Evidence for old carbon contamination in 14 C wiggle-match 1 age series for the 946 CE eruption of Changbaishan volcano 2 3

Abstract. Volcanic eruptions that are not historically attested are commonly radiocarbon dated by "wiggle matching" sequential 14C measurements of the rings of trees killed by the eruption against an accepted calibration curve. It is generally assumed that carbon laid down in the wood is uncontaminated by 14C-free ("old") carbon, although evidence for contamination is well documented. Often, ill-fitting ring ages are excluded from analysis. The ‘Millennium Eruption' of Changbaishan volcano on the China-DPR Korea border offers a valuable case study in wiggle match dating, since several independent groups reported age estimates before the determination and acceptance of a precise eruption year of 946 CE. Some of the discrepancies and incompatibilities between published dates were attributed to old carbon effects. Here, we apply a new methodology to correct for contamination levels of up to 4.5% old carbon to eight wiggle match date series for the Millennium Eruption. Without discarding ring ages, we find agreement indices as high as, or higher than, those for the published dates, and five of the eight date series yielded high-agreement-index eruption dates closer to 946 CE than the published dates. None of the five yield a best result at zero contamination. Differences between the eruption dates reveal a weak association with the direction of the sampled tree from the caldera, but no relationship with distance. Our results suggest that old carbon contamination is possible over a wide area, potentially leading to over-estimation of eruption ages by years, decades or more, cautioning against over-reliance on wiggle-match ages that are not corroborated by other lines of evidence. Our revised protocol that accounts for contamination offers a way forward in the application of wiggle match dating of eruptions and provides a platform for understanding discrepancies that exist when comparing wiggle match series.


Again for direct comparison, we followed Xu et al. (2013) in including a model incorporating a regional offset 25 of ± 10 years, using the Delta_R option in OxCal4.3.2. For the purposes of this initial study, contamination was 26 simulated by adding increments in Δt individually, but calibration algorithms could be modified to generate the 27 alternative fits and parameters for any desired range of the φ term.

28
We repeated the analyses, but without removing "low A age" samples (as above), for WM sequences 1 ages that did not reach the critical A value of 60% (for A overall and A model ) equivalent to a χ 2 test at 95% level for 2 a combination of normal distributions) and A comb (1/√n).

3
To obtain an eruption date estimate using all the WM data, we repeated the analyses for all trees using 4 the IntCal20 calibration curve (Reimer et al., 2020) to provide the most recent WM estimates for the eruption 5 date. Finally, we summed the probability distributions for the dates corresponding to the peak A index values.

6
Locations of these trees relative to the Changbaishan caldera when they were sampled are shown in

11
(2012), those at the Yengshan site, were buried together, one broken off but in growth position, and the other 12 horizontal: both had "perfect bark" and were unlikely to have been transported any distance. Both trees (DFHA, 13 DFHB-1) from the Dongfanghong site were removed from an array of logs exposed in a section through the 14 ignimbrite and could have been transported some distance.

18
(2012) concluded that the Hengshan "…trees were burned to death in situ".

19
To summarise, we modified the standard protocol for analysis and interpretation of WM age series 20 ( Fig. 3A), taking into account the potential for old carbon contamination of the wood samples (φ ≠ 0), and the 21 importance of independent correlative dating (Fig. 3B).

22
We plotted the mean tree death (assumed to be coeval with the eruption) dates and the probability 23 distributions of those dates for each WM series at φ = 0% and at the φ value corresponding to the first peak in 24 A comb with φ > 0%. For each of the three exemplar trees, we plotted A comb values for wiggle fits corresponding 25 to values of φ between 0 and 4.5%, and dates of tree death versus those A comb values, with and without regional 26 offset modelling, using the means (with standard deviation) of regional offsets (Fig. 4).

27
For all trees considered, the differences between eruption dates corresponding to φ = 0% and φ at the cosmogenic event (Fig. 5). A comb values were normalised as quotients against the actual critical value because 1 A comb varies with sample size.
2 Possible geographic effects on tree WM sequence ages were assessed by plotting the φ = 0% and first 3 A comb peak dates (as differences from the date anchored by the cosmogenic event) against distance and bearing 4 from the centre of the caldera (Fig. 6). 14 1025 CE, Fig. 4D-F) for which the A comb value was higher than at φ = 0%, could be argued to be the eruption 15 date. Even improving the fit for φ = 0% by removing the low A ring ages did not give an A comb value for the fit 16 at φ = 0% as high as that for φ ≈ 1% (Fig. 4A).

17
For tree YAB, there were two peaks in A comb , with the higher (with all low A ring dates removed) again 18 at c. φ = 1% (Fig. 4B). However, with all ring dates included, the highest peak was at φ = 0.25%, with a 19 marginally better fit than for φ = 0% (Fig. 4B).   1 change with different levels of contamination, which is consistent with their being exposed to, and 2 photosynthesising , only small amounts of old carbon (Fig. 4). The trees to the ENE of the caldera seemed to 3 have taken up consistent amounts of old carbon, whereas those to the south photosynthesised varying amounts 4 of old carbon before they were killed. 5 A "significant" wiggle match that assumes φ = 0% is not sufficient evidence to secure an eruption date 6 when there are other potential fits to the calibration curve for the wiggle match series ages which also yield 7 significant A values with φ ≠ 0%. Where such alternative fits exist, supporting evidence such as a direct 8 dendrochronological date for a cosmic ray event or a geochemically-identified tephra in a securely dated ice 9 core is needed. Indeed, if the first WM ages (Fig. 2) were the only ones available (as they were for some time),  Our conclusions on the present procedure where the contamination term φ is taken automatically to be 0% are 5 summarised in Fig. 9. It has been known for many years that carbon in vegetation, including trees, can be  (Fig. 3). We have shown that through systematic wiggle matching with the contamination term φ ≠ 0% 11 allows (1) improvement in the fit agreement A comb parameter of the age series, (2) better agreement of individual 12 wiggle match dates with the cosmogenically constrained eruption date for five out of the eight trees at different 13 locations relative to the caldera, and (3) a reduction in the number of ring age measurements that need to be 14 discarded to achieve a good fit for most wiggle match series. The trees that have systematic contamination 15 signatures could be explained by their proximity to volcano, and downwind location. We note that all trees 16 cannot provide a wiggle match eruption date close to the cosmogenically constrained eruption date even with 17 our new methodology and suggest that this may result from non-constant levels of old carbon contamination, 18 associated with proximity to locally variable CO 2 emissions.   Pasquier-Cardin, A., Allard, P., Ferreira, T., Hatte, C., Coutinho, R., Fontugne, M., and Jaudon, M. 1999.

27
Magma-derived CO 2 emissions recorded in 14 C and 13 C content of plants growing in Furnas caldera,