Introduction

The world is experiencing rising temperatures in response to human-induced global warming. However, there are few reports on the impact of anthropogenic warming on the East Antarctic Ice Sheet (EAIS), which contains the vast majority of the world’s glacial ice. The surface air temperature (SAT) variability on the EAIS is closely related to the phase of the Southern Annular Mode (SAM)1. A positive phase of the SAM is characterized by strengthened westerlies over the Southern Ocean, which reduce meridional advection of warm air into the continent, resulting in cooler temperatures over East Antarctica2,3,4,5,6. The trend of SAM towards its positive phase in recent decades has led to a cooling effect. Therefore, modest temperature changes or cooling trends during the second half of the 20th century are explained by SAM-induced cooling masking anthropogenic warming in East Antarctica2,3,4,7. For this reason, the EAIS is considered less sensitive to anthropogenic warming than other Antarctic regions8.

However, despite the continued positive SAM trend, albeit with a slight weakening in the early 21st century, a widespread warming in the EAIS has been observed in the 21st century4,7,9,10,11,12. The atmospheric circulation pattern associated with the positive SAM phase has changed to a more asymmetric pattern9,11,13,14,15. This circulation change may have attenuated the SAM-induced cooling and made the EAIS more susceptible to anthropogenic warming than previously thought. Alternatively, warming in the EAIS may be attributed to the climate change that triggered the SAM pattern change, but the mechanism of the warming is unclear.

Assessing the climate change across the EAIS remains a major challenge due to the insufficient coverage of observing networks1. The majority of the staffed stations are situated along the coast, with only two stations (Amundsen-Scott and Vostok) with long-term temperature records located in the interior of the continent16. Despite the sparsity of observing stations in the EAIS, the observed SAT records since 1999 indicate widespread warming trends during the austral spring (September to November) and summer (December to February), particularly in the Indian Ocean sector7,9,12. This is consistent with the land surface temperature trend over Antarctica, as obtained from satellite observations17. The satellite-derived surface temperature trend map from 2000 to 2018 illustrates the most pronounced warming over the interior of the Indian Ocean sector of East Antarctica in the austral spring and summer. Similarly, most reanalysis datasets also exhibit a robust warming trend in the interior region, although the spatial pattern of the SAT trend differs among the different reanalysis data12,18. These suggest that the warming in the inland regions may be occurring at a faster rate than in coastal areas. Nevertheless, there are no reliable observations that can be used to assess the climate change in this region.

Here, we present a novel SAT record for the past three decades with the objective of determining the SAT trend over the interior of the EAIS and understanding the mechanisms involved. The monthly-means of the 30-year long time series were constructed from the integration of multiple automatic weather station (AWS) records in the eastern Dronning Maud Land (DML) from 1993 to 202219 and are utilized in this study. The observations represent historical records from three AWS in the DML: Mizuho(MIZ), Relay Station(RLS), and Dome Fuji (FUJ) (Fig. 1). Nevertheless, there are several concerns that need to be addressed in order to employ the AWS observations in the investigation of climate variability and change in Antarctica. The concerns include a warm-temperature bias in summer due to intense solar radiation and insufficient ventilation of the temperature sensor within its radiation shield, non-adjustment of the height changes of the temperature sensor due to snow accumulation, and systematic errors due to changes to the equipment at the station during the long-term observations. In this study, these issues were quantified by comparing them to reliable temperature measurements and subsequently eliminating them19. Then, a rigorous quality control of the data was conducted in accordance with the protocol established by the Scientific Committee on Antarctic Research (SCAR) Reference Antarctic Data for Environmental Research (READER) database16. These careful considerations allow us to use AWS observations to study climate change. More details on the quality controls of AWS data are described in our previous study19 and Methods.

Fig. 1: Surface air temperature (SAT) changes in the Indian Ocean sector in East Antarctica.
Fig. 1: Surface air temperature (SAT) changes in the Indian Ocean sector in East Antarctica.
Full size image

Map of Antarctica (a) with an enlarged Indian Ocean sector in East Antarctica showing the locations of stations referred to in the text (b). c–g The time series of annual mean SAT are shown for Dome Fuji (FUJ, (c)), Relay Station (RLS, (d)), Mizuho (MIZ, (e)), Syowa (SWA, (f)), and Mawson (MAW, (g)). h–l Same as (c–g), but for warm half-year (October–March) mean. m–q Same as (c–g), but for cold half-year (April–September) mean. For time-series at FUJ, RLS, and MIZ, open circles indicate the parts of the dataset with more than one-third missing observations, otherwise solid circles are used (see Methods and Supplementary Methods). Plots for SWA and MAW are obtained from READER database16. Solid linear lines indicate statistically significant (p < 0.05) linear trends, otherwise dashed lines are used. Annual and half-year mean temperature trends for 1993–2022 (1993–2021 for warm half-year), with the 95% confidence intervals, are presented at stations where a statistically significant linear trend was identified. The correlation coefficient between the FUJ and the data from each site for 1993–2022 (1993–2021 for warm half-year) is shown in parentheses.

Missing observations have been filled in with bias-corrected temperature data from the ERA5 reanalysis20. ERA5 is less influenced by the assimilation of AWS data and is capable of reasonably predicting the near-surface temperature at these three AWS, despite a slight warm bias (see Methods and Supplementary Methods). In this study, the warm biases of ERA5 are quantified by comparison with observations and then removed using a linear regression model. The uncertainty of the monthly bias-corrected ERA5 data is less than 1.4 °C (Supplementary Fig. S11). This is nearly equivalent to, or less than, the measurement error of AWS temperature observation. Furthermore, it is much smaller than the standard deviations of monthly temperatures. These make it suitable for completing missing data. Details of the reason for selecting ERA5, the techniques used to correct the temperature bias in ERA5 and to fill in the missing observations are described in Methods and Supplementary Methods.

The observed variability in the annual mean SAT at each station captures broader-scale changes due to the high correlation between the stations (Fig. 1). Furthermore, the significant correlation between the annual and seasonal mean temperatures at Relay Station, located in the middle of these three AWS, and those observed at inland stations in the EAIS and other coastal stations in the Indian Ocean sector of East Antarctica, suggests that this new record may offer insight into large-scale temperature changes (Supplementary Fig. S1).

Results

Temperature trends in the inland DML

The time series of annual mean SAT records from the stations in the eastern DML over the last 30 years, along with the trends at each station, are presented in Fig. 1c–g. The Mann-Kendall trend test is used to determine whether or not there is a statistically significant trend over the past 30-years. All inland stations experienced statistically significant (p < 0.05) warming trends of 0.45 ± 0.39 °C dec−1 at FUJ, 0.50 ± 0.38 °C dec−1 at RLS and 0.72 ± 0.34 °C dec−1 at MIZ, although coastal stations have not experienced significant warming. For MIZ, however, the larger warming trend may be artificial, due to the fact that nearly half of the observations from 1993 to 2022 are missing. The warming trends observed at FUJ and RLS from 1993 to 2022 are of a comparable magnitude to those observed at the other Plateau stations during the same period (0.56 ± 0.36 °C dec−1 at Amundsen-Scott and 0.53 ± 0.41 °C dec−1 at Vostok). During the four years of the record-high annual mean SATs observed at Amundsen-Scott station (2002, 2009, 2013 and 2018)11, maxima of SATs were also observed at FUJ and RLS (Fig. 1c, d).

Seasonally, RLS and FUJ experienced statistically significant (p < 0.05) warming in austral spring of 0.61 ± 0.36 °C dec−1 and 0.56 ± 0.37 °C dec−1 and in austral summer of 0.45 ± 0.42 °C dec−1 and 0.65 ± 0.38 °C dec−1, respectively. Neither station experienced statistically significant warming in austral autumn and winter. The concurrent slight warming of coastal stations on the EAIS7 lends further support that this observed warming extends beyond inland DML. Spring warming was observed at both stations from October, and a rapid warming is evident in the half-year mean SAT from October to March, which hereafter is referred to as the warm half-year (Fig. 1h–l). In contrast, there is no significant trend over the eastern DML during the rest of the year, from April to September, referred to hereafter as the cold half-year (Fig. 1m–q). The inter-annual variability of the cold half-year mean SAT was strongly influenced by the SAM variability (Supplementary Fig. S2).

It is noteworthy that the warming trend (statistically significant at p < 0.05) in the warm half-year (0.89 ± 0.30 °C dec−1 for FUJ, 0.80 ± 0.34 °C dec−1 for RLS) appears to be more pronounced than that in the spring and summer means and corresponds to approximately twice the annual mean trend. The robustness of the clear warming trend is confirmed by the fact that the infilling of missing observations with bias-corrected ERA5 data does not contribute to the trend (Supplementary Fig. S12). These suggest that the warmth of spring persists throughout the summer months and that the recent annual-mean warming trend observed in the interior DML is attributed to the rapid warming in the warm half-year.

Identification of key drivers of warming in the inland DML

Temperatures in the interior of Antarctica are sensitive to the north-south flow of air masses that are associated with either cyclonic or anticyclonic circulation11,21,22,23,24,25. It is well known that the variability of sea surface temperature (SST) in the tropics influences the atmospheric circulation in the Antarctic region11,18,26,27,28,29. The warming trend at Amundsen-Scott and the surrounding plateau region during the 21st century has been attributed to more cyclonic conditions in the Weddell Sea, which would enhance the warm northerly flow toward the Antarctic interior. The increased cyclonic activity has been linked to the Rossby wave teleconnection triggered by anomalous convective activity due to an increase in SST in the tropical Pacific11. Interestingly, the variability of warm half-year mean temperatures at RLS exhibits a significant correlation (R=0.66; p < 0.05) with the variability observed at Amundsen-Scott (Supplementary Fig. S1). A distinct warming trend of 0.75 ± 0.42 °C dec−1 during the warm half-year has also been observed at Amundsen-Scott. These results suggest that cyclonic anomalies in the Weddell Sea, attributed to tropical Pacific climate variability, may influence temperatures at RLS in a similar way to those at Amundsen-Scott. However, the spatial distribution correlated with RLS temperature variability shows that thermal advection from the Indian Ocean plays a dominant role on the temperatures at RLS (Supplementary Fig. S1). And given that SST changes in the tropical Pacific have little effect on RLS temperatures during the warm half-year (Supplementary Fig. S3a), the tropical forcing is not a major contributor to the warming in the DML. During the austral summer, the tropical-polar teleconnection is reduced due to the absence of subtropical westerlies, which serve as the primary pathway for the propagation of the Rossby waves from the tropics to the high latitudes28,30.

An alternative source of the recent warming in the DML is variability in the strength and position of the circumpolar westerlies31,32. While the westerlies are predominantly a zonally symmetric flow, they occasionally change into an asymmetric circulation, such as the quasi-stationary zonal wavenumber 3 pattern33. The asymmetric circulation is associated with a pattern in which northward and southward flows strengthen, and enhanced meridional heat transport between mid- and high latitudes has a strong impact on Antarctic temperatures11,13,14,15. During the austral summer, planetary wave activity in the southern hemisphere reaches minimum amplitude, and thus the annular flow is dominant6,14,34. However, an asymmetric pattern is triggered by a Rossby wave source adjacent to the westerlies in the mid-latitudes. During the austral summer, the positive SST anomalies over the subtropics lead to anomalous convection, and the vortex stretching associated with enhanced convection triggers the formation of a Rossby wave train35. This suggests that SST variability in the subtropics adjacent to the westerlies influences atmospheric circulation in the Antarctic climate. The hypothesis is supported by the fact that the spatial regression pattern between SAT at RLS and the SST anomaly field (Supplementary Fig. S3, Fig. 2a) reveals the most robust signal in the southern Indian Ocean. Here we focus on the mechanism by which changes in subtropical Indian Ocean SSTs have an impact on Antarctic temperatures.

Fig. 2: Sea surface temperature (SST) frontal variability in the southern Indian Ocean contributing to Relay Station (RLS) temperature changes over the past three decades.
Fig. 2: Sea surface temperature (SST) frontal variability in the southern Indian Ocean contributing to Relay Station (RLS) temperature changes over the past three decades.
Full size image

a, b Spatial regressions of the detrended warm half-year mean RLS surface air temperature (SAT) with SST anomaly (SSTA) (units:°C) (a) and SST meridional gradient (∂SST/∂y, units:°C 100 km−1) (b) for 1993–2021. c The trend of warm half-year mean SST for 1993–2021. Only data points falling within the 90% confidence interval are plotted. The dots indicate the frontal zone with SST meridional gradient above 0.6 °C 100 km−1. d The time series of the warm half-year mean anomalies of the RLS SAT (light blue bar) and of ∂SST/∂y in the subtropical frontal zone (STFZ) (30°-35°S; 55°-80°E) (red line). The trend of ∂SST/∂y in the STFZ is shown at the upper left. The correlation of warm half-year mean RLS SAT and ∂SST/∂y in the STFZ for 1993–2021 periods is given in parentheses at the upper left. For all maps, the STFZ used for comparison with the RLS SAT is indicated by the blue box. The red star and circles are the locations of RLS and staffed stations with long-term temperature records.

The regression map in Fig. 2a shows a distinct SST anomaly with a positive-negative-positive tripole pattern along the meridian from the subtropics to the subantarctic over the southern Indian Ocean. It is notable that the region between the positive SST anomaly in the subtropics and the negative SST anomaly in the mid-latitudes coincides with the Subtropical Frontal Zone (STFZ), which is composed of multiple SST fronts with northwest-southeast alignment36,37 (Fig. 2a). Similarly, the area between the negative SST anomaly in the mid-latitudes and the positive SST anomaly in the subantarctic corresponds to the Agulhas Front and Subantarctic Front, a sharp transition zone between the southern gyre boundary and the subantarctic region38. The dipole structure of subtropical SST warming and mid-latitude SST cooling indicates an increasing SST gradient, which corresponds to the strengthening of the STFZ (Fig. 2b). In contrast, the pattern of mid-latitude SST cooling and subantarctic SST warming decreases the SST gradient, so that the Agulhas Front and Subantarctic Front weaken. This suggests that an enhanced STFZ may be a key driver of the atmospheric circulation leading to Antarctic SAT variability. Indeed, the observed RLS SAT demonstrates a robust correlation (R = 0.76; p < 0.05) with the meridional SST gradients in the STFZ region (30°-35°S; 55°-80°E) from 1993 to 2021 (Fig. 2d). The SST gradients in the STFZ are strengthened as the area of accelerated warming in the southern Indian Ocean extends north of the STFZ region (Fig. 2c). The strengthening of the STFZ may play an important factor in the warming of the interior of the DLM in EAIS.

The variability of SST fronts exerts a significant influence on the atmospheric circulation through ocean-atmosphere interaction39,40,41,42. The 300 hPa geopotential height anomalies (Z300), which are associated with the variability of the STFZ, demonstrate a meridional dipole pattern from the mid-latitude Indian Ocean to the interior of East Antarctica (Fig. 3a). This circulation pattern is almost identical to the Z300 anomalies, which exhibited a correlation with the RLS temperature (Fig. 3c). The positive (anticyclonic) Z300 anomaly over East Antarctica would enhance the warm northerly flow toward the interior of the EAIS. The three AWS (MIZ, RLS, FUJ) are located on the western flank of the anticyclonic anomaly, where warm northerly flow dominates. The increased occurrence of this dipole pattern associated with the strengthening of the STFZ may lead to a warming of inland DLM.

Fig. 3: Atmospheric response to the variability of subtropical frontal zone (STFZ).
Fig. 3: Atmospheric response to the variability of subtropical frontal zone (STFZ).
Full size image

a–f Spatial regressions of the detrended warm half-year mean ∂SST/∂y in the STFZ region with geopotential height at 300 hPa (units: m) (a) and at 850 hPa (units: m) (b), air temperature gradient at 850 hPa (∂T/∂y, units: 10−6 °C m−1) (d), outgoing longwave radiation (OLR, units: W m−2) (e), and zonal wind speed at 300 hPa (units: m s−1) (f). c Same as (a), but for regressions with the detrended warm half-year mean Relay Station (RLS) temperature. The thick black dashed lines indicate the 90% confidence level of the regression slopes. The black dots indicate the sea surface temperature (SST) frontal zone with SST meridional gradient (∂SST/∂y) above 0.6 °C 100 km−1. The STFZ region is indicated by the blue box. The red star and circles are the locations of RLS and staffed stations with long-term temperature records.

In the mid-latitudes, the negative (cyclonic) Z300 anomaly is evident above the cold SST anomalies, and its vertical structure is barotropic (Fig. 3a, b). This atmospheric structure is consistent with the response to SST anomalies in the subtropical frontal zone (Fig. 4)43,44,45,46. The enhanced oceanic front leads to a strengthening of the baroclinity in the low-level atmosphere and thus to stronger synoptic-scale transient eddy activities, represented as a storm track. These transient eddies exert an influence on atmospheric circulation in the upper troposphere by transporting heat and momentum (vorticity flux). The formation of a cyclonic anomaly with barotropic structure on the polar side of the oceanic front is a common response to an increase in the transient eddy vorticity forcing42,43,47,48. These explanations are consistent with the fact that negative Z300 anomalies in the mid-latitude Indian Ocean correspond to both positive low-level atmospheric baroclinity and negative outgoing longwave radiation (OLR) anomalies, i.e., an increase in cloudiness, associated with enhanced STFZ (Fig. 3d, e). In addition, enhanced 300 hPa zonal westerlies along the STFZ are also collocated with negative OLR anomalies which extend zonally downwind due to the transport of vorticity fluxes into the mean westerly flow (Fig. 3f).

Fig. 4: Atmospheric response to variability of sea surface temperature (SST) front.
Fig. 4: Atmospheric response to variability of sea surface temperature (SST) front.
Full size image

The sensible heat flux contrast across the oceanic fronts maintains the surface air temperature difference and thus increases the atmospheric baroclinicity, including more active transient eddy activities that increase transient eddy vorticity forcing. The increased transient eddy vorticity forcing forms an equivalent barotropic geopotential low over the cooling SST, which is located south of the oceanic fronts. The atmospheric response to the transient eddy vorticity forcing reaches into the upper troposphere and influences the westerlies in the southern Indian Ocean.

These atmospheric responses to STFZ variability are remarkably an opposite pattern to the south of the Subantarctic Front, which anchors the westerly jet axis. The weakening of the Subantarctic Front leads to a decrease in low-level atmospheric baroclinity, which in turn reduces transient eddy activities in the westerly jet region. The weakened storm track activity (positive OLR anomalies) deposits decreased cyclonic vorticity over the south side of westerly jet axis, thereby driving anticyclonic anomalies that further weaken the westerlies in the upper atmosphere.

The feature of decreasing the pressure gradient between the mid-latitudes and Antarctica, along with a weaker and equatorward shift of the westerlies, is consistent with a negative SAM pattern that increases meridional warm air advection into the interior of Antarctica2,3,5,6. The three strongly negative SAM phases (SAM index  < −0.4) in the second half of the observation period (2008-2022) correspond to the years with strong SST gradients in the STFZ (2009, 2016, and 2019) (Fig. 2d). Nevertheless, there is no statistically significant correlation between STFZ variability and the SAM index from 1993 to 2021 (p = 0.11). In addition, there are some years when a positive SAM has been associated with a positive SAT anomaly at the three inland stations (Supplementary Fig. S2a–c). This means that the enhanced STFZ serves to mitigate the cooling of DML associated with a positive SAM by modifying the atmospheric circulation over the southern Indian Ocean. Consequently, a warming in the interior of the DML has been observed in the 21st century. This is consistent with the fact that the positive SAM pattern has been accompanied by anomalous high pressure over the Indian Ocean sector of East Antarctica since the beginning of the 21st century9.

Discussion

The results of this study demonstrate that the SST gradients in the STFZ exert a pivotal influence on the warm half-yearly SAT variability in the eastern DML interior region in EAIS. The STFZ has been strengthened by the rapid SST warming at the basin scale of the Indian Ocean during the 21st century. Interestingly, the observed warming of the Indian Ocean is attributed to the heat uptake by the Pacific Ocean during the negative phase of the Interdecadal Pacific Oscillation (IPO)49,50,51. During the negative IPO phase, as shown in Fig. 5, prolonged La Niña-like conditions strengthen the Pacific trade winds, which in turn strengthens the Indonesian Throughflow49,52. An increase in the Indonesian Throughflow has led to a greater transfer of warm water to the west of the Indian Ocean via the South Equatorial Current, resulting in a rapid warming of SST at the basin scale over the past two decades (Supplementary Fig. S3b)50,53,54. Therefore, the IPO-induced SST warming in the Indian Ocean is thought to be the trigger for the observed East Antarctic warming. However, the warming of the Indian Ocean is not the only factor in the variability of the STFZ, and there is an alternative process that could be responsible for the strengthening of the STFZ.

Fig. 5: Climatological and hydrographic background of southern Indian Ocean during warm-half year.
Fig. 5: Climatological and hydrographic background of southern Indian Ocean during warm-half year.
Full size image

Sea surface temperature (SST) change from the first half of the observation period (1993–2007) to the second half (2008–2021), with Interdecadal Pacific Oscillation (IPO) induced warm water pathways (open white arrows) from the Pacific to the Indian Ocean. The black arrow represents the warm half-yearly mean surface wind during the observation period. The letter H with a solid (dashed) blue circle is the position of the Mascarene High after (before) the warming of the Indian Ocean. The dots indicate the frontal zone with a SST meridional gradient above 0.6 °C 100 km−1. The subtropical frontal zone (STFZ) region is indicated by the thick red box.

The STFZ variability is characterized by a combination of subtropical warming and mid-latitude Indian Ocean cooling. This meridional dipole pattern of SST anomalies is identical to the negative phase of the Indian Ocean Subtropical Dipole55. The negative dipole pattern is associated with a weakening or eastward migration of the Mascarene High, which represents a semi-permanent subtropical high in the southern Indian Ocean56,57,58 (Figs. 5, 6a). As the Mascarene High moves eastward, the low pressure anomaly covers the STFZ (Fig. 6b). The decrease in sea level pressure (SLP) in the STFZ results in an intensification of the meridional SST gradient (R = −0.50, p < 0.05) (Supplementary Fig. S4). This suggests that anomalous surface winds and upwelling associated with the cyclonic anomalies enhance the vertical mixing with the cooler water below, which in turn strengthens the STFZ. Figure 6c shows less westward migration of the Mascarene High and a decreasing trend of the SLP in the STFZ during the period with increasing SST in the subtropical Indian Ocean. This suggests that the negative phase of the Indian Ocean Subtropical Dipole has become dominant. Over the past 30 years, the SST gradient in the STFZ has increased by about 20%, and atmospheric circulation changes associated with Indian Ocean warming have also contributed to this rapid increase of the SST gradient.

Fig. 6: Circulation changes over the southern Indian Ocean.
Fig. 6: Circulation changes over the southern Indian Ocean.
Full size image

a The trend of warm half-year mean sea level pressure (SLP) for 1993–2021. Solid contours represent the SLP climatology for the warm half-year mean. b The spatial correlation of the detrended Mascarene High (MH) center position with SLP (shading) and the spatial regression between the detrended MH center position and surface winds (vector) for 1993–2021. c The time series of the warm half-year mean MH center position (bar), with the horizontal line indicating the mean for the whole time series (86.4°E), and SLP anomalies in the subtropical frontal zone (STFZ) region (blue line). The correlation coefficient between SLP anomalies and meridional sea surface temperature (SST) gradients in the STFZ region is shown at the upper right. For all maps, only data points falling within the 90% confidence interval are plotted. The dots indicate the SST frontal zone with SST meridional gradient (∂SST/∂y) above 0.6 °C 100 km−1. The STFZ region used for comparison with the Relay Station temperature is indicated by the blue box.

The Mascarene High center position is largely influenced by remote forcing such as the El Niño/ Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD)59,60. The fact that the interannual variability of the Mascarene High position exhibits a robust correlation with both Southern Oscillation Index (SOI) (R = −0.60; p < 0.05) and IOD index (R=0.43; p  < 0.05) indicates that the eastward migration of the Mascarene High is strongly linked to the occurrence of both El Niño and a positive IOD (Supplementary Fig. S5)56,61,62. This is in the opposite direction to the above explanation of the Mascarene High moving eastwards during the negative IPO phase. The eastward shift of the Mascarene High from the late 1990s is not an atmospheric response to the ENSO and IOD, but may reflect changes in atmospheric circulation due to the warming of the Indian Ocean. This is supported by the fact that the amplitude of the Mascarene High variability has decreased as the Indian Ocean has warmed (Fig. 6c).

The IPO-induced SST warming over the southern Indian Ocean is a climate variability on a multi-decadal scale. This highlights the importance of natural climate variability in the observed warming trends in the interior of the DML. However, it would not be reasonable to conclude that the robust warming trend in the interior of the DLM since the 1990s is attributed solely to natural decadal climate variability. A rapid warming of SST over the Indian Ocean is attributed to increased heat uptake from the atmosphere during the global warming hiatus (2003-2012)49,50,51,52. In other words, anthropogenic excess heat is the primary contributor to EAIS warming. Given that the oceans continue to absorb excess heat generated by human activities, it can be said that anthropogenic factors are altering the climate of the East Antarctic interior and making it more vulnerable to atmospheric and oceanic changes.

The impact of Indian Ocean warming on climate change in the interior of Antarctica is not yet fully understood, and thus EAIS warming is a sign that the EAIS is changing rapidly beyond our expectations. The spatial footprint of RLS temperature changes (Supplementary Fig. S1) shows that the rapid warming may be observed over the broad East Antarctic, but no clear warming trends were detected at coastal stations (Fig. 1). This is because local forcing, such as orographic blocking and sea ice variability, may attenuate the warming signal at coastal stations. The warm northerly winds that hit the coast are blocked by the steep topography of the ice sheet and deflected eastwards. They then flow parallel to the coastal slope in a strong current known as a barrier jet63,64,65,66. Cold air is entrained from the east and from the katabatic outflow into the barrier jet64. This means that the temperature does not rise as much in the coastal region as it does over the ice sheet65,67. Additionally, large sea ice cover along the coast of East Antarctica contributes to reducing the warming signal. Sea ice variability near a station can have a significant impact on the SAT, as sea ice can restrict the heat flux from the relatively warm ocean into the atmosphere. The annual mean SAT at stations on the Antarctic Peninsula is largely affected by the variability of sea ice extent near the stations1. In contrast, East Antarctic coastal stations experience much smaller annual variations in surface temperature than those on the Antarctic Peninsula as sea ice cover extends far to the north from the coast1. However, because Antarctic sea ice has decreased significantly and rapidly since 2016, reaching a record minimum in 202368, temperatures at East Antarctic coastal stations may now be more sensitive to sea ice variability than before. Changes in the extent of sea ice are associated with atmospheric circulation, with the occurrence of strong winds resulting in its displacement from or towards the coast. Thus, it can be said that temperatures at coastal stations are susceptible to the influence of Indian Ocean warming. To date, although basal melting of coastal ice is a major contributor of ice loss69, warming coastal regions could increase surface melt, leading to additional contributions to sea level rise. Careful monitoring of temperatures not only inland but also in the coastal regions is needed to accurately assess the impact of Indian Ocean warming on climate change in East Antarctica. Research is also needed to confirm the causality of the summer warming using climate model simulations and to explore potential future changes.

Methods

Data

Monthly-mean surface air temperature (SAT) data for the inland Dronning Maud Land (DML) were compiled from the historical AWS observations at Mizuho (MIZ, 70.72°S, 44.26°E, 2180 m a.s.l.), Relay Station (RLS, 74.01°S, 42.98°E, 3354 m a.s.l.), and Dome Fuji (FUJ, 77.33°S, 39.67°E, 3820 m a.s.l.). Multiple AWSs have been installed at these stations since the early 1990s, and observations have been continued to the present day. However, there are several concerns that need to be addressed prior to the integration of multiple AWS data sets into a climate change time series. The major concern is that AWS temperature measurements may have been biased warm in austral summer due to insufficient ventilation in the radiation shield70. In this study, these warm biases are quantified by comparison with temperature measurements with an aspirated shield and subsequently removed using a regression model. The other concerns resulting from changes in the sensor height due to accumulating snow and errors in the early days of the AWS system were also identified by comparing them to other temperature measurements and then removing them. The full description of these corrections is in our previous study19. After the corrections, multiple AWS records were integrated to create a time series for each station from 1993 to 2022, following the data processing steps used for the Scientific Committee on Antarctic Research (SCAR) Reference Antarctic Data for Environmental Research (READER) database16. The READER monthly temperature was calculated only if more than 90% of the 6-hourly data was available. To maximize the amount of available data, however, the cutoff percentage for computing the monthly-mean was reduced from 90% to 70% in this study. Comparison of the monthly-mean temperature calculated by randomly selecting 70% of the data with the data using a 90% cutoff percentage shows that the annual mean root mean square error (RMSE) is approximately 0.6 °C. This level of error is nearly one-third of the standard deviation of the monthly temperatures19. The percentage of time between 1993 and 2022 with available monthly-mean SAT data is 72% for FUJ, 79% for RLS, and 51% for MIZ. The detail of AWS observations and the quality-control (QC) procedure are presented in previous research19. The other monthly-mean SAT data for coastal (Syowa, Mawson) and inland (Amundsen-Scott, Vostok) stations were obtained from the SCAR READER database.

To investigate the atmospheric circulation for 1993–2022, we use monthly-mean fields from ERA5, the fifth generation global atmospheric reanalysis produced by the European Center for Medium-Range Weather Forecasts (ECMWF), which has a spatial resolution of 0.25° and 37 vertical pressure level from 1000 to 10 hPa20. Sea surface temperature (SST) variability over the southern Indian Ocean is studied using the National Oceanic and Atmospheric Administration (NOAA) Optimum interpolated (OI) SST data set, version 2 on 0.25° grid71,72. Changes in storm activity over the southern Indian Ocean are investigated using monthly-mean outgoing longwave radiation (OLR) data obtained from NOAA National Centers for Environmental Information, which has a spatial resolution of 2.5°.

Indices of climate variability modes were used to assess the influence of the observed warming in the interior of East Antarctica. For the SAM index, we used Antarctic Oscillation (AAO) index defined as the leading principal component of 850 hPa geopotential height anomalies south of 20°S73. The influence of El Niño/ Southern Oscillation (ENSO) was evaluated using the Southern Oscillation index (SOI), which is a standardized index based on the observed sea level pressure (SLP) difference between Tahiti and Darwin, Australia. We also used the Dipole Mode Index (DMI) for the Indian Ocean Dipole (IOD) intensity59.

Reconstruction of the missing SAT

The automatic weather station (AWS) observations are not continuous records, and there are some months with no observations from 1993 to 2022. To fill in the missing records, a reanalysis product was sought which would capture the month-to-month variability of observed temperatures and would not incorporate AWS data into the data assimilation process. A comparison between the monthly mean 2 m temperature (T2m) in the four reanalyses and the observations is shown in Supplementary Fig. S6. Except for JRA-55, all reanalyses show good correspondence to the observations, though lower temperatures ( < −40 °C) are overestimated by them. MERRA2 shows optimal performance in simulating the T2m with a root-mean-square error (RMSE) of 2.58 at Dome Fuji (FUJ) and 1.72 at Relay Station (RLS) (Supplementary Fig. S6b, f). ERA5 data agree well with temperature in austral summer, but the discrepancies between ERA5 and observations increase as temperatures decrease (Supplementary Fig. S6a, e). The temperature differences in the austral winter reach up to 5 °C, which results in a relatively larger RMSE for the ERA5 data. Comparable results were reported in a previous study74. Although RMSE of MERRA2 is smaller than that of ERA5, there is a concern that MERRA2 T2m at the three stations in inland DML may be affected by the AWS observations through its data assimilation system (See Supplementary Methods). This means that the MERRA2 bias computed by comparison with the AWS observations would underestimate the true model bias in the absence of observations. In contrast, the problem of circularity between the predictor and the predicted variable is avoided for ERA5 due to the fact that the AWS observations do not impact ERA5 T2m (Supplementary Methods). Therefore, ERA5 T2m data were used to fill in the missing observations to create a continuous SAT record for three stations from 1993 to 2022.

ERA5 has an overestimation of lower temperatures ( < −40 °C), but otherwise good agreement with observations (Supplementary Fig. S6). To eliminate the inherent bias in ERA5 T2m estimates, a regression model was employed, as the biases show a linear increase with decreasing temperature. It is noteworthy that more than half of AWS data from 1993 to 2009 are independent from the data assimilation system because the data is not available online. The regression model was calibrated using data from 1993 to 2009 (calibrated period) and then tested using data from 2010 to 2022 (tested period) (Supplementary Fig. S10). The RMSE of bias-corrected ERA5 T2m is less than half that of the original ERA5 T2m (Supplementary Figs. S8 and S10). The largest RMSE (1.45 °C) at FUJ, the coldest site, is much lower than that for the other reanalyses (Supplementary Fig. S6). It is worth noting that the RMSE for the tested period is substantially smaller than that for the calibrated period except for MIZ (Supplementary Fig. S10). This implies that the quality of bias-corrected ERA5 T2m is relatively homogeneous over the whole period of the observations.

Uncertainties of reconstructed temperature records

The uncertainties of the bias-corrected ERA5 T2m are assessed through the direct comparison against AWS observations for the 1993–2022 period. The correlation, RMSE, and the bias between monthly observations and ERA5 T2m data are computed at each station. The RMSE of monthly-mean bias-corrected ERA5 T2m is much smaller than the standard deviations of monthly temperatures and is also less than the uncertainty of the AWS observation (1.5 °C) (Supplementary Fig. S11). In addition, the biases of the data with respect to the observations do not depend on the AWS temperature and vary around zero. These suggest that the reliability of monthly mean corrected ERA5 T2m is sufficient to explore the interannual variability of monthly temperatures during 1993–2022. This is consistent with the smaller RMSE ( > 0.9 °C) for the mean seasonal temperatures.

To assess the influence of the infilling of the missing observations with bias-corrected ERA5 on the long-term trend, we compared the time series of the warm and cold half-year mean SAT anomalies at Relay Station, computed from AWS observations and bias-corrected ERA5 (Supplementary Fig. S12). We only focused on Relay Station because of fewer missing values. The half-year mean SAT anomalies were computed from monthly-mean AWS observations only if they were available for more than five months. For comparison, the results of the bias-corrected MERRA2 are also shown in Supplementary Fig. S12. The year-to-year variability of ERA5 captures well the observed variability and amplitude of AWS observations. In particular, the cold half-year mean ERA5 anomalies are in very good agreement with the observations (RMSE=0.18 °C). Despite a slight cold bias in the first half of the observation period (1993–2007), the warm half-year mean of ERA5 is also in good agreement in the second half of the observation period (2008-2021 for warm half-year mean; 2008-2022 for cold half-year mean). The RMSE of ERA5 (0.24 °C) is much smaller than the measurement error (0.40 °C for half-year mean). Thus, we can say that the infilling of the missing observations with bias-corrected ERA5 does not contribute to the warming trend reported in this study. It is noteworthy that the usage of bias-corrected MERRA2 does not reach the same conclusion. The peak-to-peak variability of MERRA2 overestimates the observations. In addition, there is an artificial long-term bias in the MERRA2 time series. In the first half of the observation period (1993–2007), MERRA2 data are warmer than the observations. In contrast, most MERRA2 data are colder than the observations in the second half of the observation period (2008-2021 for warm half-yearly mean; 2008–2022 for cold half-yearly mean). Thus, the warming trend could be underestimated by the infilling of missing observations with MERRA2. This is the main reason for using ERA5 for missing observations.

Statistical methods

Standard least squares linear regression is employed to calculate both trends and regression analysis. A non-parametric Mann-Kendall’s test is utilized to assess the annual, semi-annual, and seasonal trends of surface air temperature. Statistical significance is indicated by p  < 0.05. The two-tailed Student’s t-test is used to determine the statistical significance of correlation and regression analysis. The t-score, calculated as the regression model estimate divided by the standard error, was compared to the critical value from the Student’s t-distribution with n-2 degrees of freedom and the specified confidence level (P).