The mineralogy found at Site GS06-144-03 suggests several lithological sources for the IRD which may be associated with icebergs or sea ice originated from different areas. Volcanic rocks mainly outcrop surrounding Denmark Strait, in Iceland and the Geikie Plateau area on the East Greenland coast (Bailey et al., 2012; Henriksen et al., 2009). Volcanic glass can also be atmospherically transported after volcanic eruptions and be ultimately incorporated in the ice as it has been shown in Greenland ice core records (Grönvold et al., 1995). This is very likely the case of the white VG fragments found in our record because our counts of white VG (Fig. 2) do not suggest the presence of any discrete layer that could be associated with any dated Icelandic eruption (Gao et al., 2008; Sigl et al., 2015). This type of IRD was probably deposited on the top of glaciers and sea ice near Iceland and the East Greenland coast and then transported in the ice through the EGC. Although some of those volcanic shards ejected to the atmosphere could have fallen directly in the sea, the preferentially eastward dispersal pattern of Icelandic tephra follows the predominantly westerly winds in the stratosphere (Lacasse, 2001) and, hence, the amount of volcanic glass transported by winds to the study site must be rather small. Previous studies suggested the significantly low amounts of tephra transported towards Greenland prevent finding layers that can be associated with volcanic eruptions (Jennings et al., 2014). After detailed geochemical studies Jennings et al. (2014) could not recognize any specific layer that could be used as a tephrochronological event in the SE Greenland coast during the last millennium. Brown VG fragments are generally solid and not vesicular, suggesting that they are not windblown shards and were more likely to have been incorporated in the ice from outcrops in Greenland and Iceland. Similar brown VG fragments were described in Kangerdlugssuaq trough sediments and were interpreted as coming from the glaciers and sea ice from the Geikie Plateau area, based on mineralogical and X-ray diffraction analysis data (Alonso-Garcia et al., 2013).

The presence of HSGs in Eirik Drift sediments indicates drift ice (sea ice and icebergs) coming from NE Greenland and the Arctic, where red sandstones outcrop (Bond et al., 1997; Henriksen et al., 2009). Most of the glaciers in NE Greenland and the Arctic develop floating ice tongues in the fjords where semi-permanent fast ice hinders the icebergs from drifting. As a result, most of the IRD carried at the base of the icebergs is deposited in the fjords (Reeh et al., 2001). Our HSG record from the Eirik Drift shows a significant amount (up to 30 %) of this type of IRD. Therefore, despite substantial deposition of debris within the fjords, the remainder of the drifted ice still carries considerable amounts of IRD. We suggest that some of that IRD may have been wind-blown to the top of the glaciers and/or sea ice at the NE Greenland and Arctic coasts and fjords, rather than directly incorporated in the bottom layers of the glacier. Those grains were then ice-rafted southwards by the EGC when the ice was released from the fjords. A similar origin was proposed for HSGs deposited at the SE Greenland coast based on a multi-proxy study (Alonso-Garcia et al., 2013). In that study, periods of high HSG abundance were associated with strong ice export from the Arctic via the EGC.

Variations in Arctic ice export show a significant correlation with Arctic Oscillation (AO) during the last decades (Mysak, 2001; Rigor et al., 2002), with higher Arctic ice export during intervals of positive AO, although this correlation is not so straightforward because Arctic ice export also depends on the meridional wind components and the position of the atmospheric pressure centres (Hilmer and Jung, 2000), and large anomalies in ice export may have a different origin (Lehner et al., 2013). Darby et al. (2012) demonstrated that the sources of Arctic sea ice may change following the AO, and therefore we can observe changes in the mineralogy transported by the ice in sediment cores influenced by the EGC. During the negative state of the AO a strong high-pressure system dominates the Beaufort Sea restricting the Transpolar Drift to the Siberian side of the Arctic Ocean (Mysak, 2001; Rigor et al., 2002), which would bring drift ice with HSGs from the areas of Severnaya Zemlya and Franz Josef Land. The increase in HSG relative abundance and concentration at Eirik Drift after AD 1400 (Fig. 3) may be driven by an intensification in ice export from those areas in the Arctic and northern Greenland rich in HSGs, very likely favoured by atmospheric changes which promoted higher pressures in the Arctic. The increase in HSGs coincides with a shift observed in the sodium concentration (Na+, Fig. 3) in Greenland ice core GISP2 (Meeker and Mayewski, 2002), which was interpreted as an increase in storminess by ∼ AD 1400. Enhanced storminess favours the transport of icebergs and sea ice through the EGC as well as the deposition of HSGs in the sea ice and on top of glaciers, and both processes increase the number of HSGs transported to Eirik Drift. Greenland temperature also shows a decreasing trend after ∼ AD 1400 (Kobashi et al., 2010). The sedimentary record of Feni Drift (Bond et al., 2001), in the NE Atlantic, also shows an increase in HSG relative abundance during the LIA interval (Fig. 3). Colder atmospheric temperatures and the increase in ice drifted from the Arctic may have contributed to decrease subpolar sea surface temperature, favouring icebergs to reach areas further south such as Feni Drift (Bond et al., 2001).

Coal-bearing sediments are present at many areas around the Arctic such as Siberia, northern Canada, Greenland and Scandinavia (Polar Region Atlas, 1978; Petersen et al., 2013) and contribute to high-latitude IRD deposition (Bischof and Darby, 1997; McManus et al., 1996). Even though the percentage of coal fragments is rather low at our study site (under 5 %; see Fig. 2) the higher abundance of coal fragments in the Labrador Sea during the MCA may be related to an increase in drift ice from the Canadian Arctic during the positive state of NAO/AO. However, these fragments might also indicate human-related activity which increased in the area during the MCA. Further analysis should be performed to assess the linkage of those grains to any specific source.

Regardless of the mineralogy of the grains, it is worth noting the high number of lithics per gram of sediment recorded in several samples during the MCA (Fig. 2). A recent comprehensive study of the last 2 millennia (PAGES 2k Consortium, 2013) shows this interval presented sustained warm temperatures from AD 830 to 1100 in the Northern Hemisphere, including the Arctic region. The high occurrence of IRD from AD 1000 to 1250 suggests that during the MCA either a substantial amount of icebergs drifted to the study area or the drifting icebergs contained considerable amounts of IRD, or a combination of both explanations. Several studies on East Greenland glaciers and fjords point to the consistent relationship between calving rate acceleration and the presence of warm Atlantic water in East Greenland fjords, brought by the Irminger Current (Andresen et al., 2012; Jennings and Weiner, 1996). Warm atmospheric temperatures as well as the presence of Atlantic water prevent the formation of sea ice in the fjords and in front of the glacier, thus increasing the calving rate by destabilizing the glacier tongue (Andresen et al., 2012; Murray et al., 2010). When tidewater glaciers are released from the sea ice, their speed increases due to the decreased flow resistance and increased along-flow stresses during the retreat of the ice front, and rapid changes may be observed in calving rates in response to disequilibrium at the front (Joughin et al., 2008). At present, Kangerdlugssuaq and Helheim glaciers, located in the central East Greenland coast, represent the 35 % of East Greenland's total discharge (Rignot and Kanagaratnam, 2006). If conditions during the MCA were similar to or warmer than present, the calving rates of these glaciers may have been even higher than at present, delivering vast numbers of icebergs to the EGC, where they would release IRD as they melted. Moreover, during the MCA it is likely that other fjords, such us Nansen and Scoresby Sund, were also ice free during the summer, allowing them to contribute considerable numbers of icebergs to the EGC. The massive diamicton found in Nansen fjord sediments between AD 730 and 1100 demonstrates that there was continuous iceberg rafting due to warmer conditions (Jennings and Weiner, 1996). In this context, we postulate that warm temperatures were the driver of the increased iceberg calving at Greenland fjords and the high accumulation of IRD at Eirik Drift during late MCA.

After AD 1250 several spikes of high IRD abundance occurred during the intervals AD 1400–1450, 1650–1700 and 1750–1800 (Fig. 2). Because those intervals occurred within the LIA and under cold conditions, the trigger of iceberg production must have been slightly different from the drivers proposed for the MCA ice-rafting events. These intervals of high IRD accumulation during the LIA are characterized by slightly lower relative abundance of HSGs and higher relative abundance of volcanic grains and other fragments. This points to an intensification of SE Greenland production of icebergs during the LIA intervals of enhanced ice rafting. Therefore, for the LIA events, we advocate for the same mechanism that was put forward to explain rapid releases of icebergs in Denmark Strait during the last 150 years (Alonso-Garcia et al., 2013). During cold periods sea ice becomes perennial along the Greenland coast, blocking the seaward advance of glaciers and hindering icebergs from calving, thus leading to the accumulation of ice mass in the fjords. Based on model simulations, when the sea ice opens or breaks, the ice flow at the grounding line accelerates very quickly, triggering a rapid release of the grounded ice stream (Mugford and Dowdeswell, 2010). In summary, we propose that the high IRD occurrence during the intervals AD 1350–1450, 1650–1700 and 1750–1800 very likely corresponds to episodes of rapid iceberg release from SE Greenland fjords. Interestingly, the timing of these intervals of high IRD deposition coincides with the intervals of most negative volcanic–solar forcing described by the PAGES 2k Consortium (2013).

Our IRD records have been compared with other palaeoceanographic and palaeoclimatic records from Eirik Drift and other subpolar North Atlantic sites to obtain a better picture of subpolar conditions during the last millennium. The planktonic foraminifer δ18O record of N. pachyderma sin from Eirik Drift (this study) indicates slightly lower temperatures after AD 1050 (Fig. 4i). A study from the same region presented a δ18O record of Globigerina bulloides (Fig. 4i) and relative abundance of N. pachyderma sinistral (Fig. 4h) (Moffa-Sanchez et al., 2014a, b), which suggest a cooling episode during late MCA (∼ AD 1100) and a clear drop in temperature after AD 1200. The coincidence of these temperature drops with the increasing trend in total IRD concentration at site GS06-144-03 indicates that the growing iceberg production at East Greenland fjords, due to the MCA warm conditions, started to cool and freshen Labrador Sea several centuries before the LIA started. The quartz / plagioclase ratio, a bulk measure of IRD (Moros et al., 2004), also shows an increasing trend at the end of the MCA at sites in Denmark Strait (Andrews et al., 2009; see Fig. 4j) and off northern Iceland (Moros et al., 2006), providing further evidence for the intensification of iceberg calving at this time. Colder winter sea surface conditions have also been recorded off N Iceland after AD 1200 (Jiang et al., 2007; see Fig. 4f), although sea surface conditions were not cold enough to generate long seasons of severe sea ice until ∼ AD 1300 (Massé et al., 2008; see Fig. 4e), when annual SST had substantially decreased (Sicre et al., 2008). SE Greenland sea ice and SST proxies (Fig. 3a and b) indicate an increase in sea ice and SST decrease at ∼ AD 1200 (Miettinen et al., 2015). The reduction in the relative abundance of the benthic foraminifer Cassidulina teretis between AD 1000 and 1300 in Nansen fjord indicates a weaker influence of Atlantic water at the East Greenland coast (Jennings and Weiner, 1996). This decline in Atlantic water may be explained by a weakening in the northern branch of the Irminger current which would have favoured the SST decrease and sea ice formation in SE Greenland coast and in Denmark Strait and north of Iceland. Blindheim and Malmberg (2005) associated the northern Irminger current weakening with high pressure over Greenland and weaker northerly winds. In addition, the mineralogical composition and biomarker study of the last 2000 years in several sites in Denmark Strait and north of Iceland indicate a change to cold conditions at ∼ AD 1250 very likely associated with an intensification of the high pressure over Greenland and the strengthening of N and NW winds, which led to progressive presence of sea ice exported from the Arctic during winter and spring (Andrews et al., 2009).

The anomalously high Atlantic temperatures recorded during the interval ∼ AD 950–1100 (Mann et al., 2009) may indicate SPG circulation was in the strong mode during that time interval (Figs. 4a and 5c). Strong SPG circulation enhances the supply of warm Atlantic Intermediate water to the East Greenland coast, which promotes calving and, subsequently, increases the ice input in the Labrador Sea region. Switches from weak to strong SPG circulation may happen naturally due to external or internal forcings, and these changes are currently a matter of debate because of their influence on North Atlantic climate (e.g. Hakkinen and Rhines, 2004). According to model simulations, freshwater input (i.e. ice input) to the SPG may trigger weakening of SPG circulation, and this may be amplified successively by positive feedbacks resulting in further weakening and freshening of the gyre due to the attenuation of the Irminger Current (Born et al., 2010, 2016; Moreno-Chamarro et al., 2017). Specifically for this time interval, it is important that the main freshwater input reached the Labrador Sea, affecting deep water formation, because a freshwater input into the Nordic Seas may have driven the opposite effect (Born and Stocker, 2014). Our IRD record demonstrates an increase in the amount of ice transported by the EGC to the Labrador Sea from AD 1000 to 1250, with a potential main source in SE Greenland. This input of freshwater to the SPG potentially drove a slowdown of deep convection in this area and weakened the SPG circulation. A recent study also points to enhanced input of the Labrador Current to the Labrador Sea from ∼ AD 1000 to 1300 (Sicre et al., 2014), which indicates calving intensified in SW Greenland and Baffin Bay regions as well. It is likely that ice from both sources, East and West Greenland, directly affected the salinity balance of Labrador Sea water and deep convection in this region. However, even though the freshwater input started at ∼ AD 1000, the SPG circulation only started to weaken after ∼ AD 1250, as suggested by a record of deep-sea corals from the NE Atlantic (Copard et al., 2012). Moreover, our IRD data show a lag between the first temperature drops at Eirik Drift and the decrease in ice rafting (Fig. 4), indicating a delay between SPG weakening and Irminger Current slowdown. It seems the SPG entered in the weak mode, because of the reduced convection, but warm intermediate water remained in the fjords for several years, allowing continued iceberg calving. Also, the response of calving may be slower, particularly if SSTs were relatively warm and the fjords were not perennially covered by sea ice. However, simulations to reconstruct past climate changes normally are not detailed enough to characterize the impact of direct freshwater input from Greenland to the ocean, and its consequences after several years to decades, which would be very interesting to better understand past climate events as the LIA.

As the strength of Irminger Current input declined, the areas of SE Greenland, Denmark Strait and north of Iceland cooled, and coastal sea ice became perennial after AD 1450, according to the sea ice index IP25 (Massé et al., 2008). The δ18O records of N. pachyderma sinistral (Fig. 4i, this study) and Turborotalita quinqueloba (Fig. 4c) from Eirik Drift (Moffa-Sanchez et al., 2014b) indicate a shift to colder summer SST in the SPG after AD 1400 (Fig. 4), which coincides with the increase in Arctic ice export reflected by the HSGs, and the storminess intensification (Fig. 3c), recorded by the Na+ content in the Greenland ice core GISP (Meeker and Mayewski, 2002). Planktic δ18O and Mg / Ca from sites in the Norwegian Sea (Fig. 4b) display an initial decrease in temperature at AD 1200, and a subsequent distinct downward shift at ∼ AD 1400, which suggests not only SST cooling but also a decline in the stratification of the water column, very likely linked to changes in the upper-ocean conditions in this region as well (Nyland et al., 2006; Sejrup et al., 2010).

It is clear that sea surface conditions in the SPG were rather different before and after ∼ AD 1200. The freshening of the SPG and the increase in sea ice along the Greenland and Iceland coasts may have been associated with a change in atmospheric conditions, weakening winter circulation over the Arctic and promoting more storminess in the subpolar area and the development of atmospheric blocking events (Moreno-Chamarro et al., 2017). Model simulations point to the development of frequent and persistent atmospheric blocking events, induced by low solar irradiance, as one of the main drivers to develop the consecutive cold winters documented in Europe during the LIA (Barriopedro et al., 2008; Moffa-Sanchez et al., 2014a). Atmospheric blocking events derive from instabilities of the jet stream which divert or block the pathway of the westerly winds (Häkkinen et al., 2011). These events typically predominate during winter and occur linked to high pressure in the Arctic and a weak polar vortex. The cold SST events recorded at the subpolar area during the last millennium (Moffa-Sanchez et al., 2014a, b; Sejrup et al., 2010) suggest that atmospheric blocking events affected the entire North Atlantic regional climate.

It is worth noting that our IRD record shows two types of ice-rafting events: ice rafting related to warm temperatures (during the MCA), and ice rafting linked to rapid releases of the ice accumulated in the fjords due to cold conditions (during the LIA). During the LIA, the events of maximum ice rafting are coherent with the minimum values of solar irradiance (Steinhilber et al., 2009), particularly with the Wolf, Spörer, and Maunder minima (Fig. 5). Ice-rafting events in our record tend to happen during intervals of low solar irradiance and cold temperatures in the SPG, often with also significantly cold summer SST (Fig. 4c and i). The reconstruction of radiative forcing based on solar irradiance and volcanic eruptions (Sigl et al., 2015) also shows low values during the main events of high IRD occurrence (Fig. 5).

Solar irradiance has been put forward as the main trigger for the Holocene cold events because low solar irradiance induces an atmospheric reorganization in the polar region which affects not only the North Atlantic but also the mid-latitudes of the Northern Hemisphere (e.g. Bond et al., 2001). Several records from the high-latitude North Atlantic support this hypothesis, displaying cold temperatures at times of solar irradiance minima during the last millennium (Moffa-Sanchez et al., 2014a; Sejrup et al., 2010). However, the role of solar irradiance on forcing cooling events has been questioned during the last decade. A comprehensive review on the topic proposed that a combination of internal climate variability and external forcings contributed to drive Holocene cold events, including the LIA (Wanner et al., 2011). Volcanic activity is also commonly put forward as the main driver of atmospheric reorganizations which derived in cooling events. Precisely dated records of ice-cap growth from Arctic Canada and Iceland (Miller et al., 2012) showed that LIA summer cooling and ice growth, potentially linked to volcanic forcing, began abruptly between AD 1275 and 1300, followed by a substantial intensification at AD 1430–1455. Moreover, a recent study about the role of radiative forcings and climate feedbacks on global cooling over the last millennium also concluded that the volcanic forcing is the factor that contributed the most (Atwood et al., 2016).

According to our observations, the increase in Greenland calving during the MCA (Fig. 5e) took place before the ice caps started to grow, during an interval of high solar irradiance (Fig. 5f), high temperatures in the Northern Hemisphere (Fig. 5c), and low volcanic activity (Fig. 5g). This indicates that the ice-rafting events of the MCA were not related to the fluctuations driven by solar–volcanic forcing. Alternatively, we interpret these events as resulting from the acceleration of calving rates in SE Greenland glaciers, driven by warm temperatures. We postulate that the increase in calving rates during the MCA induced a decrease in the Labrador Sea salinity, which may have triggered the weakening of SPG circulation and reduced convection. A decline in Labrador Sea convection reduces deep water formation in one of the key areas of the North Atlantic, which weakens North Atlantic circulation and, in turn, decreases oceanic heat transport to this area (Born et al., 2010; Moreno-Chamarro et al., 2017). Once the SPG entered in the weak mode, this area received less heat and became more sensitive to external forcings which may have generated further cooling. This interpretation is in agreement with recent model simulations which suggest that a weakening of the SPG circulation could have induced the LIA cooling, and this shift from strong to weak circulation may have been triggered by freshwater input to the Labrador Sea (Moreno-Chamarro et al., 2017). Subsequently, low solar irradiance intervals, possibly combined with volcanic emissions, promoted atmospheric reorganizations which gave rise to a weakening of the polar vortex and promoted atmospheric blocking events, enhancing cold temperatures in the subpolar area and leading to ice sheet growth in the Arctic region during the LIA. The development of atmospheric blocking events in the North Atlantic, as suggested by Moffa-Sanchez et al. (2014a), probably propagated the atmospheric cooling across Europe and the Nordic Seas. Indeed, the first strong minimum of solar irradiance during the last millennium (Wolf, ∼ AD 1300) occurred when the Labrador Sea was already fresher and SPG circulation was weak (Fig. 5), according to our interpretations and to Copard et al. (2012) deep-sea corals record. The reconstructions of solar and volcanic forcings (Fig. 5f and g) shows a trend of lower values after AD 1450 with a first step of low values during the Wolf minimum, indicating that volcanic forcing may also have played an important role in modifying the atmospheric conditions. However, we consider that the decrease in Labrador Sea salinity prior to the Wolf minimum was crucial to produce changes in SPG circulation. Once the SPG entered the weak mode, the effects of solar and volcanic forcing possibly produced a deeper impact on North Atlantic climate. It is likely that the LIA would not have been such a cold and widespread event if the SPG circulation was strong and deep convection was active at the time.

The results of this study can be linked to the expansion and demise of the Norse colonies. According to historical data, the Norse expansion and colonization of Iceland and Greenland occurred during the warmer climate conditions of the MCA which favoured fishing and farming in these regions (Kuijpers et al., 2014; Ogilvie et al., 2000; Ogilvie and Jónsson, 2001; see Fig. 3). Our study indicates that, even though calving intensified after the settlement of the Norse colonies in Greenland, climatic conditions during the late MCA were still favourable because the strong circulation in the SPG supplied relatively warm water to SE Greenland coast. Therefore, the fjords were not perennially covered by sea ice and it is likely that a rather continuous calving may have helped hunting. However, after several decades of intense calving and melting of Greenland glaciers, the Labrador Sea became fresher and the SPG circulation started to weaken, triggering a change in oceanic and atmospheric conditions. The reduction of deep convection decreased the transport of heat to the NW subpolar area and enhanced sea ice occurrence in the fjords, which deteriorated the living conditions in Greenland. The subsequent cooling and increase in storminess brought by the shift in atmospheric conditions (increase in atmospheric blocking events) very likely favoured the abandonment of the Greenland Norse settlements at the beginning of the LIA (Dugmore et al., 2012; Ogilvie et al., 2000, Fig. 3).