Fig 1.
IntCal20 modelled 14C calibration curve (1σ) and the raw 14C data (1σ) used to compile this curve showing the period 1800–1350 BCE [41].
The notable density of data available for the interval 1700–1500 BCE is evident (most new as part of IntCal20), as also the much better definition that is therefore available for the modelled IntCal20 curve over this period. The constituent IntCal20 dataset is available from: http://intcal.org/.
Fig 2.
Relationship between 14C dating of the Thera eruption and IntCal20 14C calibration curve taphonomy [41].
(A) Calibrated calendar age probabilities (no modelling) using OxCal 4.4.4 [57] for 14C ages drawn from IntCal20 itself for calendar dates 1630, 1610, 1585, 1560, and 1530 BCE illustrate how ages around especially 1610–1530 BCE spread across the reversal-plateau in the calibration curve ~1620–1540 BCE [41] (see Materials and methods below). The only clarity is that ages ≥1630 BCE or ≤1530 BCE may be distinguished from those in between. A stated average 14C age for the Thera eruption [43] is shown at 1σ and 2σ by the yellow bars, intersecting with the calibration curve at multiple places from the early 17th century BCE to mid-16th century BCE. (B) The calibrated calendar age probabilities of each of the 14C dates in S1A Table (numbers 1–110) employed as relevant to the Thera date showing how the dating probabilities for the vast majority spread across the 17th to 16th centuries BCE given the calibration curve shape.
Fig 3.
Map showing the locations of the sites in the southern Aegean region providing the 14C dates employed in this study.
The schematic base map was produced in OxCal 4.4.4 [57] using the open access, publicly available, USGS topo map.
Fig 4.
Calibrated calendar probability distributions for each of the dates in dataset (d) from [57] using [41].
As indicated, four of the dates seem visibly too recent; and one date appears substantially too old.
Fig 5.
Results of a neutral (0 ± 10 14C years) Delta_R test for a growing season related offset (GSRO) for the tree-ring defined time-series from an oak sample from LMIA Miletos, western Anatolia.
(A) The seven weighted average 14C values [132] for 11 tree-rings (years) samples, Relative Years (RY)1000-1010 … to RY1060-1070, each spaced (mid-points of each block) 10 rings = calendar years apart (from pairs of 14C data in each case: see S1A Table nos. 20–33) show a good, ordered, fit against the IntCal20 14C calibration curve [41]. The IntCal20 curve is shown as a 68.3% highest probability band and the Miletos samples are shown with the boxes illustrating, y axis, the weighted average 14C ages plus or minus 1σ, and, x-axis, the 68.3% highest posterior density (hpd) calendar placement of each element of the time-series. (B) The Delta_R test indicates a ~0 to very negligible offset (mean difference: 1.01 ± 8.14 14C years) versus the 0 ± 10 14C years prior (see below: Bayesian chronological modelling).
Fig 6.
The prior probability distribution from the LnN(ln(3),ln(2)) constraint applied to the difference query in the OxCal [57] models discussed below (see Results).
Fig 7.
Comparison of the calendar dating probabilities for the Boundary after the datasets (a), (b), (c) and (d), and for dataset (a)+(c), and investigation of likely temporal difference between these datasets. (A) The posterior probability distributions from the models for each dataset and for datasets (a)+(c) combined with the 68.3% and 95.4% hpd ranges indicated by the upper and lower lines under each distribution. (B) The time interval (OxCal Difference query) between the posterior probability distributions–from (A)–for dataset (a) versus (c) and (a) versus (b), for dataset (c) versus (b), for datasets (a)+(c) versus (b), for dataset (a) versus (d) and for datasets (a)+(c) versus (d). The mean (μ) and median (M) of each difference is stated (in calendar years) as well as the 68.3% and 95.4% probability ranges.
Fig 8.
A re-run of the Fig 7 data and models, as datasets (e) to (h), but allowing in each case for the southern Aegean likely worst-case GSRO factor.
Fig 9.
Dating of Akrotiri stages (ii)/(iii) and the Thera eruption (stage v) from Model 1.
(A) Results for the end of Stages (ii)/(iii) Boundary and the Thera Eruption Boundary from Model 1 run 3 with log-normal, LnN(ln(3),ln(2)), constraint applied (Table 1), detailing the 68.3% and 95.4% hpd calendar age ranges. Comparison is also shown versus the approximate start date of the NK = 18th Dynasty in Egypt ~1565/1540 BCE [7–9, 25, 30, 44, 61–63]. (B) Modelled posterior (solid, cyan) probability versus the log-normal prior (hollow distribution) for the Difference constraint. As indicated by the OxCal Agreement value of 98.8%, there is a very good correspondence between modelled result and prior assumption (see also S1 Fig). (C) Fit of Miletos wiggle-match and modelled dataset (b) (stages ii/iii) individual data against IntCal20 curve (1σ ages and μ±σ modelled calendar ranges) before end of Phase Boundary (see A) from Model 1. (Note: the Miletos wiggle-match oak data are shown placed against IntCal20 in Fig 9C, where they are more easily viewed, versus in Fig 9D, although they used as relevant to the non-Thera dataset (a) samples included in Fig 9D). (D) Fit of modelled datasets (a)+(c) (Thera eruption, stage v) individual data against IntCal20 curve (1σ ages and μ±σ modelled calendar ranges) before end of Phase Boundary (see A) from Model 1.
Fig 10.
Modelled posterior probability distributions for datasets from Aegean LMIA or LCI or LHI-II contexts prior to, or around, or shortly after, the Thera eruption, and LMIB destruction datasets from the close of the subsequent archaeological period on Crete.
For the 68.3% and 95.4% hpd ranges indicated, see Table 5. The datasets offer a coherent chronological sequence.
Fig 11.
Comparison of the calendar date ranges for the close of LMIB destructions at three sites on Crete versus the Thera eruption (TE, or TE5 = stage v) and the time interval in-between.
(A) The modelled posterior probability for the Thera eruption (Fig 9) compared with the modelled posterior probabilities for the close of LMIB destructions at Chania, Myrtos-Pyrgos and Mochlos on Crete (Fig 10). Possible archaeological period relationships are indicated along with the suggestion of a potentially important absence of evidence in the earlier-mid 16th century BCE from very late LMIA through earlier LMIB, perhaps linked with the impacts and dislocation initiated by the Thera eruption. (B) The time interval (Difference query) between the Thera eruption and each of the LMIB destructions is shown. The mean (μ, ellipse) and median (M, cross) of each difference is stated.
Table 1.
Model 1 (Fig 9) results.
A. Results (bold = 68.3% hpd, and non-bold = 95.4% hpd) for the date of the Thera Eruption Boundary from 5 runs of Model 1 (Fig 9) with the Difference query with LnN(ln(3),ln(2)) constraint†, and from 5 runs of an alternative version using a Difference query with U(0,15) constraint, each without, and then with, the likely maximum southern Aegean GSRO of 4±2 14C years. B. The same but for the Akrotiri stages (ii)/(iii) Boundary. OxCal Amodel/Aoverall (Am/Ao) values are also listed for each model. Note rounding errors sometimes see the total hpd reported vary by up to 0.1%. The results show how all runs of such models are unique and results determined can vary very slightly—especially in this case in the less well-defined margins of the 95.4% probability region on the recent side, where the calibration curve plateau lacks clear discrimination (exacerbated slightly further again when the GRSO with additional error term is applied). Run 3 (*) with equal highest Am value (121.2) is illustrated in Fig 9.
Table 2.
Model 1 re-runs without the VERA-4630 TAQ (compare with Table 1).
A. Results (bold = 68.3% hpd, and non-bold = 95.4% hpd) for the date of the Thera Eruption Boundary from re-runs of Model 1 (Fig 9 and Table 1) but without the VERA-4630 TAQ. Otherwise as Table 1. The absence of the VERA-4630 TAQ leads to slightly more spread, and to minor variations at the late margins of the 95.4% ranges, in particular.
Fig 12.
Dating results for the main elements of Model 2 integrating datasets (i), (j), (l) and (m) with Model 1 (as used for Fig 9) (datasets (a)-(c)) to give a temporal sequence running from end MBA/start LCI/LHI/LMIA through to the end of LMII. (A) Model 2 with no GSRO. (B) Model 2 with the southern Aegean maximum GSRO test of 4±2 14C years. For the 68.3% and 95.4% hpd calendar age ranges indicated (upper and lower lines under each probability distribution, respectively), see Table 3. For results from re-runs of Model 2 without VERA-4630, see Table 4.
Table 3.
Results (bold = 68.3% hpd, and non-bold = 95.4% hpd) for selected elements from Model 2 (Fig 12), typical examples.
As in Model 1 (Fig 9), a LnN(ln(3),ln(2)) constraint is applied to a Difference query for the period of time between the Thera eruption (TE5 Boundary) and stages (ii)/(iii) at Akrotiri (E2/3 Boundary). Results are shown for the model run without, and then with, the likely maximum possible GSRO for the southern Aegean (see above). The clear majority range, if present, where there are two or more split ranges is underlined. Note rounding errors sometimes see total hpd reported vary by up to 0.1%.
Table 4.
Results (bold = 68.3% hpd, and non-bold = 95.4% hpd) for selected elements from re-runs of Model 2 without the VERA-4630 TAQ, typical examples.
Compare with Table 3.
Fig 13.
Dating of Akrotiri stage (v) = Thera eruption Boundary from Model 1 in Fig 9 re-run incorporating the Sofular Cave dates for likely Thera-eruption associated spikes in bromine (Br) and molybdenum (Mo) [161].
(A) with the two Sofular Cave dates treated independently, (B) with the two Sofular Cave dates combined.
Table 5.
Modelled results (bold = 68.3% hpd, and non-bold = 95.4% hpd) for the datasets shown in Fig 10.
The majority range where there are two or more split ranges is underlined.