Unveiling the role of tropical Pacific on the emergence of ice-free Arctic projections

A rapid decline in Arctic sea ice concentration (SIC; % of the surface area covered by ice) is one of the most robust features of the projected future climate. Climatologically, the sea ice melts along the southern rim of the Arctic during the boreal summer and grows back the following winter. However, in recent decades, the observed summer sea-ice loss has dominated the wintertime growth, marking a clear declining trend. Internal variability (Ding et al 2017, 2019, England et al 2019), as well as natural and anthropogenic forcing (e.g. the effect of increasing greenhouse gas concentrations, Kay et al 2011, Notz and Marotzke 2012, Notz and Stroeve 2016, Polvani et al 2020), drive the SIC change. The sea ice plays a pivotal role in modulating the regional moisture and radiation budgets, and thereby the surface air temperature (SAT), owing to its high reflectivity compared to the ice-free, darker ocean surface. In recent decades, enhanced surface warming has been observed in the Arctic, a phenomenon referred to as Arctic amplification (AA; Manabe and Wetherald 1975, Bekryaev et al 2010, Serreze and Barry 2011, Cohen et al 2014, Dai et al 2019, Previdi et al 2021) when compared to global temperature changes. When the sea surface temperature (SST) is warmer than the atmosphere above, the summertime shortwave radiation absorbed by the ice-free region of the Arctic raises the SAT of the subsequent winter months, and thus, the AA is highest in the boreal winter months (Screen and Francis 2016, Dai et al 2019). The resultant temperature amplification triggers enhanced melting of glaciers and a faster rate of sea-ice retreat.

Climate model projections show that the Arctic could turn ice-free during the summer in the second half of the 21st century (Stroeve et al 2012a, Wang and Overland 2012, Liu et al 2013, Overland and Wang 2013, Notz and Stroeve 2018, Bonan et al 2021). Under a global warming scenario, sea-ice loss peaks during the later summer months of August through October (e.g. Stroeve et al 2012a, Stroeve et al 2012b, Serreze and Stroeve 2015). However, embedded within this long-term declining trend are years marked by an increase in SIC and decades with a slower rate of sea-ice retreat, as confirmed by satellite observations (Zhang 2021). The significance of these recovery events (positive SIC anomaly years) in a warming climate is that they could eventually slow down the rate of SIC loss. El Niño Southern Oscillation (ENSO; Clancy et al 2021), Arctic Oscillation (AO; Cai et al 2021), Atlantic multidecadal variability (Cai et al 2021), and Pacific Decadal Variability (Screen and Francis 2016, Cai et al 2021), among others, modulate SIC fluctuations on interannual to multidecadal timescales.

Globally, ENSO-related anomalies form the dominant source of interannual variability (Philander 1990, Deser et al 2010). The equatorial Pacific SST anomalies interact with the northern Pacific by modulating sea level pressure (SLP; Trenberth et al 1998, Deser et al 2017), and the relationship is strongest during the winter (Clancy et al 2021). However, the effect of ENSO on Arctic sea ice is still a matter of scientific debate. El Niño events in the equatorial Pacific can lead to anomalous deep convection and upper-level divergence and generate Rossby waves that propagate to the high latitudes and strengthen the Aleutian low (AL) and vice versa (Sardeshmukh and Hoskins 1988, Alexander et al 2002, Larson et al 2022). The positive SST anomalies in the equatorial Pacific enhance the poleward transport of warmer and wetter air masses (Clancy et al 2021) by strengthening the AL and, as a result, driving enhanced sea-ice melting in the Arctic Ocean (Liu et al 2004, Svendsen et al 2018, Screen and Deser 2019). Conversely, the feedback from extratropical SLP variability can modulate the intensity, onset, and decay (or evolution) of ENSO by inducing changes in circulation pattern and the net surface heat flux (Vimont et al 2003, Yu and Kim 2011, Chen et al 2023b).

The background internal variability frequently overshadows the impact of remote forcing over the Arctic region (Ding et al 2019), suppressing the effect of ENSO. Additionally, the small sample size during the satellite era (Deser et al 2017) and the contrasting ENSO–Arctic relationships in different climate models (Ding et al 2019) make it hard to conclude their interrelationship. Recent studies suggest that the interannual ENSO–North Pacific and ENSO–Arctic relationships have undergone substantial inter-decadal changes in the past (Chen et al 2020, Aru et al 2023, Chen et al 2023b, Jahfer et al 2023). Anthropogenic warming can also modulate the strength of the tropical-extratropical relationship (Wang et al 2013).

Several studies have documented that the impacts of Arctic sea ice are not limited to the northern extratropics and have global climatic significance (Screen et al 2018), including the effect on succeeding ENSO (Chen et al 2020, Liu et al 2022), Hadley circulation (Chemke et al 2019), northward shift and intensification of the ITCZ (Wang et al 2018, England et al 2020), and intensity and frequency of cyclones over the North Pacific (Chen et al 2023a, Hay et al 2023). Therefore, understanding the driving factors is crucial for a skillful prediction of the future Arctic climate and its feedback. This study focuses on the fate of sea ice in the most vulnerable season (August to October; ASO), when the mean ice cover is minimal. Our understanding of how tropical Pacific anomalies would affect interannual sea-ice variability in the Arctic is limited (Deser et al 2017, Clancy et al 2021). This study examines the interrelationship between ocean-atmospheric conditions over the Pacific and the Arctic SIC in the Coupled Model Intercomparison Project Phase 6 (CMIP6) models. We explore how the strength of the simulated relationship determines the timing of the emergence of a projected ice-free Arctic over the Pacific sector under four different future scenarios. Further, we investigate the possible mechanism linking the fate of the Arctic SIC and its interrelationship with the Pacific. Finally, we summarize the potential impacts of the pace of sea ice loss in the future.

This study uses the climate simulations provided by CMIP6 (Eyring et al 2016) to analyze the evolution of Arctic SIC under historical and future scenarios. Since the projected future simulations are available for 86 years (2015–2100; Shared Socioeconomic Pathways [SSPs]), the last 86 years of the historical period (1929–2014; present-day [PD]) were selected for consistency. Long-term SLP, SIC, SST, and SAT data from the ERA5, provided by the European Centre for Medium-Range Weather Forecasts, were used for model validation (available for 1940–2014). The ERA5 data were compiled from various in-situ and satellite sources and interpolated to a 1° × 1° horizontal grid. Note that before the satellite era (1979), the reliability of SIC data was very low. For ease of comparison, we refer to the reanalysis product from ERA5 as observations in this study.

The future climate simulations were carried out under the SSPs that follow different forcing pathways. This study used four SSP scenarios, namely, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively, in which the radiative forcing hits 2.6, 4.5, 7.0, and 8.5 W m⁻² by 2100. The CMIP6 SIC data can be downloaded from the ocean and atmospheric component models with the CMIP6 conventional names 'siconc' and 'siconca', respectively, and siconca was used whenever siconc was unavailable. In total, 37 climate models were available in common for all the CMIP6 scenarios used in the study. All the available realizations from each of the CMIP6 climate models are used (see supplementary figure 1). The ensemble mean is computed by estimating the mean of all the ensemble members of each CMIP6 model initially, and then the multi-model mean is calculated.

Zoom In Zoom Out Reset image size

The SIC variable represents the percentage of the ocean surface area covered by ice. Since the climatological minimum of historical SIC was recorded during the late summer (ASO) and is predominantly restricted to the north of 70° N, in this study, the Arctic is defined as a region bounded by 70° N–90° N. We define the Pacific and Canadian sector of the Arctic (PC-Arctic) as the entire Arctic region encompassing 90° E–90° W. We consider the Arctic seasonally ice-free when the mean SIC during ASO is less than 15%. The El Niño and La Niña years were classified based on SST anomalies over the Niño3.4 region (5° S–5° N, 170° W–120° W), with positive anomalies representing El Niño years. To represent the North Pacific SLP time series, we use the North Pacific Index (NPI), defined as the area-weighted average of SLP anomalies over the North Pacific (30° N–65° N, 160° E–140° W). The NPI is calculated in this study as:

$$\begin{align}{\text{NPI = }}{\sum _i}{A_i} \cdot \left[ {{\text{SLP}}{{\text{X}}_i} - {\text{SLP}}{{\overline{\text{X}} }_i}} \right]/A\end{align}$$

where A_i represent the area of the grid cell i, SLPX_i is the SLP at the grid cell i, SLP${\overline{\text{X}}}$ _i is the mean SLP and A is the total area of the region bounded by 30° N to 65° N, 160° E to 140° W.

The Arctic SIC exhibits a robust declining trend in historical and future periods (Overland and Wang 2013). Therefore, to designate a recovery (growth) or loss of SIC in the Arctic, we define the term 'difference' as the interannual change in SIC magnitude. Accordingly, a positive SIC difference during ASO refers to an increase in mean SIC during the current season compared to the SIC during the same season of the previous year. Similarly, an event is classified as negative if the total SIC decreases compared to the same season of the previous year.

The strength of the Pacific–Arctic relationship in this study is estimated as the change in magnitude of the ASO SIC 'difference' (compared to the previous year) with respect to the December through February (DJF) SLP fluctuations over the North Pacific (represented as NPI index). Corresponding to 1 mb of negative NPI anomaly (NPI-; coinciding with El Niño events), if the ensemble mean of the SIC loss over the PC-Arctic region in a CMIP6 model is twice higher than the La Niña events during the historical period, the models were classified as P-models (positive). The remaining models were classified as N-models (negative). We follow the same set of P and N models for future SSP scenarios since the magnitude and region of NPI-related SIC differences may vary across different forcing scenarios (owing to anomalous SIC loss in the forthcoming decades). The SIC calculations were conducted for the ASO season, and the SAT and SLP anomalies were estimated as the DJF mean.

3.1. CMIP6 simulations of Arctic sea-ice

The SIC over the PC-Arctic sector follows a robust annual cycle with a considerable accumulation during winter and spring and a drastic loss during late summer (figure 1(i)). During the historical period, the CMIP6 multi-model ensemble mean (37 climate models) of SIC dropped from around 90% during the winter to ∼60% during the ASO season (black curve; figure 1(i)). Though the corresponding SSP scenarios replicate the seasonality of PD, the rapid retreat of summertime SIC intensifies the projected annual cycle (Carton et al 2015, Previdi et al 2021; figure 1(i)). The seasonal peak-to-trough amplitude increases from less than 40% in the historical period to >60% under the high SSP future projections. The lower SIC under high emission scenarios (figure 1(ii)), specifically during the summer, exposes most of the Arctic Ocean to the atmosphere above, warming the upper ocean. Consequently, the sea ice formation during the late spring and winter months gets delayed and depleted. In the following sections, we focus on the mechanisms of SIC changes during the most vulnerable season of ASO under different CMIP6 scenarios.

The area-weighted ensemble mean of the ASO SIC over the PC-Arctic shows a steady decline from the early 1970s during the historical period (figure 1). However, several instances of sea ice recovery events (SIC growth compared to the previous year) occur in the individual models during the historical period. Out of 37 climate models, the summer SIC was nearly absent (∼2%) in two models (CMCC-CM2-SR5 and CMCC-ESM2) during the historical period. Therefore, we omitted those models, and only 35 were considered for the remainder of the study. The averaging over all the realizations of the CMIP6 models suppresses the interannual variability. During the 86 years of the PD (1929–2014), the multi-model ensemble mean exhibits a robust declining trend in the second half of the analysis period. However, in future projections under various SSPs, this strong declining trend is apparent in the first few decades (figure 1). The PC-Arctic is projected to become seasonally ice-free in the latter half of the 21st century under SSP3-7.0 and SSP5-8.5 projections (thick yellow and red curves; figure 1). While under SSP5-8.5, the Pacific-Arctic turns seasonally ice-free by around 2050, under SSP3-7.0 and SSP2-4.5, the SIC reaches less than 15% by around 2060 and 2070, respectively.

The CMIP6 models demonstrate a huge inter-model and inter-ensemble spread (shaded area of figure 1). But the number of ensemble members available for each model varies with scenarios (supplementary figure 1). One of the biggest hurdles in assessing the ensemble mean of variables with a large inter-model spread is that much of the variability seen in the individual simulations gets suppressed while averaging across the models. To investigate the role of the Pacific on the evolution of Arctic SIC, we estimate the NPI-related SIC response separately for each of the realizations of individual models, and the ensemble mean was estimated. The multi-model ensemble mean responses were then calculated as the average of all 35 models.

Using 35 CMIP6 models, the multi-model ensemble mean of the difference composite was computed by finding the mean of the SIC 'differences' (see the Methodology Section) for all the realizations of each of the participant models corresponding to positive minus negative events, and then the mean of all the models was calculated (figure 2(i)–(vi)). In response to a positive SIC event over the PC-Arctic sector, the observed SIC increases along the edge of the Arctic in the PD. Similar to the observation, maximum SIC changes were found over the Beaufort Sea, the Chukchi Sea, the East Siberian Sea, and the Laptev Sea in the PD (a maximum increase of ∼20%; figures 2(i) and (ii)). However, in the Atlantic sector, the observed change in SIC is negligible, except for a modest decrease off the coast of Greenland, which is well captured in the multi-model mean of PD (figures 2(i) and (ii)). Similarly, in the projected simulations, the ensemble mean SIC anomalies concentrate over the central Arctic (maximum of >20%; figures 2(iii)–(vi)). The evolution of the ensemble mean of SAT over the PC-Arctic region shows anomalous warming in future scenarios (time-series; figure 2) and, therefore, results in the retreat of SIC towards the central Arctic. The warming-induced anomalous loss of SIC leads to an enhanced uptake of heat by the summertime upper ocean, and this excess heat is released to the atmosphere during subsequent winters, contributing to AA (Chung et al 2021). Therefore, the projected changes in the Arctic surface temperature are closely tied to the fate of the Arctic SIC. Below, we investigate the Pacific-mediated changes in the Arctic sea ice.

Zoom In Zoom Out Reset image size

3.2. Pacific influence on the interannual variability of Arctic sea ice

To understand the role of the Pacific on SIC growth and reduction events, the composite (positive minus negative) of the mean SIC 'difference' corresponding to NPI anomalies was analyzed using all the members of the 35 common models. Compared to the rest of the year, the DJF season shows a higher co-variability of the North Pacific SLP with the ASO Arctic SIC. When the mean wintertime (DJF) SLP change over the North Pacific (represented as NPI anomaly) turns positive, signifying a weakening of AL, the observed SIC during the historical period increases by ∼5% over the Beaufort, Chukchi, and East Siberian Seas (supplementary figure 2(A)). In contrast, the mean SIC shows a decay of similar magnitude over the Laptev Sea. However, in the CMIP6 simulations, there is a clear distinction between the eastern and western halves of the Pacific sector of the Arctic (supplementary figure 2(B)). Corresponding to an NPI+ event, the eastern half (180° E–90° W) shows a maximum increase of SIC by more than 5%. At the same time, the SIC over the western counterpart (the Arctic region bounded by 90° E–180° E) exhibits a weakening of similar magnitude (supplementary figure 2(B)). The multi-model ensemble mean shows an increase in SIC over the Atlantic sector in contrast to the observed SIC loss during the historical period (supplementary figures 2(A) and (B)). In the Pacific sector, the spatial pattern of SIC change needs to be better captured, even though the magnitude of area-weighted SIC change is comparable (supplementary figures 2(A) and (B)). These biases could be partly due to the erroneous representation of the ENSO-Arctic relationship in some of the models. In the projected future, the spatial pattern of SIC response to the NPI varies across the scenarios. However, the magnitude of the mean change over the Pacific–Arctic sector is positive and east-west asymmetry also persists in future scenarios (supplementary figures 2(C)–(F)).

During the historical period, a positive NPI anomaly centered around the Aleutian Islands is generally associated with a cooling event over the equatorial Pacific (La Niña) and vice versa (supplementary figure 3(A)). Associated with a La Niña event, the SLP anomalies turn positive to the north of 30° N, whereas the SLP over the PC-Arctic region decreases (supplementary figure 3). Under all the projected SSP scenarios, the ENSO-midlatitude teleconnection is well maintained in the future (supplementary figures 3(C)–(F)). During a DJF NPI+ event (NPI anomaly >1 mb), the observed simultaneous SST over the equatorial Pacific, centered over the Niño3.4 region, cools by more than 0.5 °C (supplementary figure 3(A)). In the last 86 years of historical simulations, the spatial pattern of SLP and SST anomalies depicts a similar picture (supplementary figure 3(B)). The central equatorial Pacific (5° S–5° N, 170° W–120° W) exhibits a robust cooling of more than 1 °C (supplementary figure 3(B)). Similar to the historical period, the spatial pattern and magnitude of the observed tropical-midlatitude covariability during the winter months are well captured in future scenarios (supplementary figures 3(C)–(F)). This suggests that despite a potential interdecadal change in strength, the observed relationship would be maintained in the projections over longer time periods. The observed tropical-midlatitude relationship during an El Niño event is well captured in all eight CMIP6 climate models, with more than 30 ensemble members representing the historical period (supplementary figure 4): except for four GISS-E2-1-G ensemble members (supplementary figure 4(D)), the NPI/AL deepens during El Niño events. However, the large inter-ensemble differences in magnitude highlight the impact of internal variability. But, the Pacific-mediated SIC variability is much weaker than the interannual variability of SIC over the Arctic region (Ding et al 2017, 2019, England et al 2019).

The CMIP6 climate models simulate the strong interrelationship between the DJF El Niño and simultaneous NPI across the scenarios, whereas the NPI relationship with the Arctic SIC varies considerably (figure not shown). We found in supplementary figure 2 that a positive NPI (coinciding with La Niña) resulted in a mean SIC loss over the eastern half of the PC-Arctic and a mean growth over the western half. Therefore, the area-weighted mean response of ASO SIC to DJF NPI is weak in magnitude. We classify the models as either P- or N-models based on the strength of NPI-related SIC loss. More specifically, a given model is termed a P-model if the 86 year average loss in PC-Arctic SIC is negative during an NPI- (corresponding to El Niño) and the loss in SIC during an NPI- is more than twice as high as the SIC gain during an NPI+. Below, we investigate how the intensity of the North Pacific–Arctic sea ice relationship plays a role in the projected timing of the emergence of an ice-free Arctic over the PC-Arctic region.

The role of tropical teleconnections in the long-term evolution of SIC in the PC-Arctic sector (70° N–90° N, 90° E–90° W) is evident in the time-series of the multi-model ensemble mean SIC (figure 3). During the historical period (1929–2014), the ensemble mean SIC in the P-models (green curve), in which the El Niño-related SIC loss is twice higher than the La Niña-related SIC growth, the magnitude of PC-Arctic SIC is slightly lower than the N-models (red curve) and the multi-model ensemble mean (black curve; figure 3(A)). From the early 1990s, the SIC simulated by the P-models reduced at a higher rate, more than 5% lower than the N-models by 2014 (figure 3(A)). The SIC evolution under SSP scenarios shows that the PC-Arctic SIC melts much faster in the P-models compared to the N-models and the multi-model ensemble mean in the 21st century. It is to be highlighted that the choice of models resulted in turning the Arctic ice-free (horizontal line in figure 3) by more than 15 years ahead in the SSP2-4.5, SSP3-7.0, and SSP5-8.5 compared to the complete set of models (figures 3(B)–(F)). Accordingly, the ensemble mean of N-models turns ice-free under the SSP3-7.0 and SSP5-8.5 scenarios with a lag of around 20 years compared to that of P-models. Therefore, while predicting the future of Arctic sea ice, the Pacific teleconnection is an important factor to be considered. Below, we discuss the possible factors leading to the faster extinction of SIC in models with ENSO-driven anomalous sea ice loss.

Zoom In Zoom Out Reset image size

The ensemble mean of the NPI-related standard deviation of SIC in the P-models and N-models in the last three decades of the historical period shows that the summertime SIC variability of more than 15%, primarily concentrated along the southern Arctic region (supplementary figure 5). With global warming, the outer edge of the Arctic slowly retreats and turns completely ice-free, and the variability shifts to the inner Arctic (supplementary figures 5(B)–(E)). Due to the higher rate of ENSO-related SIC decay in the P-models, the SIC over the central Arctic is disturbed faster (supplementary figures 5(B)–(E)). This is evident in the last 30 years of the SSP5-8.5 scenario of the P-models, where the SIC variability is nearly negligible. On the other hand, the SIC over the central Arctic exhibits more than 9% variability under the SSP5-8.5 scenario in the N-models (supplementary figure 5(J)). Note that the changes in magnitude and spatial pattern of variability in supplementary figure 5 could be the combined effects of warming, the asymmetric impact of ENSO, etc. Even though the SIC was almost 5% less in the last few decades of the historical period, which favored the P-models to maintain the lower sea ice in the future, the P and N models diverged further in the future projections (figure 3). This change in the rate of SIC loss is aided by the regional feedback mechanism in the Arctic, as discussed below.

The faster rate of SIC depletion in the P-models could be due to the contrasts in the background temperatures. The ensemble mean ASO SAT over the PC-Arctic region in the P-models (dotted line) and N-models (dashed line) is similar during the historical period but gradually increases in future scenarios (figure 4(A)). The mean SAT in the P-models and the N-models diverges after 2020, and the surface temperature in the P-models warms consistently up to 3 °C higher than the N-models under the high SSP scenarios (figure 4(A)). Simultaneously, the differences in Niño3.4 SST in the P- and N- sets of models also grows over time in the future scenarios, but at a much slower pace (figure 4(B)). We propose that this anomalously faster rate of surface warming in the Arctic could be aided by the faster rate of SIC decay and the resultant feedback. A permanent loss of SIC results in a higher rate of summertime heat absorption by the ocean (Chung et al 2021), warming the winter atmosphere and further aiding faster ice melting in the succeeding year. The Arctic SST increases by more than 3 °C in the P-models compared to the N-models by the end of this century, highlighting the significance of tropical-high latitude teleconnection in the projected future climate.

Zoom In Zoom Out Reset image size

This work analyzes the Pacific–Arctic relationship in a warming world with a rapidly declining trend in the Arctic sea ice. A complete loss of summer Arctic sea ice poses a great threat to the wintertime regrowth in the future projections as shown in figure 3. Since the Arctic SIC has global influences (Tang et al 2012, Screen et al 2018, England et al 2020, Liu et al 2022), the implications of a faster rate of decline and the early emergence of an ice-free Arctic should be investigated extensively. In the Arctic, when the seasonal ice cover retreats over the edges and gets exposed to the atmosphere (May to September), the ocean absorbs the incoming solar radiation, and subsequently, when the air above turns cooler (October to April), the stored heat energy is released into the atmosphere as latent heat and sensible heat (Franzke et al 2017, Dai et al 2019, Chung et al 2021). This oceanic source of wintertime atmospheric warming in the Arctic is important for sustaining high AA. A completely ice-free Arctic would greatly reduce the ocean-atmosphere temperature contrast and thereby the strength of AA (Dai et al 2019, Chung et al 2021). Therefore, understanding the factors affecting the timing of the emergence of an ice-free Arctic is important for a realistic long-term prediction.

Due to the nonlinearity of ENSO and the contrasts in the shifting of the sea-ice edge in each of the projected warming scenarios, it is difficult to isolate the actual effect of the tropical Pacific on the Arctic. Further, it is proposed that the background internal variability plays a bigger role in the SIC changes than the remote forcings (Ding et al 2019), and this relatively large impact of the background variability can suppress the effect of ENSO. Therefore, there are contrasting views on the effect of ENSO on the Arctic sea ice. The constraint of the small sample size of observational data (Deser et al 2017) and disagreement among the climate models (Ding et al 2019) further complicates this issue. Analyzing longer periods of data with a large number of multi-model ensembles could minimize this issue to some extent. This contrast in the warming rate could also affect the rate of Arctic SIC evolution in the projected future.

Our study confirms that Arctic sea ice variability is strongly influenced by tropical variability. Though the effect of Arctic internal variability often outweighs the remote impact interannually, the long-term impact of tropical effects cannot be disregarded. Outside the Arctic boundary, we find that the SIC differences during the ASO season have the strongest connection with the wintertime tropical Pacific Ocean and northern Pacific atmosphere. We find that the negative SST events in the equatorial Pacific favor a weakening of the AL and a consequent increase in the SIC in the PC-Arctic sector. The converse is also true. However, the future projection of a higher frequency of El Niño events (Cai et al 2014) can lower the effect of ice buildup during relatively weaker La Niña years. The persistence of the same relationship in observation and different periods of CMIP6 models under different scenarios signifies the robustness of our results. Since the rate of sea ice loss has been suggested to have global climatic significance (Chen et al 2023a, Hay et al 2023, Jahfer et al 2023, Wang et al 2018), the underlying factors need to be studied extensively. This study highlights the role of the tropical Pacific and the significance of the realistic representation of tropical-extratropical teleconnections in climate models. As the models with strong connection between the Arctic and tropical Pacific turn seasonally ice-free under SSP2-4.5, SSP3-7.0 and SSP5-8.5 scenarios by about 10–20 years earlier than the ensemble mean, we underline the importance of the simulating realistic tropical connection in future projections of the Arctic.

Though the ENSO-NPI relationship is well captured in the future SSP scenarios, the wintertime NPI-ASO Arctic SIC is not well simulated in the CMIP6 models. The models struggle to capture this observed teleconnection with the Arctic owing to the delay and complexity of the relationship. A recent study suggests that the wintertime fluctuations in North Pacific SLP affect the simultaneous atmospheric conditions over the Arctic that drive anomalous wind anomalies over the Arctic, leading to a dynamic redistribution of the SIC and thereby affecting the summer/fall SIC (Clancy et al 2021). The SIC evolution in the P and N models diverges when the ENSO anomalies turn stronger in the P-models, favoring increased SIC decay in models with stronger observed relationships (figures 3 and 4). Due to the relatively weaker impact of the remote forcing, along with the delayed response of the Arctic SIC to tropical forcing, and the complex underlying mechanism, most of the models struggle to represent the strength and polarity of the interannual relationship correctly. However, the observation and the ensemble mean of the selected models agree that the mean impact of SST anomalies in the Pacific strongly modulates the Arctic's late summer/early fall sea ice. Our results highlight the importance of realistic representation of the 'atmospheric bridge', connecting the Arctic with the tropics, for a better projection of Arctic climate and sea ice.

This work was supported by the National Research Foundation of Korea (NRF) Grant funded by the Korea government (MSIT) (Grant No. 2020R1A2C2006860) and by the Institute for Basic Science (Project code IBS-R028-D1). E-S C was supported by Korea Polar Research Institute (KOPRI) grant funded by the Ministry of Oceans and Fisheries (KOPRI PE24010). We acknowledge the World Climate Research Programme (WCRP) and Copernicus Climate Change Service (C3S) Climate Data Store (CDS) for providing the CMIP6 data and ERA5 data, respectively.

All the data used in our study can be freely downloaded from the following sources. The ERA5 data can be accessed from https://cds.climate.copernicus.eu.

The data that support the findings of this study are openly available at the following URL/DOI: https://esgf-node.ipsl.upmc.fr/search/cmip6-ipsl/.

S J and K-J H designed the study and wrote the paper with inputs from all the authors. K-J H, E-S C, C L E F and S S involved in the interpretation of the results and S S and S J prepared the figures. All authors contributed to the writing of the manuscript.

The authors declare no competing financial or non-financial interests.