Placentia Bay is located immediately south of Newfoundland, Canada, on the north-western margin of the North Atlantic Ocean. The bay is bordered by the Avalon Peninsula to the east, the Burin Peninsula to the west, and the Isthmus of Avalon to the north (Fig. 1). To the south, the seaward opening of the bay is approximately 100 km wide. Water depths exceed 400 m in the bay, which is around 130 km long. Placentia Bay is typically free from sea ice year round, although ice can form between mid-February and late April during the coldest winters. Iceberg sightings in the bay are rare. Between 1974–2003 CE sightings only occurred in 7 years (30 sightings total) (Catto et al., 1999; Mello and Rose, 2005; Vale Inco, 2008). The hydrology of the bay is strongly influenced by the inner branch of the Labrador Current, with lesser input from the Slopewater Current, a minor bifurcation from the Gulf Stream (Catto et al., 1999) (Fig. 1). The cold, inner Labrador Current flows south from Baffin Bay as a surface current and includes substantial outflow from Hudson Strait (Drinkwater, 1996). In contrast, the Slopewater Current branches north from the Gulf Stream and brings warm, saline waters from the sub-tropics, at subsurface depths, towards southern Newfoundland.

During the Last Glacial Maximum (∼ Marine Isotope Stage 2), Placentia Bay was glaciated by the Laurentide ice sheet; as a result, drumlins, moraines, and megascale lineations are present across the sea floor (Shaw et al., 2006, 2013). The bay was deglaciated prior to the Younger Dryas climate reversal (12 800–11 600 cal yr BP, Mangerud, 2021), although the precise timing of ice retreat is not well constrained (Dyke et al., 2004; Shaw et al., 2006, 2013; Pearce et al., 2013). The varying sea floor topography has resulted in heterogeneous deposition rates, and differing basal ages are reported from sediment cores across the bay, which allows for the development of palaeo-records with various temporal lengths and resolutions. The potential for differing temporal records and sensitivity to elements of both the Labrador Current and the Gulf Stream make Placentia Bay an ideal natural laboratory for studying past ocean–atmosphere interactions. As a result, numerous studies have developed palaeoenvironmental records from the bay, all of which rely on radiocarbon chronology, necessitating the adoption of marine reservoir corrections (e.g. Jessen et al., 2011; Solignac et al., 2011; Pearce et al., 2014; Sheldon et al., 2016).

Core AI07-10G measures 460 cm in length and was drilled in 2007 from 231.3 m water depth at 47.2389∘ N, 54.6140∘ W in Placentia Bay, North Atlantic Ocean. Sheldon et al. (2016) presented the results from radiocarbon dating (Table 1), Itrax X-ray fluorescence (XRF) core scanning, and benthic foraminiferal assemblage analyses. These results were combined with analyses from two other cores in Placentia Bay (12G and 14G; Sheldon et al., 2016) to form a composite full Holocene record.

To quantify volcanic glass shard concentrations in core AI07-10G (reported as shards per gram of dried sediment), we processed continuous 5 cm wide samples taken throughout the sequence, with no gaps between samples, to identify sediment intervals where cryptotephra deposits might be found (rangefinder counts). We subsequently analysed the 5 cm intervals where higher abundances of tephra grains were identified at 1 cm intervals to pinpoint the position of cryptotephra deposits (Pilcher and Hall, 1992).

We extracted glass shards from the host sediment by drying samples at 105 ∘C overnight before immersing sediment samples in 10 % hydrochloric acid and sieving them at 80 and 25 µm. Larger size fractions (>80 µm) were retained; however, given the low shard concentrations in core AI07-10G (<40 shards per gram), these were not investigated further (Abbott et al., 2018a). Following this, we used stepped, heavy-liquid (sodium polytungstate) floatation at 2.00 and 2.50 g cm-3 to concentrate volcanic glass, which was mounted on slides and counted under a high-power microscope (Turney et al., 1998).

As no basaltic glass (which is denser than rhyolitic glass) was observed in the initial counts, we extracted glass shards for electron probe microanalysis (EPMA) by sieving samples at 20 µm, followed by heavy-liquid floatation at 2.15 and 2.45 g cm-3. The extracted material was then mounted in epoxy resin within acrylic stubs and polished to expose the internal glass surfaces before carbon coating (Lowe, 2011).

The chemical compositions of individual glass shards (one analysis each) from samples taken at 195–190 and 35–30 cm were determined by EPMA, with wavelength dispersive spectrometry on a JEOL 8900 Superprobe at the University of Alberta. A suite of 10 elements (Si, Ti, Al, Fe, Mn, Mg, Ca, Na, K, Cl) were measured using a 5 µm beam diameter with a 15 keV accelerating voltage and 6 nA beam current, with time-dependent intensity corrections applied to Na to compensate for the narrow beam (<10 µm) diameter (e.g. Jensen et al., 2008, 2021). In addition, we ran two secondary standards of known compositions alongside samples from Placentia Bay to check for instrumental drift and analytical precision: (i) Lipari rhyolitic obsidian ID3506 and (ii) Old Crow tephra (Kuehn et al., 2011). The major–minor element compositions of glass shards are presented as normalised weight percent (wt %) oxides in comparative diagrams. The complete dataset and associated standard measurements are reported in the Supplement (Tables S1, S2).

To incorporate chronological information from the ultra-distal cryptotephra isochrons identified in core AI07-10G, we developed two different Bayesian age–depth models using OxCal v 4.4.4 (Bronk Ramsey, 2009a). The complete code for both models is available in the Supplement (Supplement Sect. 1.1).

Model I is conceptually the same as the age–depth models described by Sheldon et al. (2016). In our model, however, radiocarbon dates (Table 1 were calibrated with the Marine20 curve (Heaton et al., 2020) with a single reservoir correction applied to the whole core (-29 ± 45 years). In this case, we have preliminary information regarding the most probable reservoir correction range based on radiocarbon dating of near-modern marine organisms. In Bayesian statistics, these data are called a prior, e.g. a representation of the state of knowledge regarding a parameter, expressed as a probability distribution, before considering all available information (e.g. stratigraphic context). For Model I, we used a prior distribution for the reservoir correction of the weighted mean of the 20 nearest points from Reimer and Reimer's (2001) marine reservoir correction database (Table S3), updated for use with Marine20 (-29 ± 45 years). The core top was also included as an age constraint. We assume an exponential prior at zero depth, from September 2007 CE (the approximate date of collection) decaying to 1000 years earlier with a time constant (τ) of 50 years. The deposition was modelled as a Poisson process (i.e. a P_Sequence; Bronk Ramsey, 2008) with a nominal number of depositional events (k0) of 1 per cm. The k parameter was permitted to vary within a wide range (i.e. 2 orders of magnitude on either side of k0) and was selected through Markov chain Monte Carlo (MCMC) iterations (Bronk Ramsey and Lee, 2013). These settings are the default for sequences with a depth scale in centimetres.

Model II is similar to Model I; radiocarbon dates are calibrated with the Marine20 curve, the same core top constraint is applied, and the model is formulated as a P_Sequence using the same parameters. However, Model II differs from Model I in two ways: (1) it includes cryptotephra deposits to constrain the chronology further, and (2) the model calculates multiple ΔR values – varying the radiocarbon reservoir offset throughout the sequence. We used Mazama Ash (7572 ± 18 yr BP; Sigl et al., 2016, 2022) and White River Ash eastern lobe (WRAe) (1097 ± 1 yr BP; Toohey and Sigl, 2017) as age constraints, both of which are geochemically verified in core AI07-10G (see Sect. 3.1, “Tephrostratigraphy”). Shard counts from both cryptotephra deposits consist of low concentrations and do not have a clearly defined, sharp peak (Fig. 2). These variable and broader peaks are likely caused by downward translocation of shards through sediment loading or bioturbation (Griggs et al., 2015) and complicate the precise stratigraphic depth of the isochrons. In order to incorporate this uncertainty into the Bayesian models we took a conservative approach and used age uncertainties associated with the 5 cm rangefinder counts rather than the 1 cm point finder counts. To do this, we first estimated the sediment deposition rate from Model I at the central depth of both tephra samples. Then, we propagated the depth uncertainty to the tephra age by adding uniform noise in the time dimension. The prior for each cryptotephra deposit was modelled using ages derived from ice core layer counting (a normal distribution; Sigl et al., 2016, 2022; Toohey and Sigl, 2017) plus chronologic sampling uncertainty (u), where u = sampling resolution / deposition rate.

Model II also differed from Model I and earlier approaches (Solignac et al., 2011; Sheldon et al., 2016) by including multiple independent ΔR estimates (i.e. each radiocarbon date had its own ΔR estimate; Bronk Ramsey, 2009b). Each ΔR value was defined (as above) by a mean correction of -29; however, the uncertainty was expanded to ±224 years (i.e. 4× the uncertainty from Reimer and Reimer's (2001) database). This conservative uncertainty regime was adopted to permit the Markov chain Monte Carlo (MCMC) approach inherent to OxCal's sequence modelling to generate an appropriate (non-truncated) posterior estimate for the corrections. A normal -29 ± 224 prior was assumed for all radiocarbon dates except one. For the uppermost radiocarbon date (AAR-15764), we provided an even more forgiving prior (a uniform distribution centred at 0 and spanning 2000 years). This prior was selected because the WRAe mean depth is only 2 cm above this sample (Fig. 2). The MCMC is kept flexible by giving it a wide uniform prior. Therefore, the tephra age can strongly inform the ΔR for this date, providing a clearer picture of the necessary reservoir effect around the time of WRAe.

We identified two discrete cryptotephra deposits in core AI07-10G (Fig. 2) that can be robustly correlated with volcanic eruptions in North America using multiple lines of evidence. Evidence includes stratigraphic order, shard morphology, and glass major–minor elements (wt %), which were interrogated using both bi-plots, compositional principal component analysis (PCA) (Filzmoser et al., 2009; Templ et al., 2011; Vera, 2020), and similarity coefficients (Supplement Sect. 1.2 and 1.3) (Borchardt et al., 1972). In both cases, the glass EPMA data were consistent and did not include glass shards with different chemical compositions or signs of weathering that might indicate reworking (Abbott et al., 2018a).

The presence of Mazama Ash and WRAe allows the Placentia Bay marine reservoir offset to be assessed at multiple points during the Holocene. Results from Model II, which includes the cryptotephra isochrons, show that around Mazama Ash the radiocarbon offset was moderately more negative (i.e. -126 ± 151 relative to the prior ΔR of -29 ± 45 years) (Fig. 4). The large uncertainty range (relative to the offset) associated with Mazama Ash is caused by our conservative modelling approach that uses the 5 cm rangefinder results to place the isochron and the slow accumulation rate at this point in the core (∼ 0.05 cm yr-1). Around WRAe, the radiocarbon offset was larger: -396 ± 144 (Fig. 4). Around the ages of both tephra deposits, ΔR values must be more negative than previously assumed to account for the tephra ages. That is, modelled ages are made to be older than would be suggested by the original prior. Therefore generally, less old carbon is contributing to the system at the study site than modelled for the global ocean. The ΔR varies throughout the age–depth model and particularly around the WRAe isochron. At this depth, there is a large shift in ΔR near WRAe. We modelled a posterior offset of -451 ± 151 years for radiocarbon date AAR-15764 but only -10 ± 213 years for radiocarbon date AAR-17060, 42 cm lower in the core. The reservoir age for both periods of tephra deposition was lower than indicated by Reimer and Reimer's (2001) marine reservoir correction database. However, this offset appears larger in the Early Holocene than in the Mid-Holocene. This apparent difference in ΔR may be explained by either artefacts in the chronology (e.g. the radiocarbon date AAR-15764 is inaccurate) or real variance in the age of waterbodies. However, as discussed below, the two tephra isochrons were deposited during periods characterised by different hydrographical conditions, and the difference in ΔR for the two tephra deposits likely reflects real differences in the radiocarbon age of the waterbodies affecting the site.

Across the whole core, there is an average ΔR difference between the two models of 74 years, but it is as high as 416 years at 34.5 cm, not far below the WRAe, and 126 years (a secondary maximum) directly at the mean Mazama depth. In most places, changes in the more flexible Model II do not represent a departure beyond the 2σ age range of Model I. Indeed, the only portion of the core that does exceed this range and precludes the implicit null hypothesis (no change) is near the WRAe (31.8–35.7 cm) (Fig. 4). Both cryptotephra isochrons push the Bayesian model towards older values (Fig. 4). It is possible that this is caused by inaccurate placing of the position of the cryptotephra isochrons and that Model I is correct. For this to be the case, both cryptotephra isochrons would be expected to occur deeper in the core (e.g. WRAe would have had to occur at 39.3 cm depth – almost 5 cm below the observed peak at 35–30 cm depth; Fig. 2) and the observed peak in shard abundance would need to have been reworked upwards into the overlying sediments. Upward movement of the cryptotephra deposits to an extent where the position of the isochron is misplaced seems unlikely, however, as in both cases shard counts are considerably higher at the denoted isochron depth (which already includes 5 cm uncertainty) than below.

Considering the marked departure of ΔR around the time of the precisely dated WRAe from the near-modern prior, we observe that radiocarbon reservoir effects can shift rapidly because of environmental and systemic changes (e.g. carbon source and ocean circulation shifts) over time. Further, the reservoir correction uncertainty may be more substantial in marine settings than suggested by near-modern samples. We conclude that conventional means of relaying proxy records over time often fail to account for time uncertainty. A natural remedy to this failure, and one we advocate for palaeoenvironmental proxy studies, is to propagate the age uncertainty in an age model ensemble to proxy records (e.g. McKay et al., 2021).