Table 1.
14C dates for the Kyrenia Ship.
The gray-shaded data are not used in the analyses reported in this paper. The dates marked * are not employed as they clearly are either contaminated with PEG or are old wood (see main text). The two dates marked § have very low carbon content (<9% C at 8.9% for GrM-30709, and 5.5% for GrM-30714) and GrM-30714 also has a rather different (anomalous) δ13C value of -26.69‰ whereas all GrM samples on the same wood lie between -21.22‰ to 22.35‰. Thus these two samples are not used. References are given for 14C dates previously published: R13 [30] Lawn (1971); R24 = [31] Burleigh et al. (1982); R29 = [32] Ambers et al. (1987). These dates are not employed in the analysis. The other dates are published here for the first time. X dates from the ORAU are dates not given an OxA code and regarded as potentially suspect because of issues recognized in the pretreatment and dating process. We do not use them in the analysis. It is impossible to distinguish Pinus brutia from Pinus halepensis based solely on wood anatomy [33], so BM-1639R is best regarded as either Pinus halepensis or Pinus brutia. Clearly there is something amiss with the δ13C measurement reported for P-1621 –unless this is simply a typo (for e.g. -26.48‰). If this stated implausible δ13C value was however employed in calculating the reported 14C age, then the originally measured 14C age was even more recent [34] and so further discordant from the other dates on the almond samples, and this uncertainty over information renders this date suspect. OxA δ13C values ±0.3‰; Groningen δ13C values ±0.15‰.
Fig 1.
The set of 14C dates in Table 1 compared to each other in terms of 14C dates expressed in 14C years BP.
The cleaned GrM versus not entirely cleaned OxA dates on the KYR-8 samples are compared on the left (the two GrM dates with very low percent carbon, less than half of the other samples, are not used in the analysis). The other dates are shown on the right. They fall within a similar age range. The OxA-X dates and the previously run non-AMS 14C dates (P-, BM-) are not used in our analysis—see the main text and Table 1. 1σ error bars are shown.
Fig 2.
Existing and new 14C data for calibration curve construction 450–150 BCE.
A. the constituent (raw) 14C data used to construct the IntCal20 14C calibration curve [10] in the period from 450–150 BCE shown against the modelled 1σ IntCal20 curve and versus the date estimate for the Kyrenia Ship, 294–290 BCE, in [1]. There are no AMS 14C data between 350–250 BCE. B. The data in A. but adding the new AMS 14C data on known-age wood discussed and reported in this paper (see S1 Table). 1σ error bars are shown.
Fig 3.
The wiggle-match fit of the tree-ring defined time-series of GrM AMS 14C dates on the KYR-8 timber (Table 1).
A. The data after pretreatment that should have removed nearly all (most) of the PEG contaminant (as confirmed by the known-age PEG test shown in Fig 6 below) versus IntCal20 [10] (resolution 5 years). B. The data in A. but with the last extant year, RY1134, placed with uniform probability between 600–250 BCE and using the modified new AMS 14C dataset (with resolution set at 5 years): referred to as the ‘AMSAdjustedIntCal20’ calibration curve. This last model run offers more recent age ranges but still places the series before the steep slope (change) in atmospheric 14C around 400 BCE. Data from OxCal [43] version 4.4.4. The upper and lower lines under each distribution indicate, respectively, the 68.3% and 95.4% highest posterior density (hpd) ranges.
Fig 4.
A. 14C data from the same calendar year comparing values from UCIAMS on sequoia versus GrM on European oak and B. comparison of an adjusted calibration curve with all values adjusted based on the GrM European oak data values shown versus IntCal20 (contrast Fig 3B). Data in A. shown with 1σ error bars; calibration curves in B. shown as 1σ bands—with data in B. from OxCal [43] version 4.4.4.
Fig 5.
Likely range of remaining PEG-caused offset in the cleaned GrM KYR-8 samples.
A. The Delta_R offset (μ±σ) identified from neutral Delta_R tests of 0±10, 0±25, 0±50 …0±300 14C years applied to the KYR-8 wiggle-match series (against AMSAdjustedIntCal20). The offset stabilizes around 50±50 14C years. B. A wiggle-match run with a generous Delta_R of 50±50 14C years offers good agreement comfortably within assumption error margins between modelled and assumed offset. C. The μ±σ wiggle-match calendar placements of the KYR 14C ages (±σ) and modelled last extant ring (RY1134 ±σ) against the modified AMSAdjustedIntCal20 calibration curve (±σ) (Figs 2B and 3B) comparing (green) the KYR-8 data as in Table 1 with no additional Delta_R adjustment (but with RY1134 placed between 600–250 BC at uniform probability), and (pink) the data applying a generous Delta_R of 50±50 14C years (and RY1134 placed between 600–250 BCE at uniform probability). Data from OxCal [43] version 4.4.4.
Fig 6.
Test of PEG removal on a known-age tree-ring series.
A. Oak (Quercus sp.) drain construction found in a rescue excavation by the Colchester Archaeological Trust in St. Peter’s Street in Colchester in February 2008 [60]. Sample C2 came from this drain construction and was PEG treated in 2008. Photo: Martin Bridge. B. Comparison of the wiggle-matched (known tree-ring series) 14C ages on tree-ring samples from sample C2, 95.4% hpd ranges are indicated, compared with the known calendar (dendrochronological dating) of these samples using IntCal20 [10]—illustrating good correspondence and thus successful removal of the PEG contaminant. Data from OxCal [43] version 4.4.4.
Table 2.
14C dates on the Colchester (C2) oak (Quercus sp.) sample for known-age PEG removal test.
Unfortunately, a technical issue at the time meant that no reliable IRMS δ13C data are available. The known-age for each sample falls within the 14C wiggle-match 68.3% hpd range in all cases.
Fig 7.
Calendar placement of the KYR-35 sample.
A. Comparison and placement of the first two (arranged from inner/older to subsequent) dated tree-ring segments (the first a weighted average of two dates) from KYR-35 versus IntCal20 [10]. This identifies a specific fit zone in the earlier 4th century BCE. B. Comparison and placement of all the dates on the ordered tree-ring samples from KYR-35 compared against the AMSAdjustedIntCal20 curve using the data reported in S1 Table (see Figs 2B and 3B). The approximate minimum fit zone that can accommodate all the dates and in the known order is indicated and covers the early to later 4th century BCE. The upper and lower lines under the distributions indicate the 68.3% and 95.4% hpd ranges. Data from OxCal [43] version 4.4.4.
Fig 8.
Comparison of the Kyrenia Ship 14C dates versus the IntCal20 calibration curve and versus the revised AMSAdjustedIntCal20 14C calibration curve using the new data in S1 Table.
A. Comparison of the 14C measurements on short-lived samples from the contents of the Kyrenia Ship against IntCal20 [10] (with the most likely 68.3% hpd ranges indicated). With IntCal20 most of the data are not compatible with the date of 294–290 BCE proposed for the ship’s wrecking in [1]. B. As A. but instead against the AMSAdjustedIntCal20 14C dataset as in Figs 2B and 3B. The short-lived data now in all cases could include 294–290 BCE. Two possible fit areas are evident, labelled ‘A’ and ‘B’; ‘B’ offers a date range subsequent to the time period occupied by the KYR-8 tree-ring samples. Data from OxCal [43] version 4.4.4.
Fig 9.
Model 1C using the revised AMSAdjustedIntCal20 14C dataset and selected results.
A. whole model. The OxCal keywords, numerical values, and outlining indicate the structure of the model exactly. The light shaded distributions are the non-modelled calibrated calendar probabilities; the smaller dark histograms indicate the modelled probability with the lines under these indicating the 68.3% and 95.4% hpd ranges. B. Detail of the TPQ Boundary. C. Detail of the LV Boundary. D. Detail of the Difference query (time interval between the TPQ and the LV). Data from OxCal [43, 45] version 4.4.4.
Fig 10.
Model 1C using the previous IntCal20 dataset [10] and selected results—compare with Fig 9.
A. Whole model. The OxCal keywords, numerical values, and outlining indicate the structure of the model exactly. The light shaded distributions are the non-modelled calibrated calendar probabilities; the smaller dark histograms indicate the modelled probability with the lines under these indicating the 68.3% and 95.4% hpd ranges. Note, with IntCal20, the main (>90% probability range) of the overall modelled 95.4% hpd ranges of 8 of 9 of the short-lived samples from the ship contents entirely lie no later than 345 BCE. B. Detail of the TPQ Boundary. C. Detail of the LV Boundary. D. Detail of the Difference query (time interval between the TPQ and the LV). Data from OxCal [43, 45] version 4.4.4.
Table 3.
Summary of different models and date ranges for the Last Voyage (LV) Boundary.
The table shows the 68.3% and 95.4% hpd calendar date BCE (Cal BCE) ranges for the TPQ Boundary (Models 1A to 1C only) and LV Boundary from each of Models 1A-1C and 2A-2C, and the Difference query for the time period between the TPQ and the LV for Models 1A to 1C, along with OxCal Amodel (Am) and Aoverall (Ao) values across the 6 models, and comparing the values achieved with the new AMSAdjustedIntCal20 14C dataset (with curve resolution set at 5 years) versus those from IntCal20 (with curve resolution set a 1 year) in grey shading. The Am/Ao values for Model 1B are unsatisfactory (<60), indicating that this model’s assumptions are not compatible with the 14C data and the 14C calibration curve. For Model 1C, as an example, we also list the results if the AMSAdjustedIntCal20 14C dataset is used but with 1 year curve resolution (bold text). While the 1-year resolution model results overall are noisier (see S8 Fig), the findings for the TPQ and LV Boundaries are very similar to the 5-year resolution models. For Model 1C we list also the results employing the GrM and European oak adjusted overall dataset (see Fig 4B), ‘AMSAdjustedIntCal20_GrMbased’, in underlined text (see S9 Fig). Finally, for Model 1C, we list the results if the two apparently (subjectively) somewhat older 14C ages, for OxA-30953 and OxA-31033 (both have individual OxCal Agreement values <60, respectively 47 and 51: Fig 9), are excluded from the analysis (italicized text). This makes little difference. Since neither date is an outlier within the model (outlier probabilities respectively 4% and 3%, both <5%: Fig 9) we leave them in.
Fig 11.
The highest posterior density (hpd) distributions for the Last Voyage (LV) Boundary from Models 1A, 1B and 1C with the AMSAdjustedIntCal20 14C dataset with curve resolution set at 5 years and with linear interpolation and also the same but using cubic interpolation all from OxCal [43] version 4.4.4.
The indicated 68.3% and 95.4% hpd ranges, 286–272 BCE (68.3%) and 305–271 BCE (95.4%) are the upper and lower limits from each version of Models 1A, 1B and 1C. The only variation is that three models begin the 95.4% range at 305 BCE and three at 304 BCE across our runs.
Fig 12.
Dating the Mazotos ship and contents.
We show a revised version of results from the model in [15] at Fig 5 using now the AMSAdjustedIntCal20 14C dataset (as in Figs 2B and 3B) showing the placements with these data of the timber (MAZ-1) wiggle-match and dunnage and olive pit samples from the dating model for the Mazotos ship (see [15]).