Stratospheric polar vortex shapes Arctic surface climate via a radiative pathway

Abstract

The Arctic stratospheric polar vortex (SPV) is known to influence winter surface climate through dynamical coupling. Here, we demonstrate that beyond this established pathway, variations in SPV strength also modulate Arctic high clouds, inducing persistent radiative effects at the surface. This previously less well recognized mechanism shows that a strengthened SPV increases Arctic high-cloud cover, generating a positive net cloud radiative effect that amplifies Arctic Ocean warming and sea ice loss over the Barents-Kara Sea; a weakened SPV produces opposing effects. Our results establish this radiative pathway as one of the primary drivers of the winter-mean Arctic response to stratospheric variability. Numerical experiments confirm that this radiative pathway can cause up to 1.7 K of Arctic Ocean warming during the delayed phase of strong SPV events relative to climatology. Notably, the radiative influence operates more persistently than the transient dynamical response, offering enhanced predictive potential for subseasonal-to-seasonal Arctic surface conditions. These findings reveal a key radiative mechanism underlying SPV-driven Arctic climate variability, which is essential for understanding present and future winter Arctic changes.

This study discusses a stratosphere-to-surface radiative pathway in the Arctic, whereby stratospheric anomalies modulate high cloud cover through static stability, driving pronounced surface temperature and sea ice variations on subseasonal time scales.

Introduction

Surface weather and climate in the mid and high latitudes of the Northern Hemisphere are found to be closely related to the strength of the Arctic stratospheric polar vortex (SPV) in winter (December, January, and February), when SPV is most variable^1–11. On the one hand, the anomalous signals in the stratosphere, which drive a negative/positive stratospheric Northern Annular Mode (NAM), can extend downward into the troposphere and even impact the surface^1,3,12–23. On the other hand, the upward-propagating tropospheric waves can also be reflected downward by the variations of the SPV, which exerts an indirect dynamic impact on the tropospheric climate^24–34. One recent study also highlights that vertical compression or stretching of the polar air column during SPV weakening/strengthening explains observed surface pressure anomalies through potential vorticity redistribution as a kind of plunger effect³⁵. The dynamic downward effect (or plunger effect) of SPV is known to contribute to the predictability of the surface weather and climate in winter^36–42.

However, the dynamic downward impact is not always associated with the SPV anomalies^43–47. It has been found that even for sudden stratospheric warming events, which are characterized by large and rapid temperature increases in the winter polar stratosphere, approximately one third of them are not associated with a dynamic downward influence from the large-ensemble simulations^44,48. Moreover, while the dynamic downward effect typically induces Arctic warming under a weak SPV and cooling under a strong SPV, this direct dynamical forcing is found to be short-lived in our composite analysis. As illustrated in Fig. 1a, g, b, h, the downward extension of the NAM signal and its associated Arctic atmospheric warming (cooling) under a weak (strong) SPV extreme are sharply confined to a narrow time window, lasting only about 10 days before the surface response reverses sign. This reversal implies the involvement of other, slower processes beyond initial dynamical coupling. In this study, we find that the connection between the SPV and Arctic surface climate extends beyond dynamical coupling, also operating through a previously overlooked radiative downward pathway.

Fig. 1. The rapid and delayed downward effects of the Arctic stratospheric polar vortex on a daily timescale in ERA5.

Composited daily evolution of anomalies for negative (left column, a–f) and positive (right column, g–l) Northern Annular Mode (NAM) events. a, g Time-height cross-sections of the NAM index. b, h Time-height cross-sections of air temperature anomalies (Ta, K) averaged over the Arctic Ocean (70–90°N). c, i Composited time series of surface temperature anomalies (Ts, K) averaged over the Arctic Ocean. d, j Time-height cross-sections of cloud fraction anomalies (%) averaged over the Arctic Ocean. e, k Composited time series of cloud radiative forcing anomalies (Wm⁻², defined as the difference between all-sky and clear-sky net radiative fluxes at the top of the atmosphere) averaged over the Arctic Ocean. f, l Composited time series of sea ice fraction anomalies (%) averaged over the Barents-Kara Sea region (70–90°N, 20–110°E). Day 0 corresponds to the peak magnitude of the 100-hPa NAM index. In the time-height cross-sections (a, b, d, g, h, j), stippling indicates regions where anomalies are statistically significant at the 95% confidence level (Student’s t-test). In the time series (c, e, f, i, k, l), bold segments denote periods where anomalies are statistically significant at the 95% confidence level, and gray shading represents the ±2σ uncertainty (95% confidence interval).

The rapid and delayed downward effects of the SPV extremes

Figure 1 shows the downward effects of the SPV on a daily timescale using composite analysis of weak and strong SPV events in boreal winter with respect to the day of the peak magnitude in the 100-hPa NAM index (Day 0), using ERA5 reanalysis data (see “Methods” for event selection criteria). It is found that the downward extension of the NAM index and associated temperature anomalies from the lower stratosphere to the surface is sharply confined to a narrow window from around Day –5 to Day 5 (Fig. 1a, g, b, h), which is consistent with the dynamic downward extension of stratospheric NAM signals^1,46,49. Beyond Day 5, the composited dynamical NAM signal in the lower troposphere becomes weak (Fig. 1a, g), while the tropospheric and surface temperature responses after Day 5 have the opposite sign and persist much longer (Fig. 1c, i). This indicates the SPV possibly influence the Arctic Ocean through both rapid (the dynamic downward effect) and delayed processes (a less well recognized downward effect). The delayed process is likely associated with the responses of high clouds above 500 hPa (Fig. 1d, j), which are mainly caused by the changes in static stability in the upper troposphere associated with lower-stratospheric temperature changes^50–52 (Supplementary Fig. 1). These cloud responses align well with established findings in previous literature⁵³.

We find that the high clouds (above 500 hPa, Fig. 1d, j) significantly decrease/increase after Day –5 when the stratospheric anomalies reach the lower stratosphere for the negative/positive NAM events and can persistent for about 20 days, which is consistent with a previous study⁵³. Quantitatively, the peak changes in high-cloud cover reach approximately –4.5% for negative NAM events and +5.1% for positive NAM events, exhibiting a largely symmetric response to the magnitude of lower-stratospheric temperature anomalies (Fig. 1b, h).

During the polar night, when solar radiation is absent, cloud radiative forcing is dominated by longwave radiation, which is highly sensitive to the temperature contrast between the surface and cloud top^54,55. This makes variations in high clouds particularly influential on the longwave radiative budget. The decrease/increase in high clouds further leads to the negative/positive cloud radiative forcing at the top of the atmosphere (TOA) over 70–90°N (Fig. 1e, k), which can reach about –2.0 and 2.4 W m^–2, respectively. This negative/positive cloud radiative forcing persistently cools/heats the Arctic Ocean from Day –5 and results in significant Arctic Ocean cooling/warming from Day 10 to Day 30 (Fig. 1c–j). This cooling and warming associated with the negative and positive stratospheric NAM events can reach about –1.6 and 1.8 K, respectively. Such cooling and warming subsequently drive an increase or decrease in sea ice fraction by approximately 3.4% during Day 25–30 (Fig. 1f, l). Notably, this sea ice response is averaged over the Barents-Kara Sea (BKS) region (70–90°N, 20–110°E), identified as the center of the most robust thermal response. The sustained nature of this radiative downward effect underscores its potential to enhance the predictive skill for winter Arctic surface conditions and sea ice variability on subseasonal-to-seasonal (S2S) timescales.

To isolate the radiative contribution, we classified the stratospheric extremes into dynamically coupled and uncoupled cases on an event-by-event basis, inspired by previous frameworks^46,56. Specifically, we performed an event-by-event classification of the NAM extremes into dynamically coupled and uncoupled cases based on the vertical evolution of the NAM index (see “Methods” for detailed threshold criteria).

In uncoupled events, despite the complete absence of a continuous dynamical pathway to the surface, lower-stratospheric anomalies still induce an independent high-cloud response and subsequent long-lasting surface temperature anomalies (Supplementary Fig. 2), definitively proving the independent operation of the radiative pathway. In sharp contrast, coupled events (Supplementary Fig. 3) feature stronger lower-stratospheric temperature perturbations compared to uncoupled cases, which leads to a more intense cloud radiative forcing. Crucially, the actual dynamical downward influence in these coupled events is confined to a relatively narrow window (e.g., between Day –10 and Day 10). Therefore, the drastic surface anomalies observed well after Day 5 in coupled events are overwhelmingly driven by this amplified radiative effect, rather than a prolonged dynamical influence.

To further reconcile these timescales with the well-documented long-term persistence of major sudden stratospheric warmings (SSWs), we conducted an additional composite analysis of pure SSW events, categorizing them into cases with and without continuous dynamical downward coupling following Karpechko, et al.⁴⁶ (Supplementary Fig. 4). Note that Day 0 is defined as the onset of zonal wind reversal at 10 hPa here. In coupled SSW events, while tropospheric dynamical anomalies indeed persist for more than 50 days (Supplementary Fig. 4g–l), which is consistent with the previous studies^46,57,58, a remarkable “cancellation effect” occurs at the surface. The severe lower-stratospheric warming in these events strongly amplifies the high-cloud response (i.e., a reduction in high-cloud cover), generating a persistent radiative cooling that effectively counteracts the expected dynamical warming tendency over the Arctic Ocean. Consequently, the net surface temperature anomaly becomes largely insignificant. Conversely, in uncoupled SSW events, this high-cloud radiative cooling operates without dynamic interference, driving significant and sustained negative surface temperature anomalies well past Day 10 (Supplementary Fig. 4a–f). These results demonstrate that, locally over the Arctic Ocean, the radiative effect is substantial enough to either crucially offset prolonged dynamical thermal forcing or entirely dominate the surface thermal state.

Figure 2 illustrates the influence of weak and strong SPV on the Arctic surface temperature via both rapid (Day –5 to 5) and delayed (Day 10–30) processes. The rapid dynamical response is characterized by cooling/warming over northern Eurasia and concurrent warming/cooling in the Bering Sea region under weak/strong SPV conditions (Fig. 2a, b). Notably, even after the dynamical influence subsides, SPV variations continue to exert a significant delayed impact on Arctic surface climate (Fig. 2c, d). Specifically, a weak/strong SPV drives pronounced cooling/warming over the Arctic Ocean, most notably in the Barents and Kara Seas, with peak amplitudes of approximately –2.8 K and +3.5 K, respectively. The mean cooling/warming averaged over Day 10–30 reaches about –1.1 K and +1.1 K for weak and strong SPV events, respectively.

Fig. 2. The impacts of the Arctic stratospheric polar vortex on surface through both rapid and delayed processes in ERA5.

Geographic distributions of composite anomalies of 2-meter temperature for a, c negative and b, d positive Northern Annular Mode (NAM) events averaged over a, b day –5 to 5 and c, d day 10–30 based on ERA5 datasets (See “Methods”). The units are K. Regions with dots are the places where regressions have statistical significance levels higher than the 95% confidence level.

As shown in Fig. 1, high-cloud anomalies emerge around day –5, coincident with the downward propagation of stratospheric temperature signals into the lower stratosphere, and persist until approximately day 30. The spatial evolution of cloud radiative forcing and cloud fraction during this period is presented in Supplementary Fig. 5. The resulting cloud radiative forcing anomaly over the Arctic, driven primarily by changes in longwave radiation, is concentrated over Greenland and the Barents Sea, with peak values reaching approximately –7.2 W m⁻² and +6.5 W m⁻² during weak and strong SPV events, respectively (Supplementary Fig. 5a, b). These radiative anomalies are closely tied to these pronounced high-cloud changes compared with low-mid cloud changes (Supplementary Fig. 5c–h), and importantly, the spatial pattern of this cloud radiative forcing aligns closely with that of the surface temperature response (cf. Fig. 2c, d), highlighting a direct linkage between cloud-driven radiative perturbations and Arctic surface warming/cooling. Averaged poleward of 70°N, high-cloud fraction changes by about –3.2% and +1.2%, corresponding to cloud radiative forcing anomalies of –1.3 W m⁻² and +0.98 W m⁻² during weak and strong SPV periods, respectively. Supplementary Fig. 5 further reveals that, among all cloud layers, the high-cloud response is the most pronounced and dominates the radiative signal.

Overall, while the rapid dynamical response is evident in the initial phase, the subsequent delayed signal emerges as a dominant and more persistent feature in the Arctic Ocean surface evolution. This delayed response exhibits a strong spatial and temporal correspondence with high-cloud anomalies and their associated radiative effects, suggesting that cloud adjustments may play a central role in mediating this long-term surface impact of SPV variability.

The delayed radiative downward effect of SPV in winter

Having established the daily, event-by-event temporal evolution of the dynamic and radiative downward effects (Figs. 1 and 2), we now transition our focus to a seasonal-mean perspective. In this section, we provide a detailed analysis of the processes through which the SPV influences surface climate via cloud-mediated radiative pathways during boreal winter (December–January–February). For the boreal winter seasonal mean, the ST100 index³⁵ (inverse of 100-hPa polar-cap temperature anomalies) serves as a robust proxy for the broader thermal state of the Arctic lower stratosphere. As shown in Fig. 3a, variations in this single-level index effectively represent the temperature anomalies across the lower SPV, which are crucial for altering upper-tropospheric static stability. Variations in the lower-stratospheric temperature in the Arctic lead to changes in static stability in the upper troposphere and lower stratosphere (Fig. 3b), which further modify the cloud formation and cover in the upper troposphere. The correlation coefficients between the ST100 index and cloud fraction are significant above 500 hPa with a maximum value of about 0.90 at around 300 hPa over the Arctic (Fig. 3c).

Fig. 3. The cloud radiative downward effect of the Arctic stratospheric polar vortex based on satellite observations.

Upper panels: Vertical cross-section of correlation coefficients between the inverse of polar-cap average 100-hPa temperatures (ST100 index) and zonal mean a air temperature (Ta), b static stability (SS), c cloud cover (CC) in winter during 1979–2021 based on ERA5 reanalysis. The correlation coefficient of ±0.3 corresponds to the 95% confidence level for the 42 years. Bottom panels: Geographic distributions of the regression of d total CC from the Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF), e longwave (LW) cloud radiative effect (CRE, see “Methods”), f shortwave (SW) CRE, and g net CRE onto the ST100 index in winters during 2001–2021 over the Northern Hemisphere. The units are %/K in (d) and W m⁻²/K in (e–g). Regions with dots are the places where regressions have statistical significance levels higher than the 95% confidence level based on the Student’s t-test.

To further confirm the influence of the SPV on clouds, cloud fraction from the satellite observations (Clouds and the Earth’s Radiant Energy System Synoptic product) is regressed onto the ST100 index (Supplementary Fig. 6). The high and mid-high clouds (above 500 hPa) significantly increase with the increase in the ST100 index in the high latitudes (Supplementary Fig. 6a, b). The positive sensitivity of high and mid-high clouds, which is mainly located in the Arctic Ocean, Greenland, and northwestern Europe, can reach about 1.1 and 1.7%/K, respectively. In contrast, mid‑low and low clouds do not exhibit significant changes over the Arctic Ocean, but do increase with the ST100 index over northern Eurasia (Supplementary Fig. 6c, d), with maximum sensitivities of 1.0% K⁻¹ and 1.2% K⁻¹, respectively. Together, these results underscore the distinct vertical structure of cloud responses to SPV variations and highlight the dominant role of high‑level clouds in driving the associated radiative effects over the Arctic Ocean.

Satellite observations indicate that total cloud fraction over the Arctic increases significantly associated with the lower-stratospheric cooling (strengthening of the lower SPV) (Fig. 3d). The increase in total cloud fraction is mainly located over Greenland, the Barents Sea, Laptev Sea, Chukchi Sea, and Siberia, with a maximum value of 2.1%/K. Clouds can absorb and reemit the longwave radiation back to the surface. Thus, the increase in clouds leads to positive longwave cloud radiative effect (CRE, see “Methods”) over the North Eurasia, North Greenland, and the Arctic Ocean (Fig. 3e). The longwave CRE can reach about 1.9 and 1.6 W m^–2/K over the North Eurasia and the Arctic Ocean, respectively. Due to the absence of the solar radiation over the Arctic in winter, the shortwave CRE can be neglected (Fig. 3f). Therefore, the net CRE (Fig. 3g), which represents the indirect radiative effect of the SPV, is dominated by the longwave effect (Fig. 3e). This SPV’s radiative downward effect corresponds to a regional average of approximately 0.48 W m⁻² K⁻¹ over the Arctic Ocean. For the average over 70–90°N, the total cloud cover (TCC, unit: %) and net CRE (Unit: W m^–2) can be estimated by the ST100 index using the following linear regression equations:

$$TCC=0.79\times ST100+230.3$$ 1

$$CRE=0.48\times ST100+102.1$$ 2

Beyond cloud‑radiative effects, changes in stratospheric temperature, water vapor, and ozone associated with the SPV may also contribute to its downward radiative influence. Stratospheric temperature is a direct indicator of SPV strength, and its radiative impact is thus termed the direct radiative effect. Here, we first diagnose the radiative effect of stratospheric temperature perturbations during boreal winter, quantified as the regression of the stratospheric temperature radiative effect (integrated above 300 hPa using the radiative kernel method, see “Methods”) at both the TOA and the surface onto the ST100 index (Fig. 4a, d). While the TOA response is positive (Fig. 4a), peaking at about 0.43 W m⁻² K⁻¹ over the Arctic Ocean (polar‑mean: 0.32 W m⁻² K⁻¹ north of 70°N), the surface response is negative (Fig. 4d), with the strongest signal (–0.08 W m⁻² K⁻¹) over Greenland. This contrast indicates that the positive radiative effects at the TOA primarily heats the atmospheric column rather than the surface directly.

Fig. 4. The direct and indirect radiative effect of the Arctic stratospheric polar vortex in ERA5.

Geographic distributions of the regression of a, d stratospheric temperature (Tstrat) radiative effect (air temperature radiative effect integrated above 300 hPa), b, e stratospheric water vapor (SWV) radiative effect, and c, f cloud (CLD) radiative effect onto the inverse of 100-hPa polar-cap temperature anomalies (ST100 index) in boreal winter at the a–c top of the atmosphere (TOA) and d–f surface (SFC) during 1950–2021 from the ERA5. The units are W m^–2/K. Regions with dots are the places where regressions have statistical significance levels higher than the 95% confidence level based on the Student’s t-test.

Radiative‑active gases in the Arctic stratosphere, such as water vapor and ozone, also vary with SPV strength. As shown in Supplementary Fig. 7a, Arctic stratospheric water vapor increases consistently under a strengthened winter SPV. In contrast, stratospheric ozone decreases markedly (Supplementary Fig. 7b), consistent with earlier studies^59–61. Over the Arctic Ocean, SPV‑related anomalies reach approximately 0.009 ppmv K⁻¹ for water vapor and –7.6 DU K⁻¹ for total column ozone. As greenhouse agents, increased water vapor exerts a slight warming influence, whereas ozone reduction promotes cooling. The radiative effect of stratospheric water vapor, derived from radiative kernels (integrated above 300 hPa, see “Methods”), peaks near northern Greenland at about 0.01 W m⁻² K⁻¹ at the TOA (Fig. 4b), with a much weaker surface signal of only 0.001 W m⁻² K⁻¹ (Fig. 4e). Compared to other radiative contributions, the influence of stratospheric water vapor is negligible. Meanwhile, the ozone‑mediated radiative effect under a strengthened SPV tends to cool the Arctic, and thus cannot account for the delayed warming response identified above. Given its minor contribution to the net warming, we do not discuss the ozone effect further here.

Diagnostics based on ERA5 reanalysis indicate that the cloud radiative effect, both at the TOA and the surface, far outweighs the contributions from stratospheric temperature and water vapor, and likely serves as the dominant driver of Arctic Ocean warming (Fig. 4c, f). We find that the cloud radiative effect is positive across the Arctic Ocean at the TOA, reaching a maximum of about 1.3 W m⁻² K⁻¹. Similarly, at the surface, the cloud radiative effect is largely positive over the Arctic Ocean, with a comparable peak magnitude of 1.3 W m⁻² K⁻¹, consistent with satellite observations (Fig. 3g). This pattern underscores that, unlike the radiative contributions from stratospheric temperature and water vapor, the cloud effect acts primarily to heat the surface. Consequently, while TOA radiation budget analysis shows that a strengthened wintertime Arctic SPV warms the Arctic Ocean through both direct radiative effects (from stratospheric temperature perturbations) and indirect forcing (via upper‑tropospheric high‑cloud changes), the surface radiation budget reveals that cloud radiative effects dominate the net surface heating.

To quantify the impact of the SPV and its radiative effect for winter average, we performed a composite analysis of ERA5 data in winter from 1950–2021. We contrasted winters with strong versus weak SPV states, as identified by the ST100 index (19 winters per category; see “Methods”), to isolate anomalies in surface temperature, cloud radiative effects, and cloud cover.

The analysis reveals that under a strengthened SPV, associated with a pronounced cooling of up to 12.3 K at 100 hPa (Fig. 5a), surface temperature exhibits warming over Eurasia and cooling over northern North America and Greenland in winter (Fig. 5b). This pattern congruent with the positive phase of the Arctic Oscillation (AO) are consistent with the results in previous studies^{2,8,15,62–64}. However, over the Arctic Ocean, where dynamical coupling under a strengthened SPV would typically enhance polar cold-air retention and induce cooling, we instead observe pronounced warming, particularly over the Barents and Chukchi Seas, with peak anomalies reaching about 3.9 K. The Arctic Ocean mean warming amounts to 0.54 K. This warming is likely linked to cloud‑radiative heating, as the cloud radiative forcing induced by SPV strengthening is significantly positive across the Arctic, averaging 1.8 W m⁻² north of 70°N (Fig. 5c). The positive cloud forcing is dominated by longwave radiation and is closely linked to a marked increase in high cloud cover (Fig. 5d–f). Over the Arctic Ocean, high-, mid-, and low-level clouds increase by 5.7%, 2.3%, and 1.3%, respectively. In both magnitude and spatial distribution, high‑cloud changes clearly drive the cloud radiative signal, in line with the cloud response following stratospheric extreme events (Supplementary Fig. 5). During the polar night, the absence of shortwave radiation makes the surface longwave budget—and thus surface temperature—particularly sensitive to high-cloud variations, which exhibit the largest and most statistically significant response to SPV-induced stability shifts among all cloud layers over the Arctic Ocean (Fig. 5d–f).

Fig. 5. Surface climate impacts of the Arctic stratospheric polar vortex (SPV)’s downward effect in winter in ERA5.

Geographic distributions of composite anomalies (Strong minus Weak SPV events) of a temperature at 100 hPa (T100), b 2-meter temperature (T2m), c cloud radiative forcing (CRF) at the top of the atmosphere (TOA), d high cloud cover (HCC), e middle cloud cover (MCC), and f low cloud cover (LCC) over the Northern Hemisphere in winter for the period 1950–2021 using ERA5 data. The units are K, W m^–2, and % in (a, b), (c), and (d–f), respectively. Regions with dots are the places where regressions have statistical significance levels higher than the 95% confidence level. The statistical significance of the composite anomalies is assessed by the Student’s t-test.

Consistently, these seasonal-mean responses show that a strong SPV drives net Arctic Ocean warming, whereas a weak SPV promotes net cooling (Supplementary Fig. 8). Given that the established dynamic downward effect of a strong SPV typically favors local surface cooling over the Arctic Ocean basin, the emergence of this net warming suggests that the radiative heating effect ultimately outweighs the opposing dynamical cooling over the course of the winter. This resulting pattern aligns with the dominant, long‑lasting influence of the delayed radiative pathway identified over the Arctic Ocean identified in our daily analysis (Figs. 1 and 2).

It is worth noting that the geometric configuration of the SPV—such as the displacement or splitting of the vortex during sudden stratospheric warming events—can significantly influence the symmetry and location of dynamical downward coupling. While our radiative pathway is primarily controlled by the polar-cap average stability changes in the lower stratosphere, the vortex geometry might further modulate the longitudinal distribution of high-cloud responses. Future work focusing on these morphological distinctions will provide deeper insights into the spatial precision of stratospheric-forced radiative feedbacks.

The rapid and delayed downward effects of the SPV extremes in simulations

To validate the influence of the SPV on Arctic climate, we employed the nudged experiments following Zhang et al.⁶⁵ adapted from the principles of the Stratospheric Nudging And Predictable Surface Impacts (SNAPSI) protocol⁶⁶ (see “Methods”). Using the Whole Atmosphere Community Climate Model (WACCM), we nudged atmospheric conditions toward the ERA5 reanalysis dataset (four times per day) in stratosphere (above 200 hPa) over the period 1979–2018. This design isolates the stratospheric signal from unrelated tropospheric variability. Crucially, the analysis of the WACCM simulation data, including the identification of SPV events and the calculation of daily anomalies, was conducted using the exact same methodology as applied to the ERA5 reanalysis. The model results demonstrate that experiments driven solely by stratospheric variability are capable of reproducing both the rapid dynamical response and the subsequent delayed radiative response observed in reanalysis (Fig. 6). This confirmation indicates that these two-phase surface impacts originate directly from SPV extremes rather than from other confounding factors.

Fig. 6. Daily evolution of the dynamic and radiative downward effects of the Arctic stratospheric polar vortex in Whole Atmosphere Community Climate Model (WACCM) simulations.

Composited daily evolution of anomalies for negative (left column, a–f) and positive (right column, g–l) Northern Annular Mode (NAM) events. a, g Time-height cross-sections of the NAM index in the WACCM6 ensemble. b, h Time-height cross-sections of air temperature anomalies (Ta, K) averaged over the Arctic Ocean (70–90°N). c, i Composited time series of surface temperature anomalies (Ts, K) averaged over the Arctic Ocean. d, j Time-height cross-sections of cloud fraction anomalies (%) averaged over the Arctic Ocean. e, k Composited time series of cloud radiative forcing anomalies (W m⁻², defined as the difference between all-sky and clear-sky net radiative fluxes at the top of the atmosphere) averaged over the Arctic Ocean. f, l Composited time series of sea ice fraction anomalies (%) averaged over the Barents-Kara Sea region (70–90°N, 20–110°E). Day 0 corresponds to the peak magnitude of the 100-hPa NAM index. In the time-height cross-sections (a, b, d, g, h, j), stippling indicates regions where anomalies are statistically significant at the 95% confidence level (Student’s t-test). In the time series (c, e, f, i, k, l), bold segments denote periods where anomalies are statistically significant at the 95% confidence level, and gray shading represents the ±2σ uncertainty (95% confidence interval).

The WACCM ensemble simulations effectively reproduce the daily evolution of the two-way SPV influence on Arctic climate, consistent with the ERA5-based composite results. In WACCM (event selection criteria following Fig. 1), the downward propagation of the NAM index and associated temperature anomalies from the lower stratosphere to the surface is sharply confined to a narrow window from Day –5 to 5, reflecting the rapid dynamical downward effect (Fig. 6a, g, b, h). Interestingly, while a secondary, weaker NAM downward propagation appears around Day 20 during weak SPV events (Fig. 6a), this later dynamical pulse does not induce a corresponding downward propagation of temperature anomalies. Following the initial rapid phase, a delayed and persistent response emerges, characterized by opposite-sign temperature anomalies in the troposphere and at the surface after Day 5 (Fig. 6c, i), which is attributed primarily to cloud-mediated radiative processes.

The model captures the increase/decrease in high clouds (above 500 hPa) in response to negative/positive NAM events (Fig. 6d, j), with peak anomalies of approximately −7.1% and 9.0%, respectively. The simulated cloud radiative forcing at the TOA reaches about –0.83 W m⁻² for negative events and 1.1 W m⁻² for positive events (Fig. 6e, k), in qualitative agreement with the observed longwave-dominated forcing during the polar night. This forcing, in turn, drives Arctic Ocean surface cooling/warming on the order of –0.96 K and 1.7 K for negative and positive events, respectively (Fig. 6c, i), and leads to corresponding BKS ice anomalies of about 0.3% and –0.5% during Day 25–30 (Fig. 6f, l).

While the model qualitatively reproduces the timing, sign, and vertical structure of the observed response, some quantitative discrepancies are noted. For instance, WACCM tends to overestimate the high-cloud response while underestimating the magnitude of cloud radiative forcing and surface warming/cooling. Notably, both the climatological mean and the sensitivity of Arctic clouds to SPV variations in WACCM exceed those in ERA5 by roughly a factor of two to three (Supplementary Fig. 9). These biases likely reflect uncertainties in cloud microphysics and radiative schemes over the Arctic in WACCM. Nevertheless, the overall consistency between the simulated and observed temporal evolution reinforces that the delayed radiative pathway is a robust and physically plausible mechanism through which the SPV influences Arctic surface climate on subseasonal to seasonal timescales.

Figure 7 illustrates the dual-phase surface response to SPV variability in WACCM simulations, mirroring the structure of Fig. 2. The rapid response (Day –5 to 5) exhibits a dipole pattern of cooling/warming over northern Eurasia and concurrent warming/cooling in the Bering Sea region under weak/strong SPV conditions (Fig. 7a, b). In the delayed phase (Day 10–30), pronounced and persistent cooling/warming emerges over the Arctic Ocean, most notably in the Barents and Kara Seas, with peak anomalies reaching approximately –2.4 K and +3.6 K, respectively (Fig. 7c, d). Mean Arctic Ocean cooling/warming during this period amounts to –0.6 K and +1.3 K, consistent in sign, timing, and spatial structure with the observed response. Supplementary Fig. 10 presents the cloud response in WACCM during the delayed period, closely matching the ERA5 results. The simulated cloud radiative forcing, dominated by high-cloud changes, peaks over Greenland and the Barents Sea, aligning in both magnitude and spatial pattern with the results in ERA5 (Supplementary Fig. 5).

Fig. 7. The Arctic stratospheric polar vortex (SPV)’s rapid and delayed surface impacts in Whole Atmosphere Community Climate Model (WACCM) simulations.

Geographical distributions of composite 2‑meter temperature anomalies for a, c negative and b, d positive Northern Annular Mode (NAM) events, averaged over a, b days –5 to 5 and c, d days 10–30. Anomalies are derived from WACCM nudged simulations adapted from the principles of the Stratospheric Nudging And Predictable Surface Impacts (SNAPSI) protocol. Units are K. Stippling indicates regions where anomalies are statistically significant at the 95% confidence level (Student’s t‑test).

Based on the daily evolution of surface temperature anomalies in both ERA5 reanalysis (Fig. 2) and WACCM6 simulations (Fig. 7), the most pronounced warming and cooling signals are consistently centered over the BKS region (70–90°N, 20–110°E). Given this significant thermal response, we specifically focus on the sea ice fraction anomalies averaged over the BKS sector (Fig. 1e, j and 6e, j). Our results demonstrate that the SPV’s radiative pathway exerts a substantial influence on BKS sea ice. By providing an additional source of surface forcing independent of transient dynamical effects, this radiative mechanism may enhance the interannual variability of BKS ice.

Overall, the WACCM simulations capture both the rapid dynamical response and the subsequent delayed, cloud‑dominated radiative pathway. The consistency in temporal evolution and spatial coherence between simulated and ERA5-based fields reinforces that cloud adjustments represent a key mechanistic link through which the SPV influences Arctic surface climate on subseasonal to seasonal timescales.

Discussion

Our findings illustrate that the impact of SPV variability on the Arctic may involve two pathways: the traditional dynamic pathway and a less well recognized radiative pathway. This dual-pathway framework effectively explains why the SPV can exert a profound influence on surface climate even when dynamic downward coupling is absent. Specifically, the radiative downward effect of a strong SPV consistently warms the Arctic Ocean. In contrast, the dynamic downward effect typically triggers a positive AO pattern, leading to warming over Northern Eurasia but localized cooling in the Arctic Ocean. Due to the dominant and long-lasting nature of the delayed radiative pathway, it emerges as the primary driver of the Arctic-mean climate response in seasonal winter averages.

On daily to subseasonal timescales, the radiative response, though delayed by approximately 10–30 days, exerts a stronger and more persistent influence on the Arctic Ocean than the transient dynamical signal. This persistence provides a critical physical basis for improving predictions of winter surface temperature and sea-ice variability at S2S lead times. During the polar night, clouds—especially the stratospherically-sensitive high clouds—act as an adjustable thermal “blanket” over the Arctic, regulated by the strength of the SPV. Our findings reveal that the strength of the SPV directly modulates this “blanket”, either thickening it to amplify surface warming or thinning it to mitigate such effects (Fig. 8). This insight calls for a recalibration of our understanding of Arctic climate change. While S2S forecasting systems inherently include cloud-radiative schemes, our comparison between ERA5 and WACCM6 reveals a critical paradox: models may overestimate macroscopic high-cloud fraction changes while simultaneously underestimating their surface radiative impact due to microphysical biases. Recognizing this radiative pathway as a distinct delayed process offers a strategic framework for prediction. Utilizing the peak 100-hPa NAM anomaly (Day 0) provides a precise temporal anchor for statistical forecasting, while for advancing artificial intelligence (AI)-driven models, Arctic high clouds should be treated as essential training features to capture full stratosphere-to-surface predictability. Accurately representing this coupling is therefore essential for bridging the ‘gap’ in predictive skill and fostering effective climate adaptation in the Arctic.

Fig. 8. A schematic depiction of the radiative effect of the Arctic stratospheric polar vortex (SPV).

A strengthened SPV increases the high and middle cloud cover over the Arctic, which induces a positive net cloud radiative effect at the surface, amplifying Arctic Ocean warming and accelerating sea ice melt. In contrast, a weakened SPV has the opposite effect. The units are W m^–2 for the radiation flux. The arrows indicate the upward and downward longwave radiation flux. The radiation fluxes are computed based on the composite analysis in Fig. 1.

Methods

ERA5 reanalysis and observations

Monthly mean air temperature, 2-meter temperature, meridional winds, radiation fluxes, water vapor, and cloud fraction are obtained from ERA5 reanalysis over 1950–2021. ERA5, which is the fifth generation ECMWF reanalysis for the global climate and weather, combines model data with observations from across the world into a globally complete and consistent dataset⁶⁷. Compared to the 3D-Var scheme used in earlier versions (such as ERA-Interim), ERA5 employs a significantly more advanced 4D-Var assimilation scheme, which allows for a more consistent integration of observations and model physics at a 1-hourly time step. The datasets have a horizontal resolution of 0.25° × 0.25° with 37 pressure levels from 1000 hPa to 1 hPa.

To analyze the impact of SPV on the clouds and cloud radiative forcing, monthly mean clouds and cloud radiative forcing during the period March 2000–December 2021 are obtained from Clouds and the Earth’s Radiant Energy System (CERES). We use high (above 300 hPa), mid-high (500–300 hPa), mid-low (700–500 hPa), and low (below 700 hPa) clouds from CERES Synoptic (SYN1deg) product. CERES SYN1deg provides monthly mean clouds on a 1° × 1° grid^68–70. Monthly mean cloud radiative forcing, $\mathrm{\Delta }CRF$, defined as the difference in the radiative fluxes between all-sky and clear-sky conditions are obtained from CERES Energy Balanced and Filled (EBAF) edition 4.1⁷¹. All the radiation fluxes are defined to be downward positive.

Definitions of the ST100 index

The strong (weak) SPV winters with dynamic downward effect are identified using the downward phase propagation of the winter-average NAM from the lower stratosphere to surface. The strength of the SPV is defined as the inverse of polar-cap average 100-hPa temperatures (ST100 index), calculated as anomalies relative to the multi-year climatology following Baldwin, et al.³⁵. This ST100 index (Supplementary Fig. 11) is used here because the lowermost stratospheric temperature has the largest impact on the upper tropospheric static stability and high clouds⁵¹. Thus, the positive/negative ST100 index represents a stronger/weaker SPV with respect to the climatological average.

Definition of Radiative effect

To quantify the radiative effects of the SPV, we calculate the radiative forcing at both the TOA and surface using the radiative kernel method introduced by Huang et al.⁷². The radiative kernel, $K_X$, is pre-calculated by a partial perturbation method using a rapid radiative transfer model (RRTM)⁷³ for temperature (T), water vapor (q), and albedo (a). For the non-cloudy climate variables, the radiative feedback is computed as

$${\mathrm{\Delta }}_{X}R=K_X\mathrm{\Delta }X,$$

for X = T, q, or a

Stratospheric temperature is a direct indicator of SPV strength, and its radiative impact is thus termed the direct radiative effect. The direct radiative effect of the SPV is quantified as the regression of the stratospheric temperature radiative effect ${\mathrm{\Delta }}_{T}R$ (integrated above 300 hPa) onto the ST100 index.

Because of the cloud masking effects of temperature, water vapor, and albedo feedback, the cloud radiative effect is calculated following the Equation 25 in Soden et al.⁷⁴:

$${\mathrm{\Delta }}_{cld}R=\mathrm{\Delta }CRF+\left(K_T^0-K_T\right)\mathrm{\Delta }\mathrm{T}+\left(K_q^0-K_q\right)\mathrm{\Delta }\mathrm{q}+\left(K_a^0-K_a\right)\mathrm{\Delta }\mathrm{a}+\left(G^0-G\right),$$

where $\mathrm{\Delta }CRF$ is the cloud radiative forcing defined as the difference in the radiative fluxes between all-sky and clear-sky conditions, K⁰ and K are the clear-sky and all-sky kernels, and G⁰ and G are the clear-sky and all-sky forcing. The indirect cloud radiative effect of the SPV is computed by the regression of cloud radiative effect onto the ST100 index in boreal winter.

Composite analysis

In Fig. 5, the composite anomalies are computed by the difference between the strong and weak polar vortex winters over 1950–2020 using the ERA5. We define strong (weak) polar vortex winters as years when the seasonal-mean ST100 index is less than –0.8 (greater than +0.8) times its standard deviation. Under this criterion, each category comprises 19 winters (see Supplementary Fig. 11 for the specific years identified).

The downward effects of the SPV extremes on a daily timescale

To understand the temporal evolution of the downward effects, we constructed composite evolutions centered on the day of the peak magnitude in the 100-hPa NAM index⁷⁵ (defined as Day 0). Following the objective criteria established by Zhang, et al.⁴⁹, we first identified extreme stratospheric NAM events in which the 10-hPa normalized NAM index exceeds ±1.5 for at least 5 consecutive days. Notably, two events with the same sign at 10 hPa are considered as different cases if they are separated by at least 10 days. We identified a total of 34 negative NAM events and 24 positive NAM events (as detailed in Table 1 of Zhang et al.⁴⁹). Note that a total of seven events occurring in April were excluded from this list, since our study focuses exclusively on the winter season. Following the objective criteria established by Zhang et al.⁴⁹, an event is further classified as being accompanied by dynamical downward coupling if the normalized NAM index of a selected case exceeds ±1.0 at 100 hPa and is subsequently followed by an index exceeding ±1.0 for more than two consecutive days at 500 hPa; otherwise, it is classified as an uncoupled event. The composite anomalies were computed by subtracting the multi-year (1979–2020) daily climatology from the daily values of these selected positive and negative NAM events. This lead-lag composite approach allows us to effectively disentangle the rapid dynamical response from the subsequent, prolonged radiative influence on sub-seasonal timescales.

Numerical experiments

Models

The present study employed the Whole Atmosphere Community Climate Model version 6 (WACCM6), key component of the Community Earth System Model version 2.2 (CESM2.2), a comprehensive coupled climate system model. To isolate the physical and radiative impacts of stratospheric dynamics from ozone-related feedbacks, we utilized the Specified Chemistry version of WACCM6 (SC-WACCM)^76,77. In this configuration, the stratospheric ozone concentrations were fixed at the climatological levels of the year 2000, ensuring that the simulated responses were driven by the prescribed stratospheric variability rather than ozone-induced radiative perturbations⁷⁸.

WACCM6 can well simulate the atmospheric dynamic processes in the stratosphere and the mesosphere⁷⁹. WACCM6 utilizes a finite-volume dynamical core with 70 vertical layers, ranging from the Earth’s surface to approximately 140 km altitude (5.1 × 10⁻⁶ hPa). It has a horizontal resolution of 0.9° × 1.25° (latitude × longitude). The vertical resolution in the tropical upper troposphere-lower stratosphere and lowermost stratosphere is between 1.1 and 1.4 km. WACCM6 is coupled to the Parallel Ocean Program Version 2 (POP2)⁸⁰, the Community Ice CodE Version 5 (CICE5)⁸¹, and the Community Land Model Version 5 (CLM5)⁸². Thus, the sea ice responses to the SPV changes can be obtained.

Experiment design

To isolate the downward effects of the SPV, we performed stratospheric nudging experiments adapted from the principles of the SNAPSI protocol^66,83. First, a 15-year free-running control simulation (R0) was performed without nudging; the initial 5 years were discarded as spin-up, and the remaining 10 years were used to supply initial conditions for the nudged experiments. Following Zhang, et al.⁶⁵, we then performed historical nudging simulations (R1) to diagnose the influence of stratospheric processes, employing a dynamical-core-independent nudging scheme⁸⁴. Nudging was implemented according to:

$$F=-W\frac{\left(X-X_{ref}\right)}{\tau }$$

Where X_ref is reference meteorological state from ERA5 reanalysis at the next update step, τ is the relaxation timescale, and W is the nudging coefficients in the range [0.0, 1.0]. A value of W = 1.0 forces the model state to match ERA5 exactly.

In R1, the global stratospheric fields (zonal wind u, meridional wind v, and temperature T) on model levels between 1.2 hPa and 197.9 hPa were nudged toward ERA5 four times daily during 1979–2018, with W = 1.0 in these layers. Below 197.9 hPa, the nudging coefficient decreases linearly to zero at 273.9 hPa to ensure a smooth transition into the free‑running troposphere. Each R1 integration spans 8 months (October to May) and uses a distinct initial atmospheric state (1 October) taken from years 6–15 of R0, yielding 10 ensemble members. Every member simulates 39 boreal winters (1979–2018). The daily evolution of SPV downward effects was then assessed using the same composite approach applied to ERA5 reanalysis.

To verify that the 197.9-hPa nudging depth does not introduce artificial tropospheric constraints, we performed a sensitivity ensemble with nudging restricted to the middle and upper stratosphere (full strength to 50 hPa, terminating at 70 hPa). Although this shallower nudging effectively replicates the delayed radiative downward effects after Day 0—thereby validating the physical robustness of the mechanism—it substantially underestimates the initial thermal trigger in the lower stratosphere (70–200 hPa) prior to Day 0 (not shown). Given that an accurate representation of this lowermost stratospheric temperature anomaly is essential for modulating upper-tropospheric clouds, we retained the 197.9-hPa configuration to faithfully capture the observed magnitude of the forcing.

Author contributions

F. Xie and C. Zhao conceived the study and designed the research. Y. Xia, F. Xie, C. Zhao wrote the paper and analyzed the results. Y. Hu, Y. Huang, J. Bian. F. Luo, and L. Zhou interpreted the results and discussed their implications.

Peer review

Peer review information

Nature Communications thanks Marina Friedel and Ryan Williams for their contribution to the peer review of this work. A peer review file is available.

Data availability

The ERA5 reanalysis can be accessed at https://cds.climate.copernicus.eu/datasets?q=era5. The CERES EBAF Edition 4.1 is available at https://opendap.larc.nasa.gov/opendap/CERES/EBAF/Edition4.1/contents.html. The CERES SYN1deg Edition 4.1 can be accessed at https://catalog.data.gov/?q=CERES+SYN1deg + &sort=relevance. The radiative kernels used in this study are available at https://data.mendeley.com/datasets/3drx8fmmz9/1. All computer codes generated during this study are available from the corresponding authors upon request. Data available via Figshare at 10.6084/m9.figshare.29375246.

Code availability

All code used in this work is available from the corresponding authors on request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72698-w.