Introduction

Arctic sea ice has been undergoing rapid thinning and retreat in recent years due to global warming and Arctic amplification1,2. Since 1979, there has been a consistent decline in Arctic sea ice extent with an approximate decrease rate of 4% per decade3. Notably, this negative trend persists throughout all twelve months, with September experiencing the highest decline rate of up to 12.4% per decade3,4. The observed changes in the Arctic sea ice not only have substantial impacts on localized climate dynamics5, but also exert ongoing influences on the global atmosphere6, ocean systems7, ecosystems8, aerosol pollution9, and human communities10. Therefore, a more profound understanding of the factors influencing Arctic sea ice is urgently needed to enhance Arctic sea ice predictability and better adapt to climate change.

External atmospheric forcing components, such as anthropogenic greenhouse gases and aerosols, play a crucial role in driving the variability of Arctic sea ice11,12,13,14,15. It has been observed that for every 1000 kg of CO2 emitted, there was a sustained loss of sea ice area by approximately 3 ± 0.3 m2 in September11. Furthermore, the extent to which the WNP TC activities are systematically linked spatially and temporally to variations in Arctic sea ice remains uncharacterized16. Multiple studies emphasize the contribution of internal climate variability, which may have contributed up to 60% to the observed decline in summer Arctic sea ice since 197917,18,19. Recent studies concur that the observed decline in sea ice since 1979 can be predominantly attributed to alterations in Arctic atmospheric circulation, driven by teleconnection from low-frequency variability in low-latitude convection associated with sea surface temperature (SST)20,21. This is evidenced by the emergence of anticyclonic circulation anomalies at 200 hPa over the Arctic, which result in a warming and moistening Arctic atmosphere. Consequently, these anomalies intensify clear-sky downward longwave radiation (DLR) and exacerbate the summer melting of sea ice in boreal regions22,23,24.

Recent studies have brought attention to the significant impact of tropical convection on Arctic surface air temperatures during winter25,26. The Madden–Julian Oscillation (MJO), primarily driven by tropical convection, exerts influence on the atmospheric circulation, temperature, and radiation in the Arctic region27,28,29. Henderson et al. (2014) observed that MJO can cause a shift in the region with the highest variability of Arctic sea ice as it moves poleward during summer and equatorward during winter30. In addition, tropical forcing plays a crucial role in regulating ice variability in the Barents-Kara Sea31,32,33. While these studies emphasize the significance of interactions between tropical and Arctic systems on longer intraseasonal timescales, they tend to overlook the impact of short-term tropical forcing, such as tropical cyclone (TC) activities, on a scale of about 10 days.

TCs are among the most destructive and formidable meteorological phenomena34. More than one-third of the global TCs originate in the western North Pacific (WNP). As a significant driving force, TC activities can impact the interannual variability of global energy transport35,36, thereby influencing weather systems, atmospheric circulation, and precipitation distribution through dynamic and thermal feedback37,38,39,40. Recent studies have demonstrated that heightened TC activities over the southeast part of the WNP from July to September substantially intensify the ENSO’s strength from October to December by weakening the Walker circulation and enhancing the eastward propagating Kelvin waves in the tropical Pacific region40,41. Furthermore, TC activities in the WNP also exert an influence on the spatial pattern of ENSO events42,43. Collectively, these findings underscore crucial feedback effects of TC activities on climate.

The impact of synoptic-scale signals, such as TC activities on sea ice in the Arctic, has received less attention compared to climate signals, such as Arctic atmospheric circulation patterns and Pacific SST forcing. As a typical synoptic scale system generated over tropical oceans, TC activities play a crucial role in generating Rossby waves, which continuously disperse energy　during their movement44,45. Therefore, it is likely that there exists a teleconnection between TC activities and the Arctic through Rossby waves. Scoccimarro et al.46 demonstrated that TC activities over the tropical Atlantic can induce winds into the Transpolar Drift Stream, leading to increased sea ice transport from the Beaufort and Eastern Arctic into the Central Arctic. Although recent studies have identified a statistical relationship between Northern Atlantic TC activities and Arctic sea ice, this connection has been superficially analyzed in terms of dynamics without adequately revealing the important modulating effect of thermodynamic processes consistently shown by recent research20,47. Furthermore, the extent to which the WNP TC activities are systematically linked spatially and temporally to variations in Arctic sea ice remains uncharacterized.

In this work, we reveal the teleconnections between the tropics and the Arctic using observational evidence and climate model simulation experiments. It is then demonstrated that TC activity over the WNP can favor an increase in APS SIC in September by triggering the Rossby waves.

Observational relationship between western North Pacific tropical cyclones and sea ice concentration over the Arctic

TCs over the WNP primarily occur from May to November, with a peak incidence observed from late August/early September48. While sea ice variability over the Arctic, particularly over the Arctic-Pacific sector (APS) encompassing the East Siberian Sea, Chukchi Sea, and Beaufort Sea (70–80° N, 160°E–140° W), reaches its maximum during the summer months49,50. Here, monthly and daily ACE indices averaged across the WNP region (10–35° N, 110°–140° E) are constructed to evaluate TC activities. The month-to-month lead-lag correlation between the APS SIC and WNP ACE index exhibits a robust association between September APS SIC and a one-month lead (August) ACE index with a correlation coefficient of 0.43 (p < 0.05) while showing a much weaker or even uncorrelated relationship with a one-month lagged ACE index (r = – 0.02; Supplementary Fig. S1a). Considering that TCs have an average lifespan of approximately ten days, the 10d-to-10d lead-lag correlations were also calculated to investigate the temporal relationship between the APS SIC and WNP ACE index (Fig. 1). Notably, a robust positive correlation was observed between the mid-August (August 11th–20th) WNP ACE index and the lagged APS SIC spanning from 10 to 40 days, with a peak value of 0.52 (p < 0.01) occurring at a lag of 30 days (mid-September; September 11th–20th). Moreover, after removing the linear effect of the ENSO, a significant correlation between the APS SIC and the WNP ACE index remains for a lead time of 10–40 days, with a maximum correlation coefficient of 0.52 (Supplementary Fig. S1b). This result is consistent with the findings after excluding the non-linear effects of ENSO. These findings suggest that the mid-August WNP TC activities can serve as an indicator or precursor to the changes of the mid-September APS SIC.

Fig. 1: The 10-day-to-10-day lead-lag correlations between the accumulated cyclone energy (ACE) over the western North Pacific (WNP) and sea ice concentration (SIC) over the Arctic-Pacific sector (APS) for the period from 1982 to 2020.
figure 1

The symbols “*,” “**,” and “***” above the correlation value indicate statistical significance based on a two-tailed Student’s t test at confidence levels of 90%, 95%, and 99%, respectively. For example, the value \({{0.}\,{52}} ^{{***}}\) in row 5, column 13 represents the correlation coefficient between the WNP ACE in mid-August (August 11th–20th) and the APS SIC in mid-September (September 11th–20th; with a lag of 3 ten-day periods), which is statistically significant at the 99% confidence level.

Full size image

Figure 2a provides the composite anomalies of mid-September Arctic SIC between 10 strong and 10 weak ACE events years, which were identified based on the ACE index exceeding 0.8 standard deviations, as described in the Methods section. The analysis reveals a significant increase of ~ 40% in APS SIC during strong ACE events compared to weak ACE events, indicating that the mid-September APS SIC is significantly increased during strong ACE events. In addition, regression analysis demonstrates that the increase in mid-September APS SIC is primarily caused by TC activities over the western part of WNP in mid-August (Fig. 2b). Furthermore, the Maximum covariance analysis (MCA) between the WNP ACE index and the Arctic SIC illustrates the primary coupled covariance patterns between the linear detrended mid-September SIC in the Arctic and the mid-August ACE index over the WNP (Fig. 2c, d), which suggests that the positive SIC anomalies in the APS exhibit a strong correlation with positive ACE index anomalies, particularly in the western part of the WNP. At the same time, the MCA demonstrates a significant correlation between the detrended mid-August WNP ACE index time series and mid-September SIC in the Arctic (r = 0.58, Fig. 2e), that is, the WNP ACE index can explain over 30.7% of the variance contribution of Arctic sea ice change one month later. Therefore, the increases in the WNP ACE index during mid-August are likely to contribute to the subsequent increases in mid-September Arctic sea ice.

Fig. 2: The spatial correlation and maximum covariance analysis of accumulated cyclone energy (ACE) and sea ice concentration (SIC).
figure 2

a Composite anomalies of mid-September Arctic SIC (shading, unit: %) between 10 strong (1983, 1987, 1991, 1994, 1996, 1997, 2007, 2013, 2015 and 2018) and 10 weak ACE events years (1986, 1988, 1989, 1995, 1998, 1999, 2003, 2009, 2010 and 2011). b Regressed mid-August ACE (shading, unit: 104 kt2) associated with the mid-September SIC over the Arctic-Pacific sector (APS). The yellow boxes are the focus areas for this study. The leading mode of maximum covariance analysis is applied to examine the relationship between detrended (c) mid-August ACE in the western North Pacific (WNP, 10–35° N, 110°–140° E) and (d) mid-September SIC in the Arctic (70–80° N, 160° E–140° W) and (e) their corresponding time series from 1982 to 2020. The first mode accounts for 30.7% of the covariance, and the correlation coefficient between the time series of ACE and SIC is 0.58, which is statistically significant at the 99% confidence level. The stippled regions indicate statistical significance based on a two-tailed Student’s t-test at the 90% confidence level.

Full size image

Impact of TC on Arctic atmospheric circulation and sea ice

To elucidate the mechanisms underlying the impact of WNP TC activities on Arctic atmospheric circulation and sea ice, we conducted a regression analysis of atmospheric circulation with lags ranging from zero to three ten days [mid-August (August 11th–20th), late August (August 21st–31st), early September (September 1st–10th), and mid-September (September 11th–20th)] onto the WNP ACE index in mid-August. These specific lags were chosen based on the highest correlation between the APS SIC and the WNP ACE index at ~30 days lag.

It has been pointed out that changes in low-latitude atmospheric circulation can influence Arctic sea ice through modifications in temperature, specific humidity, DLR, and other related processes23,24,51. As depicted in Fig. 3, during mid-August, the WNP TC activities could be treated as a Rossby wave source situated in the WNP, driving the planetary-scale Rossby wave train primarily propagating northeastward from the WNP (Fig. 3a). Subsequently, the Rossby waves reach the North Pacific and the Bering Strait in late-August (Fig. 3b), exhibiting maximum wave activity flux in early-September before propagating into the APS region (Fig. 3c). However, by mid-September, there is a significant attenuation of Rossby waves propagation (Fig. 3d). From mid-August to early September, regression analysis of the 500 hPa and 850 hPa geopotential height against the WNP ACE index displays a persistent poleward movement, with pronounced cyclonic anomalies observed over the Arctic, while anticyclonic anomalies observed over the Aleutian Islands (Fig. 3a–c, e–g). Recently, it is found that the intensified anticyclonic circulation over the Arctic leads to a reduction in the sea ice there52,53,54, while this study reveals that the anomalous cyclonic over the APS region can enhance the sea ice formation when TCs are active over the WNP a month prior, which is also manifested through the deepening of the Arctic polar vortex in the troposphere21. Due to barotropic energy conversion, which can provide favorable background conditions for the propagation of Rossby waves55,56, continuous propagation of Rossby waves toward the APS region occurred from mid-August to early September. However, their intensity diminished by mid-September, thereby resulting in a weakening of cyclonic anomalies over the Arctic since early September (Supplementary Fig. S2). Nevertheless, regression analysis of the 850 hPa circulation indicates an increasing prominence of northerly wind anomalies over the APS region (Fig. 3g–h). Consequently, the cold advection contributes to Arctic cooling and facilitates further expansion of sea ice.

Fig. 3: Atmospheric circulation in ERA5 reanalysis dataset (1982–2020) regressed onto the first accumulated cyclone energy (ACE) mode time series.
figure 3

Geopotential height (Z500, shading, unit: m) and wave activity flux (WAF, vector, unit: m–2 s–2) at 500 hPa in (a) mid-August, (b) late August, (c) early September and (d) mid-September, respectively. Geopotential height (Z850, shading, unit: m) and wind (vector, unit: m s–1) at 850 hPa in (e) mid-August, (f) late August, (g) early September and (h) mid-September, respectively. The stippled regions and purple vectors indicate statistical significance based on a two-tailed Student’s t test at the 90% confidence level.

Full size image

The negative anomalies in geopotential height in the upper troposphere are illustrated in Fig. 4a, featuring a prominent cyclonic anomaly centered at approximately 250 hPa over the APS. It corresponds to cyclonic anomalies centered over the APS, particularly in the upper troposphere at about 250 hPa, as evidenced by the wind and pressure relationship, while negative anomalies in atmospheric temperature and specific humidity primarily manifest in the middle and lower troposphere (Fig. 4b). Due to the strong coupling between atmospheric circulation and vertical velocity, the upper-level cyclonic circulation weakens adiabatic warming, resulting in a cooling effect on the lower Arctic atmosphere. This leads to negative surface temperature anomalies that cover almost the entire Arctic, particularly exceeding 2 K over the APS region (Fig. 4d). These negative temperature anomalies result in decreased atmospheric moisture and hinder water vapor evaporation. Meanwhile, specific humidity over the APS is further diminished due to the reduction in the water vapor transport from the North Pacific to the Bering Strait (Fig. 4e). The combined effect of cooler and drier air results in rapid decreases in the local DLR at the surface51,57, with a maximum decrease exceeding 5 W m–2 (Fig. 4f), thereby favoring an increase in sea ice58,59. In addition, the composite anomalies analyses of atmospheric circulation and thermodynamic variables between strong and weak ACE events align with the results of obtained from regression analyses (Supplementary Fig. S3). Changes in cloud cover may also impact the spatial distribution of Arctic sea ice by influencing the surface energy balance. However, any warming effect caused by reduced cloud cover through increased incoming solar radiation is often compensated by its cooling effect due to decreased DLR60. To effectively demonstrate the causal relationship between TC activities and Arctic sea ice, we have depicted the spatial distribution of information flow from the WNP ACE to APS SIC using the Liang-Kleeman information flow method (refer to “Method” Section) in Supplementary Fig. S4. The analysis reveals distinct causal relationships from WNP ACE in mid-August to APS SIC from mid-August to mid-September, with the strongest association observed in mid-September. These findings provide compelling evidence that supports the assertion that mid-August WNP ACE indeed causes an increase in APS SIC during mid-September.

Fig. 4: Mid-September atmospheric circulation and thermodynamic variables associated with tropical cyclone activity over the western North Pacific in ERA5 reanalysis dataset (1982–2020).
figure 4

The meridional cross-section (averaged over 160°E–140° W) of the composite anomalies of (a) geopotential height (shading, unit: m), (b) temperature (shading, unit: K) and specific humidity (contour, unit: 10–4 g kg–1). c Sea ice motion direction (vector, unit: cm s–1) and velocity (shading, unit: cm s–1) between strong and weak accumulated cyclone energy (ACE) events years. The spatial distribution of (d) surface temperature (Ts, shading, unit: K), (e) specific humidity (q, shading, unit: g kg–1) and vertically integrated moisture flux (vector, unit: kg m–1 s–1) averaged in lower troposphere (1000–850;hPa) and (f) downward longwave radiation (DLR, shading, unit: W m–2) at the surface regressed onto the first ACE mode time series. The stippled regions indicate statistical significance based on a two-tailed Student’s t test at the 90% confidence level.

Full size image

The WNP TC activities can also impact the distribution of Arctic sea ice by disrupting its motion. Typically, following the prevailing high-pressure anticyclonic ice gyre in the Arctic Ocean, sea ice drifts transpolarly from the Chukchi Sea to the Fram Strait, facilitating both export and melting processes61. However, during the 30 days after WNP TC activities, there is a weakening of the anticyclonic circulation over the Arctic, resulting in reduced transpolar drift of sea ice into the Fram Strait (Fig. 4c), which hinders both melting and export processes for sea ice and further promotes an increase of sea ice over the APS.

The above findings hold true for the year 1994 as well, which witnessed the highest August ACE index with the greatest number of TC genesis in the past four decades. During mid-August, five TCs occurred, including two super typhoons (Doug, ID: 9414 and Fred, ID: 9417), two typhoons (Ellie, ID: 9415 and Gladys, ID: 9418), and one strong tropical storm (Harry, ID: 9419). Simultaneously, notable Rossby waves propagated from the WNP to the Bering Sea while an anomalous cyclonic circulation was established over the upper Arctic region (Supplementary Fig. S5a). The corresponding cooling (Supplementary Fig. S5b) along with drying (Supplementary Fig. S5c) effect in the lower Arctic atmosphere directly decreased the DLR fluxes (Supplementary Fig. S5d). Furthermore, the anomalous cyclone over the Arctic induced the transpolar drifts of sea ice from the central Arctic regions to the APS, leading to a remarkable increase of up to 30 percent in APS SIC after a span of thirty days (Supplementary Fig. S5e, f). It was the WNP TC activities during mid-August that led to the fact that, in 1994, there was more Arctic sea ice in mid-September than in mid-August, whereas it is typically expected for there to be less sea ice in mid-September than in mid-August.

Simulation of Arctic atmospheric circulations under ACE-like atmospheric forcing

Figure 5 depicts the simulated composite anomalies in atmospheric circulation and thermodynamic variables between SEN and CTRL (representing ACE-like atmospheric forcing and control experiments, respectively; refer to the Methods section for precise details on experimental design) from mid-August to mid-September. These simulation results replicate the driving mechanism of teleconnection between WNP and APS as mentioned above. During the subsequent 30-day period following the imposition of the ACE-like atmospheric circulation pattern onto the model, a northeastward propagation of the Rossby wave train originating over the WNP toward the Arctic is observed. This results in an anomalous cyclonic circulation at high altitudes over the Arctic, exhibiting a slightly higher amplitude compared to mean climatology observations and consistent with the findings depicted in Fig. 5a–d. This consistency implies that TC activities over the WNP have a high likelihood of inducing an atmospheric circulation pattern conducive to sea ice increase in the Arctic. Similarly, the CESM can reproduce significant decreases in surface temperature (Fig. 5e–h), specific humidity (Fig. 5i–l), and DLR (Fig. 5m–p) over the APS during the subsequent 30 days following TC activities. The model also effectively performs the zonally averaged spatial distribution of negative anomalies in temperature, geopotential height and specific humidity observed, albeit with a slight overestimation (Fig. 6). Therefore, the model provides clear evidence supporting the teleconnection mechanism between the WNP and APS as mentioned above. TC activities induce low-pressure anomalous cyclones over the Arctic by driving Rossby wave trains propagating from the WNP to APS, then cooling and drying out the lower tropospheric atmosphere and decreasing the surface DLR over there, favoring the suppression of Arctic sea ice decline during boreal summer.

Fig. 5: Ensemble mean anomalies of atmospheric circulation and thermodynamic variables simulated by Community Atmosphere Model version 6 (CAM6).
figure 5

The columns from left to right represent the anomalies between the SEN run and CTRL run in mid-August, late August, early September, and mid-September, respectively. The first row (a–d) represents geopotential height (Z500, shading, unit: m) and wave activity flux (WAF, vector, unit: m–2 s–2) at 500 hPa. The second row (e–h) represents surface temperature (Ts, shading, unit: K). The third row (i–l) represents specific humidity (q, shading, unit: g kg–1) and vertically integrated moisture flux (vector, unit: kg m–1 s–1) in the low troposphere (1000–850 hPa). The fourth row (m–p) represents downward longwave radiation (DLR, shading, unit: W m–2) at the surface. Only the anomalies that exceed the 90% confidence level using a two-tailed Student’s t test are presented.

Full size image
Fig. 6: Meridional cross-section of anomalies of thermodynamic variables simulated by Community Atmosphere Model version 6 (CAM6).
figure 6

Composite anomalies of temperature (shading, unit: K), geopotential height (black contour, unit: m), and specific humidity (red contour, unit: 10–4 g kg–1) between the SEN run and CTRL run from (a) mid-August (August 11th–20th), (b) late August (August 21st–31st), (c) early September (September 1st–10th) and (d) mid-September (September 11th–21st). Only the anomalies that exceed the 90% confidence level using a two-tailed Student’s t-test are presented.

Full size image

It is noteworthy that the anomalies in the simulated atmospheric circulation and thermodynamic variables do not reach their maximum in mid-September after a 30-day lag, but rather in mid-August and late August. This discrepancy arises due to the influence of relevant climate background fields and other atmospheric forcings on observations, while simulations only depict pure disturbances immediately following ACE-like circulation forcing, which tends to diminish or even disappear over time. The model results indicate that the impacts are not obvious after a 30-day lag compared to 10- and 20-day lags, which is consistent with the APS SIC lagging behind the WNP ACE index by 30 days.

Discussion

This study demonstrates that the active WNP TC activities in mid-August lead to an increase in the APS SIC in mid-September through the excitation of planetary-scale Rossby waves propagating poleward. The TC-excited Rossby waves generate anomalous cyclonic circulation over the Arctic, which reduces adiabatic warming over the APS region and results in negative temperature anomalies in the lower and middle troposphere. This causes a decrease in geopotential height in the upper and middle troposphere. The combined effect of cooler and drier air subsequently leads to a reduction in the local DLR at the surface, thereby favoring the increase in APS SIC. Simultaneously, the anomalous cyclonic circulation over the Arctic impedes the transpolar drift of sea ice from the Chukchi Sea to the Fram Strait, further contributing to higher SIC over the APS.

It is worth noting that our study solely examines the climatological impacts of the WNP TC activities on the Arctic atmospheric circulation and SIC. Further investigation is needed to understand how composite effects of TC activities in other oceans (e.g., the Northeastern Pacific) may differentially influence the growth and melting of Arctic sea ice by modulating large-scale dynamical and thermodynamic processes. Additional research is warranted to examine the relative contributions of TC activities in different oceans to Arctic sea ice. Furthermore, simulations driven by ACE-like atmospheric circulation anomalies merely serve as a simplified representation of the impact of TC activities on the Arctic. However, it is crucial for future studies to enhance the capability of numerical models in accurately capturing the characteristics of TC activities, encompassing intensity, track, and duration, as well as their influences on Arctic atmospheric circulation and sea ice. This enhancement will facilitate a more comprehensive investigation into interactions between the tropics and the Arctic region.

This study elucidates the impact of WNP TC activities on Arctic sea ice and its underlying mechanisms, thereby contributing to a comprehensive understanding of intricate teleconnections between the tropics and the Arctic. Considering the projected intensification of the most powerful TC activities62 and their poleward expansion in response to global warming63, it is imperative to deepen our comprehension of these teleconnections due to their increasingly remarkable and sensitive influence on the Arctic.

Methods

Data

This study focuses on the relationship between the WNP TC activities and the Arctic SIC during 1982–2020. The dataset used specifically as follows: The International Best Track Archive for Climate Stewardship (IBTrACS)64 published by the National Oceanic and Atmospheric Administration (NOAA) provides maximum sustained surface winds and locations of TC activities over the WNP, which are utilized for calculating the accumulated cyclone energy (ACE)65. Daily Sea Ice Concentrations (SIC), with a horizontal resolution of 0.25° × 0.25°, are obtained from the NOAA OI sea surface temperature (SST) V2 High-Resolution Dataset66. The meteorological parameters from ERA5 hourly data on single levels and pressure levels67, provided by the European Center for Medium-Range Weather Forecasts (ECMWF), with a horizontal resolution of 0.25° × 0.25°(interpolated to 1° × 1°), are utilized to assess the atmospheric circulation patterns and evaluate the thermodynamic variables between strong and weak ACE events. These meteorological parameters, including temperature, meridional winds, zonal winds, geopotential height, specific humidity, and downward longwave radiation (DLR), are computed as daily representative averages based on hourly measurements. The transpolar drift of Arctic sea ice is examined using Polar Pathfinder Daily 25 km EASE-Grid Sea Ice Motion Vectors, Version 4 dataset provided by the National Snow and Ice Data Center68. ERA5 hourly SST data on single levels, with a horizontal resolution of 1° × 1° 67, provided by ECMWF, are employed for the computation of the daily Niño3.4 index. This index represents the detrended SST anomaly over the Niño3.4 region (170°W–120°W, 5°S–5°N).

Accumulated cyclone energy calculation

The daily accumulated cyclone energy (ACE) in each grid cell (2° latitude × 2° longitude) is defined as the summation of the squares of the estimated maximum sustained surface winds (m s−1) during a 6-hour period for all TC activities occurring within the specific grid cell on a given day. The calculation of daily ACE in a particular grid cell is as follows,

$${ACE}={\sum}_{i}{V}_{i}^{2}$$
(1)

where i represents the ith TC occurring within the grid cell on that day, and V denotes its corresponding maximum sustained surface wind speed. A strong (weak) ACE event is characterized by an ACE index value ≥ 0.8 ( ≤ – 0.8) standard deviation from the sum of ACE values across all selected regions. From 1982 to 2020, there were 10 years with strong ACE events (1983, 1987, 1991, 1994, 1996, 1997, 2007, 2013, 2015, and 2018), and 10 years with weak ACE events (1986, 1988, 1989, 1995, 1998, 1999, 2003, 2009, 2010 and 2011).

Correlation analysis

The Pearson correlation coefficient is utilized to quantify the degree of correlation between two variables and can be computed as follows,

$$r=\frac{{COV}\left(X,Y\right)}{\sqrt{D\left(X\right)}\sqrt{D\left(Y\right)}}$$
(2)

Here, COV(X, Y) represents the covariance between variable X and variable Y, while D(X)and D(Y) denote the variances of variable X and variable Y, respectively.

Given the significant influence of ENSO on the tropical climate system and its potential to obscure the connection between WNP TC activities and Arctic SIC, we conduct partial correlation analysis to eliminate the confounding effects of ENSO, represented by the Niño 3.4 index. This analytical approach effectively disentangles the relationship between two variables from the confounding impacts caused by another or several other correlated variables69,70. The partial correlation between A and B after removing C is calculated using Eq. (3),

$${r}_{{AB},C}=\frac{{r}_{{AB}}-{r}_{{AC}}{r}_{{BC}}}{\sqrt{\left(1-{r}_{{AC}}^{2}\right)-\left(1-{r}_{{BC}}^{2}\right)}}$$
(3)

Where \({r}_{{AB}}\) represents the Pearson correlation coefficient between variable A and variable B, and so on.

The correlation between ACE and SIC, after removing the non-linear effects associated with ENSO, is examined by incorporating the empirical mode decomposition (EMD) method71,72. This method is specifically designed to handle non-stationary and nonlinear time-series data, and the results are in agreement with those of the partial correlation analysis.

Diagnosis of Rossby waves

To diagnose the propagation of Rossby waves, this study utilizes the T-N wave activity flux, a diagnostic tool well-suited for detecting large-scale quasi-stationary Rossby waves in a propagating manner73,74. The wave activity flux is defined as follows,

$$W=\frac{P}{2|U|}\left( \begin{array}{c}{\bar{u}\left({\psi }_{x}^{{\prime} 2}-{\psi }^{{\prime} }{\psi }_{{xx}}^{{\prime} }\right)+\bar{v}\left({\psi }_{x}^{{\prime} }{\psi }_{y}^{{\prime} }-{\psi }^{{\prime} }{\psi }_{{xy}}^{{\prime} }\right)} \\ {\bar{u}\left({\psi }_{x}^{{\prime} }{\psi }_{y}^{{\prime} }-{\psi }^{{\prime} }{\psi }_{{xy}}^{{\prime} }\right)+\bar{v}\left({\psi }_{y}^{{\prime} 2}-{\psi }^{{\prime} }{\psi }_{{yy}}^{{\prime} }\right)}\end{array}\right)$$
(4)

where U = (u, v) represents the horizontal wind, while ψ denotes the stream function. The horizontal and vertical lines in Eq. (4) indicate the climatological mean and perturbation values correspondingly.

Maximum covariance analysis

Maximum covariance analysis (MCA), also known as Singular Value Decomposition, is employed to identify the dominant covariance patterns between the ACE index and Arctic sea ice concentrations. The MCA analysis involves performing singular value decomposition on the temporal covariance matrix derived from the mid-August ACE index and mid-September Arctic sea ice concentrations75,76. The resulting pairs of singular vectors exhibit spatial patterns and time series that optimally capture coupling between these two fields.

Liang-Kleeman information flows

Lead-lagged correlation is commonly employed in atmospheric sciences to substitute for the causal relationship between two time series; however, it should be noted that correlation and causation are distinct concepts, and a strong correlation does not necessarily imply causation. To further elucidate the impact of ACE on SIC, this study utilizes the Liang-Kleeman information flow theory77,78 to depict the causal relationship between WNP ACE and APS SIC. Causality refers to the temporal rate at which information flow from one time series to another. Assuming a linear model, we estimate the maximum likelihood of the information flow (uncertainty transfer) from X2 to X1 is estimated as follows79:

$${T}_{2\to 1}=\frac{{C}_{11}{C}_{12}{C}_{2,d1}-{C}_{12}^{2}{C}_{1,d1}}{{C}_{11}^{2}{C}_{22}-{C}_{11}{C}_{12}^{2}}$$
(5)

where X2 and X1 represent two time series, and \({T}_{2\to 1}\) denotes the information flows from series X2 to X1. Cij is the covariance of time series Xi and Xj. Ci,dj is the covariance of time series Xi and {\(({X}_{j,n+1}-{X}_{j,n})/\partial t\), and ∂t is the time step. If \({T}_{2\to 1}=0\), then X2 is not the cause of X1, otherwise, it is causal (for either positive or negative information flow).

Model description and experimental design

To validate the teleconnection and causality between WNP TC activities and Arctic atmospheric circulation, we executed atmospheric general circulation model experiments using the Community Atmosphere Model version 6 (CAM6)80, which is the atmospheric component of the state-of-the-art fully-coupled climate model CESM2 (Community Earth System Model 2)81. CAM6 can be configured with various spatial resolutions. For our model setup, we utilize a global horizontal resolution of 0.9° × 1.25° and employ 32 vertical layers from the surface up to an altitude of 3.6 hPa.

Two experiments are performed to investigate the observed linkage between anomalous WNP TC activities and Arctic atmospheric circulation, and the underlying mechanism. One is the control experiment (CTRL), which is executed for 40 days starting from August 11, 2000, with default initial conditions for the F2000climo component set, and the external forcing at the level of the year 2000 is employed, which roughly represents the 40-year climatological mean. Another experiment is a sensitivity run, referred to as SEN, which replicates the CTRL experiment but incorporates sea level pressure anomalies in the WNP region (10–35°N, 110°–140°E) derived from the differences between strong and weak ACE events in mid-August during 1982–2020. By comparing the SEN and CTRL experiments, differences in the Arctic atmospheric circulation can be characterized between active and inactive TC activities years. Both experiments consist of 20 ensemble members with slightly perturbed initial conditions, and experimental results are based on ensemble mean. To generate different initial conditions, the parameter of perturbation limit was set in CESM2, and then 20 initial temperature perturbations were generated with the order of 1, 2, 3, …, 20 × 10–14, respectively.