Experimental results

Here, we present 93 new high-precision ¹⁴C measurements at annual-resolution covering a total span of 153 calendar years. Due to practical constraints, such as very narrow tree rings, it was not possible to obtain a result for every single year. The list of the ¹⁴C measurements from all tree samples is given in the Supplementary Material Tables S1–S4. All of these data are plotted in Figure 1 in comparison with IntCal20 (Reimer et al. Reference Reimer, Austin, Bard, Bayliss, Blackwell, Ramsey, Butzin, Cheng, Edwards, Friedrich and Grootes2020).

Figure 1. All Groningen (GrM) measurements on tree rings from Bodegraven, the Netherlands and Garry Bog and Sentry Hill, Northern Ireland, UK (Q1933, Q8046, Q5392). Different colors represent different samples and tree-ring chronologies.

Comparisons with IntCal and other laboratories

To analyse our dataset for any potential systematic offsets, we compared it across laboratories, tree species, and based on the latitude of the sample origin. To maintain a controlled approach, only measurements from Northern Ireland were used (the data from the Netherlands were excluded), ensuring that all comparisons were conducted using the similar latitude (55°–54°N) and species (Quercus sp). In order to create one continuous dataset, we calculated the weighted mean of multiple measurements from the same calendar year for all the wood samples (i.e. the weighted mean of measurements on samples Q8046 and Q5392). There were 12 such averaged results and each passed the χ² test (see Supplementary Material Table S5) (Ward and Wilson Reference Ward and Wilson1978). After applying the weighted mean, a total of 70 single-year measurements were used in the offset comparisons.

Initially, we compared our dataset directly to IntCal20. As seen in Figure 1, our data closely agrees with the 1σ band of this reference dataset. The GrM data does not show any statistically significant offset (0 ± 3 ¹⁴C years). However, the calibration curve is a statistically smoothed dataset that averages out fluctuations in ¹⁴C concentration; certain details may be obscured in the process. We therefore shifted our focus to the composite datasets for a more comprehensive analysis. We tested for differences between GrM and the single-year measurements from the Curt-Engelhorn-Centre for Archaeometry, Mannheim (MAMS; oak from Knetzgau, Germany), the Swiss Federal Institute of Technology (ETH; oak from Timahoe West, Ireland), and the University of Arizona (AA; bristlecone pine from Sheep Mountain, California, USA, and oak from Timahoe West, Ireland). Additionally, we tested other relevant datasets from the same time period that are not incorporated into IntCal20. These provide alternative parameters for comparison, such as different latitude and species (AA; juniper from Gordion, Turkey, and kauri from Gibson’s Farm, Dargaville, New Zealand) (Pearson et al. Reference Pearson, Salzer, Wacker, Brewer, Sookdeo and Kuniholm2020a, Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b). The weighted mean offsets between the various datasets are presented in Table 2.

Table 2. Comparisons between the GrM annual dataset on oak from Northern Ireland, UK (55°–54°N) corresponding to IntCal20, ETH, MAMS, and AA measurements. Juniper data from Turkey (Pearson et al. Reference Pearson, Salzer, Wacker, Brewer, Sookdeo and Kuniholm2020a) and kauri data from New Zealand (Pearson et al. Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b) are not incorporated in the IntCal20 dataset. All the datasets listed below were prepared using holocellulose extraction, except for GrM, where α-cellulose was the preferred extract.

The offset of –35 ± 5 ¹⁴C years with the kauri data confirms the difference in the ¹⁴C concentration in the atmosphere between the Northern and Southern Hemispheres (NH, SH) (Hogg et al. Reference Hogg, Heaton, Hua, Palmer, Turney, Southon, Bayliss, Blackwell, Boswijk, Ramsey and Pearson2020; –36 ± 27 ¹⁴C years). Otherwise, GrM data shows statistically significant offsets relative to the AA measurements on bristlecone pine from Sheep Mountain (–10 ± 3 ¹⁴C years) and juniper from Gordion (–17 ± 4 ¹⁴C years), as the 2σ ranges for these offsets do not overlap zero. In contrast, the GrM offsets relative to ETH oak (5 ± 3 ¹⁴C years) and MAMS (–4 ± 3 ¹⁴C years) datasets are not statistically significant.

The variances obtained are unlikely to result from inter-laboratory offsets, as GrM does not show a significant difference with AA measurements on oak (1 ± 5 ¹⁴C years). However, it should be noted that the absolute age of the juniper sample is based on a wiggle-matched sequence, even though the calendar year match is quite precise (± 4 years for the sequence’s final ring). Thus, our offset calculations might not reflect a direct year-to-year comparison.

The offset in ¹⁴C years observed between GrM and AA datasets from Gordion and Sheep Mountain (approximately 15°–16° and 17°–18° difference with Northern Ireland datasets respectively) accord with the latitudinal gradient proposed by Büntgen et al. (Reference Büntgen, Wacker, Galván, Arnold, Arseneault, Baillie, Beer, Bernabei, Bleicher, Boswijk and Bräuning2018). Their research has highlighted that the ¹⁴C concentrations are higher at latitudes above 60°, identifying a “meridional north–south gradient of declining average ¹⁴C values” (Büntgen et al. Reference Büntgen, Wacker, Galván, Arnold, Arseneault, Baillie, Beer, Bernabei, Bleicher, Boswijk and Bräuning2018). The atmospheric ¹⁴C levels can vary across different atmospheric circulation zones, which are defined through carbon cycle modeling and ¹⁴C dating over the Bomb Period (Hua et al. Reference Hua, Barbetti and Rakowski2013, Reference Hua, Turnbull, Santos, Rakowski, Ancapichún, De Pol-Holz, Hammer, Lehman, Levin, Miller and Palmer2022). Such factors are particularly relevant to dating of the Thera eruption, as the island is located in NH Bomb Zone 2, much farther south than the trees that make up the calibration curve, which predominantly grew in NH Bomb Zone 1. Any latitude variation in ¹⁴C levels on the calibration curve could therefore influence the results.

To assess this effect, we plotted the mean Δ¹⁴C (‰) values in relation to latitude, following a similar approach to Büntgen et al. (Reference Büntgen, Wacker, Galván, Arnold, Arseneault, Baillie, Beer, Bernabei, Bleicher, Boswijk and Bräuning2018, p.4), as shown in Figure 2. We used 66 ¹⁴C measurements from 1610 to 1510 BCE across the five NH latitudes listed in Table 2. This subset was selected as it represents the common calendar years present across all datasets. Our results reveal a similar pattern to their analysis with a strong positive correlation between latitude and the mean Δ¹⁴C values, higher latitudes exhibiting higher ¹⁴C concentrations. It should be acknowledged that our dataset is limited to only five groups.

Figure 2. Scatter plot of the mean ∆¹⁴C values by latitude. The mean value of 66 ¹⁴C measurements on tree rings between 1510–1610 BCE. Datasets are from Knetzgau, Timahoe West and Garry Bog/Sentry Hill (NH1); Sheep Mountain and Gordion (NH2).

However, when comparing samples from different latitudes, several lurking variables could not be controlled for. These include variations in local growing season, tree physiology and even differences in laboratory pretreatment. At higher latitudes, trees tend to grow primarily in mid-to-late summer, whereas at lower latitudes, growth occurs earlier, between spring and early summer (Kromer et al. Reference Kromer, Manning, Kuniholm, Newton, Spurk and Levin2001; Manning Reference Manning2024a). This coincides with the seasonal fluctuations of atmospheric ¹⁴C levels, reaching a minimum in late winter and a maximum in late summer for NH (Kitagawa et al. Reference Kitagawa, Mukai, Nojiri, Shibata, Kobayashi and Nojiri2004 ; Levin and Kromer Reference Levin and Kromer2004; Manning et al. Reference Manning, Griggs, Lorentzen, Bronk Ramsey, Chivall, Jull and Lange2018, Reference Manning, Wacker, Büntgen, Bronk Ramsey, Dee, Kromer, Lorentzen and Tegel2020b). As a result, trees from higher latitudes, such as Irish and German oaks, take in carbon during this maximum, potentially making their ¹⁴C ages slightly younger. In contrast, trees from lower latitudes, such as Turkish juniper, tend to absorb carbon earlier in the year, leading to slightly older ages (Dee et al. Reference Dee, Brock, Harris, Ramsey, Shortland, Higham and Rowland2010; Kromer et al. Reference Kromer, Manning, Kuniholm, Newton, Spurk and Levin2001). This is consistent with what we present in Table 2 and Figure 2, with measurements on higher latitude trees exhibiting younger ages. Bristlecone pine from California, however, has a particularly short and distinct growing period in summer because of the high altitudes at which it grows (Beasley and Klemmedson Reference Beasley and Klemmedson1980; Pearson et al. Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b). It should thus primarily record the summer maximum in ¹⁴C values and should align more closely with boreal wood rather than with trees from similar latitudes, such as Turkish juniper. Yet, the ¹⁴C levels in bristlecone pine appear more comparable to Turkish juniper, indicating that additional factors may be driving the observed offsets.

Another such parameter is differences in laboratory approaches and pretreatment protocols. For example, the observed offsets could relate to differences between the earlywood and latewood fractions of trees. Due to seasonal atmospheric fluctuations and carbon storage dynamics, it has been suggested that NH latewood tends to contain more ¹⁴C, and hence appears younger, than earlywood (McDonald et al. Reference McDonald, Chivall, Miles and Ramsey2019). While this difference was not statistically significant within 2σ, McDonald et al. (Reference McDonald, Chivall, Miles and Ramsey2019) report a 96% probability that latewood is more enriched. This observation is qualitatively consistent with a recent study on tree physiology using samples from the post-bomb era by Kromer et al. (Reference Kromer, Wacker, Friedrich, Lindauer, Friedrich, Bitterli, Treydte, Fonti, Martínez-Sancho and Nievergelt2024). They claim that earlywood in most species is formed from stored older carbon, whereas latewood is primarily derived from products of more recent photosynthesis. Among the datasets listed in Table 2, only the Timahoe West samples are separated by wood fraction, with only latewood used, while the rest are measurements on full tree rings. Consistent with the aforementioned research on ring fractions, as shown in Figure 2, this dataset exhibits the highest mean Δ¹⁴C (‰) values, appearing younger than the rest. Such differences could contribute to small but consistent offsets, and further controlled studies are needed.

Finally, variations in the different pretreatment methods should also be taken into account. For example, the protocols used by Friedrich et al. (Reference Friedrich, Kromer, Wacker, Olsen, Remmele, Lindauer, Land and Pearson2020) and Pearson et al. (Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b) involve holocellulose, while our GrM dataset is the only one analysed using α-cellulose. Even though it requires a longer and harsher procedure than simple holocellulose extraction (Khumalo et al. Reference Khumalo, Svarva, Zurbach and Nadeau2024) and therefore may be problematic for poorly preserved samples (Capano et al. Reference Capano, Miramont, Guibal, Kromer, Tuna, Fagault and Bard2018; Jędrzejowski et al. Reference Jędrzejowski, Michczyńska, Kłusek, Michczyński, Pawełczyk, Piotrowska, Heeb and Hajdas2024), α-cellulose is a more discerning protocol, and results in higher removal of lignin and other potential contaminants (Hoper et al. Reference Hoper, McCormac, Hogg, Higham and Head1997; Rinne et al. Reference Rinne, Boettger, Loader, Robertson, Switsur and Waterhouse2005). One previous research found no statistically significant difference between these two pretreated fractions, although the ¹⁴C measurements on α-cellulose from the same tree rings were consistently older than those on holocellulose in their dataset (Lange et al. Reference Lange, Nordby, Murphy, Hodgins and Pearson2019). It therefore remains unclear which fraction yields the most representative age.

In Table 2, there is no significant inter-laboratory offset between our data and MAMS, ETH or AA measurements on the same wood species (oak) from northern Europe (Knetzgau, Timahoe West). This was expected due to the minimal differences in latitude between the sampling locations and the use of the same species. However, for the remaining comparisons, multiple variables, such as species and latitude differences, growing seasons, atmospheric ¹⁴C fluctuations, and pretreatment techniques, make it challenging to isolate the exact cause of the observed offsets. Nevertheless, the offsets between these different datasets exert only a limited influence on calibrations using IntCal, since the curve is constructed as a statistically smoothed average of all these records.

Short-term features in mid-16th century BCE

To investigate the veracity of the possible Miyake Events in 1557 and 1528 BCE, we first zoom in to our dataset and extract 50 years of measurements over the period 1560–1510 BCE. Once again, we compare the data obtained by our laboratory with some of the IntCal20 composite datasets over this window (see Figure 3). Our GrM measurements align more closely with those of ETH than AA or MAMS. We do not observe the specific anomaly reported by Pearson et al. (Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b) around 1558-1557 BCE, similar to the ETH, AA oak and MAMS datasets. However, the GrM, ETH and AA bristlecone pine data all show a short-lived but comparatively sharp decrease between 1557-1556 BCE (GrM 8‰; ETH 6‰; AA bristlecone pine 9‰).

Figure 3. Single year measurements between 1560–1510 BCE from GrM (this study), along with ETH, AA (Pearson et al. Reference Pearson, Brewer, Brown, Heaton, Hodgins, Jull, Lange and Salzer2018, Reference Pearson, Wacker, Bayliss, Brown, Salzer, Brewer, Bollhalder, Boswijk and Hodgins2020b), and MAMS (Friedrich et al. Reference Friedrich, Kromer, Wacker, Olsen, Remmele, Lindauer, Land and Pearson2020) showing (a) ¹⁴C data (b) corresponding Δ ¹⁴C values.

Turning to the proposed anomaly around 1528 BCE, there is a very similar pattern between all laboratories around 1534–1515 BCE, where a steep rise in ¹⁴C levels is indeed observed. This increase begins in 1533 and peaks between 1530 and 1528 BCE and subsequently declines in all the datasets. The peak rise amounts to approximately 5–6‰ for GrM; 7‰ for ETH; 8‰ for AA bristlecone pine; 6–7‰ for MAMS; and 8‰ for AA oak in Figure 3b. While the value is lower than the confirmed Miyake Events listed in the Introduction; it is comparable to the proposed events of 1052–1053 CE (5.9‰) and 1279–1280 CE (6.5‰) (Brehm et al. Reference Brehm, Christl, Knowles, Casanova, Evershed, Adolphi, Muscheler, Synal, Mekhaldi, Paleari and Leuschner2022), the existence of which has latterly been disputed by Scifo et al. (Reference Scifo, Abi Nassif, Conti, Bayliss, Doeve and Dee2024). Further, the values show an exponential increase over 4–5 years rather than an abrupt rise or a spike like a Miyake Event, similar to the rise between 541–546 CE (6‰) reported by Scifo et al. (Reference Scifo, Abi Nassif, Conti, Bayliss, Doeve and Dee2024).

While the gradual rise in ¹⁴C levels ca. 1530 BCE lacks the characteristics of a Miyake Event, there seems to be a particular pattern evident in all the datasets. The shape of the data between 1534 and 1515 BCE becomes visually apparent after applying data smoothing, as shown in Figure 4. We used the Loess nonparametric method of data smoothing since the GrM, MAMS and AA oak datasets do not have evenly spaced x-axis values, as some calendar years were skipped. Although the intensity of the ¹⁴C concentration varies over this period, for instance AA bristlecone pine and MAMS show elevated levels between 1525–1520 BCE, the overall wiggle pattern is consistent across all labs (GrM, AA, ETH, and MAMS) and latitudes (Ireland, Germany, California). This pattern may still offer a time marker of similar value, provided the sample in question has the right characteristics (i.e. a series of single-year samples) for its identification. For instance, if the eruption had occurred after 1530 BCE, this undulation would almost certainly be evident in the data.

Figure 4. The ¹⁴C data shown in Figure 3 smoothed using the Loess nonparametric method. The red lines mark the previously reported anomalies on 1557 and 1528 BCE. A similar wiggle fluctuation pattern is visible across all datasets between 1534 and 1515 BCE.

Potential candidates for Miyake Events can also be examined using carbon box models (see Scifo et al. Reference Scifo, Abi Nassif, Conti, Bayliss, Doeve and Dee2024; Brehm et al. Reference Brehm, Pearson, Christl, Bayliss, Nicolussi, Pichler, Brown and Wacker2025). We investigated the proposed anomaly in 1528 BCE using the open-source Python package ticktack on the extracted data over the period 1560–1510 BCE on Garry Bog only (Zhang et al. Reference Zhang, Sharma, Dennis, Scifo, Kuitems, Büntgen, Owens, Dee and Pope2022; Sharma et al. Reference Sharma, Zhang and Dennis2023). This modeling library is designed to analyse ∆¹⁴C (‰) values, detect potential anomalies and summarize ¹⁴C production data. Detailed explanations of this package and its application to different datasets are provided in Zhang et al. (Reference Zhang, Sharma, Dennis, Scifo, Kuitems, Büntgen, Owens, Dee and Pope2022). We used the 22-box model of Brehm et al. (Reference Brehm, Bayliss, Christl, Synal, Adolphi, Beer, Kromer, Muscheler, Solanki, Usoskin and Bleicher2021), already implemented within ticktack, and applied spike only command to test whether there is a sudden rise on ¹⁴C production on 1528 BCE (see Supplementary Material Table S6 for model specifications). This method is specifically designed to identify potential events within the input dataset.

The results of the modeling are presented in Figure 5. First, we display the model estimates in terms of ∆¹⁴C (‰) in Figure 5a. The key posterior parameters from this analysis, the Start Date, Duration, and Q, with the latter representing integrated ¹⁴C yields over the observed rise, are shown in Figure 5b. A small potential increase in ¹⁴C production of approximately 1.6 ¹⁴C atoms/cm²/s per year is estimated to occur around 1532.7 ( $^{+1.6}_{-1.1}$ years) BCE. In comparison, the 775 event produced 6.4 ¹⁴C atoms/cm²/s per year (Scifo et al. Reference Scifo, Abi Nassif, Conti, Bayliss, Doeve and Dee2024), so our data shows an increase in production rate that is roughly four times lower. Our result is comparable with the 1052 CE (1.4 atoms/cm²/s), 1279 CE (1.1 atoms/cm²/s), and 540s CE (1.7 atoms/cm²/s) event candidates. All these latter features are ambiguous and characterized by low production rates and unresolved start dates. Indeed, they may simply represent normal variations in solar modulation, as reported by Scifo et al. (Reference Scifo, Abi Nassif, Conti, Bayliss, Doeve and Dee2024). Furthermore, we do not have a ¹⁴C measurement for the year 1532 BCE, as our data skips from 1533 to 1531 BCE, which may have influenced the model’s interpretation of a possible rise in ¹⁴C production. Nevertheless, the models exclude 1528 BCE as a starting date for a Miyake Event at 95.4% probability (Figure 5b).

Figure 5. Focus on 1560–1510 on new GrM ¹⁴C data, all dates are in BCE. (a) The possible Δ ¹⁴C (‰) estimates of the event from spike only model results shown in green lines (b) Corner plot of the posterior parameters from the ticktack model showing 68.2% (dark blue) and 95.4% (light blue) probability ranges.

In summary, there is only weak evidence to suggest a significant and sudden ¹⁴C production event comparable with the previously confirmed, globally replicated Miyake Events. Hence, there is no specific tie point here for chronology, particularly for the Thera eruption.

Modeling of the archaeological datasets related to the eruption of Thera

The smoothed curves on Figure 4 reveal elevated ¹⁴C levels in the AA bristlecone pine and MAMS datasets after 1530 BCE, with measurements reaching as much as 3300 BP. In contrast, the other datasets do not show such high ¹⁴C ages post-1530 BCE. Such features could also influence the ¹⁴C calibrations for the Thera eruption. To test this, we prepared two new calibration curves: one that incorporated single-year IntCal20 data with the new GrM data (GrM inc.), and another based solely on AA bristlecone pine and MAMS data (AA BCP & MAMS-only). These two test curves are simply the weighted average of the measurements for each calendar year from the single-year datasets.

Using IntCal20 and these two test curves, we employed OxCal software (Bronk Ramsey Reference Bronk Ramsey2009a, version 4.4) to rerun the R_Combine analysis of 24 ¹⁴C dates on short-lived plant samples from the Akrotiri volcanic destruction level (VDL) at Thera, excluding four problematic measurements with uncertain contexts, as discussed by Manning (Reference Manning2024a). The resulting calibrations using each curve are not significantly different from each other (Figure 6). The highest region of probability still falls between 1650–1600 BCE. The only difference is the GrM data introduces a slight probability for the eruption date to be earlier than the 1700s, which is not archaeologically feasible.

Figure 6. Calibration of Thera Akrotiri VDL seeds dataset using R_Combine using (a) IntCal20 (b) single-year IntCal20 data incorporating the new GrM measurements (c) AA bristlecone pine and MAMS-only data.

Apart from using R_Combine with the VDL dataset, we also tried modeling them simply by placing the 24 measurements under a Phase within two uniform Boundary functions. The resulting probability for the end Boundary for this model is given on Figure 7a. Since this model is very simple, with no specific constraints, the resulting probabilities are wider. The highest region of probability still falls in the 17th century BCE.

Figure 7. Modeled results for (a) the end Boundary of Akrotiri VDL seeds, (b) the Eruption Boundary of the model by Manning (Reference Manning2024a) incorporating Thera olive sequence, Therasia shrub sequences and the VDL dataset, using different calibration curves and Delta_R adjustment. The models are run with three different curves: IntCal20 (blue), single-year IntCal20 data incorporating the new GrM measurements (green), and a curve using AA bristlecone pine and MAMS-only data (orange). The 68.3% and 95.4% probability ranges are shown below the graphs, following the respective color coding.

Additionally, we tested the applicability of the Delta_R adjustment to account for a potential offset for the Thera dataset in comparison to the calibration curve, in light of the results of our analysis in the previous section of this paper. As discussed earlier, the precise cause of these observed offsets could be a combination of factors, including differences in growing season, tree physiology, laboratory pretreatment, and potentially latitude. The latitudinal gradient proposed by Büntgen et al. (Reference Büntgen, Wacker, Galván, Arnold, Arseneault, Baillie, Beer, Bernabei, Bleicher, Boswijk and Bräuning2018) also correlates with differences in growing season and offers a practical proxy for modeling purposes. We explore this by applying a Delta_R correction of 10 ± 3 which is the offset between the GrM oak vs. AA bristlecone pine measurement. This offset approximately reflects the difference between the trees that predominantly contribute to the calibration curve between 1700–1500 BCE (Irish and German oak) and the latitude of Thera. We recognise that this might be an oversimplification of multiple overlapping effects; however, latitude is the only variable we can control using the available data. With this compensation included the probability for a 16th century date for the eruption increases (Figure 7a). However, there are 882 ¹⁴C measurements in total between 1700 and 1500 BCE in IntCal20 and 198 measurements come from AA bristlecone pine, representing 22% of these data. Thus, just under a quarter of the IntCal20 data used to calibrate ¹⁴C dates from the Theran eruption come from the same latitude as the volcano, compared to 33% from German oak and 42% from Irish oak. So, while it seems highly likely that an offset exists, it would be significantly less than 10 ± 3 in the case where all the IntCal20 data are included. Applying this offset is less justified when using the AA BCP & MAMS-only curve, where 50% of the data in this period comes from the same latitude and would not be affected by a latitude-related offset, and we have therefore not included this version in the figures.

Since the ¹⁴C calibration curve over this period is affected by a plateau (Manning Reference Manning2024a), it is customary to make use of Bayesian modeling to incorporate constraints. To address this issue, we reran the most extensive model proposed by Manning (Reference Manning2024a) which integrates the Therasia olive shrub sequences (Pearson et al. Reference Pearson, Sbonias, Tzachili and Heaton2023), Thera olive sequence (Friedrich et al. Reference Friedrich, Kromer, Friedrich, Heinemeier, Pfeiffer and Talamo2006) as well as the VDL dataset (Manning Reference Manning2014; Manning et al. Reference Manning, Ramsey, Kutschera, Higham, Kromer, Steier and Wild2006) (see Manning Reference Manning2024a for the OxCal code and the underlying rationale). However, we have adapted the code with a uniform Boundary instead of Tau Boundary and also included a General Outlier model (Bronk Ramsey Reference Bronk Ramsey2009b). These models were run both with and without the aforementioned Delta_R adjustment. While this approach introduces more constraints to the model, it allows us to overcome the impact of the plateau. The Boundary labelled as Eruption is plotted on Figure 7b. The 95.4% probability ranges are narrower than those of the VDL model as a result of the additional constraints. There is less probability allocated to 16th century BCE, with the highest probability again clustering around 1600 BCE. Yet, despite the elevated ¹⁴C levels of AA bristlecone pine and MAMS datasets, the modeled results still do not support a timeline for the eruption as late as 1500 BCE.