The new 10Be samples come from the composite sediment profile MFM09 (Martin-Puertas et al., 2012a) which was retrieved at MFM, a deep crater lake situated in the Eifel region of western Germany that was annually laminated (varved) throughout most of the Holocene (Brauer et al., 2000). Around 0.25 g of dried and crushed material was taken for each sample with a temporal resolution of 3 and 10 years (see dataset), and 0.5 mg of 9Be carrier was added. 10Be was extracted from the sediment samples at the 10Be laboratory of the Earth Sciences Department of Uppsala University, Sweden, following the methodology described by Berggren et al. (2010). All samples were measured using the accelerator mass spectrometer (AMS) of the Tandem Laboratory in Uppsala. The 10Be concentration (in atoms per gram) of each sample is calculated based on the 10Be counts R to 9Be counts Rst ratio and, taking into consideration the NIST SRM 4325 reference standard (10Be/9Be=2.68×10-11), on the weights of the carrier WC and of the sample WS, as well as the Avogadro constant NA and atomic weight Ar of beryllium: 10Beconc.=RRst⋅2.68×10-11⋅WCWS⋅NAAr.

The paleoclimate data investigated herein come from different studies with different records and thus different chronologies. The new sediment 10Be concentration data come from MFM, the chronology of which (MFM2012) was established using mainly microscopic varve counting fixed on an absolute timescale via tephrochronology, as well as radiocarbon dating, with a maximum varve counting error of up to 110 years (Brauer et al., 2000; Martin-Puertas et al., 2012a). A more recent chronology (MFM2015) exists which includes the identification and age of the Vedde Ash, although it remains unchanged for the Holocene part (Lane et al., 2015), which is the period of focus in this study. We also use published 10Be flux data (Adolphi et al., 2014) from the GRIP ice core in central Greenland and within the Greenland Ice Core Chronology 2005 (GICC05) framework (Rasmussen et al., 2006; Vinther et al., 2006; Svensson et al., 2008; Seierstad et al., 2014). Finally, we use 14C production rate data (Muscheler et al., 2014) inferred from the IntCal13 14C calibration curve (Reimer et al., 2013) as the anchoring record for our synchronization. That is to say that we synchronize the MFM2012 timescale (using our 10Be concentration data) and the GICC05 timescale (using the GRIP 10Be flux data) with IntCal13 (using the 14C production rate data).

The synchronization of the different radionuclide records was computed following the methodology described in Adolphi and Muscheler (2016). This method employs the same Bayesian approach as is used for wiggle matching tree ring 14C sequences to the radiocarbon calibration curve (Bronk Ramsey et al., 2001). It exploits the fact that the spacing between samples is precisely known from varve and layer counts and that, hence, the probability density functions from individual samples can be combined using Bayes' theorem. Analogous to radiocarbon wiggle matching of tree rings, we use one record (for example 14C production rate) as our “calibration curve”, while the other record serves as our “tree rings” (e.g., MFM and GRIP 10Be). By shifting one relative to the other, we can thus estimate a probability density function of the timescale difference between the two records. Further details can be found in Adolphi and Muscheler (2016) and Bronk Ramsey et al. (2001). For these calculations, we linearly detrend all radionuclide records between 11 800 and 11 000 years BP and assume a production rate uncertainty of 20 % for all records, which corresponds to the root mean square error between the records after synchronization.

The new 10Be concentration measurements from MFM are displayed in Fig. 1 alongside 10Be flux data from the GRIP ice core in central Greenland (Finkel and Nishiizumi, 1997; Muscheler et al., 2004; Adolphi et al., 2014) and older 10Be concentration data from MFM for the late glacial–Holocene transition (Czymzik et al., 2016). Each dataset is plotted on its original timescale, which is the MFM2012 chronology (Brauer et al., 2000; Martin-Puertas et al., 2012a) and the GICC05 chronology (Rasmussen et al., 2006; Vinther et al., 2006; Svensson et al., 2008; Seierstad et al., 2014). The most striking feature of these datasets is the approximately 250-year-long period of increased 10Be concentration around 11 150 years BP. The most likely explanation for this increase is a decrease in the intensity of the heliomagnetic field (solar activity), leading to an increased impingement of Earth by galactic cosmic rays and thus an increased atmospheric production rate of 10Be and 14C nuclides. It was also shown that meteorological and catchment influences on 10Be deposition are likely small at MFM (Czymzik et al., 2016). The high resolution of our 10Be measurements allows us to observe finer structures within this period of increased 10Be concentration. One example is the double peak structure at 11 200 and 11 040 years BP, which is also present in 14C atmospheric production rate data (Muscheler et al., 2014; Fig. 2) but not expressed well in the GRIP 10Be data. Finally, it is of importance to note that although the increased production around 11 150 years BP is observed in all these radionuclide records, there is an apparent chronological offset at its onset around 11 300 years BP (Fig. 1). More specifically, the 10Be flux data from GRIP begin to increase around 11 320 years BP, whereas a similar increase is seen in the 10Be concentration from MFM around 100 years later, although some short-term features are not visible in all records such as a 40-year-long trough in the MFM 10Be data (ca. 11 290–11 250 years BP).

The Greenland ice core timescale is characterized by an accumulating layer counting uncertainty back in time (Rasmussen et al., 2006) as are chronologies based on sediment varve counting such as MFM. In comparison, tree ring chronologies, underlying the 14C calibration record, are considered accurate with virtually no dating uncertainty for the Holocene period (Reimer et al., 2013). Considering the different timescale uncertainties, it is challenging to compare the timing of short-lived climate oscillations such as the PBO/11.4 ka event. Here we use the global signature common to all cosmogenic radionuclide records as a synchronization tool (Muscheler et al., 2008, 2014). More specifically, we use the large fluctuations in both the MFM and GRIP 10Be data to synchronize these records with the chronologically more accurate and precise IntCal13 timescale (Czymzik et al., 2018). It was previously shown that GICC05 increasingly overestimates age during the Holocene compared to IntCal13 (Muscheler et al., 2014) and that this timescale difference is estimated to increase to 67 (±6) years at 11 000 years BP (Adolphi and Muscheler, 2016). We use the same Bayesian wiggle matching approach as in Adolphi and Muscheler (2016) but here for the period 11 800–11 000 years BP to synchronize both the MFM sediment and Greenland ice core records with IntCal13.

Figure 2 shows both the ice core and sediment core 10Be data once synchronized with the IntCal13 timescale using the 14C production rate from Muscheler et al. (2014), with the corresponding probability density functions displayed in panel (b). We find that the MFM 10Be data fit best with 14C by adding 20 years to MFM2012 (+6/-19 years uncertainty with a 95.4 % confidence interval), whereas the GRIP 10Be data fit best with 14C by shifting GICC05 78 years towards the present (+32/-8 years uncertainty with a 95.4 % likelihood interval). When comparing GICC05 directly to MFM2012, we find that the best fit occurs by shifting GICC05 72 years towards MFM2012 (+4/-8 years with a 95.4 % likelihood interval). There is thus a difference of 26 years (72 +4/-8 years versus 98 +33/-21 years) when comparing GICC05 and MFM2012 directly rather than synchronizing them with IntCal13 first, which illustrates the uncertainties inherent to this exercise. In the following, we will compare GICC05 and MFM2012 when synchronized with IntCal13 as it is the more robust timescale, and then consider the combined chronology offset of 98 (+33/-21) years. Another uncertainty from these estimates arises from the influence of climate on the cosmogenic signal of all radionuclides (Adolphi et al., 2014; Muscheler et al., 2008; Pedro et al., 2012). For instance, 14C oxidizes to form 14CO2 and enter the carbon cycle, while 10Be readily attaches to aerosols and is thus influenced by precipitation. Even though 10Be deposition is not expected to have strong environmental influences at MFM (Czymzik et al., 2016), this was taken into account within the 20 % uncertainty since these effects are difficult to quantify objectively.

If we correct the GICC05 and MFM2012 timescales for their respective offsets to IntCal13, we can compare early Holocene climate in Greenland to data from MFM with a high chronological precision. Figure 3 displays a selection of climatic proxy data from both Greenland ice cores and the varved MFM record on the IntCal13 timescale as per Fig. 2. In addition, both 14C atmospheric production rate and GRIP 10Be flux data are shown as a general indicator of changes in solar activity (Fig. 3a). The stack of δ18O anomalies from four Greenland ice cores (DYE-3, GRIP, NGRIP, and Renland; Fig. 3b) can be related to surface air temperature around Greenland (Rasmussen et al., 2007; Vinther et al., 2009) and shows one negative fluctuation between 11 400 and 11 250 years BP. Following this oscillation, the Greenland δ18O anomaly record remains largely constant and positive. In addition, we also use the accumulation rate anomaly stack (Fig. 3c) from the DYE-3, GRIP, and NGRIP ice cores (Rasmussen et al., 2007) to illustrate changes in snow accumulation rates over Greenland. Here again, a negative fluctuation is observed between 11 400 and 11 250 years BP. Then, we make use of the MFM δD records of n-alkanes (Fig. 3d) that have been interpreted as being a proxy for precipitation δD (Rach et al., 2014) which, similar to δ18O in Greenland, can thus be regarded as indicative of distance from and temperature and/or humidity at the moisture source (Dansgaard, 1964), as well as fractionation related to air temperature. In contrast to the Greenland stack, the δD data show no fluctuations between 11 400 and 11 250 years BP with δDaq remaining constant and δDterr showing an increasing trend. Then at 11 250 years BP, both δD series depict a 20 % drop that persists until 11 100 years BP. To test the spatial scale of which the δD record from MFM can be representative, we have investigated the spatial relationship between surface air temperature (SAT) in the NOAA–CIRES 20th climate reanalysis V2c (20CR; Compo et al., 2011) and δD in precipitation at the Trier meteorological station (about 50 km SW of MFM). It can be seen in Fig. 4 that there is a significant relationship (p<0.1) between annual precipitation δD from the Trier station (IAEA/WMO, 2006) and annual SAT over most of western Europe. In addition, Fig. 4 also points to a relationship between annual SATs over Greenland and Iceland and annual δ18O at Summit (central Greenland) (Steig et al., 1994; White et al., 2009). Finally, we also show varve thickness changes at MFM that were primarily controlled by runoff from the catchment. After a period of low varve thickness, a sharp increase occurred at 11 250 years BP followed by a gradual decrease and a second but very small increase around 11 080 years BP. Titanium-centered log ratio data (Ticlr), determined by micro X-ray fluorescence (μ-XRF) from the same MFM sediment composite profile (Martin-Puertas et al., 2017), confirm the interpretation that the variance in varve thickness at the time was mostly controlled by detrital supply to the lake (Fig. 3e). It is important to mention that in a longer time perspective, the changes described above in the sediments of MFM (Martin-Puertas et al., 2017; Rach et al., 2014) do not exceed other fluctuations in varve thickness and Ticlr.

In Greenland, a cold and dry climate episode occurred around 11 400–11 250 years BP known as the 11.4 ka event (Rasmussen et al., 2007). This is evidenced by a significant drop in the signal of the Greenland ice core δ18O stack, as well as in the accumulation stack (Fig. 3b, c). By shifting GICC05 78 years towards the present, the central part of the 11.4 ka event (lowest value in δ18O) is dated to around 11 372–11 272 (+32/-8) years BP, which is consistent with GICC05 within the combined uncertainty of our synchronization and the maximum counting error in GICC05. When looking at the temperature proxy and varve thickness data from MFM (Fig. 3d, e), we do not find any event that is coeval with the 11.4 ka event in Greenland. Interestingly though, Ticlr data (Fig. 3e) gradually decreased from ca. 11 490 years BP only to be interrupted by a small increase around 11 300 years BP. The low Ticlr data suggest less runoff probably related to drier conditions, which shows some commonality with the conditions in Greenland at that time (11.4 oscillation) evidenced by the lower accumulation rate. Therefore, a possible link to the dry Rammelbeek Phase described in the Borchert peat sequence in the Netherlands (van der Plicht et al., 2004; Bos et al., 2007) may be tentatively put forward, although chronological uncertainties hinder proving this. We can now also confidently deduce that the termination of the δ18O and accumulation anomalies in Greenland (the 11.4 ka event) is synchronous with a large decrease in solar activity (Fig. 3a–c). More specifically, high levels of solar activity prevailed throughout the occurrence of the 11.4 ka event in Greenland. Then, as solar activity started to decrease (ca. 11 250 years BP) into a grand solar minimum that lasted for around 250 years, the climate in Greenland switched back to warmer and wetter conditions with higher δ18O values and a higher accumulation rate. This is in accordance with the suggestion of an abrupt warming (4∘±1.5∘) in Greenland following the event based on δ15N in the GIPS2 ice core (Kobashi et al., 2008). The rapid transition towards positive accumulation anomalies occurred over a few decades only.

While climate over Greenland following the 11.4 ka event returned rapidly to warmer and wetter conditions, all proxies from MFM sediments (Fig. 3d, e) show fluctuations around 11 250 years BP (henceforth MFM oscillation). In particular, aquatic δD data from small-chain alkanes (Rach et al., 2014) show a clear oscillation with a 20 % drop around 11 250 years BP (Fig. 3d), while terrestrial δD data show a decrease reaching levels seen around 11 500 years BP. This deuterium depletion in the alkanes most likely mirrors a depletion of deuterium in precipitation which can be explained, in part, by lower air temperatures over western Europe in view of Fig. 4. Simultaneously, varve thickness and Ticlr show a rapid increase at 11 250 years BP (Fig. 3e), denoting a likely increasing detrital contribution to this varve thickening. When considered in a longer time perspective (Martin-Puertas et al., 2017), this varve increase reaches the level of other fluctuations that are unrelated to known early Holocene oscillations in North Atlantic climate. Nevertheless, this shift at 11 250 year BP does correspond to a change in the composition of the sediments as Martin-Puertas et al. (2017) defined a compositional boundary of MFM varves at 11 230 years BP (11 250 years BP on the IntCal13 timescale) based on μ-XRF scanning analyzed with Ward's clustering methods. By synchronizing MFM2012 with IntCal13 (Fig. 2), we find that this compositional boundary is also coeval with the onset of the grand solar minimum (Fig. 3), although the cause of this change is difficult to assess. In fact, Ticlr, as well as ln⁡(Si/Ti) and ln⁡(Ca/Ti), which is generally regarded by Martin-Puertas et al. (2017) as indicating relative changes in biogenic silica concentrations and authigenic calcite precipitation, is significantly correlated with the new 10Be concentration measurements and also with the GRIP 10Be data and with the 14C atmospheric production rate (Figs. 5 and S1 in the Supplement). Because GRIP 10Be data and the 14C atmospheric production rate are unaffected by environmental changes at MFM, we suggest that the catchment area of MFM was likely influenced by the substantial changes in solar activity that characterized this period rather than 10Be concentration at MFM being affected by this sediment compositional change. In support of this assumption, Czymzik et al. (2016) also reported negligible climate influences on 10Be deposition at MFM even across distinct climatological boundaries. It can also be seen that the second and smaller increase in varve thickness and Ticlr is coeval with a second dip in solar activity shortly after 11 100 years BP (Fig. 3a and e). Finally, it is worthwhile to note that the percentage values of Pinus pollen and biogenic silica, as well as pollen concentrations in MFM, all decreased at 11 250–11 230 years BP while percentage values of Betula increased (Brauer et al., 1999). Although not interpreted by the authors, these changes echo the findings of Björck et al. (1997), who defined the PBO in terrestrial records of Sweden with a similar decrease in pollen concentrations and more notably of Pinus pollen percentages, interpreted as a setback of tree vegetation in southern Sweden. It should be stressed here that we cannot directly compare the palynology of MFM to these Swedish lakes because of the challenging interpretation of the former record, as well as the chronological uncertainties and the vicinity to the retreating Fennoscandian Ice Sheet (FIS) of the later records.

In summary, the radionuclide-based synchronization of the GICC05 and MFM2012 timescales indicates a combined timing offset of up to 98 (+33/-21) years during the earliest part of the Holocene. Correcting for this offset, we observe that cold oscillations at both locations and inferred from water isotopes did not occur simultaneously between 11 450 and 11 000 years BP. We further note that this pattern appears to be coupled with large changes in solar activity, which leads us to suggest a causal link. More specifically, the cold and dry climate oscillation in Greenland (the 11.4 ka event) occurred under a period of high solar activity between ca. 11 370 and 11 270 years BP but did not leave a discernable imprint in either varve thickness or biomarker δD from MFM. Subsequently, solar activity dropped to a grand minimum that lasted for as long as 250 years. This change was coeval with the termination of the 11.4 ka event (Greenland) and the onset of the MFM oscillation with colder conditions inferred from δD data (Figs. 3d and 4). The ostensible link with solar activity which we infer in view of Fig. 3 resembles what has been described substantially in the recent literature and is discussed in the following section.

Our suggestion of a causal sun–climate link during the earliest part of the Holocene can be further supported by the spatial patterns of the 11.4 ka event in Greenland followed by a cold period at MFM starting at 11 250 years BP (MFM oscillation). Based on our synchronization of the different paleoclimate records, we find an asynchronous relationship between the Greenland and European climates, characterized by cold and dry conditions over Greenland but with no evidence of it at MFM under high solar activity and a warm and wetter Greenland climate, as well as colder conditions at MFM for low solar activity (Fig. 3).

This pattern is consistent with a number of, but not all, climate modeling studies that find a top-down influence of solar activity on North Atlantic and European atmospheric circulation patterns. This forcing mechanism involves the increase in UV radiation during solar maximum years (Haigh et al., 2010; Lockwood et al., 2010), which enhances the production of stratospheric ozone and leads to stratospheric heating through the increased absorption of longwave radiation (Haigh et al., 2010), especially at the Equator. This increases the stratospheric temperature gradient between the Equator and poles (Simpson et al., 2009), leading to an acceleration of the polar night jet (Kodera et al., 2002), which eventually propagates down to the troposphere via wave refraction (Matthes et al., 2006; Ineson et al., 2011). In turn, this leads to patterns in surface pressure and temperature which mimic those of the positive phase of the North Atlantic Oscillation (NAO) in winter (Woollings et al., 2010; Ineson et al., 2011). The opposite mode applies during periods of solar minima. It should, however, be stressed that there is no consistent correlation between the North Atlantic Oscillation and solar forcing for the past centuries (Gray et al., 2013; Ortega et al., 2015), although a solar influence on the region is not necessarily related to the NAO (Moffa-Sánchez et al., 2014; Sjolte et al., 2018). Even though the spatial pattern we observe agrees well with a top-down solar forcing, other mechanisms cannot be excluded as lying behind the different North Atlantic response patterns. Overall, it has to be kept in mind that different time periods with different climate boundary conditions could lead to shifting atmospheric patterns.

In the following we explore the solar hypothesis further by investigating a modern analog with climate reanalysis data. Figure 6a shows the surface air temperature (SAT) anomalies in the North Atlantic region for periods of solar maxima compared to periods of solar minima in 20CR (mean±1σ of the sunspot group numbers from Svalgaard and Schatten, 2016, between 1946 and 2011 CE; see Fig. S2). It can be seen from the SAT anomalies that a distinct antiphase pattern between Greenland and Europe is coincident with highs and lows in solar activity. That is to say that Greenland experiences lower SATs during winters of solar maxima compared to winters of solar minima, whereas lower SATs are observed across Europe for winters of solar minima compared to winters of solar maxima. This highlights the correspondence between the solar influence on North Atlantic climate, which has been proposed to have been active during the 20th century, and the synchronized climate proxy records during the early Holocene in terms of spatial distribution of SAT anomalies. Furthermore, this correspondence can also be qualitatively described by comparing the mean annual temperature anomalies at both Summit (central Greenland) and MFM (Fig. 6c, d) through an average of all 11-year solar cycles of the 20th century (Fig. 6b). Decadal temperature changes in 20CR at both Summit (blue curve in Fig. 6c) and MFM (red curve in Fig. 6d) agree qualitatively well with centennial δ18O and δD changes observed in Greenland ice cores and in MFM sediments during the period ranging from 11 450 to 11 000 years BP (black curves in Fig. 6c, d; note the different time axes). Of specific interest here is the average transition from high to low solar activity that is coincident with an annual temperature rise or drop of ca. 1 K at Summit and/or MFM. Assuming changes in water isotopes to be, in part, indicative of regional temperature changes (Dansgaard, 1964; Masson-Delmotte et al., 2005; Rach et al., 2014; Fig. 4), this decadal pattern between Summit and MFM in climate reanalysis data mimics the centennial-scale climate changes that prevailed in Greenland and Europe throughout the period 11 450–11 000 years BP. Water isotopes are often dominated by a particular seasonal signal. It is therefore of interest to note that the spatial patterns observed in climate reanalysis are also present during the summer, although to a lesser degree (Fig. 3).

It should be noted that the efficiency of the top-down mechanism remains largely unexplored for centennial timescales. For instance, previous studies have proposed a top-down solar influence on atmospheric circulation on similar timescales for both Greenland (Adolphi et al., 2014) and MFM (Martin-Puertas et al., 2012b), leading to a similar spatial pattern in reanalysis data. The modeling results in these studies, however, only investigate the effect of decadal (11-year) changes in solar activity. In contrast, it was also shown more recently that the centennial response of North Atlantic atmospheric circulation to solar forcing is correlated with the second mode of atmospheric circulation, the East Atlantic pattern, rather than to the first mode, the NAO (Sjolte et al., 2018). The latter study consequently does not find a similar pattern in SAT anomalies between Greenland and western Europe.

For the same reasons, another uncertainty arises from the relevance of using 20th century climate reanalysis as an analogy of early Holocene conditions. In particular, the Laurentide Ice Sheet (LIS) is known to have played an important role in the position of the North Atlantic eddy-driven jet by accelerating and displacing it southward (Merz et al., 2015). However, it is also known that the LIS waned to the point of separation with the Cordilleran at around 14 000 years BP (Dyke, 2004). According to a study based on a transient climate simulation from the Last Glacial Maximum (LGM) (Löfverström and Lora, 2017), this separation led to a shift in the dominant topographic stationary wave source in North America. This, in turn, induced a transition from a strong and subtropical jet stream to a weaker and more meridionally tilted jet stream and storm track as observed for present conditions. This suggests that similar atmospheric processes could have been at play during the earliest part of the Holocene, relative to today, in spite of different boundary conditions. Furthermore, the results in Fig. 6 arise from an 11-year solar cycle forcing which is considerably weaker and less persistent than the potential solar forcing that the 11 400 years BP solar maximum to 11 200 years BP grand solar minimum could have provoked, leading to possibly different reactions due to feedback processes. In fact, both the 14C data and GRIP 10Be data shown in Fig. 2 depict one of the most prominent increases in the Holocene record (Vonmoos et al., 2006) in terms of both amplitude and the duration of the grand solar minimum. In comparison, its duration represents twice the length of the longest grand minimum known from sunspot observations (Svalgaard and Schatten, 2016) and is called the Maunder Minimum (1645–1715 CE).

The PBO has also been associated with an increase in freshwater supply hampering the Atlantic meridional overturning circulation (AMOC) possibly from the Baltic Ice Lake drainage and the rapidly waning Fennoscandian Ice Sheet (Björck et al., 1996; Hald and Hagen, 1998). It was next proposed by Fisher et al. (2002) that an outburst of Lake Agassiz could represent the trigger of the PBO through an increased thickness and extent of Arctic Ocean sea-ice pack. This would have resulted in an increased albedo and a slowdown of North Atlantic Deep Water (NADW) formation due to increased freshwater delivery to the North Atlantic. However, the timing of the outburst event to which they attribute the PBO (11 335 years cal BP) has rather large uncertainties (±130 to 230 years) due to the Δ14C age plateau in this period. More recently, it was suggested that even small changes in the prevalence of the AMOC can influence atmospheric circulation with couplings to the NAO, with an intensification of the former resulting in a negative index of the latter (Frankignoul et al., 2013).

To further investigate the potential spatial distribution of SAT anomalies due to a slowdown of the AMOC, we again investigate 20CR for winters with a negative reconstructed AMOC index (Duchez et al., 2014) compared to winters with a positive reconstructed AMOC index for the period 1961–2005 CE (Fig. 7a). Interestingly, SAT anomalies similar to those for solar forcing subside. That is to say that an amplified meridional temperature gradient with a colder Greenland and a warmer western Europe is favored in winters in which the AMOC is weaker relative to winters in which it is stronger. Although it is difficult to obtain direct evidence of an AMOC slowdown during the early Holocene, it is conceivable that the waning Fennoscandian Ice Sheet would have routinely released enough freshwater to weaken and condition the AMOC for the onset of the 11.4 ka event in Greenland. This result could also be explained by the influence of the NAO on the AMOC index as it is difficult to disentangle these tightly coupled processes (McCarthy et al., 2015). In this case, the persistent high levels of solar activity, which can also favor such temperature and pressure patterns, could represent a potential trigger for these climate oscillations. Figure 7b depicts the large temperature differences for winters in which both high solar activity and a weak AMOC prevailed during the period 1961–2005 CE with up to a -4 K anomaly in western Greenland. This, however, needs to be treated with caution due to the relatively short period of observation that results in having very few years when such solar activity and AMOC conditions existed in parallel (Fig. 4).

In addition, a coupling between solar and freshwater forcing could also explain the lack of significant climate responses to subsequent grand solar minima which were also large in amplitude but did not yield an unequivocal impact on North Atlantic climate. It is indeed notable that the following changes in solar activity occurred while the influence of freshwater release by the FIS was diminishing, and therefore the North Atlantic was not conditioned as it was during the PBO. For instance, a similar but weaker event was found in the δ18O signal of the GRIP ice core around 10 300 years cal BP, coinciding with a low in Δ14C (high solar activity) and a cooling in the Faroe Islands (Björck et al., 2001). In contrast, the subsequent grand solar minimum which occurred around 9500 years BP (Vonmoos et al., 2006), at a time during which the FIS had completely vanished (Stroeven et al., 2016), did not coincide with any evident climate oscillation in Greenland.