The proxy used is the NGT stacked stable oxygen isotope record δ18O obtained from 21 ice cores drilled mainly from the 16 different locations in central and Northern Greenland overlapping on a common period . In particular, the stacking methodology adopted for computing the NGT stacked series is the average of δ18O from all different ice cores considered . The NGT stacked δ18O series is provided as anomalies with respect to period 1961–1990. The NGT stacked δ18O series is available from 1000 to 2011 with annual resolution. However, the series used in this study is limited from the 1602 due to the paleoreanalysis dataset described in the last paragraph of this section. For more details on the stacking method and data sources see .

The atmospheric circulation during boreal winter months (December, January and February; hereafter DJF) in the observational period (1920–2011) is investigated using the third version of the 20th Century reanalysis (20CRv3) , which provides daily data and a spatial resolution of 1.0°×1.0°. In particular, the atmospheric variables used in this study are the geopotential height at 500hPa, Z500, along with the zonal u and meridional v components of the wind at the same pressure level. The specific humidity q, the zonal, and meridional components of the wind at different pressure levels (300–1000 hPa) are used to investigate the water vapor transported by the atmosphere. The pressure levels are limited between 300–1000 hPa because most water vapor is in the lower troposphere. In particular, the water vapor transport is determined by the Integrated Vapor Transport (IVT) , 1IVT=1g∫1000300qudp2+1g∫1000300qvdp2, where g is the gravitational acceleration in ms-2, u, and v are the zonal and meridional components of the wind in ms-1, q the specific humidity, and dp is the pressure difference between two adjacent pressure levels.

The extreme conditions regarding temperature and precipitation over Europe in boreal winters are assessed in the observational period using the E-OBS dataset . In particular, two ETCCDI extreme indices are considered: TN10p and PRCPTOT. The TN10p index is defined as the percentage of days when the minimum temperature is below the 10th percentile over a given period, while the PRCPTOT is the cumulated rainfall in days when at least 1 mm of rainfall occurred. In the case of this study, the TN10p and PRCPTOP are computed for each single month DJF in the observational period. The two extreme indices are computed using the Python package icclim . A comparison is also made with the 20CRv3 dataset and a new dataset of reconstructed extreme indices named CRAI . The main goal of the CRAI dataset is to reconstruct climate extreme indices, particularly for periods and regions with sparse observational data. This dataset uses a deep-learning algorithm that combines U-net architecture and partial convolutional layers . This dataset is used in comparison with E-OBS to provide a more complete overview of extreme temperature and precipitation conditions across Europe, considering the data sparsity present in some regions of the E-OBS dataset, especially in the period 1920–1950.

The Ensemble Kalman Filter (EKF) v2 Paleoreanalysis is used to investigate atmospheric circulation patterns, temperature and precipitation in the boreal winters in the long-term perspective period (1602–2003). The EKF dataset has monthly time resolution and spatial resolution of 1.85°×1.85°. This data set provides an optimized combination of model outputs and paleoclimate reconstructions obtained by different proxies through data assimilation techniques. In particular, the technique adopted in this dataset uses the ensemble-based Kalman filtering to assimilate historical and instrumental measurements of temperature, precipitation and pressure together with paleoclimate records from tree-rings and corals. The model used for modeling the atmosphere is the atmospheric-only module of the general circulation model ECHAM5.4 . It is worth noting that prior 1850, the paleoclimate records from tree rings play a major role in this paleoreanalysis dataset. Even though, the use of primarily tree rings may introduce some biases due to the different growth rates for each species, tree line sampling, water availability , the contribution from tree rings is mainly focused in mid-latitude in summer time. Indeed, as stated in , the climate reconstruction is mainly driven by the model output. See for details on the data assimilation techniques and proxies adopted.

The principal method of investigation in this paper is the composite map analysis . The aim of the composite analysis is to identify relations between two physical variables during the occurrence of particular extreme conditions in one of the two variables. In this case, we identify the years when the NGT stacked δ18O series exceeds two threshold values considered extreme and then analyses the atmospheric variables to characterize the average atmospheric circulation during those years. The choice of these threshold values is a trade-off between capturing years when δ18O series takes extreme values and having a congruous number of observations available to carry out the composite analysis. The two threshold values chosen are ±1σ, where σ=(n-1)-1∑i=1n(xi-x¯)21/2 is the standard deviation and x¯ the average of the NGT stacked δ18O series xi during the period considered, respectively. The values of the NGT stacked δ18O series exceeding these two thresholds represent the 30 % most extreme values and permit to have enough number of years to analyze. It follows that positive years are whenever the NGT stacked δ18O series assumes values larger than 1σ. Similarly, negative years are whenever δ18O series is lower than the -1σ. Throughout the paper, the years where δ18O series takes values above (below) the thresholds are named positive (negative) years. The standard deviation for both the observational and long-term periods is then calculated and used to identify the positive and negative years of the NGT stacked δ18O series in each period.

To assess the presence of atmospheric blocking, the 1D-index , hereafter TM90, and the geopotential height modification of the potential-vorticity-based 2D-index , hereafter S04, are computed with the Mid-Latitude Evaluation System (MiLES) software . A description of the modified TM90 and S04 index is provided below as implemented in . For details on TM90 index, see , and for details on S04 index, see .

In the TM90 index, for each longitude λ, the geopotential height gradients GHGS and GHGN are computed as: 2GHGS(λ,Δ)=Z500(λ,ϕ0)-Z500(λ,ϕS)ϕ0-ϕS,3GHGN(λ,Δ)=Z500(λ,ϕN)-Z500(λ,ϕ0)ϕN-ϕ0, where ϕN=80°N+Δ, ϕ0=60+Δ, ϕS=40°N+Δ, and Δ={-5°,-2.5°,0°,2.5,5°}. A longitude λ is labeled as blocked if GHGS(λ,Δ)>0 and GHGS(λ,Δ)<-10m(°lat)-1, for any values of Δ.

In the S04 index, for each grid point at longitude λ and latitude ϕ, from 30 to 75°N, the daily anomaly of the geopotential height is computed. A grid point is labeled as blocked if the geopotential height anomaly exceeds the 90th percentile of the geopotentail heigth anomalies between 50 and 80°N. In order to detect areas of large-scale blocking, further requirements are needed. It is then required that the grid points must be blocked together, along with any grid points marked as blocked in a box of 10° longitude by 5° latitude, for at least five days.

To asses the effect of the large-scale atmospheric circulation on boreal winter temperature and precipitation in the long-term perspective, spatial index of the geopotential height at 500 hPa, average temperature and precipitation are computed, weighting each grid point by cos⁡(ϕ), where ϕ is the latitude. The spatial indices are then standardized for a better comparison. A probability density function is estimated using the non-parametric kernel estimator for identifying changes in the distribution. To quantify those changes, the ratio between the probability density functions in the negative and positive years is computed. Along with the probability density function, the 90 % confidence intervals are computed using the simple resampling with replacement bootstrap technique. The simple replacement can be used in this application because there is no serial correlation between monthly data.

All daily atmospheric variables were averaged to derive their corresponding monthly values. A linear detrend is performed by each month and each grid point to remove the trend and seasonality. The NGT stacked δ18O series has also been detrended, but each single detrending has been carried out for the observational and long-term perspective periods.

The first noticeable aspect is that average atmospheric circulation over Europe during positive years shows no well-defined anomaly pattern (Fig. a), whereas during negative years, a more pronounced pattern emerges (Fig. b). The average pattern in negative years features a high-pressure system extending from the Azores Islands to the Baltic Sea and low-pressure system over Greenland. The atmospheric circulation pattern in the negative years resembles the atmospheric circulation in the positive phase of NAO. Before discussing the relationship between the high-pressure system over Europe in the negative years and NAO, additional analyses focusing on this high-pressure system are conducted from a more dynamical perspective, particularly in relation to atmospheric blocking.

To assess whether the high-pressure system over Europe is a sign of atmospheric blocking, a 1D and a 2D blocking index are used. The TM90 and S04 indices associated with negative δ18O years show an increase in frequency of atmospheric blocking occurring over Europe (Fig. a) compared to positive years. Specifically, the TM90 index indicates an increase of blocking frequency at around 60°N and between 10°W and 35°E, with its peak around 15–25°E. The S04 index (Fig. b) permits also to analyze the spatial pattern of the atmospheric blocking, which matches the extent of the high-pressure system (Fig. b). It corroborates an increase of blocking around the same region, the Baltic Sea.

The variability of δ18O found in Greenland ice cores is affected by the NAO and the dipole structure in the geopotential height observed during the negative years (Fig. b) resembles the pattern typically associated with the positive phase of the NAO. However, two aspects should be considered before classifying such an atmospheric circulation pattern as the NAO in the positive phase. First, the atmospheric configuration during the positive NAO phase depicts a high-pressure system located over the Azores islands e.g., whereas the pattern illustrated in Fig. b exhibits a broader spatial extent over continental Europe. Moreover, presented evidence that different stages of the NAO are associated with varying occurrences of atmospheric blocking over Europe. In particular, the high-pressure system observed during the negative years in the NGT stacked δ18O ice cores (Fig. b) presents similar southwest/northeast direction of blocking occurrence during the decaying phase of the positive phase of NAO . Second, as discussed before, in the northern Greenland ice cores are found to be less affected by the NAO compared to central and southern Greenland ice cores due to their high-latitude location and topography. In conclusion, it is complex to classify the dipole structure in the geopotential height observed during the negative years of the NGT stacked δ18O series (Fig. b) as a pure atmospheric configuration of the positive phase of NAO.

Years characterized by particularly low δ18O values in NGT stacked series are associated with an increased frequency of atmospheric blocking events that prevent water vapor from reaching Greenland. The increased presence of atmospheric blocking over Europe makes the jet stream deviate northward but eastward of Greenland. This jet stream diversion reduces the transport of O18-enriched moisture toward Greenland, resulting in colder condensation temperatures and therefore more depleted δ18O precipitation . Such blocking events favor the development of extreme events e.g.:. Indeed, the frequency of occurring of atmospheric blocking over Europe is approximately four times higher than in years showing positive δ18O values. The relationship between negative δ18O years and the occurrence of atmospheric blocking events has been tested by selecting the years with extreme low values of NGT stacked δ18O and assessing the frequency of occurrence of atmospheric blocking events. This relationship is further examined by analyzing the distribution of NGT stacked δ18O series during years with increase and reduced atmospheric blocking events. Years in which the TM90 atmospheric blocking index exceeds +1σ(-1σ), where σ represents the standard deviation of atmospheric blocking frequency, and at least for 12° longitude between 10°W and 20°E are marked as years with increased (reduced) blocking events. The distribution of δ18O during blocking events exhibits a shift in the mean and a skew toward negative values, supporting the proposed linkage (Fig. ).

Consequently, the relationship between negative δ18O values and atmospheric blocking is supported by the presented evidence, which demonstrates a physical connection between δ18O variability and changes in atmospheric circulation that may contribute to an increased frequency of extreme events. For these reasons, the subsequent analyses investigate the influence of such atmospheric blocking structures on temperature and precipitation over Europe during negative δ18O years.

The increase of frequency of atmospheric blocking over Europe entails a change in the regional pattern of temperature and precipitation (Fig. ). The position and spatial extent of atmospheric blocking facilitate the poleward transport of warm, moist air from the Atlantic Ocean (Fig. a and b) toward the Norwegian coast, leading to higher temperatures (Fig. c) and increased precipitation (Fig. d). The temperature increase results from warm-air advection, while the precipitation increase is driven by heat loss and the orographic uplift of these relatively warm and moist air masses along the Norwegian coast, which boosts condensation and precipitation formation. The resulting loss of moisture on the windward side is likely what facilitates the subsequent adiabatic compression and warming over Sweden (Fig. c), a process characteristic of föhn winds.

Concurrently, the atmospheric blocking also facilitate advection of cold air masses from the Nordic region toward mid-latitudes, which leads to colder and drier conditions over southern Europe (Fig. c and d). As a result, temperatures drop in western Turkey, Greece, and, to a lesser extent, in southern Italy and the Iberian Peninsula. Additionally, moist air masses from the Atlantic, which would typically influence southern Europe, are diverted toward Scandinavia, reducing precipitation in this region. Furthermore, central Europe shows a pattern of increased temperature and precipitation. This is likely due to a portion of Atlantic air masses reaching the region, contributing to both warming and wetter conditions.

It is also worth noting that the spatial distribution of TN10p over Scandinavia appears irregular. This irregularity is likely a result of the sparse coverage of the E-OBS dataset before 1950, which affects the computation of temperature indices and leads to irregularities in the observed spatial patterns. On this purpose, the same analysis is carried out using CRAI dataset for TN10p, which shows consistent results with the results obtained using E-OBS dataset. Therefore, the sparsity of E-OBS before 1950 does not affect much the spatial patterns of temperature (Fig. ). Similarly, the same plot is carried out with the 20CRv3 dataset (Fig. ) that show similar patterns obtained using the CRAI dataset.

The atmospheric circulation in the negative years during the long-perspective period (Fig. a) follows closely the atmospheric circulation in the negative years of the observational period (Fig. b). The high-pressure system is present over Europe and in particular over the northern France and southern Great Britain. The particular circular shape of the high-pressure system has to be ascribed on average conditions as discussed in the previous section. In this case, the presence of the atmospheric blocking is not possible to assess by computing atmospheric blocking indices since the EKF dataset provides only monthly data, and therefore, the TM90 and S04 indices cannot be computed. On the other hand, the association between high-pressure patterns and the increased frequency of synoptic-scale blocking circulation has been demonstrated by .

The physical explanation outlined in the previous section for the observational period aligns well with the effect of the atmospheric blocking on temperature and precipitation over Europe observed in the long-term perspective period. In particular, it is observed a rise in temperature and increase in precipitation in Norwegian coastal areas due to atmospheric transport of moist and relative warmer air masses. On the contrary, in southern Europe a decrease in precipitation due redirection of the moist air masses toward the Nordic regions and a decrease in temperature due to cold advection of Nordic air masses is observed as well. The observed patterns in temperature and precipitation in the long-term perspective (Fig. b and c) result to show more homogeneous patterns than the observational period due to the use, in this case, of a reanalysis product.

Even if the geopotential height anomalies associated with the high-pressure system over Europe and the low-pressure system (see Fig. a) differ statistically from zero, the main effects of the high-pressure system on temperature and precipitation are not. The consistency of the results between the observational and long-term perspective periods and the statistical significance of the geopotential height pattern suggests a non-significant shift in the mean.

To further investigate whether the observed atmospheric blocking affects temperature and precipitation, the distribution of weighted averages of temperature and precipitation in the following selected grid boxes (i)

“central Europe” for the geopotential height (38–55°N and 5°W–20°E);

The distribution of the indices derived from these four boxes corroborate the hypothesis of a change in the distribution of the temperature and precipitation (Fig. ). The geopotential height index over the central Europe region during negative years is not only statistically different from zero but also presents positive anomalies ranging from +1σ to +2σ. These anomalies occur between 52 % and 93 % more frequently than in the distribution of positive years, with peaks reaching 120 % around +1.5σ (Fig. a). The temperature index over the Baltic region indicates warmer conditions from +1σ, occurring approximately 15 % more often than during positive years and increasing up to 58 % around +1.5σ, although the signal appears visually less pronounced (Fig. c). In contrast, the precipitation index over Scandinavia shows a clear shift toward heavier rainfall between +0.5σ and +1.5σ, with frequencies increasing from about 15 % to 38 %, and peaks reaching 70 % around +1σ, compared with positive years. Similarly, the precipitation index over southern Europe reveals drier conditions from -0.5σ to -2σ, with increases of 20 % to 42 % and a peak of 75 % around -1.5σ, relative to positive years. The changes in the precipitation index distributions over Scandinavia and southern Europe are statistically significant compared with their corresponding distributions during positive years.

Since EKF dataset provides monthly data, it is possible to investigate further if the change in distribution of precipitation occurs among all winter months or only throughout some particular months. Investigating the monthly distribution may help determine if opposite monthly changes hide variations at the seasonal scale.

The probability density function of the indexes is then estimated for the monthly values of the indexes (i)–(iv) computed from the boxes “central Europe”, “Baltic region”, “Scandinavia” and “southern Europe”, as defined above (Fig. ). The distribution of geopotential height over the “Central Europe” region exhibits a higher occurrence of positive anomalies, ranging from +1σ to +2σ, during January and February, with peak frequencies of 141 % and 88 %, respectively, relative to the positive years. Consistently, the precipitation distributions over “Scandinavia” and “Southern Europe” show distinct changes toward wetter conditions (from +1σ to +2σ, with peak occurrences of 78 %) and drier conditions (from +1σ to +2σ, with peak occurrences of 100 %), respectively, compared to the positive years. In contrast, the temperature index over the “Baltic region” shows no notable difference from that of the entire season (Fig. b). However, positive anomalies exceeding +2σ occur more frequently in December during the negative years in the “Baltic region” than in the positive years.

The composite analysis presented in this study suggests more frequent blocking activity in the northeast Atlantic/western Europe during 1602–2003 period (Fig. a). This pattern suggests a northward shift in the jet stream in this region during low δ18O periods as a direct consequence of the blocking. Consistent with our results, reported that an equatorward shift of the North Atlantic jet stream is associated with lower blocking frequency over Europe and more frequent floods in western Europe and the Alps. Indeed, the correlation between the index of the North Atlantic jet latitude and the NGT stacked δ18O time series, although small (r=-0.15), is significant at 95 % confidence level, during this period.

The analyses presented demonstrate that negative years in the NGT stack δ18O series reflect a large-scale atmospheric circulation pattern that favors the recurrence of extreme weather events, supporting the use of δ18O as a proxy for such events. Furthermore, this relationship can be further linked to external forcings, such as solar variability and volcanic eruptions, and their influence the occurrence of extremes, possibly through similar mechanisms that favor atmospheric blocking over Europe, which are also recorded in the NGT stacked δ18O series.

Previous studies e.g. suggest that, during the observational period, low solar winters are associated with more (less) frequent blocking over western Europe (Greenland), a pattern that resembles our NGT blocking pattern (Fig. b). Indeed the correlation between our NGT δ18O record and total solar irradiance over the period 1610–2003, although low, is significantly positive (r=+0.15%, 95 % level). However, the blocking pattern associated with low solar forcing during the observational period , extends more northeastward leading to extreme cold conditions over most part of Europe. This is not inconsistent with the NGT temperature pattern in this study (Fig. b) showing mild conditions over much part of northwestern Europe for low NGT years during 1602–2003.

Volcanic eruptions from tropical and extratropical regions are related to a northward shift in the Atlantic jet stream as well as to positive NAO e.g.. However, the atmospheric response to particular volcanic events presents specific characteristics . For instance, the Krakatoa eruption in 1883 caused changes in atmospheric circulation during the following winter, favoring a high-pressure system over Europe . In 1884, a minimum is observed in the NGT stacked δ18O series. By averaging the effects of twelve major volcanic eruptions, identified a cooling signal over northern Africa and southern Europe, and a warming over northern Europe. The associated mean sea-level pressure anomalies show a pattern consistent with the geopotential height anomalies observed during negative years in the NGT stacked δ18O series. Furthermore, the eruptions of El Chichón in 1982 and Pinatubo in 1991 are also associated with warming across northern Europe and very low values of δ18O in the NGT stacked series. Although the warming observed in the Northern Hemisphere, particularly over Europe and Asia, occurred after the Pinatubo eruption, it has been attributed to internal variability rather than to a causal effect of the eruption .