3.1 The SAM-PSA1 in-phase relationship and RICE δ18O
Strong positive δ18O correlations are found with the ASL (rmax = 0.52, p < 0.01, 1979–2011; Figs. 3b–c), with high Z500 being associated with high δ18O and low Z500 with low δ18O. Indicating that, atmospheric circulation in the ABS/Ross Sea Z500 (namely the ASL), is the driving force governing the extended-winter RICE isotopic signal (Fig. 3c). This is also reflected in the SAT and SIC correlation patterns (Figs. 3d, e). Positive δ18O anomalies are associated with both positive SAT and negative SIC anomalies in the eastern Ross Sea (Figs. 3d, e). This pattern corresponds to the western flank of the anticyclonic circulation, where strong poleward meridional winds prevail. Resulting in airmasses that are isotopically enriched by the nearby open ocean north of the sea ice edge. Changes in sea ice also modify the sensible heat flux (the conductive heat flux from the ocean to the atmosphere) (Noone and Simmonds 2004). For example, when sea ice recedes (mechanically by wind and from melting associated with warm air mass intrusions), ice is replaced by open ocean leading to an increase in sensible heat flux followed by higher SAT.
As both SAM and PSA1 project onto the ABS Z500 field (Turner et al. 2013), we next examine whether SAM and PSA1’s phase relationship is preserved in δ18O. The difference between SAM+/PSA1− and SAM−/PSA1+ years (marked by asterisk and circle in Fig. 4a, respectively) for Z500, SAT, and SIC is shown in Figure 4b–d. There is a striking similarity between the δ18O correlation panels (Figs. 3c–e) and the in-phase composite differences (Figs. 4b–d). These results show that δ18O captures the SAM PSA1 in-phase relationship during the satellite era (Figs. 3c and 4b) as well as that the associated δ18O SAT and SIC patterns (see the Figs. 3d, 4c and 3e, 4d pattern pairs). Here positive δ18O anomalies are associated with SAM−/PSA1+ and negative δ18O anomalies are associated with SAM+/PSA1− (Figs. 3a and 4a). The 2010 SAM+/PSA1− event was an exception; it was associated with a positive δ18O anomaly value.
In addition to the δ18O-Z500 correlation, we depict regions where the SAM and PSA1 overlap in Figure 3c. The significant areas (encircled by black dashed contours in Figs. 2a and 2b) are used to find regions of overlap. Here, the overlaid contours enclose regions where both patterns are active (significant at the p < 0.05 level). The contours in Figure 3c depict regions impacted by both the SAM and PSA1 modes, with the magenta contours showing regions where both modes display the same sign when the modes are in-phase (or opposing signs when out of phase). Note that SAM and PSA1 are in-phase [SAM−/PSA1+ (El Niño) or SAM+/PSA1− (La Niña)] when they are associated with the same sign of the Z500 anomaly over the Amundsen Sea (Figs. 2a and 2b). This is consistent with Z500 PC1 (SAM) being negatively correlated and Z500 PC2 (PSA1) being positively correlated with δ18O (Table 1). Positive geopotential height anomalies over the Amundsen Sea/the eastern Ross Sea region [weak ASL and positive δ18O anomaly (Emanuelsson et al. 2018)] tend to be associated with SAM’s negative phase and/or the PSA1 patterns positive phase (Fig. 3c). The region of SAM PSA1 overlap in the ABS/Ross Sea largely corresponds to the ASL region, as defined by Hosking et al. (2013) (60°–75°S, 170°E–70°W). Further evidence that the characteristic Rossby wave teleconnection pattern in the Pacific and Atlantic sectors is strengthened when SAM and PSA1 are in-phase (Fig. 3c).
Regions of significant δ18O-Z500 correlations are characterized by either: (1) a region where both EOF patterns are significant and strengthen the correlation by showing the same anomaly sign when in-phase., e.g., over the Ross and ABS (magenta contours, Fig. 3c); or (2) only one pattern is active (non-stippled and non-contoured areas), e.g., SAM over East Antarctica.
Furthermore, we have demonstrated that the EOF patterns SAM and PSA1 (that, are independent of RICE δ18O) can explain the δ18O-Z500 correlation pattern (see the overlain contours in Figure 3c). Suggesting that the leading EOF patterns are real distinct large-scale dynamical modes and not degenerates, as some studies have cautioned (Dommenget and Latif 2002; Monahan et al. 2009).
The extended-winter RICE δ18O-SIC correlation pattern shows high significance in the Amundsen/eastern Ross Seas region (Fig. 3e). The ABS SIC physically affects δ18O, by dictating the distance to the open ocean. The positive correlation with SIC in the Bellingshausen Sea and the Weddell Sea reflects the Antarctic Dipole Pattern (ADP) (Yuan and Martinson 2000; Renwick 2002; Yuan 2004; Turner et al. 2009; Thomas and Abram 2016), caused by the ASL creating opposing SIC anomalies between the Pacific and Atlantic sectors.
3.2 Decadal-scale variability
Next, we examine the temporal variability of the relationship between Z500 and SIC with δ18O for the extended-winter period. Figure 5a shows the 11-year running correlation between δ18O and the Z500 PCs. The δ18O-SAM correlation is consistently negative until the end of the correlation interval (middle year 2000) when the correlation strength is dramatically reduced. The δ18O-PSA1 correlation is stronger than the SAM correlation. The moving δ18O-PSA1 correlation is significant at the p<0.05 level until ~1999 (middle year 2005).
The effective sample size (neff) is larger than the sample size for the δ18O-SAM correlation (Table 1a). This occurs when the lag-one autocorrelation coefficient of one time series is negative, which is indicative of a blue noise process. We investigate this further by calculating moving average neff values for the δ18O correlation with SAM and PSA1 (Figs. S3a, b). The 11-year mowing window δ18O-SAM neff values become larger than the sample size around the year 2000 (Fig. S3a). This appears to be related to an increased magnitude towards higher frequencies in the SAM PC (Figs. S3c). A similar but subdued increase in neff towards more recent years is also apparent for the δ18O PSA1 correlation (Fig. S3a). Higher frequency blue noise can occur in δ18O records at shallow depths, where the signal has not yet been attenuated by diffusion (Fisher 1985; Fisher et al. 1996). This is ruled out for RI δ18O because it shows low magnitudes of higher-frequency variability close to the snow surface (Fig. S3b). We conclude that the 1990–1991 δ18O increase in magnitude during the extended-winter period is driven by the ASL (Fig. 3b).
The loss in correlation corresponds to an apparent breakdown of the in-phase relationship with δ18O (Figs. 3a and 4a). In 2010 a SAM+/PSA− (La Niña) event is associated with a positive δ18O anomaly. Additionally, PSA1 is neutral in 2011, yet δ18O displays the most positive anomaly during the satellite era. Thus, the running correlation between δ18O and SAM and between δ18O and PSA1 is lost. As the PSA1 teleconnection is related to ENSO, the running correlation with Niño-4 SST is also lost (Figs. 5d). Furthermore, as SAM and PSA1 are the dominant drivers of ADP SIC variability the δ18O running correlation with ADP is lost. Note, the δ18O moving correlation with Niño-4, PSA1, and ADP curves are almost identical. The 11-year mowing window δ18O-Niño-4 and δ18O-ADP neff values are smaller than the sample size for the (middle-year) 1995–2004 period (Fig. S3f), indicative of red noise.
The scatter plots in Figure 5 show the linear regression between δ18O and the two leading Z500 PCs (Figs. 5b and 5c) and with ADP and central tropical Pacific SSTs (Niño-4; Figs. 5e and 5f). If years 2010 and 2011 are excluded, the δ18O correlations with SAM and PSA1 remain strong (Figs. 5b, c). Suggesting conditions during 2010–2011 are causing a reduction in correlation strength. The loss of correlation is also evident in the spatial correlation patterns, which bear an even clearer resemblance with the Z500, SAT, and SIC in-phase composites when the correlation interval is limited to 1979–2009 (Figs. S4 and 4b–d).
The change in intercept (but not slope) implies that SAM and PSA1 still have the same effect on δ¹⁸O during the extended winter of 2010 and 2011 (red asterixis Figs. 5b, c), but a polynya and/or decadal variability (IPO, change in teleconnection) may have caused the offset in the intercept (red dashed lines in Figs. 5b, c). Polynyas and open ocean adjacent to RICE can provide a source of local maritime air, which can enrich the isotopic signal. This will be a topic for future research.