2.1 Temperature records

The length of observational records and locations of meteorological stations are crucial for a robust correlation between ice cores and meteorological observations. The Pituffik station is the only observation station in the northwestern part of Greenland (NW Greenland) and the closest one to the NEEM site (Fig. 1; Cappelen, 2017). Although the temperature record only covers the period back to 1948, the Pituffik station is the best source of information on the weather and climate in NW Greenland. As the ice core data span the last 150 years, we also test our δ¹⁸O record against longer-term temperature observations from the southwestern part of Greenland (SW Greenland). The SW Greenland temperature record is a merged temperature data set based on 13 observational records along the southwestern Greenland coastal area spanning the period 1784–2005 (Fig. 1; Vinther et al., 2006). This data set covers the complete period of seasonally resolved ice core isotope data from NEEM, facilitating an extended comparison period. The changes in NW Greenland coastal temperatures are regionally consistent around western coastal Greenland (Hanna et al., 2012; Wong et al., 2015). Therefore, some consistency of the SW Greenland temperature record with temperatures closer to NEEM can be expected.

2.2 Twenty Century Reanalysis data

The Twenty Century Reanalysis (20CR; Compo et al., 2011) data set is selected to investigate the relationship between NEEM isotope records and atmospheric circulation patterns and temperature. The 20CR data are a global atmospheric 2 by 2^∘ gridded climate model data set only assimilating surface observations of synoptic pressure and using sea surface temperature and sea ice concentration as boundary conditions (Compo et al., 2011). This data set provides estimates of global atmospheric variability spanning 1851 to 2012 at 6-hourly resolution. However, there are very few stations delivering pressure data over the Greenland area until 1922 after which the number of observation stations increases significantly (Compo et al., 2011). This leads to a less well-constrained reanalysis data set for Greenland for the period before 1930. To test the results for the early period, we divide the whole period into two sub-periods (1855–1930 and 1931–2004) and examine correlations with ice core data within these sub-periods. The aim is to investigate the influence of temperature and atmospheric circulation on NEEM seasonal δ¹⁸O signals.

2.3 Indices of climate patterns

Previous analyses have related the variability in the Greenland ice core stable water isotopes to changes in the atmospheric North Atlantic Oscillation (NAO; Barlow et al., 1993; Vinther et al., 2003) and the oceanic Atlantic Multidecadal Oscillation (AMO; Chylek et al., 2012). In this study, these two indices are extracted from the 20CR data set. We choose the principal component (PC)-based NAO (NAOPC) index, which optimally represents the NAO pattern spatially and temporally (Hurrell and Deser, 2009). To obtain the monthly NAOPC index, we performed the empirical orthogonal function (EOF) on monthly pressure anomalies over the Atlantic sector (20–80^∘ N, 90^∘ W–40^∘ E). The leading mode of EOF is used as the monthly NAOPC index. For the AMO index, we first average the sea surface temperature anomalies over the sector 0–60^∘ N, 0–80^∘ W and then subtract the average sea surface temperature anomalies between 60^∘ S–60^∘ N from it (Trenberth and Shea, 2006). By calculating indices from the 20CR data, both indices can cover the period 1855–2004.

2.4 Baffin Bay ice concentration

Steen-Larsen et al. (2011) suggested a strong link between annual sea ice cover in Baffin Bay and NEEM annual δ¹⁸O signal. To test this hypothesis, we selected the COBEsic sea ice data set to compare with the NEEM seasonal δ¹⁸O data. The COBEsic record (Hirahara et al., 2014) is a combination of monthly globally complete fields of sea ice concentration on a 1 by 1^∘ grid based on satellite observation starting after 1979 and historical data provided by Walsh and Chapman (2001). The mean Baffin Bay area sea ice concentration was calculated by averaging the values over the area between 65–80^∘ N and 80–50^∘ W (Tang et al., 2004).

3.1 The NEEM shallow ice core data

The annual δ¹⁸O data from four shallow NEEM ice cores (NEEM07S3; NEEM08S2; NEEM08S3; NEEM10S2) have been published by Masson-Delmotte et al. (2015). The shallow cores cover depths ranging from the surface down to between 52.6 and 85.3 m. A back-diffusion calculation following Johnsen et al. (2000) was applied to the δ¹⁸O records to restore the original variability and hence improve the identification of individual years. The annual dating of those records was performed by counting the seasonal cycles in δ¹⁸O and verified by identifying signals of volcanic eruptions in the electrical conductivity measurements (Masson-Delmotte et al., 2015). The four shallow cores share a common period from 1855 to 2004, which is the focus in this study.

3.2 Greenland seasonal δ¹⁸O data

The NEEM seasonal δ¹⁸O data are also compared with other seasonal records obtained from 13 sites in central and southern Greenland over the period 1778–1970 (Fig. 1, Vinther et al., 2010). There are no other seasonal δ¹⁸O records from northern Greenland. Most records originate from single ice cores while some are stacked records from multiple cores (GRIP, n=6; DYE3-71/79, n=2). The first principal component (PC1) of these ice core data is considered as representative of the seasonal δ¹⁸O signal of central and southern Greenland. Vinther et al. (2010) divided the Greenland seasonal δ¹⁸O data into summer and winter season corresponding to May–October and November–April, respectively.

5.1 NEEM records and signal-to-noise ratio

For low accumulation sites like NEEM, it is important to examine the mean signal-to-noise variance ratio (SNR) of four seasonal δ¹⁸O series. The SNR can be calculated as

(more details can be found in Vinther et al., 2006). Here, ${\stackrel{\mathrm{‾}}{V}}_{\mathrm{i}}$ is the mean variance of the records going into this analysis, N is the number of records and V_a is the variance of the average record.

The SNR for the δ¹⁸O data in NEEM cores is 0.64 for the winter and 1.28 for the summer. The winter δ¹⁸O is more strongly influenced by noise than the summer signal, possibly due to windier conditions that lead to a more disturbed signal by sastrugi formation and less snow accumulation than during summer. These two SNRs are in line with a previous study by Masson-Delmotte et al. (2015) that found a SNR of 1.3 for the annual NEEM δ¹⁸O. Note that the seasonal SNRs observed here are higher than the level obtained for six ice cores from the GRIP project (0.57 for winter and 0.89 for summer; Vinther et al., 2010). Therefore, we conclude that the set of these four ice cores is sufficient to extract a robust seasonal δ¹⁸O at NEEM.

5.2 Comparison with other seasonal Greenland ice core records

Figure 3Correlation coefficients between NEEM seasonal δ¹⁸O and Greenland seasonal δ¹⁸O records for the period 1855–1970 (a for summer and b for winter). The PC1 of seasonal central/southern Greenland δ¹⁸O records is shown within the black dashed rectangle. The ice divide is marked by dotted black line. The significant correlations at the 95 % confidence level are filled with a black dot (t test).

Download

Figure 3 presents the correlation between seasonal stacked NEEM δ¹⁸O and other seasonal ice cores in Greenland, including the Greenland δ¹⁸O PC1. All data are detrended before correlation. The NEEM summer δ¹⁸O data are significantly correlated with the summer Greenland ice core isotope data from locations in southern Greenland and to the west of the central ice divide (Fig. 3a; with correlation from 0.3 to 0.46). However, summer δ¹⁸O from cores located to the east of the central ice divide (Renland, Site E, G and A) do not correlate significantly with the NEEM summer δ¹⁸O data. This is in accordance with the fact that moisture pathways are different for snow accumulation to the east and west of the central ice divide (Vinther et al., 2010). Therefore, having ice core records from both the east and west side of the ice divide facilitates the identification of regional-scale atmospheric variability. The correlation between NEEM summer δ¹⁸O and the Greenland summer PC1 record is significant both in inter-annual (r=0.54) and 11-year smoothed data (r=0.67, Fig. 4a and c). The correlations of 11-year averaged data are tested using the “random-phase” method introduced by Ebisuzaki (1997). The correlations are consistent with the correlation between annual NEEM δ¹⁸O and Greenland δ¹⁸O PC1 found by Masson-Delmotte et al. (2015). NEEM winter δ¹⁸O shows no significant correlation with most winter Greenland δ¹⁸O records and weak negative correlation with three southern ice core records (DYE3-71/79, 18C, 20D; Fig. 3b). No correlations are observed for the comparison with Greenland winter δ¹⁸O PC1 at an inter-annual and decadal scale (Fig. 4b and d). The results indicate a rather different winter climatic fingerprint archived in northwestern Greenland, suggesting one needs to be careful when interpreting the NEEM winter δ¹⁸O records. Such poor correlations between NEEM winter δ¹⁸O and winter Greenland δ¹⁸O PC1 are obscured in the annual correlation with Greenland δ¹⁸O PC1 due to the dominance of summer accumulation (Masson-Delmotte et al., 2015).

6.1 Association with the temperature and atmospheric circulation

Figure 5a and b show the spatial correlation maps between NEEM seasonal δ¹⁸O and surface air temperature (SAT) retrieved from the 20CR data set. All data are detrended before correlation. NEEM summer δ¹⁸O is significantly positively correlated with May–October averaged SAT over all of Greenland, Baffin Bay and the open water to the east of Greenland. This significant correlation also occurs as far south as 35^∘ N in the North Atlantic, where a previous study suggests the possible moisture source for precipitation at NEEM (Steen-Larsen et al., 2011). For winter δ¹⁸O and November–April averaged SAT, no correlation is displayed over Greenland or nearby consistent with the results from observations. Winter δ¹⁸O correlates significantly with the SAT near 35^∘ N in the North Atlantic and the Canadian Archipelago. But the correlation coefficients are only up to 0.25. Due to less reliable data in the early stage of 20CR data, we also examine the correlations within two sub-intervals 1855–1930 and 1931–2004 (Fig. S3). The strong extended correlations between NEEM summer δ¹⁸O and May–October averaged SAT are consistent within two sub-intervals. For winter correlations, both show no correlations over Greenland or nearby. The correlations with the SAT data from the reanalysis data support the conclusion that summer δ¹⁸O from NEEM has a better correlation with temperature than winter δ¹⁸O.

Table 1The correlations of seasonal NEEM δ¹⁸O records with seasonal averaged different atmospheric circulation indices. The bold text is significant at the 95 % confidence level, and the text marked with underline is significant at the 90 % confidence level (t test).

Download Print Version | Download XLSX

Figure 4(a, b) The NEEM seasonal δ¹⁸O identified in this study. The dashed line and bold line show annual and 11-year averaged data, respectively. (c, d) The Greenland seasonal δ¹⁸O PC1 extracted from ice cores in central/southern Greenland (Vinther et al., 2010). The dashed line and bold line show annual and 11-year averaged data. (e) The NAO indices calculated from 20CR reanalysis data using principal component analysis. (f) The AMO indices calculated from 20CR reanalysis data based on the method by Trenberth and Shea (2006). (g) The averaged SIC over Baffin Bay extracted from COBEsic (Hirahara et al., 2014). The dashed line indicates the start year of satellite observation (1979). All red lines show summer, and blue lines show winter. All data are normalized and detrended.

Download

Figure 5The spatial correlation map between NEEM seasonal δ¹⁸O and SAT (a, b) and SLP (c, d) from 20CR reanalysis data for the period 1855–2004. The winter data are averaged for November–April and the summer data are averaged for May–October. The red solid lines indicate significant correlation at the 95 % confidence level (t test).

Download

Figure 6The spatial correlation map between NEEM seasonal δ¹⁸O and SIC over Baffin Bay for the period prior to satellite observations (a, c) and the satellite observation period (b, d). The winter data are averaged for November–April, and the summer data are averaged for the May–October. The red solid lines indicate significant correlation at the 95 % confidence level, and the green solid lines indicate this at the 90 % confidence level (t test).

Download

The NEEM seasonal δ¹⁸O is also compared with the sea level pressure (SLP) from 20CR data for the same time intervals as temperature (Fig. 5c, d and Fig. S4). There is no obvious NAO-like pattern (the seesaw structure over the North Atlantic Ocean) for the comparison between summer δ¹⁸O and May–October averaged SLP for the whole period. A NAO-like pattern emerges for the second sub-period (1931–2004) (Fig. S4b), but the northern node is limited suggesting a rather weak summer NAO footprint on δ¹⁸O at NEEM. There is a seesaw structure when correlating winter δ¹⁸O with November–April averaged SLP over the last 150 years (Fig. 5d) and within the sub-period 1855–1930 (Fig. S4c). However, the correlations with SLP are also rather weak for these periods. The absolute values of correlation coefficients are less than 0.33 both for 1855–1930 and for the whole period. Furthermore, it should be noted that there is an absence of the NAO-like pattern for the second 75-year period (Fig. S4d) when observations are generally more reliable due to the increased number of assembled observations around Greenland. Hence, care should be taken when interpreting inconsistent correlations in the sub-intervals. Another interesting feature in Fig. S4c and d is the consistent negative correlation between NEEM winter δ¹⁸O and November–April averaged SLP over North America and the Canadian Archipelago within the two sub-periods. This suggests that NEEM winter δ¹⁸O is more likely influenced by the pressure over North America and the Canadian Archipelago.

As the circulation indices are the simplified indicators of circulation patterns, we here further investigate the possible connections to AMO and NAO patterns with the seasonal NEEM data (Table 1). Both indices (Fig. 4e and f) and NEEM seasonal data are detrended before correlation. The summer δ¹⁸O signal correlates weakly with May–October averaged AMO (r=0.22) over 1855–2004. The correlations are also consistent within the two sub-intervals. There is no correlation between winter δ¹⁸O and November–April averaged AMO. Summer δ¹⁸O correlates weakly with May–October averaged NAO over the whole period and for 1931–2004, but no correlation is seen in the 1855–1930 period. It should be noted that for the whole period the summer correlation with NAO is significant, but no NAO-like pattern is seen in correlation with SLP for 1855–2004 (Fig. 5c). This may be attributed to the rather weak correlation with NAO which is only −0.16. NEEM winter δ¹⁸O has no correlation with the November–April averaged NAO in 1931–2004. Although the correlation map with SLP shows a NAO-like pattern for the 1855–1930 and the 1855–2004 period, the correlation coefficients with November–April averaged NAO indices are also rather weak (r=0.217 for 1855–1930 and r=0.191 for 1855–2004). Furthermore, it should be noted that even if there are correlations between seasonal NEEM δ¹⁸O and AMO and NAO, those circulation patterns can only explain less than 7 % of the variance of NEEM δ¹⁸O. We conclude that both patterns exert a weak influence on NEEM δ¹⁸O even though we calculate the correlation between seasonal circulation indices and seasonal NEEM δ¹⁸O. The weak correlations with NEEM δ¹⁸O are likely due to a larger distance from the Atlantic Ocean and a much lower snow accumulation at NEEM than in other ice cores in central and southern Greenland (Chylek et al., 2012; Steen-Larsen et al., 2011). The weak correlations can also explain why NEEM annual δ¹⁸O is highly correlated with annual Greenland δ¹⁸O PC1 but surprisingly weakly correlated with annual and winter NAO (Masson-Delmotte et al., 2015), which leave a strong footprint in most ice cores in central and southern Greenland (Vinther et al., 2003, 2010). The seasonal precipitation bias at NEEM, which is dominated by summer precipitation, skews the NEEM annual average δ¹⁸O towards summer. Therefore, the NEEM annual δ¹⁸O presents a summer-biased signal which has strong correlation with Greenland δ¹⁸O PC1. Furthermore, irrespective of the weaker winter signal in the annual δ¹⁸O, we also find that the isolated NEEM winter δ¹⁸O correlates poorly with winter NAO. This weak correlation between winter NEEM δ¹⁸O and winter NAO is in contrast with the finding of a strong winter NAO footprint in the winter δ¹⁸O records in central/southern Greenland. This is important to know when considering NEEM δ¹⁸O for use in circulation reconstructions using emerging reanalysis techniques (e.g., Hakim et al., 2016), where a strong seasonality can both be a caveat but also be exploited for climate reconstructions.

6.2 Comparison with sea ice concentration

In this section, NEEM seasonal δ¹⁸O is compared with the SIC record in Baffin Bay for 1901–2004 (Fig. 6). The period is further divided into the period prior to satellite observations (1901–1978) and the satellite observation period (1979–2004) for comparison. The year 1979 is the onset year of the satellite observations, which are regarded as the more reliable data source. Prior to the satellite period, the data are mainly calculated by the compilation of historical data (Walsh and Chapman, 2001). SIC data are linearly detrended before correlations (Fig. 4g). The NEEM winter δ¹⁸O correlates significantly with November–April averaged SIC extent over Baffin Bay in 1979–2004 with correlation coefficients of up to −0.62 (Fig. 6d). The correlation coefficient between NEEM winter δ¹⁸O and averaged SIC over the whole of Baffin Bay is −0.53. Prior to the satellite period, the correlation between NEEM winter δ¹⁸O and averaged SIC over Baffin Bay is also significant ($r=-\mathrm{0.27}$). Summer δ¹⁸O correlates well with May–October averaged SIC in 1979–2004 with correlation coefficients of up to −0.59 along the Greenland western coastal area (Fig. 6b). The correlation between NEEM summer δ¹⁸O and averaged SIC over Baffin Bay is also significant ($r=-\mathrm{0.46}$). However, in contrast with the good correlation in the late 20th century, there are limited significant correlations over the southern part of Baffin Bay for summer in the 1901–1978 period. There is no correlation between NEEM summer δ¹⁸O and averaged SIC over Baffin Bay ($r=-\mathrm{0.04}$) for this period. One possible explanation for the weaker correlations both for winter and summer in 1901–1978 may be due to less reliable historical data sources. Furthermore, the reconstructed summer SIC can be underestimated sometimes due to the lower concentration along the coastlines (Titchner and Rayner, 2014). The correlations with SIC in 1901–1978 are expected to be re-examined in the future, possibly leading to an improved sea ice extent data set (Titchner and Rayner, 2014). However, both winter and summer δ¹⁸O are still strongly negatively correlated with Baffin Bay ice extent for 1979–2004, sharing more than 22 % variance, which is in agreement with the relationship between the annual Baffin Bay sea ice anomaly and NEEM annual δ¹⁸O data as illustrated in Steen-Larsen et al. (2011).

A possible explanation for the sea ice effect on δ¹⁸O is that a reduced sea ice cover may amplify regional temperature changes and favor enhanced storminess and enhanced precipitation (Noël et al., 2014; Sime et al., 2013), thus bringing more local moisture. By contrast with the long-distance transport of moisture from the North Atlantic, the local source leads to less depleted δ¹⁸O in the clouds and thereby increases NEEM δ¹⁸O. However, this mechanism cannot explain the good correlation with winter δ¹⁸O as NEEM winter δ¹⁸O is poorly correlated with SAT over Baffin Bay (Fig. 5b). One hypothesis of this significant winter correlation with SIC is based on the wind over Baffin Bay. Changes in the wind strength/direction over Baffin Bay may modulate the moisture transport from Baffin Bay to the NEEM site. However, we find no correlations between NEEM winter δ¹⁸O and November–April averaged wind speed/direction at 850 and 200 mb altitude (jet stream) over 1901–1978 and 1979–2004 (not shown), which may exclude this hypothesis. Another possible hypothesis could be that, instead of the direct coupling of precipitation to local moisture sources at NEEM resulting in the high winter correlation, it is merely a climatic connection between sea ice extent and the clouds' temperature, thereby influencing the isotopic composition of the precipitation at NEEM (Steen-Larsen et al., 2011). Future work can focus on investigating the possible driving factors for this strong winter correlation, which is also consistently significant for the early 20th century. The strong correlations with SIC indicate the possible strong influence of sea ice changes on the variability in stable isotope ratios in northern Greenland. It was found that high δ¹⁸O values during the last inter-glacial period (the Eemian period) could not be achieved in interglacial simulations driven by orbital forcing alone (Sime et al., 2013). Sime et al. (2013) suggest that sea ice reduction may be the most likely cause of high interglacial δ¹⁸O in Greenland ice cores. This explanation is supported by our study showing that changes in sea surface temperatures (SSTs) and sea ice cover are indeed key to understanding the past changes in Greenland water isotopes.