1 Introduction

The control exerted by large scale atmospheric circulation modes on the oxygen isotopic composition of precipitation (δ18Op) has been utilised to infer past atmospheric circulation states using proxies that capture δ18O at a wide range of locations (e.g. Smith and Hollander 1999; Feng et al. 2007; Fischer and Mattey 2012). Such reconstructions typically rely on the oxygen isotopic composition of terrestrial archives such as ice-cores, tree rings, speleothems and lacustrine carbonates (Jones et al. 2009) and are underpinned by assumptions about the links between atmospheric circulation modes and the ocean source regions of atmospheric vapour and its transport paths (e.g. Lawrence et al. 1982; Friedman et al. 2002; Baldini et al. 2010; Liu et al. 2010, 2013).

However, since δ18Op data are available for the last few decades only and are not spatially or temporally coherent (International Atomic Energy Agency's Global Network of Isotopes in Precipitation, GNIP; IAEA/WMO 2015), palaeoclimate reconstructions are typically based on the assumption of long term stationarity of the influence of the atmospheric teleconnection pattern of interest on δ18Op, derived from linear relationships between δ18Op and the proxy records δ18O for some overlapping calibration period.

In the N. Atlantic/ European region, substantial efforts have been made to reconstruct past North Atlantic Oscillation (NAO) states, the main driver of European winter climate variability (Hurrell 1995; Hurrell et al. 1981; Wanner et al. 2001), in order to provide the necessary perspective to assess its long-term variability, and to evaluate the potential impact of anthropogenic climate change on the NAO (Luterbacher et al. 2002; Trouet et al. 2009).

Optimal locations of natural archives for δ18O-based NAO reconstructions can be identified with the δ18Op sensitivity to the NAO using correlation distribution maps (Baldini et al. 2008), principal component (PC) or canonical correlation analysis (Fischer and Mattey 2012). However, recent studies have shown that migrations of the geographical positions of the NAO dipole relative to those used to define the classical station-based NAO indices (Hurrell 1995; Jones et al. 1997, Fig. 1a) are linked to shifts in the winter temperature and precipitation anomaly patterns in European mid-latitude regions, particularly around 45–50°N (Comas-Bru and McDermott 2014; Raible et al. 2014). Most of these migrations can be rationalised through the concomitant state of the second main atmospheric circulation mode affecting the North Atlantic European region, namely the East Atlantic pattern (EA) (Moore and Renfrew 2012; Moore et al. 2013; Comas-Bru and McDermott 2014). This second mode of sea-level pressure (SLP) variability in the North Atlantic region was first described by Wallace and Gutzler (1981) as anomalously high 500 mb height anomalies over the North Atlantic (near 55°N ; 20–35°W) and low heights over the subtropical Atlantic and eastern Europe when in positive mode. This atmospheric pattern is also prominent during winter (Barnston and Livezey 1987) and is known to affect precipitation and air temperature patterns over Europe (Rodriguez-Puebla et al. 1998; Comas-Bru and McDermott 2014). Recent studies (Moore and Renfrew 2012) have used an EA index back to 1870 based on SLP data from Valentia Island (Ireland), but this index has also been defined in the literature as the second principal component of regional gridded SLP (Moore et al. 2013; Comas-Bru and McDermott 2014, Fig. 1b).

Fig. 1
figure 1

Time series of the NAO and EA indices. a Instrumental NAO index using Gibraltar and SW Iceland stations (red solid line; Jones et al. 1997), PC-based NAO index using the 20CRv2 reanalysis dataset (Compo et al. 2011) as used in Comas-Bru and McDermott (2014, dashed grey line), and the PC-based NAO index used in this study (black solid line). b mean winter (DJF) SLP data from Valentia Observatory (in mbar; Ireland) used to derive an instrumental EA index as in Moore and Renfrew (2012, red solid line), PC-based EA index using the 20CRv2 reanalysis dataset (Compo et al. 2011) as in Comas-Bru and McDermott (2014, dashed grey line), and the PC-based EA index used in this study (black solid line). The indices used here (solid black lines) are calculated using ERA40 (Uppala et al. 2005) and ERAinterim (Dee et al. 2011) data for the periods December 1957 to February 1979 and December 1979 to February 2013, respectively. Each winter mean is ascribed to the year of January (i.e. Dec 1981 to Feb 1982 mean is reported as 1982) following the common definition for the NAO index (Hurrell 1995)

Full size image

Migrations of the geographical positions of the NAO dipole mentioned above involve shifts in the southern NAO centre of action toward the southwest when the NAO/EA indices have the same sign, and shifts towards the northeast when the indices have opposing signs (Moore et al. 2013; Comas-Bru and McDermott 2014). As a result, the axis over which the polarity of the NAO climate correlations changes, migrates to higher latitudes when the southern centre of action is closer to the European continent (Comas-Bru and McDermott 2014). Crucially, these migrations of the NAO-winter climate relationship might also affect δ18Op patterns, potentially weakening their ability to accurately reflect the NAO at some locations in Europe.

Fig. 2
figure 2

Spearman correlation coefficients between winter δ18Op and a NAO; b EA using the GNIP data (coloured circles; 1961–2013) and the ECHAM5-wiso gridded model output (1959–2013). Significance level at 95 % (white circles denote non-significant correlations)

Full size image

It is well known that a complex set of factors influence δ18Op (Yurtsever 1975; Rozanski et al. 1993), but no study has so far addressed the impact of the EA on the spatial δ18Op patterns associated with the NAO, or its implications for palaeoclimate reconstructions. Here, for the first time, systematic spatial patterns in δ18Op linked to effects of the EA on NAO-δ18Op relationships during the observational period are identified. A better understanding of these effects is important to improve the interpretation of natural δ18O archives in terms of past NAO states in Europe.

2 Data and methods

2.1 δ18O in precipitation

Precipitation isotopic data (δ18Op) for the winter months December to February (DJF) were obtained from the International Atomic Energy Agency's Global Network of Isotopes in Precipitation (GNIP; IAEA/WMO 2015), which provides a network of stations over Europe covering the period from March 1960 to December 2013. The 26 stations that have more than 25 complete winters of data and are located in the area 11°W–15°E; 40°N–60°N have been selected (Table 1 and Fig. S1). The stations with the longest records are Vienna (Austria), Groningen (Netherlands) and several stations in Switzerland. A matrix map showing the missing values for all 26 stations is shown in Fig. S2.

Fig. 3
figure 3

Spearman correlations between winter δ18Op and the NAO index (a–c); δ18Op-winter air temperature (wTmp; d–f) and δ18Op-winter precipitation amount (wPre; g–i); for the period 1959–2013 (a, d, g), years with NAO/EA of the same sign (b, e, h; n = 36) and years with NAO/EA of opposite sign (c, f, i, n = 19). All correlations shown are significant at 95 % level

Full size image

In addition, the mean δ18Op in winter (DJF) at the surface was also obtained from the output of the isotope-enabled version of the atmospheric general circulation model ECHAM5 (Roeckner et al. 2003, 2006), namely ECHAM5-wiso (Werner et al. 2011), for the N. Atlantic sector (100°W–40°E; 10–80°N) in the period January 1958 to December 2013. For this study, the model has been run at a grid spacing of approximately 1° (T106 spectral mode; comparable to that of Langebroek et al. 2011) in a so-called nudged mode, forced with ECWMF, ERA40 and ERAinterim data (Butzin et al. 2014). As discussed below, annual and seasonal ECHAM5-wiso simulation results are in good agreement with available observations of isotope ratios in precipitation from the GNIP stations (IAEA/WMO 2015) on both global and European scales (Langebroek et al. 2011; Werner et al. 2011).

2.2 Climate variables

Total winter (DJF) precipitation amounts (wPre; mm) and surface air temperatures (wTmp; °C) have been compiled for the North Atlantic European region from the CRU-TS3.1 global observational dataset (Mitchell and Jones 2005) for the period 1959–2009. The horizontal grid interval of this dataset is 0.5° latitude × 0.5° longitude. When available, wPre (mm) and wTmp (°C) have also been obtained at the selected European GNIP stations listed in Table 1. These climate variables are also provided by the ECHAM5-wiso model output (Werner et al. 2011).

Fig. 4
figure 4

Map showing the areas where the δ18Op-NAO (a) and the δ18Op-climate (b, c) relationships are robust. Grey areas indicate the grid cells (\(\sim 1^\circ \times 1^\circ\)) where Spearman correlations are significant at 95 % for both the OP and EQ subsets.

Full size image

2.3 NAO and EA indices

The NAO and EA indices used here are the two leading vectors of the cross-correlation matrix calculated from monthly (December to February) SLP anomalies over a confined Atlantic sector (100°W–40°E; 10–80°N) using ERA40 (Uppala et al. 2005) and ERAinterim (Dee et al. 2011) re-analysis data for the periods December 1957 to February 1979 and December 1979 to February 2013, respectively. This NAO index based on our Principal Component (PC) analysis (Fig. 1a) is similar to the instrumental NAO index calculated as the difference between the normalised SLP over Gibraltar and that over SW Iceland (Jones et al. 1997), updated by the Climate Research Unit of the University of East Anglia (www.cru.uea.ac.uk/~timo/datapages/naoi.htm; \(\rho\) = 0.98; p value \(<\) 0.01; Fig. 1a). Similarly, the PC-based EA index used here is equivalent to that derived from the instrumental December to February SLP data from Valentia Observatory, Ireland (\(\rho\)= 0.81; p value \(<\) 0.01; Fig. 1b). SLP data from this meteorological station was used in previous studies as an instrumental EA index (Moore and Renfrew 2012).

Fig. 5
figure 5

δ18Op time series at the location of four GNIP stations extracted from the ECHAM5-wiso model output (dashed grey) and from the GNIP data (solid black). The stations are a Valentia (Ireland), b Groningen (Netherlands), c Grimsel (Switzerland) and d Vienna (Austria). Their exact locations are shown in Table 1. Each winter mean is ascribed to the year of January (i.e. Dec 1981 to Feb 1982 mean is reported as 1982)

Full size image

A positive NAO index indicates a strong meridional SLP gradient in the North Atlantic region between the Icelandic low and the Azores high. A positive EA index (consistent with usage by Comas-Bru and McDermott (2014), with the SLP series from Valentia Observatory (Ireland) used in Moore and Renfrew (2012) and with the original Wallace and Gutzler (1981) definition, but of opposite sign to that of Woollings et al. (2010), Woollings and Blackburn (2012) and Moore et al. (2013)), indicates the presence of positive SLP anomalies over the subtropical North Atlantic during winter, with a centre of action west of Ireland, near 55°N ; 20–35°W. Thus, a positive EA index, as used in this study, is associated with a poleward migration of the North Atlantic jet stream in winter, and the implications of this for observed and modelled δ18Op values are discussed below.

2.4 Data selection

Following the approach of previous studies (Moore et al. 2013; Comas-Bru and McDermott 2014), the available monthly δ18Op and climate data have been divided in two subsets according to the relative polarity of the NAO and EA indices. The first group (hereafter “EQ”) comprises winters with NAO and EA indices of the same sign (n = 36 winters), and corresponds to the subset (NAO/EA)S defined by Comas-Bru and McDermott (2014). The second group (hereafter “OP”) is equivalent to the (NAO/EA)O subset of Comas-Bru and McDermott (2014), and it includes winters with NAO and EA indices of opposite sign (n = 19 winters).

Fig. 6
figure 6

Scatter plots of δ18Op during OP versus EQ winters. a Using the GNIP data; b using the ECHAM5-wiso model output for the same locations. To aid interpretation, only NAO \(>\) 0 winters are considered here. Colour bar refers to the latitude (°N) of each station. Symbols within a box in b correspond to the modelled output at Vienna and Graz Universität (Austria). See the text for more information on these locations

Full size image

To aid interpretation, only the NAO \(>\) 0 winter sub-set (n = 24) is considered here when calculating absolute differences of δ18Op, winter surface air temperature and precipitation amount values between OP and EQ winters in Figs. 6 and 7. It is important to note, however, than the same conclusions are valid if only NAO < 0 winters are considered.

3 Results

3.1 Spatial sensitivity of δ18Op to atmospheric circulation modes

Consistent with Baldini et al. (2008), broadly positive correlations are observed between winter δ18Op and the NAO index for the majority of GNIP stations (Fig. 2a), with non-significant correlations found in the southeast of the study area, and weak correlations observed at the Valentia station (Ireland) and at some stations close to the Baltic Sea. Similar results are obtained using the δ18Op modelled with the ECHAM5-wiso model (Fig. 2a). On the other hand, the relationship between δ18Op and the EA is much weaker, with few stations showing a significant correlation (Fig. 2b). In this case, δ18Op in the southern stations (<∼48°N ) is negatively correlated to the EA, whereas positive correlation coefficients occur in the northern locations. A similar change in the sign of the correlation coefficient is observed when the ECHAM-5 wiso model δ18Op output is used to calculate the correlations (Fig. 2b), although in most of southern Europe modelled δ18Op is not significantly correlated with the EA index.

Fig. 7
figure 7

Differences between OP and EQ winters for a number of winter variables obtained from the GNIP dataset: δ18Op (a), winter surface air temperature (wTmp; b) and winter precipitation amount (wPre; c). To aid interpretation, only NAO \(>\)0 winters are considered here. Red bars are stations located at elevations \(>\)500 m. Rho ( \(\rho\)) values in b and c are Spearman correlation coefficients with respect to a. The correlation with * is significant at 95 %; the not significant correlation is indicated with N.S

Full size image

The spatial δ18Op-NAO correlation patterns in Europe (Fig. 3a–c) exhibit shifts linked to the relative polarity of the NAO/EA modes that are similar to those reported previously by Comas-Bru and McDermott (2014) for NAO-climate relationships. An important result is that when the NAO and the EA have the same sign (Fig. 3b), the area showing positive δ18Op-NAO correlations expands over most of north central Europe, whereas when the atmospheric teleconnection patterns have opposite polarity, the positively correlated area is shifted distinctly southwards (Fig. 3c) and the correlation in the Iberian Peninsula improves. Similar patterns are observed in the correlation between δ18Op and winter air temperature (wTmp; Fig. 3e, f). In terms of winter precipitation amount (wPre), only the UK and Ireland and some circum-Mediterranean regions are significantly correlated with δ18Op (Fig. 3g). However, only in the circum-Mediterranean region the correlations with wPre hold for any combination of NAO and EA polarities (Fig. 3h–i).

To better identify regions that exhibit temporal stationarity in their sensitivity to the NAO in terms of δ18Op, Fig. 4a illustrates the grid cells where the correlation between δ18Op-NAO remains significant regardless of the shifts observed in the correlation distribution maps discussed above (Fig. 3b, c). These areas include most of central Europe, the northern coast of the Iberian Peninsula, the southern tip of Sweden and Norway, Iceland and southwest Greenland (Fig. 4a). On the other hand, the relationship between δ18Op and wTmp is robust regardless of NAO/EA polarities between 45 and 60°N in Europe, including most of the UK and Ireland, as well as the majority of Greenland (Fig. 4b). Moreover, the correlations between δ18Op-wPre are robust for the OP and EQ subsets considered here, but are significant at a 95 % level, only in the Italian Peninsula, the Mediterranean islands and parts of Greenland (Fig. 4c).

3.2 Observed and modelled δ18Op

To assess how the ECHAM5-wiso model output compares with the GNIP observations at specific locations, time series of modelled δ18Op have been extracted at the grid-cell locations of four GNIP stations (Fig. 5) selected to cover four distinct regions of Europe: Valentia (S.W. Ireland), Grimsel (Switzerland), Groningen (Netherlands) and Vienna (Austria) (Fig. S1 and Table 1). All stations have more than 38 complete winters of data.

Although an offset to higher δ18Op values is modelled in Grimsel, the ECHAM5-wiso model adequately reproduces the temporal δ18Op variability observed at the four selected GNIP stations (Fig. 5). Probability density functions (PDF) of rainfall δ18Op computed for each of these four locations reflect a similar δ18Op behaviour during OP and EQ winters in the observations and the ECHAM5-wiso model output (Fig. S3). The δ18Op frequency distributions at high latitudes (Valentia and Groningen) show a substantial shift towards higher δ18Op during EQ winters (Fig. S3a, b, e, f). On the other hand, a shift towards lower δ18Op during EQ winters is observed and modelled at the continental station of Vienna (Fig. S3d, h). Different distributions are evident in the observed and modelled results for the alpine station of Grimsel, however (Fig. S3c, g). While a slight offset towards lower δ18Op values during EQ with respect to OP winters is observed in Grimsel (Fig. S3c), the model output shows the opposite behaviour (Fig. S3g) highlighting potential deficiencies in the model output at this grid spacing in regions with steep orography as discussed below. Similar results are also found for other Alpine stations such as Meiringen and Gutannen (not shown).

3.3 Latitudinal response of δ18Op to atmospheric circulation mode combinations

To investigate if the North–South trends described for the four selected stations reflect a wider geographical effect at the European GNIP stations listed in Table 1, OP vs. EQ scatterplots of δ18Op for NAO > 0 winters are illustrated in Fig. 6a. In addition, the equivalent ECHAM5-wiso δ18Op output at those locations is shown (Fig. 6b). Because of the geographical distribution within Europe of the GNIP stations considered in this study (Fig. S1), northerly stations are closer to the coast compared with stations to the south of the area of interest. In Fig. 6, data from the high latitude stations are characterised by higher observed and modelled δ18Op values, in contrast to what might be expected if a latitudinal effect only was considered. In essence, the “continental effect” in which progressive distillation of moisture and preferential removal of 18O occurs as storm tracks move eastwards dominates the dataset, with the lower latitude stations in south/central Europe exhibiting lower δ18Op values because they are more continental. Our results show that δ18Op at low latitude stations is offset to higher values during OP winters compared with EQ winters (Fig. 6), whereas δ18Op values at high latitude stations are typically lower during OP winters compared to EQ winters. This is consistent with the PDFs described above for the 4 selected stations (Fig. S3). In contrast, only two stations (Vienna and Graz Universität) show a modelled δ18Op higher during OP winters (symbols within a box in Fig. 6b). No clear trends in δ18Op between OP and EQ winters are observed or modelled as a function of altitude (Fig. S4).

The absolute differences in several winter parameters between OP and EQ winters at the GNIP stations during NAO > 0 winters, expressed as “VariableOP – VariableEQ”, are shown as a function of latitude in Fig. 7. The trend described above for δ18Op (Fig. 6a) is also shown in Fig. 7a, where positive (negative) bars indicate that δ18Op is higher (lower) during OP winters than during EQ. The latitudes where δ18OOP > δ18OEQ (positive bars in Fig. 7a; <∼48°N ) correspond to positive differences in air temperature of about 1–1.5 °C (Fig. 7b). On the other hand, at latitudes >50°N , the temperature difference is close to 0.5 °C , with the exception of three stations: Bad Salzuflen (Germany) and Groningen (Netherlands), with wTmpOP − wTmpEQ > 0.5°C, and Braunschweig (Germany) with wTmpOP − wTmpEQ < 0 °C. The differences described above between OP and EQ winters have also been assessed for the N. Atlantic/European region using the ECHAM5-wiso and the CRU-TS3.1 gridded datasets for the periods 1959–2013 and 1959–2009, respectively (Fig. S6). According to the ECHAM5-wiso model output, the largest negative differences in δ18Op for OP and EQ winters occur in the North Sea coast and the British Islands, while the most pronounced positive differences are modelled south of the Alps, in northern Italy and Eastern Europe (Fig. S6a). Overall, warmer conditions (about 1–1.5 °C higher) are observed during OP winters over continental Europe (Fig. S6b, d), while wetter conditions occur in Western Europe (Fig. S6c, e).

4 Discussion

Although the temporal correlations between δ18Op and the EA index are generally weak in continental Europe for the period 1959–2013 (Fig. 2b), the polarity of this atmospheric mode of winter climate variability in Europe exerts an important influence on the δ18Op-NAO spatial correlation patterns (Fig. 3a–c). The shifts observed for different NAO/EA scenarios (Fig. 3) are linked to the spatial structure to the SLP dipole in the N. Atlantic region (Comas-Bru and McDermott 2014; Moore et al. 2013) that results in changes of the latitude and speed of the jet stream (Woollings et al. 2010). A positive EA index (Comas-Bru and McDermott 2014) coupled with a positive NAO index results in a northward migration of the jet stream (Woollings and Blackburn 2012). As a result, storm tracks are more likely to enter Europe from the north-west in this condition, having traversed the North Atlantic at higher mean latitudes compared with the south-easterly sourced storm tracks that characterise negative EA conditions. Since only positive NAO conditions are considered for this discussion, positive EA conditions correspond to the EQ winters defined here, and negative EA conditions correspond to the OP winters. Because the northerly stations are closer to the NW coastline of Europe (Fig. S1), storm-tracks arriving predominantly from the north-west (EQ winters) undergo less rainout and Rayleigh distillation, thereby producing rainfall with higher δ18O values compared with stations in central and southern Europe that are further from the NW European coastline. This effect is clearly observed in the GNIP data (Figs. 6a, 7a) in which δ18OEQ > δ18OOP in the high-latitude stations (>48°N), but δ18OEQ < δ18OOP in the more continental stations to the south. δ18OOP values higher than δ18OEQ values in the south likely reflects rainout from moisture bearing air-masses that have been steered more directly into Europe from the south-west by the southerly-displaced jet stream (Woollings et al. 2010) and that have consequently undergone less isotopic distillation compared with their north-westerly derived counterparts. It is acknowledged that whilst this study groups winters according to their NAO/EA phases, some scatter is observed in δ18Op values within a group. This results from the complexity of factors that govern δ18Op variability and may imply that the source of the moisture-bearing air-masses could derive from different latitudes in winters that have been grouped together in this study (i.e. OP and EQ winters).

Similar results, although with the zero-line of the δ18Op anomalies slightly south of ∼48°N, are observed in the ECHAM5-wiso model output (Fig. S6a). This minor disagreement between observed and modelled data could arise from the limitation of the model to reproduce steep orographic effects in the Alps as a consequence of the model grid interval of 1° × 1°. This limitation is indeed reflected in the offset found when comparing GNIP data from three Alpine stations at different altitudes that lie within the same model grid-cell, with the time series extracted from the model output at that location (Fig. S5).

The results presented here have important implications for current efforts to reconstruct past NAO states from δ18O in natural archives. The non-stationarities between NAO and proxy δ18O described here have the potential to degrade the δ18O-NAO correlations observed during the calibration period. Consider, for example, a proxy collected in the UK or Ireland covering the instrumental period and an earlier period of interest where the ratio of OP to EQ winters is greater than the value of 0.35 observed during the instrumental period (1959-2013). In this example, the proxy δ18O -NAO relationship during the period with higher persistence of OP winters, would be weaker than that of the instrumental period, or perhaps even non-significant, as a result of the non-stationarities presented here for UK and Ireland (Fig. 3b, c). Similarly, a stronger δ18O-NAO relationship would be expected for a proxy collected from southern Europe if OP winters were more common than in the calibration period (Fig. 3c).

Hence, collection of natural proxies for NAO reconstruction purposes should be focused on those areas where the correlation between δ18O-NAO is stationary and robust regardless of the observed changes in the NAO and EA patterns. We identify these areas to be central Europe (north of the Alps), the Cantabrian coast in Northern Spain, the coastal areas surrounding the Baltic sea, Iceland and western Greenland (Fig. 4a). On the other hand, optimal regions for reconstruction of past states of surface air temperature from natural δ18O archives include the areas surrounding the North and the Baltic Sea, as well as some regions in central Europe and most of inland Greenland (4b). Similarly, if the aim is to reconstruct precipitation amounts using a proxy for δ18Op, only Italy and a few coastal sites in the circum-Mediterranean region would be suitable (4c).

5 Conclusions

Ideally, δ18Op-NAO relationships should remain stationary through time in order to use δ18O variations in natural archives as proxies to reconstruct past NAO states. Identification of regions subject to non-stationarities of the δ18Op-NAO relationship is crucial so that resources can be focused in the areas least affected by such non-stationarities.

Table 1 List of GNIP stations used in this study ordered by their latitude in descending order
Full size table

Here we show that it can be anticipated that natural δ18O archives will exhibit different degrees of stationarity in their δ18Op response to the NAO in many regions of Europe simply as a function of temporal changes in the EA pattern. This observation can be rationalised through changes in the SLP structure in the N. Atlantic region as a result of the concomitant states of the NAO and EA patterns, which affect the trajectories of the air-masses carrying moisture onto Europe and ultimately the δ18Op signal. Proxy calibration (an overlapping period with instrumental data) could be compromised or biased by different NAO/EA combinations within the calibration dataset. Depending on the proxy location, the slopes of δ18Oproxy/climate variable data trends could be affected by different persistence patterns of the NAO and EA modes within the instrumental calibration period, compared with that expected over longer timescales of the proxy record. We have identified regions where δ18Op shows a robust stationary relationship with the NAO regardless of the concomitant state of the EA during the instrumental period. These regions are central Europe (north of the Alps), the northern coast of the Iberian Peninsula, Iceland and west Greenland (Fig. 4).