The Magdalen Islands (47°26^′54^′′ N, 61°45^′08^′′ W) are an archipelago of eight rocky islands situated on the shallow (< 100 m) Magdalen Shelf in the Gulf of the St. Lawrence (Fig. 1a) (Owens and McCann, 1980). Six of these islands are connected by a system of barrier beaches, tombolos, and lagoons (Bernatchez et al., 2012b). The islands are organized around basaltic hills surrounded by karstic foothills and sandstone platforms (Brisebois, 1981; Rémillard et al., 2013; Hétu et al., 2020). Previous work on the Late Quaternary history of the archipelago has provided insights into the pre- and post-glacial coastal environments of the area (Prest et al., 1976; Dredge et al., 1992). Deglaciation occurred approximately 20 ka before present (BP), leaving the islands largely submerged until about 15 ka BP. Subsequent glacial isostatic rebound caused the islands to emerge, resulting in a relative sea level (RSL) low-stand at 9–10 ka BP (Rémillard et al., 2016). This was followed by eustatic sea-level rise, which brought RSL up to about 20 m below present 6 ka BP. Since then, RSL has risen at a decelerating rate, eventually reaching modern levels (Vacchi et al., 2018; Barnett et al., 2019). Foraminifera, testate amoeba, and plant macrofossils from a salt marsh at Havre-Aubert Island show that RSL has risen at a rate of 1.3–2.0 mm yr⁻¹ for the last two millennia, a rate substantially slower than the > 4.0 mm yr⁻¹ observed during the 20th century (Barnett et al., 2017). This Late Holocene sea-level stabilization provided conditions favourable for the development of modern coastal landforms, including dunes, tombolos, spits, and beaches.

The coastal geomorphology of the Magdalen Islands reflects the combined influence of marine transgression and coastal erosion. Today, approximately 28 % of the coastline consists of rocky cliffs (Bernatchez et al., 2012b). In the southern islands, including Havre-Aubert Island where our research sites are located, these cliffs are composed mainly of relic coastal sands, loose unconsolidated sediments (colluvium), and a thick, porous, and highly erodible layer of red sandstone, known as the Drift des Demoiselles, all of which are predominantly composed of quartz (Brisebois, 1981; Dredge et al., 1992; Rémillard et al., 2013). Due to the friable nature of these materials, cliff retreat occurs at an average rate of 0.3 m yr⁻¹, supplying a steady stream of sediment to adjacent depositional systems (Bernatchez et al., 2012b). These sediments are redistributed alongshore by prevailing currents through longshore drift, contributing material to the intertidal zone and beaches (Morin, 2001). This sediment supply supports a dynamic coastal dune system, whose morphology varies with accumulation rates. In areas of high sand supply, dunes evolved into large-scale coastal features such as tombolos and sand spits, some of which have continued to prograde despite ongoing sea level rise (Rémillard et al., 2015). This has led to the formation of extensive beach ridge systems, highlighting the sustained and abundant sediment inputs from the surrounding cliffs. In contrast, most beaches are bordered by a single foredune that remains less stable due to more limited sand availability (Morin, 2001). These foredunes are frequently eroded by storm waves, forming dune scarps that expose sand to wind transport. Aeolian processes often carry this sand inland, where it accumulates in back-dune environments, sometimes resulting in landward dune migration.

The beach ridges located landward of tombolos and spits provide valuable insights into Late Holocene shoreline dynamics. Optically stimulated luminescence (OLS) dating indicates that many of these features developed over the last 2000–3000 years (Rémillard et al., 2015). The oldest known beach ridge system, located on the Havre-aux-Maisons Island, was dated to at least 2600 years ago. On Havre-Aubert Island, near our study area, a beach ridge complex in the north of the island formed between approximately 550 and 400 years ago.

The region is characterized by a strongly maritime climate, with cool summers and mild winters (Hétu et al., 2020). Daily maximum temperature is highest in July, reaching 21 °C for the period 1981–2010, and is lowest in February at 4.3 °C for the same period (Environment and Climate Change Canada, 2024). Monthly precipitation is relatively stable throughout the year, although it is highest in November at 109 mm and lowest in late winter (February, March, April) and July at around 75 mm (Environment and Climate Change Canada, 2024). Sea ice cover in the Gulf of St. Lawrence normally begins in December and extends until April each winter (Canadian Ice Service, 2022), although over the past two decades sea ice around the Magdalen Islands has been diminishing (Derksen et al., 2019).

Prevailing winds in the Gulf of St. Lawrence and around the Magdalen Islands are predominantly westerlies, with a distinct seasonality. Winds are primarily from the south southwest in spring and summer (40–50 km h⁻¹ for 15 % of the time), and from the northwest in the fall and winter, which are occasionally much stronger (70–132 km h⁻¹ for 15 % of the time) (Fig. A1). These westerlies contribute to more pronounced coastal erosion and narrower beaches on the islands' western shores (narrower beaches), while the eastern coasts remain relatively sediment-rich (Morin, 2001; Forbes et al., 2004).

We studied two peatbogs, Tourbière-du-Lac-Maucôque (TLM) (47°14^′30.9^′′ N, 61°55^′24.7^′′ W, ∼ 14 m a.s.l.) and Tourbière-de-l`Anse-à-la-Cabane (TAC) (47°12^′54.0^′′ N, 61°57^′48.4^′′ W, ∼ 20 m a.s.l.), located on Havre-Aubert Island, the southernmost island of the Magdalen Islands archipelago (Fig. 1a–c). TLM, a circular peatland with an area of 40.2 ha and an elevation 14 m a.m.s.l., is situated approximately 2 km from coastal beaches on the south and northeast side of Havre-Aubert Island (Fig. 1d) (Ministère de l'Environnement, de la Lutte contre les changements climatiques, de la Faune et des Parcs, 2021b). TAC lies parallel to the coast, covers 9.0 ha, has an elevation 20–22 m above mean sea level, and has its centre 250 m from the nearest south-facing red sandstone cliffs (Fig. 1d) (Ministère de l'Environnement, de la Lutte contre les changements climatiques, de la Faune et des Parcs, 2021a). Both sites are designated as protected areas by the Gouvernement du Québec due to their endangered populations of northern dwarf huckleberry (Gaylussacia bigeloviana [Fernald] Sorrie and Weakley), a small coastal shrub present at both locations. The dominant vegetation in the peatlands is Sphagnum spp. moss (Lindb.) and shrubs from the Ericaceae family, with black spruce (Picea mariana [Mill.] Britton, Sterns & Poggenburg) growing around the perimeter.

In August 2020, two peat cores were collected from each of TLM and TAC using a Russian peat corer in 50 cm × 5 cm drives, near the centre of each peat dome to target the deepest and oldest peat (Fig. 1d–e). Coring was conducted until dense bottom layers were reached that no longer permitted penetration of the corer. The two sequences at each site were less than 1 m apart and vertically offset by 25 cm. Due to the low peat density at the surface, the top 0–50 cm consisting of monoliths were extracted using a shovel. Cores and monoliths were stored at 4 °C at the Climate and Environmental Change Laboratory at Bishop's University. Continuous composite cores for each site were assembled by integrating core sequences with the surface monoliths (details in Appendix B).

Twenty-five radiocarbon (¹⁴C) dates (9 for TLM and 16 for TAC) were obtained from the A.E. Lalonde AMS Laboratory at the University of Ottawa, following the method described in Crann et al. (2017), and calibrated with the IntCal20 curve (Reimer et al., 2020). Most samples submitted for AMS dating were terrestrial macrofossils, with some bulk peat samples included. In addition, total lead-210 (²¹⁰Pb) activity was measured from 15 samples each taken at regular intervals from the surface monoliths of TLM and TAC at the GEOTOP Radiochronology Laboratory at the Université du Québec à Montréal (UQAM), using an acid-digestion sequence and alpha spectrometry, adapted from De Vleeschouwer et al. (2010).

Age-depth models were constructed using the Bayesian Plum algorithm from the rplum package version 0.5.1 in R (Aquino-López et al., 2018; Blaauw et al., 2024), which integrates ¹⁴C and ²¹⁰Pb data. While the ²¹⁰Pb profile from TAC reached the supported level, the profile for TLM did not extend deep enough to clearly identify it. However, because Plum uses prior information to estimate supported ²¹⁰Pb activity, we applied the supported level measured in TAC (supported mean = 17.5 Bq kg⁻¹) as a prior for TLM (Aquino-López et al., 2018; Chartrand et al., 2023; Cwanek et al., 2025). This approach is appropriate given the similar sedimentary and environmental conditions at the two nearby sites and allowed us to derive a reliable ²¹⁰Pb-based chronology for TLM despite the data limitation.

Peat type was identified using the Troels-Smith method (Troels-Smith, 1955). Samples of 1 cm³ were extracted every 8 cm (TLM) and 4 cm (TAC) and examined under a stereoscopic microscope (at 4–40× magnification) to estimate the relative abundance of each peat type. Four main peat types were identified: bryophytic peat (dominated by Sphagnum spp. moss remains), herbaceous peat (dominated by sedge remains), ligneous peat (dominated by twigs and bark fragments), and humous peat (highly degraded peat with indistinguishable components). Mineral-rich layers and some macrofossils (seeds, leaves, needles) were also noted using the Continental Paleoecology Laboratory collection from GEOTOP at UQAM as reference (Garneau, 1995).

Organic and mineral content in TLM and TAC were measured using loss-on-ignition (LOI) (Dean, 1974; Bengtsson and Enell, 1986; Heiri et al., 2001). Wet sediment samples (5 cm³) were extracted every 2 cm (TLM) and every 1 cm (TAC), dried during 15 to 17 h at 105 °C (dry weight, DW₁₀₅) to calculate dry bulk density, then combusted at 550 °C (LOI₅₅₀) in crucibles with lids, with weights recorded at each step. Mineral content (%) was calculated as the ratio of LOI₅₅₀ to DW₁₀₅.

The Aeolian Sand Influx (ASI) index was used to estimate inputs of coarse minerals transported by wind into the peatlands. ASI represents the mass accumulation rate of the sand-sized mineral fraction (g m⁻² yr⁻¹) and has been used as a storm proxy in several peat-based paleoclimate studies (Björck and Clemmensen, 2004; Orme et al., 2016a; Kylander et al., 2020; Vaasma et al., 2025). To isolate this fraction, subsamples previously analysed for LOI₅₅₀ were treated with 10 % hydrochloric acid (HCl) to remove carbonate and recalcitrant organic matter, and 10 % potassium hydroxide (KOH) to deflocculate sediments (Vaasma, 2008). Samples from TLM were then wet-sieved through a 63 µm sieve, and from TAC through a 125 µm sieve. Sediments retained on the 63 and 125 µm sieves correspond to the sand-silt boundary and fine/very fine sand boundary, respectively, based on the Wentworth (1922) classification. Grain-size thresholds were selected based on established aeolian transport theory and the geomorphological context. The initiation of sand transport by wind depends on grain size, surface moisture, vegetation, and local topography (Pye and Tsoar, 2008; Goslin et al., 2019). At TAC, the > 125 µm fraction would require minimum wind speeds of ∼ 22.5 m s⁻¹ (∼ 80 km h⁻¹) for entrainment and transport in saltation or suspension, while the > 63 µm fraction used at TLM has a lower transport threshold. Although the TLM site is farther from the coastline, justifying our choice of a lower mineral threshold, both peatlands are adjacent to unconsolidated sediment sources (e.g., sandstone cliffs, beaches), and the sparse vegetation likely facilitates aeolian input under storm conditions. ASI was calculated by multiplying the dry bulk density of the sieved sand fraction (g cm⁻³) by the peat accumulation rate (cm yr⁻¹), yielding a sand mass accumulation rate (g cm⁻² yr⁻¹), which was then converted from cm² to m² to give the final unit g m⁻² yr⁻¹.

We measured the inorganic geochemical compositions of TLM and TAC using an ITRAX Core Scanner (ITRAX–CS) from the Laboratoire de géochimie, imagerie et radiographie des sediments (GIRAS) at the Institut national de recherche scientifique (INRS) in Quebec City. Typically applied to marine and lacustrine sediments (Longman et al., 2019), the use of µ-XRF has recently been expanded to peat cores (Kylander et al., 2011; Orme et al., 2016b; Kern et al., 2019). Scans were acquired at 2 mm (TLM) and 1 mm (TAC) intervals using a Cr tube set to 40 kV, 10 mA and an acquisition time of 20 s. Prior to scanning, the core surfaces were smoothed to minimize surface roughness and reduce measurement variability. We applied quality control procedures on the resulting dataset based on count rate (CPS) and mean squared error (MSE) to exclude outliers affected by poor data quality. For each element, measurements with CPS values exceeding ±3 standard deviations from that element's mean, or exhibiting anomalously high MSE, were removed prior to analysis (Bishops, 2025). Our analysis used a subset of well-measured elements linked to terrigenous mineral inputs: K, Ti, Si, Fe, Mn, Ca, and S. Downcore variations are expressed as the proportion of each element relative to the total of these lithogenic elements (e.g., Ti (%) = 100 ⋅ Ti/[Ti + K + Si + Fe + Mn + Ca + S]). This normalization reduces the influence of variations in sample density, water content, and organic matter on geochemical trends (Jansen et al., 1998; Bertrand et al., 2023).

Two independent paleo-storm reconstructions were performed, one for each of TLM and TAC. Storm reconstructions were based on the ombrotrophic portions of the peat cores, excluding the early developmental stages and sections where ombrotrophic conditions could not be confidently established. These zones likely reflect mixed sedimentary processes, including fluvial or groundwater transport, rather than purely aeolian input (Sjöström et al., 2020). We selected a paleo-storm proxy by identifying µ-XRF elemental signals that co-varied with ASI in ombrotrophic peat. To do this, we conducted a PCA on the ombrotrophic sections of the TLM and TAC cores using the subset of seven elements, with ASI included as a passive variable.

While µ-XRF data were CLR-transformed prior to performing the PCA, the paleo-storm identification method employed a µ-XRF storm proxy normalized to the sum of terrigenous elements, or Proxynormalized= Proxy/[Ti + K + Si + Fe + Mn + Ca + S]. This approach mitigates matrix effects while preserving the natural variability and peak values essential for detecting extreme storm events, features that CLR transformation tends to smooth out.

To identify intervals where the µ-XRF storm proxy exceeded background levels, we used an approach adapted from Donnelly et al. (2015) and Lane et al. (2011). First, to account for variations attributed to long-term lithogenic changes, we detrended the storm proxy using a running average: 1Proxydetrendedz=Proxynormalizedz-RAproxyz,w, where RAproxyz,w is the running average of the storm proxy normalized at depth z, calculated over a window length w of 30 years. The threshold for identifying outlier proxy values was defined with the Tukey Rule of outlier detection (Tukey, 1977), following Winkler et al. (2022): 2Proxythreshold=Q0.75Proxydetrended>0+x⋅IQR(Proxydetrended>0), where Q0.75Proxydetrended>0 is the third quartile of detrended storm proxy values above zero, and x⋅IQR(Proxydetrended>0) is the interquartile range of detrended storm proxy values above zero, multiplied by a factor x of 0.75 for TLM and 1.5 for TAC, respectively, to account for differences in proxy values between the two cores. Detrended proxy values exceeding the threshold were considered storm events. To avoid overestimating the number of events due to consecutive storm proxy measurements associated with a single storm proxy peak, consecutive measurements were grouped together, and only the maximum value within each group was retained in the final paleo-storm record. Finally, the number of events per century was calculated using a 100-year running-sum.

Age–depth models indicate that peat initiation began around 5000 years ago (Fig. 2). The basal sands in both cores are interpreted as marking the onset of peat accumulation and there is no evidence to suggest that the full organic sequences were not recovered. These layers likely formed in a marine-influenced environment during a period of submergence before ca. 10.7 ka PB (Rémillard et al., 2017), though further data would be needed to confirm the depositional setting. The overlying organic-rich sediments are characteristic of early peatland development. At TLM, the presence of Menyanthes trifoliata seeds within mineral-rich ligneous peat suggests that the site initially formed through the terrestrialization of a shallow water body (Hewett, 1964; Kuhry and Turunen, 2006). In contrast, the basal peat at TAC, composed of mostly herbaceous remains and lacking Menyanthes trifoliata seeds, suggests that the peatland formed directly on a mineral substrate in a well-drained environment, likely through primary paludification – a direct shift from bare land to peatland, without an intermediary wetland step (Kuhry and Turunen, 2006).

The fen-to-bog transition in TLM and TAC was characterized by the increasing dominance of Sphagnum mosses and a positive correlation between mineral content and lithogenic elements such as Ti and K. However, the timing of this transition differed substantially, with TAC transitioning around 2050 BCE and TLM following over a millennium later, around 1000 BCE. This temporal discrepancy may reflect differences in the initial conditions of peat formation at each site. Additionally, the smaller size of TAC, which currently extends over 9 ha compared to TLM's 40 ha, likely rendered it more sensitive to climate and hydrological changes, such as the sharp increase in mineral content around 2600 BCE, as has been found in a study from Sweden (Sjöström et al., 2020). This interpretation is supported by TAC's lower peat accumulation rate, shallower ²¹⁰Pb profile, and lower ²¹⁰Pb influx compared to TLM, suggesting that TAC is more exposed, has reduced snow accumulation, and, consequently, has more unstable hydrological conditions (Perrier et al., 2022).

A notable aspect of peat development at TAC are the two distinct mineral-rich layers in Zone 3, which are absent in the TLM record. The timing of these mineral intervals should nevertheless be interpreted with caution due to the very low resolution of the TAC age model in this zone. While the older mineral-rich layer is fairly well constrained by two radiocarbon dates spanning from 970–475 BCE, the younger layer is only constrained by one upper radiocarbon date at 945 CE (Table 1). Although their origin remains uncertain, they are unlikely to represent single depositional events. TAC is situated ∼ 20 m a.s.l. and, based on sea-level indicators, was even higher during the late Holocene (Rémillard et al., 2016; Barnett et al., 2017), making marine overwash an improbable explanation. Likewise, an aeolian origin seems unlikely, as it would presumably have left a similar signal at TLM, which is not observed.

Although these layers are not reflected in TLM's mineral content or ASI records, their timing coincides with increased Ti extremes in TLM's storm reconstruction (Fig. 6a), suggesting that mineral inputs at TAC and Ti extremes at TLM may be linked to broader climatic variability. Indeed, the onset of slow peat accumulation at TAC coincides with the 2.8 ka event, a centennial-scale period generally characterized by cool and often dry conditions in the circum-North Atlantic (van Geel et al., 1996) that was recorded in nearby peatlands from Prince Edward Island (Peros et al., 2016), Anticosti Island (Perrier et al., 2022), and the north shore of the Gulf of St. Lawrence (Magnan and Garnau, 2014; Primeau and Garneau, 2021). The more pronounced response at TAC compared to TLM may reflect greater sensitivity to climate-driven hydrological shifts at this site, potentially due to TAC's smaller size and orientation perpendicular to a downslope gradient, which may have made it more vulnerable to groundwater fluctuations.

TLM and TAC exhibit distinct patterns of aeolian sediment input (Figs. 5–6), reflecting differences in sediment availability and local geomorphology. In the modern period (post–1851 CE), ASI values at TLM are an order of magnitude lower than those at TAC (Fig. 4), even after adjusting the grain-size threshold (63 µm for TLM vs. 125 µm for TAC). This disparity likely reflects TLM's greater distance from active sediment sources: TLM lies ∼ 2 km from the nearest beach, while TAC is situated only ∼ 210 m from eroding south-facing sandstone cliffs (Fig. 1d–e). Although the precise sources cannot be confirmed without sedimentological or geochemical fingerprinting, proximity strongly suggests that the closest sources contribute most of the aeolian material to each site.

At TAC, the nearby sandstone cliffs contain a highly erodible unit known as the Drift des Demoiselles, a friable red sandstone rich in hematite (Fe₂O₃)-coated quartz (SiO₂), K-feldspar (KAlSi₃O₈), and Ti-bearing minerals derived from underlying basaltic rocks (Brisebois, 1981; Dredge et al., 1992; Rémillard et al., 2013). These cliffs retreat at an average rate of 0.3 m yr⁻¹ (Bernatchez et al., 2012b), supplying a steady stream of sediment to coastal systems, which is then redistributed by longshore drift to local beaches (Morin, 2001). Importantly, the hematite coatings are typically removed during beach transport, meaning that Fe-rich grains are more likely to represent relatively non-weathered, locally sourced sediment from cliffs rather than beach-derived material.

While both cores show a strong association among ASI, Ti and K (Fig. 5), elements common in sandstone and beach-derived sands, the geochemical signals diverge in the modern period, suggesting different sediment sources. At TAC, positive correlations among ASI, Si, and Fe (Table E1) support a dominant input from nearby sandstone cliffs. Silicon is enriched in coarser grain-size fractions (Liu et al., 2019), which are more likely to be transported over short distances during storm events. The Fe enrichment further points to the presence of hematite-coated grains, characteristic of cliff-derived sediment.

In contrast, the TLM record suggests a primary source from local beaches, which are depleted in hematite due to longer transport pathways and reworking. This is consistent with TLM's greater distance from any major cliff exposures. However, beaches can still serve as significant aeolian sediment sources under storm conditions. Local observations show that under normal conditions, sand from exposed dune scarps is often transported inland and deposited in back-dune environments, contributing to landward dune migration (Morin, 2001). This supports the interpretation that, during storms, aeolian transport carries beach sand even farther inland and reaches the TLM bog, especially given its proximity to extensive beach ridge systems to the east (Fig. 1).

The TLM and TAC records together offer a 4000-year chronology of storm activity for the Magdalen Islands, adding critical insights to the limited records available from the northern extent of the North Atlantic hurricane track. Despite being aeolian reconstructions, the Magdalen Islands records show notable similarity with marine overwash records from lakes and blue holes spanning the past 2000 years from eastern Canada, the US, and the Bahamas (Fig. 9). This north-south transect across the western North Atlantic reflects modern storm tracks from the Magdalen Islands (Fig. 8a) and enables broader regional comparisons.

During the instrumental period (1851 CE–present), the storm records from the Bahamas to the Magdalen Islands show significant heterogeneity (Fig. 9). The only temporally overlapping record from eastern Canada, at Harvey Lake, New Brunswick, also indicates elevated storm activity after ∼ 1895 CE (Patterson et al., 2022). While the Harvey Lake study does not calculate events per century and is thus excluded from Fig. 9, it corroborates trends in TLM and TAC. In contrast, the record from Salt Pond, Massachusetts, does not show a comparable increase (Donnelly et al., 2015) (Fig. 9d), though a nearby record from Mattapoisett Marsh, just 18 km south of Salt Pond, documents numerous 20th century hurricanes strikes (Boldt et al., 2010). Donnelly et al. (2015) attribute this discrepancy to Salt Pond's avoidance of several modern hurricanes. Similarly, the Bahamian compilation, aggregating five blue hole storm records across the archipelago, shows a quiet 20th century (Fig. 9e). This trend reflects mostly inactivity at the southern sites (South Andros, Long Island, and Middle Caicos), while northern sites (Grand Bahamas and Abaco) exhibit increased storm activity (Wallace et al., 2021a; Winkler et al., 2023).

The Little Ice Age (LIA, ∼ 1300–1850 CE) (Magnan and Garneau, 2014) stands out as an active hurricane period across the western North Atlantic. At TLM, increased storm activity persisted between 1510–1785 CE (Fig. 9a). Similarly, TAC shows heightened activity 1320–1715 CE, with a calm period between 1400–1435 CE, and a peak between 1500–1600 CE (Fig. 9b). These trends align with Robinson Lake, Nova Scotia, which shows increased storm activity between 1475–1670 CE (Oliva et al., 2018) (Fig. 9c), and Harvey Lake, New Brunswick, where coarse silt beds were deposited between 1630–1640 CE (Patterson et al., 2022). Salt Pond also experienced elevated storm activity with 35 events between 1500–1600 CE (Donnelly et al., 2015) (Fig. 9d). In the Bahamas, storm activity peaked between 1570–1670 CE (Fig. 9e), matching patterns observed in the Magdalen Islands and nearby sites (Winkler et al., 2023). Notably, none of these records document storm activity during the early LIA (∼ 1200–1400 CE), as seen in the Magdalen Islands.

During the Medieval Climate Anomaly (MCA), which lasted between ∼ 900–1300 CE in the Gulf of St. Lawrence (Magnan and Garneau, 2014), storm activity decreased across the Magdalen Islands, consistent with calmer periods at Salt Pond and in the Bahamas (Fig. 9d–e). However, brief upticks between 1000–1100 CE at Salt Pond and 980–1030 CE in the Bahamas contrast with calm conditions in the Magdalen Islands. In the Bahamas, the increase is attributed to southern sites like Southern Andros (Wallace et al., 2019) and Middle Caicos (Wallace et al., 2021a), while northern sites, such as the Grand Bahamas, recorded a decrease in storm frequency from 550–1090 CE (Winkler et al., 2023) (see Fig. 8a for location). Between 500–750 CE, storm frequency at TLM aligns with increased storm activity in the Bahamas (720–820 CE), while Salt Pond documented an earlier active period between 300–600 CE, partly overlapping with the TLM record (Fig. 9c). During this period, sediment delivery at TAC was also elevated (Fig. 6b). Unfortunately, the lack of pre-LIA eastern Canadian records limits our understanding of storm dynamics earlier in the record, while the shorter duration of the Salt Pond and Bahamas reconstructions (< 2000 years) restricts longer-term comparison with the Magdalen Islands data. Overall, the similarities between our Magdalen Islands reconstruction and records from the eastern seaboard and the Bahamas lend additional confidence to our interpretations of ASI and Ti as storm proxies.

Storm records from the Magdalen Islands, eastern Canada, New England, and the Bahamas reveal consistent patterns suggesting that our Magdalen Islands records capture storms originating in the Main Development Region (MDR), located in the tropical Atlantic off West Africa. These storms typically track through the Bahamas before recurving towards the US east coast and eastern Canada (Kossin et al., 2010), a pattern consistent with the IBTrACS storm tracks observed in the Magdalen Islands since 1851 CE (Fig. 8a).

Hurricane activity in the MDR is strongly influenced by the Atlantic Meridional Mode (AMM), a coupled ocean-atmosphere system driven by variations in tropical Atlantic sea surface temperatures (SSTs) and vertical wind shear (VWS) (Kossin and Vimont, 2007; Vimont and Kossin, 2007; Klotzbach and Gray, 2008; Kossin, 2017; Ting et al., 2019). Positive AMM phases, characterized by elevated SSTs and reduced VWS, create favourable conditions for storm formation and intensification (Kossin, 2017). Although long-term AMM reconstructions are unavailable, the AMM is positively correlated with the Atlantic Multidecadal Variability (AMV) (Fig. 7d), and variations in the AMV are thought to indirectly modulate long-term hurricane activity by influencing AMM trends (Kossin and Vimont, 2007; Vimont and Kossin, 2007). The annually resolved AMV reconstruction from Lapointe et al. (2020) identifies a strong negative AMV during the LIA (not conducive to hurricane formation) and a more positive phase during the MCA (conducive to hurricane formation) (Fig. 9g). This trend is corroborated by other coarser-resolution AMV reconstructions (Mann et al., 2009b; Wang et al., 2017).

Nevertheless, storm records from eastern Canada (including that from the Magdalen Islands), New England, and the Bahamas show the opposite trend: increased storminess during the LIA and reduced activity during the MCA. For instance, Mann et al. (2009a), in a statistical reconstruction of paleo-hurricane frequency across the North Atlantic, estimated low hurricane counts during the LIA and higher counts during the MCA. A recent synthesis of proxy-based paleo-storm reconstructions spanning from Belize to New England (Yang et al., 2024) supports this, showing elevated storm frequency from 900–1200 CE, followed by a calm period from 1250–1700 CE, interrupted only by a peak in activity between 1375–1475 CE (Fig. 9f). This contrasts with widespread evidence from eastern Canada, New England, and the Bahamas showing frequent storms throughout the LIA. This persistent antiphase with the tropical Atlantic suggests that regional factors, such as storm track shifts or local intensification, may dominate at higher latitudes during periods of low tropical storm activity.

Several studies support this view. Patterson et al. (2022) and Oliva et al. (2018) proposed that elevated storm activity during the early LIA might reflect local warmer SSTs in the Gulf of Maine and the Sargasso Sea, despite cooler temperatures in the MDR. Dickie and Wach (2024) provide a concise summary of these mechanisms, highlighting that while the LIA experienced colder winter and lower annual mean temperatures, warmer summers may have been sufficient to fuel local storm intensification in mid-latitude North America. Ting et al. (2019) documented a dipole pattern over the past 150 years, where positive AMV conditions in the MDR coincide with unfavourable conditions along the US east coast – and vice versa – possibly shifting storm activity northward during negative AMV phases.

The Magdalen Islands records add significant value to western North Atlantic storm reconstructions. They extend evidence suggesting that mid- to high-latitude storm activity is governed less by conditions in the MDR and more by regional controls such as SST gradients, storm steering, and local intensification. As a long, uninterrupted record from a previously underrepresented region, our data help bridge the spatial gap between well-studied tropical sites and the North American coast, offering new insights into the climatic drivers of storm variability over centennial to millennial scales.