Learning from the past: Impact of the Arctic Oscillation on sea ice and marine productivity off northwest Greenland over the last 9,000 years

Abstract Climate warming is rapidly reshaping the Arctic cryosphere and ocean conditions, with consequences for sea ice and pelagic productivity patterns affecting the entire marine food web. To predict how ongoing changes will impact Arctic marine ecosystems, concerted effort from various disciplines is required. Here, we contribute multi‐decadal reconstructions of changes in diatom production and sea‐ice conditions in relation to Holocene climate and ocean conditions off northwest Greenland. Our multiproxy study includes diatoms, sea‐ice biomarkers (IP25 and HBI III) and geochemical tracers (TOC [total organic carbon], TOC:TN [total nitrogen], δ13C, δ15N) from a sediment core record spanning the last c. 9,000 years. Our results suggest that the balance between the outflow of polar water from the Arctic, and input of Atlantic water from the Irminger Current into the West Greenland Current is a key factor in controlling sea‐ice conditions, and both diatom phenology and production in northeastern Baffin Bay. Our proxy record notably shows that changes in sea‐surface conditions initially forced by Neoglacial cooling were dynamically amplified by the shift in the dominant phase of the Arctic Oscillation (AO) mode that occurred at c. 3,000 yr BP, and caused drastic changes in community composition and a decline in diatom production at the study site. In the future, with projected dominant‐positive AO conditions favored by Arctic warming, increased water column stratification may counteract the positive effect of a longer open‐water growth season and negatively impact diatom production.


| INTRODUC TI ON
Climate change is causing an accelerating decline in the seasonal duration and thickness of Arctic sea ice (Serreze & Stroeve, 2015), with important implications for marine primary production (e.g., Bergeron & Tremblay, 2014;Comeau, Li, Tremblay, Carmack, & Lovejoy, 2011;Tremblay et al., 2012). For the Arctic Ocean, satellite-based measurements suggest that annual net primary production has increased by 30% between 1998 and 2012 (Arrigo & van Dijken, 2015). This increase is largely attributed to thinning sea ice and more abundant and larger melt ponds that allow greater light transmittance and earlier onset of seasonal sea-ice melt, thereby enhancing both the under-ice productivity and the length of the growing season (Arrigo et al., 2012;Mundy et al., 2009). However, both satellite observations and in situ measurements indicate pronounced spatial and interannual heterogeneity in the response of primary producers to ongoing Arctic warming owing to local and remote factors that can have a synergistic impact on nutrient supplies and sea-surface conditions (e.g., ocean currents, vertical mixing, upstream biological processes; Hopwood et al., 2018;Tremblay et al., 2015). Determining future changes in Arctic primary production in a scenario of continued climate warming is therefore challenging, especially since our understanding of how primary production responded to climate forcing on longer timescales is limited by the paucity of comprehensive and highly resolved paleoarchives.
In addition to radiative forcing, variability in hemispheric-scale atmospheric circulation patterns can impact ecosystem functioning (Post & Forchhammer, 2002). The Arctic Oscillation (AO), also referred to as the Northern Annular Mode, is described as the difference in sea-level pressure between the Arctic Ocean and the middle latitudes (Cohen & Barlow, 2005;Rigor, Wallace, & Colony, 2002;Thompson & Wallace, 1998). The region defined for AO overlaps with the closely related North Atlantic Oscillation, which measures the regional variability in the atmospheric highand low-pressure systems over the Azores and Iceland, respectively (Wallace, 2000). Instrumental time series capture a dynamic relationship between the AO and interannual variations in storminess, sea-surface temperatures and sea-ice motion and conditions (e.g., Ogi & Wallace, 2007;Rodwell, Rowell, & Folland, 1999;Thompson & Wallace, 2001;Weckström et al., 2013;Zweng & Münchow, 2006). During a positive phase, lower-than-normal pressure over the Arctic traps cold air masses in the Arctic, but wintertime surface winds and thermodynamic processes can lead to an overall thinning of the sea-ice cover over the Arctic Ocean.
Therefore, although positive winter AO is associated with a cold anomaly in the surface atmospheric temperature over the Arctic Ocean and Greenland, it counterintuitively results in an earlier Arctic Ocean ice melt and increased export of meltwater and drift ice via the Transpolar Drift through Fram Strait (Rigor et al., 2002) and other gateways such as Nares Strait (Figure 1b). In the F I G U R E 1 (a) Greenland Ice sheet and historical median sea-ice extent   the WGC (e.g., Flatau, Tally, & Niller, 2003;Morley, Rosenthal, & deMenocal, 2014;Sarafanov, 2009). By contrast, a weaker EGC can be associated with a contracted subpolar gyre and increased northwestward flow of warm Irminger Water into the WGC (e.g., Holland, Thomas, De Young, Ribergaard, & Lyberth, 2008;Jennings et al., 2011). Thus, the dominant mode of atmospheric circulation can indirectly influence the strength and heat advection by the WGC, and amplify changes in surface-water properties and ocean-ice sheet interactions along the West Greenland margin. At the same time, the AO polarity index influences the consolidation or collapse of the ice arches in Nares Strait (e.g., Georgiadis et al., 2020), thereby also modulating the outflow of Arctic freshwater and drift ice through this gateway.
Given the key role of northern Baffin Bay as mediator between the Arctic and North Atlantic oceans, a number of studies documenting changes in regional ocean conditions in relation to Holocene climate fluctuations have been conducted (e.g., Caron, Rochon, Montero-Serrano, & St-Onge, 2019;Caron et al., 2018;Giraudeau et al., 2020;Hansen, Massé, Giraudeau, Pearce, & Seidenkrantz, 2020;Knudsen, Stabell, Seidenkrantz, Eiríksson, & Blake, 2008;Levac, de Vernal, & Blake, 2001;Mudie, Rochon, & Levac, 2005;Saini et al., 2020;St-Onge & St-Onge, 2014). However, the link between atmosphere-ocean forcing as expressed by swings in the dominant mode of the AO and changes in sea-ice seasonality and primary production in northern Baffin Bay has never been investigated. Here, we examine long-term changes in sea ice and primary production off northwest Greenland, northeastern Baffin Bay, using a series of biogenic proxies applied to a 7.34 m-long sediment core spanning the last c. 9,000 years. We also provide a continuous and high-resolution record of changes in diatom production and sea-ice conditions which can serve as a reference baseline to interpret recent changes and better predict the response of primary producers to future climate change.

| MATERIAL S AND ME THODS
The marine sediment core AMD14-204 was collected using a Calypso Square gravity corer, in the cross-shelf trough north of the

| Age model
The core chronology is based on 11 accelerator mass spectrometry 14 C dates from 15 samples of planktic and mixed benthic foraminiferal assemblages, a few also including ostracods (Table 1; Figure 2; Hansen et al., 2020). All dates were calibrated using the Marine13 radiocarbon calibration curve (Reimer et al., 2013) and an additional reservoir age correction (∆R) of 140 ± 30 years was applied (Lloyd et al., 2011). For the age-depth modeling, a depositional P_sequence model was used with a k value of 0.68 (Ramsey, 2008 Assuming that diagenetic enrichment (Robinson et al., 2012)

| Diatoms
Diatoms account for an important proportion of oceanic primary production and usually dominate the arctic microplankton community during spring blooms (e.g., Lalande, Nöthig, & Fortier, 2019;Tréguer et al., 2018). Their silicified cell walls (i.e., frustules) exhibit a large morphological diversity (size, shape, level of silicification, etc.), which gives them differential sinking and sedimentary preservation potential. Their short generation time and largely species-specific response to environmental conditions make them useful indicators of changes in sea-surface conditions. While some diatoms thrive in the bottom layer of sea ice, a seasonal succession of taxa typically follows sea-ice breakup leaving a unique, though fragmentary, environmental signature in the underlying sediment. Following nutrient exhaustion, fast-growing genera, including Chaetoceros, can produce a rain of small, highly silicified and fast-sinking resting spores that preserve well in the sediment and attest to past productivity levels (Abelmann, Gersonde, Cortese, Kuhn, & Smetacek, 2006). Subfossil diatom abundance and assemblage composition can thus be used to infer information about past primary production and its link to climate-sensitive parameters, including sea ice.
Samples for diatom analyses were prepared using c. 0.3 g of dry sediment and following the standard methodology described in Crosta et al. (2020;Data S1
IP 25 is produced by at least three pan-Arctic diatom species (Haslea spicula, Haslea kjellmanii and Pleurosigma stuxbergii var. rhomboides; Brown, Belt, Tatarek, & Mundy, 2014, Limoges, Massé, et al., 2018 that thrive in the bottom horizon of sea ice. This source-specific molecular compound is produced within the sea-ice matrix during the spring bloom and deposited in seafloor sediments following ice melt. In marine settings, high-sedimentary IP 25 contents are typically interpreted as indicating enhanced sea-ice concentrations, whereas the absence of IP 25 in sediments can either reflect perennial seaice cover, ice-free conditions or insufficient production of source diatoms (e.g., Belt, , 2019. HBI III is known to be biosynthe- . The seasonal sea-ice biomarkers were analyzed at a 1 to 10-cm sampling interval, which corresponds to an effective age resolution ranging from c. 12 to 165 years (mean: ~33 years).

| RE SULTS
3.1 | TOC, TN, δ 15 N and δ 13 C TOC contents remain low and range between 0.59 and 1.52 wt% (Data S2). The lowest values were measured in the lowermost F I G U R E 2 (a) Age-depth model for core AMD14-204. The light-blue envelope represents the modeled 1 sigma range, and the blue line marks the modeled median age. The light-shaded areas for each radiocarbon date indicate the probability distribution prior to modeling, whereas the darker areas indicate the posterior probability distribution. The color code for the dates indicates the material used: Blue: mixed benthic foraminifera, red: mixed planktonic foraminifera, gray: mixed planktonic and benthic foraminifera, green: mixed ostracods, planktonic, and benthic foraminifera.

| Diatoms
A total of 60 diatom taxa were identified (Data S3). Diatom concentrations range from <1 to c. 6 × 10 7 valves/g dry sediment, which translates into fluxes between c. 0 to 4 × 10 9 valves surface area −1 year −1 (Data S4). Our data reveal a strong temporal variability in Holocene diatom fluxes, with a 24× difference between the lowest and peak diatom fluxes.  borealis (resting spore), which largely dominates the assemblages.
After 4.4 kyr BP, a stepwise decrease in total diatom fluxes is observed. At the same time, the assemblages show a decrease in the relative abundances of T. antarctica var. borealis (resting spore) and a strong increase in the marginal ice zone taxa. At 3 kyr BP, an abrupt and pronounced decline in the total diatom fluxes is accompanied by a rapid shift to a new regime of diatom productivity: a sharp decline in the marginal ice zone assemblage concurs with a marked increase in the share of the "drift-ice/pack-ice" and "summer subsurface" taxa.
Although still a dominant contributor to the assemblages, T. antarctica var. borealis (resting spores) shows lower relative abundances from 3 kyr BP. Relatively low diatom fluxes, with slight oscillations, prevail for the rest of the core record. However, fluxes were particularly low between 3 and 2 kyr BP. From 2 kyr BP, a progressive increase in Rhizosolenia spp. is observed, whereas the contribution of T. longissima, occupying the same ecological niche, decreases to moderate values. The share of both the "drift-ice/pack-ice" and "summer subsurface" assemblages remains high for the rest of the core.

| Sea-ice biomarkers
The sea-ice biomarker IP 25 was detected throughout the entire record, attesting to the presence of a seasonal sea-ice cover in the study area over the past 9.2 kyr BP. Fluxes of IP 25 are moderately high at the bottom of the core, increase starting from 7.8 kyr BP, peak at around 7.4 kyr BP, and then progressively decrease from 6.6 until to 4.4 kyr BP (Figure 3; Data S5). A slight increase in the IP 25 fluxes is also observed between 3.9 and 3.2 kyr BP. Between 3.2 and 1.5 kyr BP, IP 25 reaches its lowest values. Following this interval, IP 25 rises again to reach a maximum around 0.2 kyr BP.
The earliest peak in IP 25 is followed by an increase in HBI III fluxes starting around 7 kyr BP. HBI III remains relatively high until 6.1 kyr BP and declines for about 600 years before increasing slightly again at 5.5 kyr BP (Figure 3; Data S5). Generally low HBI III fluxes after  Figure 3).

| Primary production and sea-ice dynamics on the Northwest Greenland shelf
As illustrated by previous studies based on dinoflagellate cysts, foraminiferal assemblages, and geochemical data ( The arrows indicate that ice particulate OM is more enriched in 13 C compared to pelagic particulate OM values due to limited atmospheric CO 2 exchange in the sea ice (e.g., Rau, Takahashi, Desmarais, Repeta, & Martin, 1992), and land-derived OM yields higher TOC:TN ratios (20-100; Meyers, 1994) than aquatic OM (typically between 4 and 10, Meyers, 1994). Our data indicate a pelagic-dominant signature, with varying mixing ratios between pelagic, sympagic, and terrestrial organic matter. In general, the elemental and isotopic signature of the organic matter shows that samples younger than 4.5 kyr BP (indicated by the gray box), and especially between 3 and 1.2 kyr BP, contain higher ratios of ice-derived organic material populations in nutrient-depleted waters (Booth et al., 2002). The high δ 15 N values, along with low TOC and diatom fluxes, are coherent with a scenario of limited surface-water nitrate availability and late sea-ice breakup. At the closely located core GeoB19927, the IP 25 and HBI III trends are comparable to ours, and brassicasterol and dinosterol fluxes show relatively high but decreasing values from c.
9.5 to 7.8 kyr BP (Saini et al., 2020). While our proxy signals may have been somewhat diluted by elevated clastic input during this interval, all proxies nonetheless point toward a low to moderately productive environment. This agrees with the extensive sea-ice cover, marked meltwater and detrital sediment discharges also documented by Caron et al. (2019), and low entrainment of Atlantic water into the WGC inferred from foraminiferal assemblages (Hansen et al., 2020), during an interval of rapid thinning of the Melville sector of the Greenland Ice Sheet . It is also between c. fluxes accompanied by a peak contribution of A. curvatulus, a species strongly associated with heavy pack-ice in northern Baffin Bay (Williams, 1986(Williams, , 1990, likely further suggest an episode of heavier sea-ice cover centered at c. 7.2 kyr BP. this study); (h) relative contribution of the "summer subsurface" assemblage (%; this study); (i) relative contribution of the "marginal ice zone" assemblage (%; this study); (j) relative contribution of the "drift-ice/pack-ice" assemblage (%; this study); and (k) total diatom fluxes (valves unit surface −1 year −1 , dark green includes Chaetoceros spores and light green excludes Chaetoceros spores (this study). The gray areas indicate intervals of persistent positive AO phases (Darby et al., 2012;Funder et al., 2011  interval (Hansen et al., 2020;Jennings et al., 2011Jennings et al., , 2014Jennings, Knudsen, Hald, Hansen, & Andrews, 2002;Perner et al., 2013). It is worth noting that low amplitude peaks in HBI III were always coeval with events of reduced Atlantic benthic foraminifera species at the study site (Hansen et al., 2020), possibly indicating episodes of decreased advection of Atlantic Water, and increased water column stratification or polar inflows.

| From primary production optimum with increased
After 4.4 kyr BP, the long-term decline in IP 25 ceases and diatom production declines stepwise. The same declining trend is observed in the dinosterol content of core GeoB19927 (Saini et al., 2020). A gradual increase in the contribution of the marginal ice zone diatom group (Figure 6) hints at a progressive return to colder conditions, while higher δ 13 C values suggest increased contribution of icederived organic matter to the sediments (Figure 4). These changes coincide with the inception of a stable and recurrent North Water polynya dated to c. 4.4 cal kyr BP (Davidson et al., 2018). This timing also coincides with a snowline lowering across 15 different ice caps in central West Greenland, which has been interpreted to reflect persistent summer cooling (Schweinsberg, Briner, Miller, Bennike, & Thomas, 2017). During the entire interval, short-term increases in HBI III fluxes were synchronous with low amplitude rises in IP 25 fluxes, suggesting that the orbitally driven long-term Neoglacial cooling (Vinther et al., 2009) associated with decreasing summer insolation in the Northern Hemisphere was interrupted by short events of fluctuating sea-ice conditions at the study site.
Starting around 3.5 kyr BP, the major increase in the contribution of the marginal ice zone group, and especially F. reginae-jahniae, suggests the re-establishment of more persistent seasonal sea ice ; Figure 6). Interestingly, this is synchronous with an increase in the dinoflagellate productivity reported at core GeoB19927 (Saini et al., 2020), and marked increase in the dinofla-

| Primary production decline, extensive sea-ice cover, stratified water column, and weakened Atlantic water signature of the WGC (c. 3.0-1.2 cal kyr BP)
We record an abrupt change in the diatom regime taking place at 3 kyr BP. IP 25 , HBI III, and diatom fluxes (including Chaetoceros spores) decline to minimum values at the beginning of this interval.
A steep decline in the contribution of the marginal ice zone group is shortly followed by a pronounced increase in the "drift-ice/pack-ice" and "summer subsurface" assemblages ( Figure 6). The drift-ice/packice assemblage includes M. arctica, a species known to form dense aggregates under sea ice, which can be abundant under drift ice and has been observed to form free-floating filaments in the meltwater layer of the Arctic Ocean (Poulin, Underwood, & Michel, 2014;von Quillfeldt, Ambrose, & Clough, 2003). It secretes gelatinous extracellular polysaccharide substances that facilitate its anchoring to the ice matrix (Krembs, Eicken, & Deming, 2011) and has been linked with widespread spore formation following nutrient depletion and grazing activities toward the end of the spring bloom in Baffin Bay (Lafond et al., 2019). In surface sediments and sediment traps from Greenland, M. arctica is strongly associated with spring/summer sea ice (Krawczyk et al., 2017;Luostarinen et al., 2020). The drift-ice/pack-ice assemblage also includes A. curvatulus, a coldwater species that is associated with high sea-ice concentrations (Oksman et al., 2019) and heavy pack ice in northern Baffin Bay (Williams, 1986(Williams, , 1990. The summer subsurface assemblage includes the genus Coscinodiscus, containing species known to thrive at low light irradiance levels under pack ice (Duerksen et al., 2014;Kemp, Pike, Pearce, & Lange, 2000), and contributing substantially to the early spring bloom in the North Water polynya (Lovejoy, Legendre, Martineau, Bâcle, & von Quillfeldt, 2002). Outside of the polar regions, Coscinodiscus spp. (such as the dominant Coscinodiscus species in our study) are an important constituent of the subsurface chlorophyll maximum (Odebrecht & Djurfeldt, 1996;Weston, Fernand, Mills, Delahunty, & Brown, 2005). This assemblage also  (Knudsen et al., 2008;Oksman et al., 2019), and marine to brackish conditions (Pearce, Weckström, Sha, Miettinen, & Seidenkrantz, 2014). Albeit distinct from the rest of the core, this diatom assemblage has a strong sea-ice fingerprint.
Additionally, Thomas, Briner, Ryan-Henry, and Huang (2016) reported a steady decrease in winter snowfall in western Greenland from 4.4 to 2 kyr BP, implying a more extensive sea-ice cover that limited not only moisture availability but also primary production during this interval. In this context, our low IP 25 flux values appear puzzling. A similar low IP 25 signal from the neighboring core GeoB19927 has been interpreted to reflect low sea-ice concentrations under the influence of a persistent WGC (Saini et al., 2020).
However, this interpretation is difficult to reconcile with the overall low primary production, relatively high organic matter δ 13 C, sudden decrease in the contribution of planktonic and benthic Atlantic Water foraminiferal species at the study site (Hansen et al., 2020), and low dinocyst fluxes dominated by cysts of the spring bloomer P. dalei (Caron et al., 2019). Our proxy record suggests that conditions other than sea-ice presence may have affected the growth of IP 25 -producing species or IP 25 biosynthesis during this interval.
Changes in nutrient availability can exert a major control on total primary production and influence the cellular HBI synthesis in diatoms . Considering that the assemblages are composed of large diatoms that typically benefit from higher nutrient levels (i.e., larger diffusion boundary layer), the hypothesis of a decrease in the nutrient availability is unlikely. Alternatively, a change in the ice matrix linked to reduced sea-surface salinity may have impacted the settlement efficiency and growth of IP 25 -producing sympagic algae (Gosselin, Legendre, Therriault, Demers, & Rochet, 1986;Poulin, Cardinal, & Legendre, 1983) as seen in coastal sediments and sea ice from Greenland fjords (Limoges, Massé, et al., 2018;Ribeiro et al., 2017). This could partly explain the presence of sub-ice colonial species such as M. arctica that have developed strategies to mitigate the effect of unfavorable sea-ice microstructure (Krembs et al., 2011). To the north, evidence of instabilities in the Kane Basin ice arch   Furthermore, we argue that the drastic changes in the diatom species composition and production reflect an important oceanographic reorganization during this interval (see Section 4.2).
From 2.3 to 1.2 kyr BP, HBI III increases sharply and the diatom assemblages show a progressive increase in Rhizosolenia spp. and F.
cylindrus, and a decrease in T. longissima. The proxy signal suggests enhanced surface-water cooling and freshening during this interval, likely partly associated with increased meltwater input. This is supported by the observed increase in IRD between 2.3 and 2 kyr BP at our study site , and is coherent with the retreat of the Qangattaq ice cap on the Nuussuaq peninsula between c. 2.5 and 1.9 kyr BP (Schweinsberg et al., 2017) and the Greenland ice-sheet margin in Northern Nunatarssuaq between 2.1 and 1.6 kyr BP (Farnsworth et al., 2018), at the transition to a period of significant boreal atmospheric warming that is commonly referred to as the Roman Warm Period (~2.0-1.65 kyr BP, Wanner et al., 2008).
Biomarker and foraminiferal datasets further suggest the potential complete collapse of the Kane Basin ice arch in Nares Strait during this interval , which would have further promoted the export of freshwater and (multiyear) ice into northern Baffin Bay.

| Low primary production, extensive but fluctuating sea-ice conditions (c. 1.2-0.2 cal kyr BP)
The last 1.2 kyr BP were characterized by a marked increase in IP 25 and generally stable HBI III fluxes, high contribution of the packice/drift-ice and summer subsurface assemblages, moderate contribution of the marginal ice zone assemblage, and relatively high organic matter δ 13 C values, suggesting the presence of an overall extensive, but fluctuating sea-ice cover (Figures 4-6). While diatom fluxes remain low for the complete interval, the signal is overprinted by high frequency events of low amplitude increase and decrease in diatom production. Saini et al. (2020) also reported low brassicasterol and dinosterol fluxes. Collectively, the diatom, brassicasterol, and dinosterol signals contrast with the high primary productivity reconstructed using the dinoflagellate cyst-based modern analogue technique (MAT; Caron et al., 2019). We hypothesize that the MAT signal is steered by the dominance (50%-60%) of cysts of the phototrophic species P. dalei, which often occurs in large numbers in the productive waters of the early season bloom, but whose intense, transient pulses of cyst production rather seem to be favored by the highly stratified waters resulting from ice melt (e.g., Heikkilä et al., 2014;Howe, Austin, Forwick, & Paetzel, 2010). During this interval, low primary production would have resulted from a relatively extensive seasonal sea-ice cover and comparatively short open-water season during an interval of generally cold atmospheric temperatures (Lasher et al., 2017;Lecavalier et al., 2017). A re-advance of the Upernavik Isstrøm was dated to c. 0.6 kyr BP (Briner, Hakansson, & Bennike, 2013). At the top of the core, the sharp increase in HBI III and Rhizosolenia spp. hint at a possible return to more stratified conditions and increased freshwater influence.

| Impact of millennial-scale AO variability on primary production
Millennial and centennial variations in the dominant atmospheric mode have been identified in various studies. These are based on, to name a few, indicators of sea-ice drift patterns in the Arctic Ocean, such as ice-rafted iron in the Beaufort Sea (Darby et al., 2012) and the abundance and origin of driftwood on the Northern Greenland shores (Funder et al., 2011), as well as tracers of exceptional rainfall events in North American lakes (Noren, Bierman, Steig, Lini, & Southon, 2002). While a dominantly positive AO phase was suggested for the northern Greenland Holocene Thermal Maximum between 8.5 and 6 kyr BP (Funder et al., 2011), it is only by ~6.8 kyr BP that modern-type ocean surface circulation presumably was established in the North Atlantic (Van Nieuwenhove et al., 2018).
Given this and the lower resolving power of the diatom analyses (due to low diatom valve concentrations) in the lower part of our core (9.2 to 6 kyr BP), we refrain from linking our proxy record to a predominant atmospheric circulation pattern during this interval.
We however reconstruct moderate diatom production associated with increased Arctic water and drift-ice influence, and a stratified water column. From 6 kyr BP and as solar irradiance decreases, we note that the onset of a 3,000-year phase of dominantly negative AO (Darby et al., 2012;Funder et al., 2011;Staines-Urías et al., 2013) is associated with sustained higher diatom fluxes and abundance of Atlantic benthic foraminiferal taxa. The proxy signal suggests a longer duration of the open-water growth season and generally warmer and strengthened WGC during this interval. The initial phase of the Neoglacial cooling was then marked by the establishment of progressively more extensive spring sea ice, which translated into a stepwise overall decrease in primary production starting at 4.4 kyr BP and sharp increase in the dominance of spring marginal ice zone taxa.
This important cooling event was further associated with a slowdown of the Atlantic meridional overturning circulation around 3.1-2.4 kyr BP (Oppo, McManus, & Cullen, 2003). As summarized by Morley et al. (2014), this shift is thought to be partly associated with a strengthened EGC provoking a southeastward movement of the subarctic front and longitudinal stretching of the subpolar gyre.
This cooling event coincides with a swing from persistently negative to positive AO at c. 3 kyr BP (Darby et al., 2012;Funder et al., 2011; Figure 6), reinforcing the hypothesis that millennial-scale AO-driven changes in ocean conditions may have prompted changes in sea ice and primary production off northwest Greenland during the mid-to-late Holocene. While positive winter AO is typically associated with colder than normal atmospheric temperatures in West Greenland (Box, 2002), the long-term decreased contribution of the warm Irminger Current into the WGC (e.g., Flatau et al., 2003;Morley et al., 2014;Sarafanov, 2009) may have exacerbated atmospheric cooling along the West Greenland margin from 3 kyr BP (Figure 7). Such a reduction of Atlantic-sourced water to the WGC from 3 to 2.3 kyr BP would have contributed to later sea-ice melt. This hypothesis is supported by the major decline in both the Atlantic water foraminiferal species (Hansen et al., 2020)

| Future evolution of marine primary production
Our proxy record suggests that over the last 9,000 years, mod-  (Cohen, Pfeiffer, & Francis, 2018), neither the highly productive interval from 6 to 4.4 cal kyr BP, nor any other Holocene interval can be regarded as a direct analog for the future. Instead, earlier and rapid seasonal sea-ice melt may affect the timing and production of fast-growing sea ice-associated taxa (e.g., Tedesco, Vichi, & Scoccimarro, 2019).
Additionally, our long-term record clearly shows that thermal and freshwater-induced upper water column stratification resulting from atmospheric warming, increased glacial meltwater input and polar water/drift-ice throughflows such as those taking place during positive AO intervals and predicted to increase in a warming Arctic, have a strong impact on community structure and are generally detrimental to diatom productivity. Hence, future diatom productivity in the area will be modulated by the balance between the effects of a longer open-water growth season versus the effects of an enhanced water column stratification. This will certainly result in changes in microalgal community structure and phenology.

| CON CLUS IONS
Our multiproxy record supports the hypothesis of a pervasive effect of the dominant AO phase on sea-ice conditions and diatom production on the northwest Greenland shelf. Most notably, we show that an important and rapid decline in primary production starting around 3 cal kyr BP was coeval with a shift from low to high AO polarity values.
Changes in sea ice and diatom productivity over the past 9,000 years on the northwest Greenland shelf can be summarized as follows: 1. Extensive seasonal sea-ice cover from 9.2 to 7.7 cal kyr BP associated with low to moderate diatom production during an interval of significant meltwater influence and deglacial opening of the Nares Strait gateway.
2. Polar water outflow event centered at c. 6.6 cal kyr BP associated with increased HBI III fluxes and a moderately high diatom production.
3. Progressive decline in seasonal sea-ice concentrations from 6 cal kyr BP culminating with primary production optimum.

F I G U R E 7
Simplified schematic representation of surface circulation under persistent low and high AO polarity index. The areas colored in red and blue reflect the milder and colder atmospheric winter temperatures during negative and positive AO phases, respectively. (a) During AO(−), a strong Beaufort gyre promotes the formation of multiyear sea ice and is linked to generally reduced export of freshwater and ice from the Arctic Ocean via Fram Strait (Rigor et al., 2002). The transpolar drift stream follows the Lomonosov Ridge (Steele et al., 2004;Talley, Pickard, Emery, & Swift, 2011). The East Greenland Current, an extension of the transpolar drift, is relatively weaker. Multiyear ice and weaker winds may facilitate the formation of ice bridges in Nares Strait, restricting export fluxes through this gateway . In the North Atlantic, a relatively contracted subpolar gyre facilitates the entrainment of warm and salty Irminger waters into the WGC (e.g., Morley et al., 2014;Sarafanov, 2009). (b) During AO(+) a weaker Beaufort gyre is associated with thinner Arctic Ocean sea ice, especially in the eastern sector (Rigor et al., 2002), which enhances sea-ice mobility and favors the export of solid and liquid freshwater via the main Arctic gateways. The transpolar drift stream flows nearly directly from the Bering Strait to the northern side of Greenland (Steele et al., 2004). This translates into increased EGC strength. Thinner Arctic sea ice and strong winds may contribute to weaken the Nares Strait ice bridges , promoting freshwater and drift-ice exports through this gateway. Export across the Canadian Arctic Archipelago channels is mainly in the form of liquid freshwater (Serreze et al., 2006). Atmospheric temperatures over Labrador, Greenland, and the western subpolar North Atlantic decrease (Talley et al., 2011). AO(+) is associated with strong westerlies, cold sea-surface temperatures in the Labrador Sea and subpolar gyre, and expansion of the subpolar gyre, with consequent decreased northwestward entrainment of Atlantic waters, which are confined to the easternmost part of the North Atlantic (e.g., Morley et al., 2014;Sarafanov, 2009)

ACK N OWLED G EM ENTS
Core AMD14-204 was collected as part of the GreenEdge project funded by ANR and the Total Foundation. Ship-time was funded by the ERC-STG-ICEPROXY project. We acknowledge the help of Drs. Guilmette and Van Nieuwenhove for biomarker analyses and in picking foraminifera. We thank Drs. Darby and Caron for providing IRD and dinocyst data. This project was funded by FRQNT (188947)

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that supports the findings of this study are available in the supplementary material of this article. Subfossil diatom counts are also available from the corresponding author upon request.