In the present study, we have used a new coastal ice core record (IND-33/B8; hereafter IND33, 101.4 m long) retrieved using an electromechanical drill during 2013–14 from the central DML region (location: 71° 30′ 36″ S and 10° 09′ 36″; elevation: 1,470 m) as shown in (Figure 1). To check the suitability of the drilling site, a systematic ground penetrating radar survey was carried out over the region, which revealed an undisturbed ice stratigraphy. Ice cores were sealed in high-density polyethylene core bags, packed in expanded polypropylene boxes, and kept at −20°C until they arrived at the National Centre for Polar and Ocean Research in Goa, India, stored at the Ice Core Laboratory at −20°C. The ice cores were processed inside the sample processing room maintained at −15°C. Prior to subsampling, the ice cores were cleaned properly using ceramic knives and prewashed microtome blades. The ice core was subsampled at 5 cm resolution for various chemical and stable isotopic analyses to reconstruct climate records at seasonal resolution.

The stable isotopic ratios of melted ice core samples were measured using an “Isoprime” isotope ratio mass spectrometer (GV Instruments, United Kingdom). The analytical details are already discussed elsewhere (Ejaz et al., 2021). Both major ions (Na⁺ and Cl⁻) and δ¹⁸O ice core records were used to establish a robust and reliable chronology. A first-order chronology was established using manual counting of summer maxima in δ¹⁸O values, which was further improved by the tuning of the seasonal signal N a + and C l − and the tie points based on known volcanic events identified in nssSO₄ ²⁻ (nonsea-salt sulfate) anomalies and Tritium peak of nuclear bomb testing. To reduce the uncertainty related to the method of manual counting, a MATLAB-based StratiCounter program (Winstrup et al., 2012) was used for automated annual layer counting. Based on such robust constraints, the chronology for the IND33 core was established for the past approximately 200 years (1809–2013 CE) with an uncertainty of ±1 year. The details of the establishment of chronology are discussed elsewhere (Ejaz et al., 2021).

In order to have a regional perspective on the temperature variability, in addition to the IND33 core, we have used three ice core δ¹⁸O records from the DML region (DMDL05, DML07, and DML17); (Graf et al., 2002) (Figure 1). These cores are situated 190–550 km away from the coast at elevations varying from 1,470 to 3,160 m; thereby, representing a broad range of topographic and snow accumulation patterns under a similar climatic regime (Table 1). The selected ice cores have accumulation rates varying from 46–62 kg m⁻²a⁻¹ (DML ice cores) to 288 kg m⁻² a⁻¹ (IND33 ice core) (Table 1). Earlier studies have suggested that ice cores having accumulation rates above 40 kg m⁻² a⁻¹ would represent seasonal cyclicity at shallow depths (Landais et al., 2017). This suggests that annual signals can be extracted from ice cores selected for the present study.

In this study, ERA5 reanalysis of surface air temperature data available from 1979 at a resolution of 0.5° × 0.5° latitude-longitude was used to investigate the role of temperature in controlling δ¹⁸O variability in these ice core records and to reconstruct the temperature beyond 1979 as discussed in Section 3.1 (Hersbach et al., 2020). To examine the ENSO influences, we have used long-term Southern Oscillation Index (SOI) data available from NOAA (https://www.ncdc.noaa.gov/teleconnections/enso/indicators/soi/) (Ropelewski and Jones, 1987).

We also employ a long-term IPO index available online (https://psl.noaa.gov/data/timeseries/IPOTPI/tpi.timeseries.ersstv5.data) (Henley et al., 2015). The Marshall SAM index (Marshall, 2003) is combined here with the reanalysis 20CRV2c SAM index (Gong and Wang, 1999) to have a longer SAM record comparable to our target time window (1809–2019 CE). The DML05, DML07, and DML17 ice core data used in this study are available at the PAGES 2k (http://pastglobalchanges.org/science/wg/2k-network/data) repository.

The temporal variations of δ¹⁸O in the IND33 ice core are shown in (Figure 2A). This new ice core from the coastal DML region offers a high-resolution (seasonal–annual scale) record of δ¹⁸O; higher values are observed in summer and lower values in winter. The δ¹⁸O values of the IND33 ice core range from 27.6‰ to 32.70‰ with a mean value of 30.47 ± 0.89‰. In order to have a better understanding of regional climate variability on a longer time scale, we have selected three ice core records (DML05, DML07, and DML17) from the inland region of DML which are in close proximity to the IND33 record (Figure 1). The annual δ¹⁸O variability in DML ice cores is shown in (Figures 2B–D). Although these ice cores show large variability, they do not show a discernible common trend. The three inland ice cores from DML are close to each other within a few 100-km distances and show mean δ¹⁸O values ranging from −45‰ to −46.7‰ (Graf et al., 2002). The ice cores in this study were selected based on the following criteria: 1) ice core records should have a common time interval of at least past two centuries with a time resolution of at least annual or better, 2) the top part of the record should have a reasonable number of overlapping years with the available instrumental/reanalysis data for calibration with confidence, and 3) the region from where the cores were selected should come under the significant influence of ENSO and SAM.

Several studies have demonstrated that δ¹⁸O records in ice cores are influenced by several factors such as temperature (Petit et al., 1999; Stenni et al., 2017), sea ice variability (Ejaz et al., 2021), precipitation (Fujita and Abe, 2006), and postdepositional changes related to exchange of water vapor between surface snow and atmosphere through diffusion (Landais et al., 2017). As a result, the δ¹⁸O records represent a composite signal of various meteorological parameters and/or postdepositional processes and hence cannot be directly used to decipher climate variability. A previous study from this region has observed that the selected δ¹⁸O ice core records are only weakly correlated with ERA5 annual surface air temperature (Ejaz et al., 2021). Therefore, principal component analysis (PCA) was performed on these four δ¹⁸O records within their common time window (Supplementary Table S2) to deconvolute different climate signals embedded in the δ¹⁸O records and finally to extract a common temperature signal, which is representative of a large region of DML.

Temperature is the dominant factor among the meteorological parameters that explain maximum variability in δ¹⁸O ice core records from Antarctica (Masson-Delmotte et al., 2008; Rahaman et al., 2019). The first component (PC1), derived from PCA explains maximum variability (∼32%) whereas the second principal component PC2 explains ∼27% variability of the δ¹⁸O record (Supplementary Table S3). The other principal components such as PC3 and PC4 are not significant (eigenvalues <1), and therefore, they were not considered for further investigation. In a previous study, Ejaz et al. (2021) demonstrated that PC2 derived from PCA of these records is strongly related to annual sea ice variations in the western Indian Ocean sector of Antarctica. Based on the linear relationship, they have reconstructed past sea ice variability and trend in the western Indian ocean sector of Antarctica during the past two centuries. In this study, we have carried out a more rigorous investigation based on Pearson correlation analysis between PC1 and annual/seasonal surface air temperature shown in Supplementary Figure S1. We have observed that PC1 is weakly related to ERA5 annual surface air temperature averaged across the grid box over the DML region (71.5°S to 75.5°S; 11.0°E to 5°W), from 1979 to 1993 CE (r = 0.33, p > 0.05) (Figure S1 A). Correlation between PC1 and temperature is not found to be significant during summer (DJF) and autumn (MAM) (Figure S1 B, C). However, the ERA5 surface air temperature shows a strong correlation with PC1 (r = 0.73, p < 0.05, lag = 1 year) during the late spring to summer (NDJ) season for the period 1979–1993 CE (Figure S1 F). It is interesting to note that such a strong temperature correlation is not observed during winter (JJA) and spring (SON) seasons (Figure S1 D), possibly because of strong cyclonic activity and blowing wind (Figure S2 B) causing reworking and redistribution of accumulated snow (Mott et al., 2018; Mills et al., 2019). It is clearly shown in our ERA5 seasonal temperature PC1 correlation analysis that significant correlation is mainly observed during the season of reduced surface wind speed, whereas insignificant correlation is observed during the season of high surface wind speed (winter), because of strong reworking of freshly deposited snow in cyclonic activity. The maximum correlation is observed during the late spring to summer (NDJ) when accumulation was reasonably higher (Figure S2 A) and wind speed was lower to moderate (Figure S2 B) and thereby, suitable for better preservation of climate signals in the deposited snow/ice sheet. Furthermore, we have used the spatial correlation between PC1 and surface air temperature during the NDJ season (Figure 3A). The spatial correlation is found to be significant over the DML region during 1979–1993 CE (Figure 3A). Therefore, for the reconstruction of the DML temperature, we have used the regression equation derived from the correlation between the PC1 and NDJ temperature anomaly.

We have reconstructed the temperature anomaly record (1809–1993 CE) of the DML region using the linear regression equation derived from PC1 and the ERA5 NDJ temperature correlation (Figure 3B). Furthermore, to determine the recent trend in the temperature record, we have combined the ERA5 surface air temperature anomaly for the DML region (1994–2019 CE) with our reconstructed ice core temperature anomaly record. The combined record of reconstructed and reanalysis ERA5 temperature anomaly records for the past two centuries (1809–2019 CE) enable us to determine the long-term trend in the DML temperature anomaly (Figure 4) and the influence of climate modes on temperature variability. To discern the significant long-term trend, a Mann–Kendall test (Kendall, 1975) was conducted and determined the breaking points of the trends in the temperature record. Therefore, two significant trends have been identified in the DML temperature record (Figure 4). The reconstructed temperature record revealed a significant cooling trend during 1809–1907 CE (n = 98, p < 0.05) with a rate of −0.164 ± 0.045°C decade⁻¹, followed by a significant warming trend with a rate of +0.452 ± 0.056°C decade⁻¹ during 1942–2019 CE (n = 77, p < 0.05). A recent study has shown that during the past two decades (1998–2016 CE), a more rapid increase in the temperature (+1.1 ± 0.7°C decade⁻¹) is observed in western DML (Medley et al., 2018). Our study is also consistent with such a rapid warming trend in recent decades. Therefore, it is important to identify the potential drivers and their role in controlling the DML temperature trends and variability in the past two centuries.

Tropical Pacific oscillations such as ENSO, IPO, and their teleconnections to Antarctica in space and time, particularly in the recent time of global warming, are not well understood. To identify dominant periodicities associated with the climate modes and their temporal evolutions during the past two centuries, we first performed a spectral analysis (Schulz and Mudelsee, 2002) of the temperature anomaly record (Figure 5A) (1809–2019 CE). The spectral analysis revealed significant (95% χ² level) periodicities of 5.2 and 15 years (Figure 5B). Furthermore, to understand the influence of ENSO and IPO on DML temperature during the past two centuries and their temporal evolution, we performed a continuous wavelet analysis of the temperature record (Figure 5C). We used the Morlet wavelet (Torrence and Compo, 1998) to analyze nonstationary climate signals in the temperature record, which enable us to identify significant (at 95% significance level) periodicities and their links to ENSO and IPO. This highlights the maximum power of lower frequencies in a decadal band (12–18 years) during 1830–1870 CE and 1910–1950 CE, whereas after the 1930s, high-frequency ENSO signals within the (2–8 years) band become more prominent. The variance at the ENSO band shows an increasing trend in the last century (Figure 5D), which is consistent with the earlier report from East Antarctica (Rahaman et al., 2019).

A cross-wavelet analysis (Grinsted et al., 2004) between the DML temperature anomaly record and SOI reveals the highest common power at the 2–8 years band (Figure 6A). This clearly shows a systematic increase from lower to higher frequencies at the ENSO band in the 20th century. This is consistent with the earlier reports of increasing ENSO activity and its influence on Antarctic temperature (Schmidt et al., 2012; Otto-Bliesner et al., 2016; Yeh et al., 2018; Rahaman et al., 2019). Cross-wavelet analysis between DML temperature and IPO shows a strong antiphase relation during 1920–1950 CE (Figure 6B). In order to deconvolute the signals of ENSO and IPO, bandpass filters were applied at 2–8 years and 12–18 years bands (Figures 6C,D). The bandpass filtered temperature record at 12–18 years band shows antiphase relation with that of IPO index throughout the record (except for 1890–1920 and 1960–1980 CE). This antiphase relation between temperature and IPO was found to be more prominent from 1920 to 1950.

In order to understand the role of ENSO in modulating the temperature over the DML region, it is important to investigate the spatial variability of temperature during the El Niño (warm) and La Niña (cool) phases. Therefore, we have used the ERA5 2m temperature anomaly for strong El Niño (1987, 1997, 2002, and 2016) and La Niña phases (1989, 1999, 2000, and 2010) to highlight the temperature variability during ENSO events. During the El Niño events, the spatial anomaly plots show an increase in surface temperature over the DML region (Supplementary Figures S3A–D). By contrast, it shows reduced surface temperature over the DML region during the La Niña events (Supplementary Figures S3E–H). The spatial composite anomaly record of mean surface ERA5 temperature during El Niño years also shows a strong positive anomaly over the DML region, indicating a significant increase in temperature during these events (Figure 7A). In a similar manner, during the La Niña years, the composite temperature anomaly record shows significant cooling across the DML region (Figure 7B).

The teleconnections between the tropical Pacific and Antarctica are determined using the well-established dynamics of Rossby wave trains (Hoskins and Ambrizzi, 1993; Lachlan-Cope and Connolley, 2006). The atmospheric signature of this wave train is reflected in the Pacific South American pattern during ENSO events (Mo and Higgins, 1998; Li et al., 2021). Rossby wave trains are caused by convective heating of the tropical atmosphere (Fogt et al., 2011; Ding et al., 2012; Li et al., 2021) and propagate with alternating highs and lows in the atmosphere, curved poleward and eastward toward Antarctica (Clem and Fogt, 2013). Once reaching Antarctica, they interact with Amundsen sea low and further combine into interannual variability, which is influenced by SAM (Clem et al., 2016). Nonetheless, it is also important to understand the heat transfer during the El Niño and La Niña years to better understand the mechanism behind the temperature variability in the DML region. Toward this, we have used the vertical integral of the northward heat flux during El Niño and La Niña years. The anomaly plots of vertical heat flux (W m⁻²) show strong positive anomalies over the DML region during the La Niña years (Figure 8A), suggesting increased northward transport of heat (Isaacs et al., 2021). On the contrary, during the El Niño years, we have observed a strong negative heat flux over the DML region (Figure 8B), suggesting a strong tropical heat transfer toward the DML region during El Niño events. This highlights the role of ENSO and its atmospheric teleconnections in modulating the temperature variability in the DML region of East Antarctica.

SAM is the primary climate mode in the Southern Hemisphere, and hence, it is important to understand how SAM influenced DML temperature variability. In addition to the Rossby wave trains, ENSO signals are transmitted to the Antarctic region through a coupled interaction with SAM (Clem and Fogt, 2013). It is, therefore, necessary to understand the SAM–ENSO coupled interaction and its possible impact on surface air temperatures over the DML region. SAM is derived as the leading mode of empirical orthogonal function (EOF) of mean sea level pressure (MSLP) between 20°S and 75°S (Marshall et al., 2012; Lim and Hendon, 2015). Therefore, to understand the role of SAM–ENSO coupled interaction and its influence on surface air temperature of the DML region, we have used MSLP during the positive and negative SAM (pSAM and nSAM) years and correlated with ERA5 surface air temperature during the El Niño and La Niña years. During the pSAM and El Niño years, MSLP is inversely related to ERA5 surface air temperatures, whereas a strong positive correlation is observed during the La Niña years (Figures 9A,B). Furthermore, during the nSAM and El Niño years, MSLP is strongly correlated with the surface air temperature, whereas it is inversely related during the La Niña years (Figures 9A,B). In the above case, we have observed two modes of SAM–ENSO interactions: 1) pSAM/La Niña and nSAM/El Niño are in-phase (Figures 9B,C), and 2) pSAM/El Niño and nSAM/La Niña are antiphase (Figures 9A–D). There is a possibility of stronger teleconnections between ENSO and the Southern Pacific, based on the above in-phase pattern observed in case 1), which is found to be consistent with earlier studies (Fogt and Bromwich, 2006; Fogt et al., 2011). By contrast, the teleconnections became significantly weaker in case 2) when there was an antiphase relation (Fogt et al., 2011). In previous studies, it was observed that periodic break out of anticyclonic waves occurs more frequently during the La Niña phase, generating high zonal wind anomalies that propagate the low-pressure anomalies poleward (60°S) during the pSAM years (Clem and Fogt, 2013). Whereas, reverse conditions are often prevalent during El Niño events, which lead to positive pressure anomalies over the higher latitude and nSAM phases (Clem and Fogt, 2013). Therefore, our study supports the additional role of coupled interaction of SAM–ENSO in regulating the surface air temperature over the DML region.

The Marshall SAM index (Marshall, 2003) regressed with ERA5 reanalysis of annual surface air temperature for 1957–2019 CE highlights a positive correlation across the western and central DML (Figure 10A). To understand the influence of the SAM–temperature relationship on a longer time scale, we have combined the Marshall SAM index with the reanalysis 20CRV2c SAM index for an extended SAM record and compared it with the DML temperature record (Figure 10B). It is noteworthy to observe that this extended SAM record shows its highest values and increasing trend in recent decades, similar to the temperature anomaly record (Figure 10C). This indicates that the warming trend since the 1940s is concomitant with the shifting of SAM to a positive phase. Such a positive shift could be attributed to the warming trend in the eastern DML for the last 7 decades. Furthermore, a wavelet coherency between SAM−SOI records (1851–2019 CE) shows significant antiphase relation between SAM at the decadal band since ∼1940, which exactly coincides with the warming trend in the DML region (Figure 10D). Compared to this, significant in-phase relation at 12–18 years band during 1880–1920 CE coincides with the cooling trend (Figure 10E). Therefore, we suggest that the SAM−ENSO phase relation influences the DML surface air temperature variability at annual to multidecadal scales.