Cold season warming in the North Atlantic during the last 2,000 years: Evidence from Southwest Iceland

Temperature reconstructions from the Northern Hemisphere (NH) generally indicate cooling over the Holocene which is often attributed to decreasing summer insolation. However, climate model simulations predict that rising atmospheric CO2 concentrations and the collapse of the Laurentian ice sheet caused mean annual warming during this epoch. This contrast could reflect a bias in temperature proxies, and particularly a lack of proxies that record cold (late fall-early 15 spring) season temperatures, or inaccuracies in climate model predictions of NH temperature. We reconstructed winterspring temperatures during the Common Era (i.e. the last 2,000 years) using alkenones, lipids produced by Isochrysidales haptophyte algae that bloom during spring ice-off, preserved in sediments from Vestra Gíslholtsvatn (VGHV), southwest Iceland. Our record indicates cold-season temperatures warmed during the last 2,000 years, in contrast to NH averages. Sensitivity tests with a lake energy balance model show that this warming is likely driven by increasing winter-spring 20 insolation. We also found distinct seasonal differences in centennial-scale, cold-season temperature variations in VGHV compared to existing records of summer and annual temperatures from Iceland. Sustained or abrupt cooling in VGHV temperatures are associated with the cumulative effects of solar minima and volcanic eruptions, and potentially ocean and sea-ice feedbacks associated with cooling in the broader Arctic. However, multi-decadal to centennial-scale changes in cold season temperatures were strongly modulated by internal climate variability, i.e. the North Atlantic Oscillation, which can 25 result in winter warming in Iceland even after a major negative radiative perturbation. https://doi.org/10.5194/cp-2020-84 Preprint. Discussion started: 26 June 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Temperatures in the Northern Hemisphere (NH) are generally thought to have cooled over the past 2,000 years, culminating in the Little Ice Age (LIA, c. 1450-1850 CE) (Kaufman et al., 2009;Seppä et al., 2009;PAGES 2K Consortium, 2013, 201930 McKay and Kaufman, 2014). However, the majority of NH temperature reconstructions are based on proxies that respond to climate change during the warm season and may not capture trends in annual or winter and spring temperatures (Liu et al., 2014;PAGES 2K Consortium, 2019). This limits our understanding of major atmospheric phenomena in the NH, such as the North Atlantic Oscillation (NAO) which dominates wintertime variability, as well as changes in ocean circulation and other phenomena driving variability in the extent of Arctic sea ice. 35 Many oceanic and atmospheric processes that influence surface climate in the Atlantic and the broader NH are centered in the high North Atlantic region, making it an important location to study changes during the cold season (Hurrell, 1995;Yeager and Robson, 2017). Terrestrial paleoclimate records from Iceland, for instance, have the potential to resolve temperature changes during the cold season as this region is sensitive to the NAO and sits near the southern limit of Arctic 40 sea ice (Hurrell, 1995;Hanna et al., 2004Hanna et al., , 2006. The high sedimentation rates in Icelandic lakes, along with well-known volcanic eruptions that can be used as age constraints, make this an ideal location and archive to test how winter and spring temperatures evolved over the past 2,000 years (Axford et al., 2007(Axford et al., , 2009Geirsdóttir et al., 2009Geirsdóttir et al., , 2019Gathorne-Hardy et al., 2009;Larsen et al., 2011;Langdon et al., 2011;Holmes et al., 2016). However, existing terrestrial records of temperature from Iceland are limited due to their sensitivity to the warm season, low temporal resolution and length, or compounding 45 effects on proxies from human land-use or precipitation over the past 2,000 years.
Here we present a reconstruction of winter-spring temperatures developed using well-dated lake sediments from southwest Iceland to assess seasonal temperature changes in the North Atlantic climate over the past 2,000 years. We take advantage of alkenone-production by Group I Isochrysidales (i.e. haptophyte algae) during the spring season to develop a record of 50 winter-spring temperatures and investigate the forcings responsible for cold-season temperature changes using a lake energy balance model.

Lipid analyses
Sediments were freeze-dried and extracted using a Dionex TM accelerated solvent extraction (ASE 350) system at 120 °C and 75 1200 psi. All of the extracts were separated by silica gel (40-63 µm, 60 Å) flash chromatography to obtain alkane (hexane; Hex), ketone (dichloromethane; DCM), and polar (methanol; MeOH) fractions. Saponification was used to remove wax esters by dissolving the dried ketone fraction in a 1 molar potassium hydroxide solution with MeOH:H2O (95:5, v/v) and heating the samples for 3 hrs at 65 °C. 5 % NaCl in H2O and 50 % HCl in H2O were added to the samples and the lipid faction was extracted using Hex (100 %). Ketone fractions were further purified using silver nitrate columns (D'Andrea et 80 al., 2007) with DCM (100 %) followed by ethyl acetate (100 %) to elute the alkenones. If additional cleaning was needed, a modified procedure from Salacup et al. (2019) was used. The alkenone fraction was dried under N2 gas and re-dissolved in 1.5 mL of DCM:Hex (2:1, v/v). To this, a 1.5 mL solution of 100 mg/mL urea in MeOH was added. The resulting crystals were dried under N2 gas, and the urea addition was repeated two more times. The dried urea crystals were cleaned with Hex (100 %) and extracted as the non-adduct. Milli-Q water was added to the vial to fully dissolve the urea crystals, and the 85 adduct was extracted using Hex (100 %). The samples were then analyzed for alkenones. For several samples, co-eluting compounds were still visible, or concentrations were too low for reliable quantification. These samples were not included in our final reconstruction. The resulting alkenone fraction was analyzed using an Agilent 6890N gas chromatography (GC) and flame ionization detector (FID) system with an Agilent VF-200ms capillary column (60 m x 250 µm x 0.10 µm). Samples were injected into a CIS-PTV inlet in solvent vent mode (6.9 psi at 112 °C). The oven program was set to 50 °C and increased to 235 °C at 20 95 °C/min and ramped to 320 °C at 1.39 °C/min where it was held isothermally for 5 min. For additional verification or identification of co-eluting compounds, samples were run on an Agilent 6890N GC system coupled with an Agilent 5793 N quadrupole mass spectrometer (MS). All samples were injected with pulsed splitless injection mode (20 psi at 315 °C) and run on an Agilent VF-200ms capillary column (60 m x 250 µm x 0.10 µm). The oven program was started at 40 °C for 1 min, ramped up to 255 °C at 20 °C/min, increased again to 315 °C at 2 °C/min, and then held isothermally for 10 min. The 100 MS ionization energy was set to 70 eV with a scan range of 50 to 600 m/z.

Alkenones as a proxy for lake water temperatures
Alkenones are long-chain ketones produced by Isochrysidales haptophyte algae in both marine and lacustrine environments.
Numerous marine-based culture and core-top studies show that variations in alkenone saturation (i.e., changes in C37:3Me and C37:2Me production) are inversely correlated with temperature, and can be linearly calibrated to temperature using either the 105 U 37 K or U 37 K' index (Brassell et al., 1986;Prahl and Wakeham, 1987;Prahl et al., 1988;Müller et al., 1998;Conte et al., 2006). Similarly, culture studies, core tops, and in situ measurements in lakes show that changes in alkenone saturation are also correlated with temperature (Zink et al., 2001;Sun et al., 2007;Toney et al., 2010;D'Andrea et al., 2011D'Andrea et al., , 2012D'Andrea et al., , 2016Wang and Liu, 2013;Nakamura et al., 2014;Longo et al., 2016Longo et al., , 2018Zheng et al., 2016). The U 37 K index can be applied to lacustrine environments and is calculated as follows: 110 Despite successful application of the U 37 K index in lakes (D'Andrea et al., 2011(D'Andrea et al., , 2012van der Bilt et al., 2018), regional variability in the relationship between the U 37 K index and temperature often requires the development of local temperature 115 calibrations (Wang and Liu, 2013;D'Andrea et al., 2016;Longo et al., 2016). Unfortunately, there is currently no local calibration for Icelandic lakes. We further discuss this issue below.

Seasonal temperature sensitivity of Group I alkenones in lakes
In Greenland and Alaska, Group I Isochrysidales bloom during the early spring in the photic zone as lake ice starts to melt 130 (D'Andrea and Huang, 2005;D'Andrea et al., 2011;Longo et al., 2016Longo et al., , 2018. Alkenone production remains high as the lake undergoes isothermal mixing and decreases when thermal stratification begins to develop in late spring/early summer (Longo et al., 2018). This holds true for other Group I-containing lakes in the NH, including lakes in Iceland, as evidenced by the positive correlation between the U 37 K index and mean spring air temperatures (Longo et al., 2018).

135
We investigated the controls on spring lake water temperatures and the timing of ice-melt in VGHV using a lake energy balance model (Dee et al., 2018). The model was initialized using ERA-Interim daily data ( The perturbation experiments focused on the effects of changes in seasonal air temperatures and shortwave and longwave radiation on lake surface temperatures and ice-off dates. We used instrumental data from Hella, Iceland (1958 to determine the magnitude of seasonal air temperature changes (Icelandic Meteorological Office). In southwest Iceland, the average temperature range for each season is about ± 7 °C, whereas the interannual variability is about ± 3 °C. Based on this 145 we perturbed the ERA interim seasonal air temperature values by -7 °C, -3 °C, 0 °C, +3 °C, and +7 °C and re-ran the lake model with the adjusted parameters. We repeated these experiments, but instead perturbed surface incident shortwave radiation to test how external forcings can drive changes in temperature. Incoming (top of the atmosphere) insolation at 63° N has increased in winter-spring (DJFMAM) by 2.5 W m -2 and in spring (MAM) by 3.7 W m -2 over the past 2,000 years (Laskar et al., 2004). We therefore tested insolation forcing by perturbing seasonal changes in surface incident shortwave 150 radiation by -4 W m -2 , -2 W m -2 , 0 W m -2 , +2 W m -2 , and +4 W m -2 . The effects of volcanic eruptions on temperature and ice-off dates were also tested by changing shortwave radiation by -30 W m -2 , -10 W m -2 , 0 W m -2 , +10 W m -2 , and +30 W m -2 . These values were based on regional radiative feedback studies from the 1783 CE Laki (Oman et al., 2006) and 2010 CE Eyjafjallajökull (Hirtl et al., 2019) eruptions in Iceland. It should be noted that Iceland receives minimal light during the winter months, and values were set to 0 W m -2 if a negative perturbation decreased shortwave radiation below 0 W m -2 . To 155 assess the effects of longwave radiation on lake water temperatures and ice-off dates, we decreased and increased incoming longwave radiation by -0.2 W m -2 , 0 W m -2 , +0.2 W m -2 . These values reflect the forcing from well-mixed greenhouse gas (GHG) radiation during the pre-industrial period (Schmidt et al., 2011).

U 37 K index: corrections for species-mixing 160
Our U 37 K index from VGHV suggests that there was substantial variability in temperature during the last 2,000 years, but there was also variability in the community of alkenone-producers (Fig. 3a). Alkenones in VGHV surface sediments have a RIK37 value of 0.60 and genetic analyses confirm that Group I Isochrysidales is the main alkenone-producer (Longo et al., 2018;Richter et al., 2019). However, the RIK37 values increase slightly above the Group I cut-off of 0.63 about c. 500 CE, and then show a more sustained increase after human settlement in Iceland (c. 870 CE), suggesting that Group II alkenone-165 producers were also present in the lake (Fig. 3b).

Controls on spring lake water temperature
Results from the lake energy balance model show that seasonal perturbations can have a strong influence on spring lake water temperatures and ice-off dates in VGHV ( Fig. 5; Tables S1 and S3). The control run yields an average ice-off date of April 15 th with water temperatures on May 1 st about 6.6 °C. Air temperature perturbations during the winter (DJF) and spring  (MAM) alter the timing of ice-off and how rapidly surface water temperatures warm, with warmer air temperatures leading to warmer water temperatures and earlier ice-off dates ( Fig. 5a-b). In addition, an increase in shortwave solar radiation during the spring season (MAM) leads to earlier ice-off dates and warmer lake water temperatures ( Fig. 5c-d, Table A2).
Shorter days during the winter months (DJF) limits the amount of shortwave radiation reaching Iceland, and therefore has a minimal influence on Icelandic temperatures ( Fig. 5d and f). Shortwave radiative perturbations from volcanic eruptions 205 during the winter season result in small changes in spring lake water temperatures and ice-off dates, while eruptions during the spring lead to much colder spring water temperatures and later ice-off dates (Fig. 5e-f, Table A2). The increase in longwave radiation from GHGs during the pre-industrial period is relatively small and has no significant influence on either lake water temperatures (change from control = 0.01 ± 1.02 °C) or ice-off dates (change from control = -0.03 ± 8.77 days) ( Fig. A1, Table A3). Thus, the timing of the alkenone bloom and the water temperatures recorded by the alkenones are most 210 likely responses to changes in air temperature and temperature changes driven by shortwave solar radiation during the late winter and spring season (DJFMAM).

Long-term trends and short-term variability in the U 37 K record and temperature
The U 37 K index can provide temperature estimates using linear relationships that are calibrated in lakes with Group observe temperature swings close to 10 °C during the spring transitional season (Fig. 5b). Nevertheless, the U 37 K index is known to be highly sensitive to temperatures in NH lakes, suggesting that the extreme amplitude of reconstructed temperatures stems from the lack of a local calibration. Therefore, we use the U 37 K index to infer and evaluate qualitative changes in temperature trends and variability during the past 2,000 years. 225 The U 37 K record from VGHV, corrected for species mixing, exhibits a long-term trend towards warmer spring lake water temperatures over the past 2,000 years as well as strong multi-decadal to centennial variability (Fig. 6). In particular, a warmer period from the start of our record to c. 200 CE roughly coincides with the Roman Warm Period (RWP) in Northern Europe (Seppä et al., 2009). Cooler temperatures from c. 250-600 CE correspond to the Dark Ages Cold Period (DACP). 230 More variable temperatures, but on average warmer temperatures between c. 850-1100 CE could be associated with the Medieval Climate Anomaly (MCA). The time period usually associated with the LIA is characterized by a warming trend with an abrupt decrease in the U 37 K index from c. 1800-1900 CE. However, caution should be used when interpreting results after c. 1400 CE because of low sampling resolution (c. 50 yrs between each sample).

Long-term seasonal climate trends in North Atlantic paleoclimate records
In global climate model simulations, rising GHG concentrations and retreating ice sheets during the Holocene lead to warming of global mean annual temperatures, including mean annual temperatures in the NH (Liu et al., 2014). However, temperature reconstructions from the NH typically exhibit a long-term cooling trend during the Holocene (Fig. 6b) (Kaufman et al., 2009;Seppä et al., 2009;PAGES 2K Consortium, 2013, 2019McKay and Kaufman, 2014). This cooling is 240 often interpreted as a response to decreasing summer insolation. This discrepancy could highlight important deficiencies in climate models and/or feedbacks to NH summer temperatures that influence the temperatures in other seasons or suggest that existing NH temperature reconstructions are biased towards the warm season (Liu et al. 2014). The latter interpretation would imply that insolation has a large influence on seasonal temperatures, with summer cooling and winter warming controlled by increasing and decreasing seasonal insolation, respectively (Fig. 6a). In this case, mean annual temperatures 245 could still be primarily controlled by rising greenhouse gases, particularly at the global scale (Liu et al., 2014).
Winter-spring temperatures from VGHV warm over the past 2,000 years (Fig. 6), and our lake energy balance modeling results show that increasing shortwave radiation and air temperatures during the winter and spring season result in warmer water temperatures and earlier ice-off dates. In contrast, mean annual long-wave radiative forcing, i.e. GHGs, over the pre-250 industrial period have a minimal influence on water temperatures and ice-off dates. This suggests that the long-term warming trend in VGHV is driven by air temperature changes and solar insolation during the winter and spring season.
Long-term warming trends over the Holocene were also observed in other records of cold-season temperatures in the NH.
For instance, pollen records of cold-season temperatures from North America and Europe (Mauri et al., 2015;Marsicek et 255 al., 2018) and an alkenone reconstruction of winter-spring temperatures from Alaska (Longo, 2017) suggest that increasing winter orbital insolation over the Holocene drove warming during the winter season. Winter warming over the Holocene, including the Common Era, was also inferred from chrysophyte cysts in Spain (Pla and Catalan, 2005). An increase in winter sea surface temperatures (SSTs) and a decrease in seasonality over the Holocene is observed in a marine record directly south of Iceland from the North Atlantic subpolar gyre (Van Nieuwenhove et al., 2018). In each of these studies, insolation is 260 proposed as the primary mechanism that is driving changes in seasonal temperatures over the Holocene, supporting the results of our energy balance model.

years. In addition, (b) a compilation of Arctic temperature reconstructions (McKay and Kaufman, 2014) is shown for comparison with the (e) winter-spring temperature reconstruction from VGHV (the thick black lines for panels (b) and (c) are lowpass filters with data resampled to every 25 years to capture the low-frequency variability of the datasets).
Although there are existing reconstructions of NH cold season temperatures during the Common Era, the high-resolution 270 reconstruction of winter-spring temperatures from VGHV is one of the few sites in the Northern high latitudes where the effects of seasonal insolation on winter temperature can be tested without having to account for the influence of compounding factors on proxy records. For instance, varve thickness records have been interpreted to reflect winter temperatures but might be influenced by human activities in the catchment area, variations in snow accumulation, the timing of spring melt, and changes in precipitation (Ojala and Alenius, 2005;Haltia-Hovi et al., 2007). Varve thickness records 275 from the Arctic that record changes in snow or glacial melt are also used to infer long-term cooling during the melt season; however, the melt season in the high Arctic often extends well into the summer months (Cook et al., 2009;Larsen et al., 2011) and can be affected by Arctic summer hydrology. Water isotope records from ice cores from Svalbard and a stalagmite from the Central Alps are sensitive to winter air temperatures during the instrumental period, but changes in the moisture source and seasonality of precipitation over time can alter long-term temperature interpretations (Isaksson et al., 2005;280 Mangini et al., 2005;Divine et al., 2011a, b). In Iceland, records of increasing drift ice along the North Icelandic shelf are sensitive to cold season temperature but are also strongly influenced by different ocean currents and are in direct contrast to records from Western Iceland (Ogilvie et al., 1984;Hopkins, 1991;Moros et al., 2006;Andrews et al., 2019). Although our record lacks a local calibration, we argue that the U 37 K record from VGHV provides a robust, albeit qualitative, record of cold season temperature given the unique seasonal growth ecology of Group I haptophytes in NH lakes. 285 The warming trend observed in our new record from Iceland and other records of cold-season temperatures from the NH suggest that long-term temperature changes are driven by a common forcing, mostly orbitally-driven changes in winter and spring insolation (Pla and Catalan, 2005;Mauri et al., 2015;Longo, 2017;Marsicek et al., 2018;Van Nieuwenhove et al., 2018). Our lake model results confirm that water temperatures are sensitive to perturbations in shortwave radiation, 290 indicating that the long-term warming trend observed in the VGHV record is most likely being driven by increasing winterspring insolation.

Seasonal multi-decadal to centennial climate variability in the North Atlantic
The VGHV reconstruction of winter-spring temperatures provides an opportunity to test how changes in internal climate variability and external forcings (volcanic eruptions and total solar irradiance, or TSI) influence temperature changes during 295 the cold season in the North Atlantic region. This is important for understanding multi-decadal to centennial climate changes during the cold season and the role of the Atlantic Multi-decadal Variability/Oscillation (AMV/AMO), a low-frequency basin-wide North Atlantic SST anomaly that varies in response to external forcings and internal variability, in driving and/or responding to changes during the cold season (Kerr, 2000;Wang et al., 2017;Yeager and Robson, 2017). The NAO is most active during the winter months, and therefore our record should be particularly sensitive to the NAO and could provide 300 insight into its variability (Hurrell, 1995).
Whether forced or unforced, variability in winter atmospheric circulation, including the NAO, and sea ice extent are often linked to multi-decadal and centennial climate change in the North Atlantic region, particularly over Iceland (e.g. Hanna et al., 2006;Massé et al., 2008;Wang et al., 2017;Yeager and Robson, 2017). In the VGHV record of winter-spring 305 temperatures there is a sharp decrease in the U 37 K index c. 250 CE and a corresponding increase in drift ice along the North Icelandic shelf (c. 400-900 CE), which coincides with the RWP to DACP transition ( Fig. 7; Cabedo-Sanz et al., 2016).
Cooling during the DACP is typically attributed to volcanic eruptions and a minimum in solar activity (c. 400-700 CE; Fig.   7a-b); however, the DACP was not uniform across the NH and records differ as to when peak cooling occurred (Steinhilber  Sigl et al., 2015;Helama et al., 2017). For instance, there is no distinct or prolonged cooling in Iceland SST 320 records during the DACP (Fig. 7c and d), whereas terrestrial records indicate that temperatures were on average cooler between c. 450-700 CE in the Arctic and Northern Europe or c. 500-650 CE in Europe (Sicre et al., 2008(Sicre et al., , 2011Kaufman et al., 2009;Miettinen et al., 2012;Jiang et al., 2015;Sigl et al., 2015;Helama et al., 2017). The heterogeneous temperature response is commonly attributed to a prolonged negative phase in the NAO, which would result in more pronounced cooling in Icelandic and Northern European winter temperatures (Helama et al., 2017). Our record would support this interpretation. 325 A negative winter NAO phase does not cause colder NH summer temperatures, which might explain why cooling is not observed in summer temperature reconstructions from tree rings or SSTs before 500 CE (Sicre et al., 2008(Sicre et al., , 2011Miettinen et al., 2012;Jiang et al., 2015;Sigl et al., 2015;Helama et al., 2017).
Winter and spring temperatures in VGHV were on average higher between c. 880-1100 CE than temperatures during the 330 DACP but were not stable or particularly warm, unlike the climate usually associated with the Norse settlement of Iceland between c. 870-1100 CE (Fig. 7) (Ogilvie et al., 2000). This time period corresponds to a peak in TSI and weak volcanic activity, which could have triggered a positive phase in the AMV/AMO and warmer summer and annual SSTs near Iceland (Sicre et al., 2008(Sicre et al., , 2011Justwan et al., 2008;Ran et al., 2011;PAGES 2K Consortium, 2013;Otto-Bliesner et al., 2016;Wang et al., 2017). Observational data and model results suggest that a positive AMV/AMO can result in more frequent 335 negative NAO phases during the winter, which could explain the cold excursions observed in the VGHV reconstruction and reports of several severe winters in historical records (Ogilvie et al., 1984;Omrani et al., 2014Omrani et al., , 2016Peings and Magnusdottir, 2014).
The 14 th -15 th centuries mark the start of the LIA and are often associated with a colder-than-average climate in Iceland 340 (Ogilvie et al., 1984;Ogilvie and Jónsson, 2001). In contrast, the VGHV record indicates there was no prolonged cold period during the winter and spring season between c. 1300-1800 CE. Rather, temperatures inferred from the VGHV record varied on multi-decadal to centennial timescales. Cooling during this time period is associated with major volcanic eruptions and decreases in TSI that most likely resulted in a negative AMV/AMO phase and temporary increases in sea ice (Fig. 7;Moros et al., 2006;Massé et al., 2008;Steinhilber et al., 2009;PAGES 2K Consortium, 2013;Cabedo-Sanz et al., 2016;Otto-345 Bliesener et al., 2016;Wang et al., 2017). In some marine records from Iceland this led to cooling of summer/annual SSTs between c. 1400-1900 CE (Fig. 7c ande;Jiang et al., 2005Jiang et al., , 2015Sicre et al., 2008;Ran et al., 2011).
In the VGHV record, multi-decadal variability and an inconsistent temperature response to major radiative forcings during the LIA suggest that temperature anomalies during the cold season are driven by both forced and unforced variability. For 350 instance, the strong negative radiative forcing after the Samalas eruption (1258 CE) and the Wolf solar minimum (c. 1280-1350 CE) correspond to an increase in drift ice along the North Icelandic shelf and cooling in the VGHV temperature record ( Fig. 7; Massé et al., 2008;Steinhilber et al., 2009;Sigl et al., 2015;Cabedo-Sanz et al., 2016). Similarly, the cumulative https://doi.org/10.5194/cp-2020-84 Preprint. Discussion started: 26 June 2020 c Author(s) 2020. CC BY 4.0 License. effects of the Dalton solar minimum c. 1790-1830 CE and multiple major volcanic eruptions (i.e., Laki 1783 CE, unidentified 1809 CE, and Tambora 1815 CE) in the late 18 th and early 19 th century could have resulted in enhanced sea ice 355 feedbacks and cooling in VGHV temperatures between c. 1800-1900 CE (Massé et al., 2008;Steinhilber et al., 2009;Zanchettin et al., 2012;Sigl et al., 2015;Cabedo-Sanz et al., 2016;Toohey and Sigl, 2017). Enhanced cooling in the Arctic or colder initial conditions associated with a solar minimum could have dampened the positive NAO response, i.e. winter warming in Iceland and Northern Europe, that is usually observed for two to five winters after volcanic eruptions (Yoshimori et al., 2005;Zanchettin et al., 2012;Ortega et al., 2015;Smith et al., 2016;Sjolte et al., 2018). The inconsistent response in 360 VGHV temperatures to volcanic eruptions and solar minima between c. 1450-1750 CE could be associated with a return to a positive NAO phase that counteracted the effects of negative radiative forcings and led to winter warming over Iceland.
On multi-decadal to centennial timescales, changes in the VGHV record do not consistently correspond to major temperature anomalies observed during the summer months. The differences in seasonal climate responses to external forcings imply that 365 the regional manifestation of these events depends on the initial state of the atmosphere and ocean but is also modulated by internal climate variability (Yoshimori et al., 2005;Zanchettin et al., 2012;Otto-Bliesner et al., 2016;Anchukaitis et al., 2019). For instance, a negative NAO phase coupled with a solar minimum and/or major volcanic eruptions could result in amplified cooling (e.g. Zanchettin et al., 2012;Moffa-Sánchez et al., 2014;Helama et al., 2017;Anchukaitis et al., 2019).