Introduction

The Tibetan Plateau (TP, often called the third pole) covers an area of about 2.5 million square kilometers with an average altitude of over 4000 m1. It holds thousands of glaciers and is the source of many Asian rivers such as the Yellow River and Yangtze River2,3. Serving as the water tower of Asia4,5, the rivers originating from the TP provide water for about 40% of the world’s population. The unique topography of the TP exerts a significant influence on the Asian monsoon and even the global climate through the dynamic and thermal effects6,7,8,9. In the context of global warming, the annual warming rate in the TP is 0.34 °C/decade during 1979–2020 based on the ERA5 reanalysis10, which is faster than the global average and second only to the Arctic among all regions in the world10,11,12. The observed warming trend exists in all seasons but the strongest warming occurs in winter, followed by autumn10,11,12. As a result, the frequency and intensity of extreme weather and climate events in the TP have increased significantly13, and the glaciers have melted intensively3,14. The reasons behind the enhanced TP warming are complex, and some mechanisms have been proposed, including the increase of greenhouse gas emissions9,15 and atmospheric water vapor16,17, snow and ice albedo feedback17,18, the weakened winter monsoon19, increased lower clouds20,21 and brown clouds22,23, and the land use24,25. However, it is not yet clear which process is dominant.

Meanwhile, owing to the unique environmental characteristics and rich natural resources, the Arctic has attracted much attention from scientists and governments worldwide26,27,28. The climate in the Arctic has undergone unprecedented changes. Particularly, the near-surface temperature of the Arctic has risen at a rate more than twice the global average29 and even more than four times in winter during the recent decades. This phenomenon is known as the Arctic amplification (AA), which has unneglectable regional and global influences30,31. The mechanisms responsible for the AA have been widely discussed32,33,34,35,36,37, and some studies36,37 indicate that the Arctic sea ice loss38,39 plays an important contribution.

So far, however, there have been no reports connecting the TP warming with the rapid change in the Arctic sea ice to our knowledge. It is suggested that the reduction of the Arctic sea ice can intensify aerosol transport to the TP40, increase the frequency of Indian Summer Monsoon Rainfall extremes41 and heatwaves in southwestern China42, and deepen the East Asian Trough in late winter43. This study aims to detect the possible contribution of the sea ice loss in the Arctic to the TP winter warming quantitatively by data analysis and numerical experiments from an atmospheric general circulation model (AGCM).

Results

Changes in surface air temperature and Arctic sea ice concentration

As indicated in Fig. 1a, the linear trend of winter surface air temperature (SAT) during 1979–2018 presented by the ERA5 reanalysis can reach above 2 °C/decade in the northeastern BKS. Meanwhile, a significant warming trend also exists in the TP. Along with the dramatic Arctic warming, the sea ice concentration (SIC) in the northeastern BKS has declined rapidly by more than 20% per decade (Fig. 1b). A numerical sensitivity experiment with prescribed observed SIC and surface sea temperature (SST) in the BKS (named as BKS experiment; see details in Methods) demonstrates that the SIC change in BKS could enhance the winter TP warming. Note that the warming magnitude in the Arctic derived from the BKS experiment is similar to the ERA5 (Fig. 1c). Moreover, the decrease of BKS SIC induces a significant cooling trend in the mid-high latitudes of the Eurasia continent and obvious warming in the TP (Fig. 1c). The maximum warming rates (above 0.5 °C/decade) are located in the western TP, only slightly smaller than the observed warming.

Fig. 1: Linear trends of the winter SAT and SIC during 1979–2018.
figure 1

a Linear trend in the winter SAT derived from ERA5 reanalysis (K/decade); b Linear trend in the winter SIC derived from HadISST dataset (%/decade); c The same as in a, but for BKS experiment results (K/decade). Stippled areas indicate regions exceeding the 90% statistical confidence level. The areas of the Tibetan Plateau (TP) with altitude above 2500 m are shown by green lines in a and the region of BKS (65°–80°N, 0°–80°E) is outlined by red lines in b.

Full size image

Two leading modes of SAT over Eurasia

As seen from Fig. 1a and c, the striking warming in the Arctic and TP is accompanied by the cooling in mid-high latitudes of the Eurasia continent, although the cooling areas are much smaller in the observations than the sensitivity experiment simulations. In fact, numerous studies have revealed the mechanisms responsible for the Eurasia continent winter cooling. Some of them44,45,46,47 indicated that the cooling in the Eurasia continent is caused mainly by the loss of Arctic sea ice, whereas others48,49 argued that the contribution of internal atmospheric variability is dominant. To clarify the atmospheric response driven by the Arctic sea ice reduction and the role of internal atmospheric variability in the TP warming, the EOF analysis of the observed SAT is conducted for winter SAT (27°–90°N, 0°–180°E) in the period of 1979–2018. Results show that the first mode (EOF1) and the second mode (EOF2) of winter SAT account for 29.3% and 20.9% of the total variance, respectively. The first mode is a monopole pattern with consistent warming over the Eurasia continent (Fig. 2a). In addition, the first principal component (PC1) is highly correlated with Arctic Oscillation (AO) index, with a correlation coefficient of 0.77, passing the significance test at the 99% level (Fig. 2c). It implies this uniform warming pattern is closely related to the internal atmospheric variability of AO, which is consistent with Mori et al.46. The second mode shows a tripole pattern of warm Arctic, cold Eurasia, and warm TP (named as the AET pattern; Fig. 2b), with a high pattern correlation (0.8, passing the 95% significance level) to the BKS experiment as shown in Fig. 1c. The correlation coefficient between PC2 and SIC anomaly averaged over BKS is as high as −0.79, passing the significant test at the 99% level (Fig. 2d). It implies that the AET pattern, independent of AO, is closely connected with the change of BKS SIC. Moreover, the two leading modes of winter SAT are almost the same in different reanalysis datasets (Supplement Figs. 1 and 2), indicating that the dominant patterns of winter SAT are independent of dataset selection.

Fig. 2: The two leading modes of winter SAT anomalies (27°–90°N, 0°–180°E).
figure 2

a, b EOF1 and EOF2 of winter mean SAT in ERA5. c, d Standardized PC time series (blue lines) and AO index (red line), SIC anomaly (green line, unit: %) in the BKS (axis reversed).

Full size image

Mechanisms of the TP warming connected to the BKS sea ice loss

The first two EOFs of winter SAT from the ensemble mean of 20 members in the BKS experiment are very similar to the observed patterns (Supplementary Fig. 3a, b.vs. Fig. 2a, b). However, the AET pattern becomes the first mode in model simulations, accounting for ~55% of the total variance, while the AO pattern turns into the second mode. To further investigate the potential impact of the BKS sea ice loss on the TP warming, 10 years of lowest and highest BKS SIC anomaly in the observation are selected as shown in Supplementary Fig. 3c, named as low-ice (LICE) and high-ice (HICE) years, respectively. Obviously, much more positive phases of the AET pattern appear in LICE years and a reverse situation can be found in HICE years (Supplementary Fig. 3d). In contrast, there is barely a difference in AO phases in LICE and HICE years (Supplementary Fig. 3e). Note that the EOF analysis on each member of BKS experiment indicates that most members can reproduce the AET pattern (Supplementary Fig. 4). Therefore, the AET pattern is an internal mode of atmospheric variability, and the reduction of the BKS sea ice is in favor of the occurrence frequency of the AET positive phase. In order to investigate the mechanisms of the influence of the BKS SIC on the TP warming, we calculated the trends of atmospheric circulation caused by the decrease of BKS SIC in the experiment (Fig. 3a, b). Two anticyclone responses can be seen over the Arctic and TP at both 200 and 500 hPa, accompanied by a cyclone response over Eurasia in between. The anticyclonic circulation over the BKS can be explained by the increased upward turbulent heat fluxes related to the BKS sea ice loss (Supplementary Fig. 5), as mentioned in the previous studies40,50. It further enhances the propagation of a wave train from the Arctic via Eurasia and then to the TP, corresponding to the negative vorticity response over the Arctic and TP and a positive vorticity response over the Eurasia continent (Supplementary Fig. 6). As a result, the strengthened southwesterlies and warm advection towards the TP (Supplementary Fig. 7a, b) lead to the TP winter warming. It is worth noting that the circulation response to the BKS sea ice loss in the model simulation resembles the regression wind fields against the AET mode in the ERA5 reanalysis (Fig. 3c, d). This means that the reduction of BKS sea ice, as an external forcing, could induce an atmospheric circulation response resemblance to the AET mode, hence enhancing the TP winter warming.

Fig. 3: Observed and simulated wind fields.
figure 3

a, b Linear trends of horizontal wind fields (m s−1 decade−1) at 200 hPa (a) and 500 hPa (b) obtained by the BKS experiment; c, d Regressed horizontal wind fields (m s−1) at 200 hPa (c) and 500 hPa (d) against the normalized PC2 of winter SAT anomalies in the region of Eurasia from 1979 to 2018 derived from ERA5. Blue streamlines indicate the regions exceeding the 90% statistical confidence level. The TP domain is outlined by green curves with the surface altitude above 2500 m.

Full size image

The above results indicate that the sea ice loss in BKS region could strengthen the TP winter warming. Then, we quantitatively analyze the contribution of the BKS sea ice loss to the TP winter warming through the ratios of warming trends between the BKS experiment and multiple datasets. It is indicated that the BKS SIC loss could contribute ~18–32% of the TP winter warming during the period of 1979–2018 (Fig. 4).

Fig. 4: Contribution of BKS sea ice loss to the TP winter warming.
figure 4

Ratios of warming trends between the BKS experiment and multiple datasets during the period of 1979–2018.

Full size image

Discussion

Based on the observational data analysis and ensemble experiments from an AGCM, this study for the first time proposes and verifies that the dramatical reduction of BKS SIC is an important factor in the TP winter warming. We find that the second mode of winter SAT over Eurasia (27°–90°N, 0°–180°E) links the BKS SIC and the TP warming. The mode (i.e., the AET mode) is characterized by a tripole pattern with warm centers over the Arctic and the TP, and a cold center in the mid and high latitudes of the Eurasian continent. Moreover, the BKS SIC can efficiently regulate the phases of AET, and the reduced BKS sea ice benefits for the occurrence of the AET positive phase. The underlying mechanism could be summarized as that the BKS SIC reduction results in upward turbulent heat fluxes, which promotes the Rossby wave train extension from the Arctic through the Eurasian continent to the TP. Subsequently, the intensified southwesterlies toward the TP reinforce the warm advection and the winter warming over the TP.

This study focuses on the linear trend of winter SAT. Does the physical connection also exist between the BKS SIC and TP SAT anomalies on the interannual time scale? To address this issue, we calculated the correlation between the observed winter BKS SIC anomaly and winter SAT anomaly after removing the linear trend (Supplementary Fig. 8a). It is shown that the less BKS SIC is accompanied by higher SAT over the TP. In addition, the correlation coefficient between the AET and BKS SIC anomaly after removing the linear trend is −0.65, passing the significance test at the 99% significance level. A similar result can also be found in the BKS experiment (Supplementary Fig. 8b–d), despite the anomalies are not significant over the TP and the Eurasia continent. Therefore, the influence of BKS SIC may exist in both variability and trend. Since the decline of BKS SIC can account for 18–32% of the winter warming over the TP, other drivers for the winter SAT anomaly over TP need to be explored further. Moreover, the future loss of Arctic sea ice may exacerbate the TP winter warming suggested by the simulations of a coupled global climate model51. Thus, the model capability in simulating the AET pattern might affect the projected warming amplitude over TP, and this issue will be addressed in the future.

Recent studies indicate that internal variability plays an important role in the BKS sea ice loss in spring52. Given the similar decreasing sea ice pattern between the spring and winter over the Arctic, whether the winter BKS sea ice loss is a forced response or driven by internal variability needs to be further investigated in our future study.

Methods

We chose the period of 1979–2019 in the above datasets for diagnoses. The winter season includes December in the current year, January and February in the next year. Taking 2018 winter as an example, it is defined as the average value of December 2018 and January–February 2019, and the similar for other years.

Model simulation

The AGCM, i.e., Community Atmosphere Model version 5 (CAM5)53 with 2.5° × 2.5° horizontal resolution was used to explore whether the observed reduction in the BKS sea ice has an impact on the TP winter warming from 1979 to 2019. We performed an ensemble experiment with 20 members by perturbing the initial condition of each ensemble member.

We first performed a control run lasting for 30 model years. The model states of the beginning (January 1st) of each year from year 11 to year 30 have been used as the initial condition of each ensemble member in the experiments.

We used the same initial condition of the sea ice thickness for each ensemble member, based on the default initial condition of the model, while the sea ice thermal dynamic module is active in this simulation. Each ensemble member starts on January 1st of 1978. After a 1 year spin-up of the atmospheric model and the sea ice thermal dynamic module, the sea ice thickness achieves an equilibrium state. We then start our model experiments for each ensemble member, from Jan 1st of 1979.

In each member, the atmospheric model was driven by the observed monthly sea ice and the sea surface temperature over the BKS (65°–80°N, 0°–80°E) from 1979 to 2019 in the Hadley SST and SIC datasets, whereas the SST and SIC forcings over the other ocean areas were the climatological means with a seasonal cycle. There is a linear buffer zone between the forcing areas and other areas, whose width is 10-degree latitudes (and longitudes). We calculated the ensemble mean of the simulation results of these 20 ensemble members, which is considered to be the response of the model to the observed BKS SIC.

Statistical analysis

The two-tailed Student’s t-test was used when the statistical significance was assessed. The least-squares linear trend was removed at each grid point when discussing the interannual variability.