evidence can be easily reconciled. The eruption, and preceding earthquake, led to the abandonment of the late Minoan site of Akrotiri which was buried by volcanic ash. The problem lies in the perceived discrepancy between the traditional archaeological chronology, based on historical Egyptian imports, and the absolute dating of other stratigraphic layers thought to represent the eruption. A re-evaluation of the archaeological chronology (Betancourt 1987; Betancourt and Michael 1987; Michael and Betancourt 1988) has suggested that the traditional date (c. 1500 BC or a little earlier) for the end of LM IA (and thus the Theran eruption) may be too late, and they have proposed a revised date of c. 1610 BC. However, this revised chronology has been disputed (Warren 1987, 1988). Radiometric radiocarbon dates from the destruction layers at Akrotiri have been cited as evidence for the earlier chronology by Betancourt et al., but Warren (1987) has argued that the Theran dates are spread over too long a period to be useful and emphasizes the ceramic evidence. Aitken (1987, 1988), in recent resumes, has concluded that the radiocarbon dates currently available give, for the most part, a consistent result which after calibration can be made to support either an 'early' end to LM IA, or an end not much earlier than the traditional date of c. 1500 BC.

There are two other forms of stratigraphic evidence on which absolute dating has proved possible, although in neither case is the evidence directly linked with the Theran eruption. Firstly, there is the occurrence of frost damage and other signs of severe climatic stress in dendrochronologically dated tree rings, both Californian bristlecone pines (LaMarche and Hirschboeck 1984) and Irish oaks (Baillie and Munro 1988). Both of these tree-ring chronologies dated the (presumed volcanically induced) severe frost damage to 1628-1626 BC. The second piece of evidence consists of the presence of an acidity layer in the Camp Century ice core in North Greenland (Hammer et al. 1980), and again, more importantly, in the Dye 3 ice core in South Greenland (Hammer et al. 1987).

The acidity layer is datable largely through counting back an annual signal from the present. The latter core clearly indicated a date of 1645 ± 7 BC, but while the presumed volcanic eruption is consistent with the scale and location of the Theran eruption, there is no direct evidence to equate this date with Thera. On the other hand, it is important to note that there is no evidence in either of the ice cores or the tree-ring record for a major volcanic eruption at the archaeologically indicated date, i.e. around c. 1500 BC. Given the discrepancy between the ice-core and tree-ring evidence, and the traditional archaeological chronology, much has been made of the need for radiocarbon dating to resolve this issue. However, because of the form of the calibration curve for the period c. 1620-1525 BC, stringent conditions are demanded of any radiocarbon dating in terms of accuracy of measurement, and in terms of the material used for dating, since this must be reliably free from contaminants and strictly related stratigraphically to the eruption.

Why have the present Thera dates been perceived as inadequate? It has been alleged that the radiocarbon dates obtained from samples from the destruction levels at Akrotiri associated with the volcanic eruption 'do not make sense' (cited in Aitken 1988, 165), and it has been suggested (Olsson 1987) that carbon dioxide emanations from the volcano, deficient in carbon-14, were the cause. The main reasons cited why the Akrotiri dates are unreliable have been: (i) that they exhibit an unacceptable degree of scatter, and, (ii) that they cannot be reconciled with the well-established Minoan chronology, established by archaeological linkages to the Egyptian astronomically-anchored chronology (Weinstein and Betancourt 1978).

Aitken (1987) has argued that the radiocarbon dates from Thera, excluding the long-lived samples and the two early deviants (P-2560 and P-2561), when compared to the expected experimental error of the laboratory, show the degree of scatter to be expected within the average error limit. The two deviant short-lived samples - about 600 years earlier than the rest of the group - have been excluded as perhaps being caused by the 'dead' carbon effect, the result of plants absorbing carbon dioxide from volcanic gas emanations, leading to reported 'ages' 1000 to 1400 years too old (Bruns et al. 1980; Olsson 1987).

Turning to the point that the Akrotiri dates cannot be reconciled with the well-established Minoan chronology, the use of the new high precision calibration curves recommended by the 12th International Radiocarbon Conference (Pearson and Stuiver 1986), has made this objection less tenable. The new calibrations are in general concordant with previous curves, but for this particular period the new curves have had the effect of making calibrated dates more recent by about half a century, thus bringing the Theran dates more in line with the traditional chronology. However, there is still sufficient discrepancy to pose problems for Aegean archaeologists, to the extent that in 1986, at the Archaeometry Conference in Athens, Professor Doumas challenged the archaeometric community to provide unambiguous radiocarbon dating for the eruption. This paper presents the results of our response to this challenge.

The work described was done during the same period as, and in co-operation with, the Simon Fraser University AMS group, who also report their results in this volume. AMS dating was used in preference to radiometric counting because the small sample size allowed for the possibility to make multiple dates from the same archaeological sample by different laboratories. This allowed both groups to use the same assemblages of carbonized grain samples, supplied by A. Sarpaki, the palaeobotanist studying the plant remains from Professor Doumas's excavations. We had expected the two groups to produce similar results which, taken together, might provide unequivocal dating. In the event, despite considerable similarity in the dates, the results obtained demonstrate rather clearly the present limitations of radiocarbon dating and unfortunately do not constitute the unequivocal dating sought.

THE MATERIAL DATED

Any new radiocarbon dates from Akrotiri would only be of value in this argument if they were from short-lived species of plants, and were firmly associated with the destruction layers of the site. The chosen samples came from rooms 5 and 6 in the West House at Akrotiri, and consisted of caryopses of barley (Hordeum), two vetch species (Lathyrus clymenum and L. cicera), plus two samples of mixed legume fragments. By choosing different genera and species, probably grown in different localities on the island, it was hoped that the potential problem of take-up of volcanic CO₂ in photosynthesis could be avoided.

The dated seed assemblages were all recovered from the contents of storage jars (Sarpaki, pers. comm.). The three assemblages from room 6 were thought to be associated with stages 1/2, i.e. with the pre-earthquake habitation and reoccupation, whereas the seeds from room 5 were from stages 2/3, the pre-eruption occupation and destruction phases. Each bulk seed assemblage was divided into three subsamples: one third for processing and dating at Oxford, one third for processing and dating at SFU and one third being kept back as a reserve.

THE METHOD OF DATING

At Oxford, the seeds were dated in two separate series. The first series (I) of 11 dates underwent a standard pretreatment (see below), the resulting dates (Table 3) being in broad agreement with the ice-core and tree-ring chronology. This was followed by a second series (II) of dates, carried out directly in response to results obtained by SFU, who kindly made them available ahead of publication. SFU's evidence suggested the seeds contained a removable fraction with an older carbon-14 age. On removing the 'contaminant', they obtained an age for the 'cleaned' seeds of 3218 ± 18 BP (weighted mean of 13 dates). Using this date and the 20 year average curve of Pearson and Stuiver (1986), they calculated a calendrical age range of 1520-1500 and 1481-1459 cal BC for the last human occupation of the site prior to the eruption, thus lending support to the traditional archaeological date. The fact that these results disagreed with the recent re-evaluations of previous radiocarbon dates they explained as being due to an apparent failure, in pretreatment, to remove the 'contaminant' by the laboratories which undertook the previous dates. Given the discrepancy between our series I dates and SFU's, we therefore attempted to reproduce their cleaning method as closely as possible before determining a second set of dates (series II - Table 5).

• Pretreatment procedures:

Series I.

Seeds weighing about 10 mg were cleaned using the laboratory's standard procedures, described in Batten et al. (1986), except that the 0.1 M NaOH treatment was reduced to a rinse of a few seconds because of its effect in solubilizing the seeds. About 30 % by weight of the sample was lost by dissolution in this procedure. Following cleaning, the samples were combusted and converted to graphite using standard procedures, broadly outlined in Gillespie et al. (1985). The quantity of carbon used in the dating was about 4 mg.

Series II.

Six groups of seeds, after being weighed, were steeped in 0.1 M HCl (2 cm³ per sample) overnight, followed by ultrasonication in the same acid solution for a 15 minute interval. This treatment released very little colour from the samples. Excess acid was removed from the samples by repeated rinsing, centrifuging, and decanting, until a near-neutral pH was recorded on narrow-range indicator paper. It was observed that the 'colour' released from the seeds increased as the amount of residual acid decreased. At this point, 2 cm³ of distilled water was added to each sample, and these were then ultrasonicated for 30 minutes. This treatment released a significant amount of coloured material into solution from all samples, and this was removed (after centrifuging at 3500 rpm) by pasteur pipette, for combining with the acid extracts. In accordance with the method outlined by the SFU group, the samples were each rinsed with distilled water (1.5 cm³) with ultrasonication (10-15 minutes), in an attempt to remove all of the extractable 'contaminant'. After five such rinses, no significant decrease in the amount of coloured material extracted in each step could be observed, and so the combined rinses for each sample were stored as discrete fractions. A series of five more rinses was carried out with ultrasonication as described above. The optical density of the extracts obtained, when diluted ten-fold with distilled water, was measured on a Shimadzu UV-240 spectrophotometer over the range 500 to 350 nm, using narrow-section plastic cells with a 1 cm path-length.

This confirmed that the loss of colour from the samples was virtually constant over the last four wash steps, suggesting that the treatment given was simply a mild way of dissolving the whole sample, rather than a procedure which was removing a specific contaminant from the surface of the seeds. The sequential washes and ultrasonication were therefore terminated, and the last set of five rinses were combined to produce a second fraction for each sample. To ensure that the wash samples were not contaminated by fine particles of the 'charcoal' fraction, the solutions were centrifuged at 3500 rpm for fifteen minutes then twice filtered through 0.7 μm porosity Whatman GF/F glass-fibre filters. Similarly, to prevent the 10x-extracted 'charcoal' fractions from becoming contaminated by incompletely extracted material, the small amounts of solids removed with each water wash were isolated, combined, and kept separate from the bulk samples. When dried and weighed it was found that the combined values for the two corresponding fractions were similar in magnitude to the weights reported by SFU; i.e. about 10 % by weight had been extracted as 'humic' material (although it is not clear in the SFU procedure if this was 10 % by weight of seed or 10 % by weight of carbon -  see later discussion). Samples for combustion were prepared from the acid and water-leached solid residues ('charcoal'), and from the contaminant ('humic acids'). Pretreatment yields, burn yields, and graphite conversions are presented in Table 1.

Fractions from some seeds were isolated and dried for analysis by infra-red (IR) spectroscopy. If the suspected contaminant was of a humic or fulvic nature, then the water extracts should show absorption bands in the infra-red which are characteristic of these compounds; e.g. at 1720 cm^-1 (C=O of -COOH) and 1660 cm^-1 (-C=O). The 'cleaned' seeds, on the other hand, should exhibit featureless spectra, if they consist of charcoal only. IR spectra were obtained by mixing sufficient amounts of each sample (0.3 mg) with 40 mg of KBr, pressing the mixtures into discs, and analysing these on a Perkin-Elmer 1420 ratio-recording spectrometer. Characteristics of the IR spectra obtained for both fraction types are listed in Table 2. The IR analysis on 'humic' and 'charcoal' fractions showed that the intensity of absorbance due to -OH, -COOH, and -C=O functional groups were of a similar magnitude for all fractions, arguing that the 'humic contaminant' is largely solubilized 'charcoal' and not an intrusive material of different chemical composition and origin which has been adsorbed by the charred seeds before excavation. We note that SFU found the δ¹³C values for the residue and the contaminants to be indistinguishable, and this supports the conclusions drawn from the IR data.

For the purpose of dating, the contaminant material extracted by the two five-wash treatments was combined for each seed. In general, this material made up 10 % by weight of the untreated seeds, but contained 20 % of the carbon. The weight yield of carbon obtained from the combustion of the cleaned seed residues averaged 4 mg.

DATING

The series I dates (Table 3) form a set consistent with a single date. If the estimated errors are all random (we have taken the average SFU δ¹³C value, -23‰, as the correction for isotopic fractionation), the weighted mean date and error comes to 3357 ± 21 BP for 11 dates. A number of the samples shared the same provenance as the dated Philadelphia and Copenhagen samples, and it is interesting to compare the equivalent results (Table 4), being careful to bear in mind that the laboratories did not always use the same type of material from a particular context. Excluding the undersize samples, there is, in general, good agreement except that the two previously obtained deviant dates (P-2560 and P-2561) did not produce such early dates this time.

Since for series II (Table 5), there is a systematic difference in sample size between contaminant and residue, it is necessary to consider carefully possible sources of error which could lead to a systematic difference in date. For example, beam currents are often smaller for graphite targets of less than 1 mg. In particular, the 'background' contribution from laboratory-derived modern carbon is much more significant for the smaller samples. We have re-calibrated this contribution for different size samples and find, for the conditions under which these samples were prepared, that the 1 mg ('contaminant') samples have a background contribution of an extra 70 years to their age.

As Table 6 shows, the 'contaminant' samples are consistently older than the 'residue' samples, in accord with the SFU findings. But the difference we find is much smaller than that reported by SFU. The series II dates for 'contaminant' and 'residue' form distinguishable sets of internally consistent values, with a mean date and error for the 'residue' of 3300 ± 30, and for the 'contaminant' of 3442 ± 38 BP. These differences are only barely significant, nevertheless they are different in the same direction, although with less than one quarter of the magnitude reported by SFU.

If the series I dates are pooled with the series II 'residue' dates, on the argument that the pretreatment in either case is chemically to be expected to have the same effect, the dates obtained by the two methods are found to be consistent with a weighted mean date and error of 3338 ± 17 BP. This error is unrealistically low, since any systematic error of less than 20 years or so would not be detectable in our system. Therefore we believe the date for the mean of the combined series I and II dates is 3338 ± 30 BP (errors of this magnitude have been checked with other AMS labs in the dating of the Shroud of Turin; Damon et al. 1989). We note that this date (3338 ± 30) is 120 years older than that proposed by SFU (3218 ± 18) for the 'cleaned' seeds.

DISCUSSION

Is there an older contaminant present? By repeating the SFU protocol for treating the seeds, we have obtained evidence suggestive of the presence of a removable older fraction. While this lends support to the finding by SFU, the effect we measure (120 years on 20 % of the carbon) is much smaller than that reported by them. SFU measured the contamination on one batch of seeds (sample identification no. 2055) with three related cleanings.

Our dates are on the 'contaminant' and the 'residue' on three seed sets, including group 2055, as well as an unrelated 'residue' (2061) and 'contaminant' (2054). However, it is noteworthy that the greatest effect we observed was on the same seed batch that SFU measured, viz. 2055, where the difference in age measured for the pair was 200 ± 100 (difference between 3280 ± 60 and 3480 ±  70).

This compares with the three SFU values of 550 ± 130 for an 18% carbon component, or 700 ± 140 for a 9% component, or 1380 ± 240 for an 8% component. Thus we estimate the effect of the 'contaminant' to be less than half that reported by SFU for the seed assemblage which had the largest effect. Averaging over the four assemblages measured, the effect is less than one quarter, so that, whereas SFU results calculate an apparent increase of age due to the contaminant of about 100 years (comparable with the difference in disputed age of the Theran eruption), our results imply a difference in age of only about 25 years. It could be argued that we have failed to isolate the contaminant measured by SFU; but since SFU's results on the 'contaminant in seed assemblage 2055 are fairly consistent internally, despite sonication in media ranging from water to 6 N HCl, it is difficult to see where our treatment has failed.

Our experience in cleaning the seeds by ultrasonication in very dilute acid was not quite the same as that reported by SFU. There was no obvious distinction in visual appearance, in reaction with cleaning reagents, or in infra-red spectrum, between the contaminant and the residue. We also note that SFU and other laboratories have reported the ease with which the seeds dissolve in NaOH solution, implying a humified rather than fully developed charcoal as composing the bulk of the material. We are not therefore persuaded of the chemical or physical distinctness of the contaminant, nor are our dates on the contaminant so different from those on the residue that they provide compelling evidence for the contaminant as a distinct entity with a different date.

An explanation of the occurrence of the contaminant would increase confidence in its reality. We agree with SFU that this is rather a puzzle. Volcanic CO₂ is an obvious candidate as a source of 'dead' carbon, but apart from the problematic physical and chemical mechanisms requiring it to be incorporated into humic material, its distinct δ¹³C signal should show up in the contaminant; incorporation via photosynthesis is doubtful.

CONCLUSIONS

What then is the age of the seeds? We have two sets of dates (series I and series II, residues) for the seeds. Series II have been cleaned according to the SFU protocol. Series I, in our view, should give equally well cleaned samples, and indeed the dates for the two series (3357 ± 21 and 3300 ± 30) are not distinguishable. Furthermore, both sets of dates are as consistent as if they had been made on a single sample. However, if we assume the contamination has not been removed from series I, we might expect, on the basis of our dates from the isolated 'contaminant', that the series I dates are too old by 25 years. Therefore, if we take this worst case, (i.e. 'true' date is 3335 ± 20), and combine it with the series II dates, the resultant age is 3325 ± 17. We feel that a final error of 30 years more truly reflects the possibilities of systematic error. This date (3325 ± 30) is about 100 years older than that obtained by SFU (3218 ± 18). When calibrated, two age ranges are obtained, 1674-1606 cal BC and 1554-1534 cal BC at the one sigma (68% confidence) level. Considering these two respective age ranges, the probability that the result lies in the 17th century BC age range is c. 70% against a c. 30% probability that it lies in the 16th century. Although this gives a greater measure of support to the ice-core and tree-ring data for a 17th century BC eruption, the result is hardly conclusive. It is interesting to note that, of the previously published Akrotiri dates, six (P-1889, P-1890, P-1892, P-1894, K-3228, and K-4255) had been treated with NaOH and their weighed mean value is 3332 ± 23 BP; whereas the samples not pretreated with NaOH average 40 years younger (S.W. Robinson, pers. comm. to Aitken 1988). It is important to note the critical part the calibration curve plays in converting the radiocarbon result to a calendrical date. The distinction between the two historical dates of, say, 1500 BC and 1640 BC, requires a resolution of better than 40 years (since 3330 calibrates largely in support of 1640 BC and 3290 BP calibrates largely in support of a 16th century BC date). It is unlikely that radiocarbon measurements, if they fall into this region of the calibration curve, can resolve such a small difference, even when sample contamination is not an issue.

Where do we go from here? Until we have a better understanding of the nature of the 'contaminant', there is little point in undertaking any further dating, since small residual traces will affect any new dates, including high precision ones. The absence of a distinctive δ¹³C signal, and infra-red absorbance characteristic, introduces difficulties when attempting to pin down just what the contaminant is, where it is coming from, and how it is being taken up by the charred seeds. A programme of analyses is therefore needed to characterize the chemical and physical properties of this particular fraction.

An alternative approach, which is actively being pursued at Oxford, is to date LM IA sites, other than Akrotiri, in the hope that, by getting away from the problem of the volcanic ash and its associated 'humics', the contamination problem can be circumvented. This will not solve the problem of the calibration curve though, and unless there are drastic improvements in the accuracy of radiocarbon dating, this will remain a difficulty. One solution would be to date material of a specific known relative age at the time of the eruption (such as an inner growth ring of a long-lived tree which was killed by the eruption). By knowing that the growth ring was laid down, say, 150 years before the death of the tree, the poor resolution of the calibration curve for the eruption period could be circumvented, allowing the two hypothetical dates for the eruption to be distinguished.

Whether a suitable tree stump is available remains to be seen. A more promising alternative would be to date the adjoining archaeological periods (MM III, LM IB, and LM II), thus erecting a framework whereby the imprecision of the calibration curve in the LM IA period can be minimized.

By dating these adjoining periods, the effect would be to define the temporal boundaries between which the LM IA period must lie and thus minimize the disruptive effect the present controversy is having on the Late Bronze Age chronology of the Aegean. With this aim in mind suitable samples are being sought and it is hoped that results from this work can be presented in the future.

Addendum

Some additional comment is needed to clarify the relationship between the samples and results of Oxford and Simon Fraser University AMS groups.

Our paper was written when the SFU results from runs A and B were available to us and their publication was in prospect. The results of run C had not yet been obtained. In our view, the results of run C do nothing to strengthen the assertion that the seeds contain a significantly older contaminant which, on removal, would make the radiocarbon dating of the seeds significantly younger.

It is also worth making clear that the batches of seeds we dated were divided between SFU and Oxford. Of the 20 different batches available (sample code: 2054 to 2073 inclusive), Oxford dated only those which we knew were most clearly related to the latest period of occupation. Five batches (2065 to 2069 inclusive) were thought to be of Middle Bronze Age date and we did not date them. To what extent SFU dated these five batches is unknown (since individual results are not identified except for run B). The palaeobotanist who recovered, sieved, and supplied us with the seeds, A. Sarpaki, asserts that the seeds had no preservative of any kind applied, and we are entirely satisfied of this from our own examinations.

It should be noted that in our case a single date was produced from, in general, three-four individual seeds, whereas we believe SFU typically produced one date from one seed. This may be relevant to explaining the greater spread in SFU's results. If such variation in SFU's results is wholly due to the seed radiocarbon content, such variation would be reduced by nearly one half in our results. The variation in our results is of the magnitude expected from the known and understood errors in the measurement, and does not lend support to the suggestion that there is very significant variation in ¹⁴C content between individual seeds for all the batches analysed.

In our paper we suggested it would be worthwhile dating short-lived samples from adjoining archaeological periods (MM III, LM IB, LM II) to bracket the LM IA period. At the Congress three AMS dates from the LM II layer in the Unexplored Mansion at Knossos in Crete were reported:

OxA-2096 charred barley grain, δ¹³C = -23.3‰     3070 ±  70

OxA-2097 charred barley grain, δ¹³C = -23.6‰     3190 ±  65

OxA-2098 charred barley grain, δ¹³C = -22.9‰     3220 ±  65

As with the Akrotiri dates, these results are uncalibrated in radiocarbon years BP using the half life of 5568 years. The pooled mean date is 3177 ± 44 BP and the calibrated age ranges are c. 1510-1430 cal BC (1 sigma) and c. 1590-1390 cal BC (2 sigma). Betancourt (1987, 47) suggested that the LM II period should be placed between c. 1550 and 1490 BC, whereas Cadogan (1978) has argued LM II dates to between c. 1450 and 1400 BC. The calibration curve in this period means that the Knossos dates can accommodate either hypothesis, and thus the issue remains unresolved.

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

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

(For figures and tables please refer to book)

Source: "Thera and the Aegean WOrld III, Vol 3" (pp. 207 - 215)

Authors: R.A. Housley, R.E.M. Hedges, I.A. Law, and C.R. Bronk (Radiocarbon Accelerator Unit, Research Laboratory for Archaeology, 6 Keble Road, Oxford OXI 3QJ, England)