We first characterize the SSW in the stratosphere based on the MERRA-2 reanalysis data output every 3 hours and compare with two non-SSW winters. Figure 1 shows the evolution of the December 2020–January 2021 major SSW at 10 hPa (30 km) in terms of the (a) zonal mean temperature T ̄ and (b) zonal wind u ̄ (red lines). They are compared with those for two non-SSW winters, December 2019–January 2020 (black lines) and December 2021–January 2022 (gray lines). The temperature is plotted at 60^◦N and 90^◦N and the zonal wind at 60^◦N.

Figure 1 demonstrates that, after the onset of the warming on 1 January 2021 (day zero; vertical dashed lines), the Northern polar temperature increases by 40 K–from about 210 K to 250 K, peaking on 5 January 2021. During the ascending phase of the warming, u ̄ at 60^◦N gradually changes to westward–from ∼ 30 m s⁻¹ at the onset of the SSW to about −10 m s⁻¹ at the peak phase, demonstrating a reversal of the mean flow direction. The recovery phase of the SSW is relatively long, during which the temperatures remain elevated and u ̄ is westward compared to the pre-SSW period in December 2020 and during the non-SSW seasons in January 2020 and 2022. It is noticeable that the December 2019 (non-SSW winter) and December 2020 (pre-SSW winter) exhibit some minor differences in mean winds, owing, partially, to interannual variations in planetary wave activity and behavior of large-scale internal waves.

In order to study thermospheric variations observed by ICON and GOLD during the major SSW, we selected a period centered around the onset of the SSW, i.e., from 6 December 2020 to 26 January 2021 (hereafter called “SSW winter”). The results are compared to those for the non-SSW winter (6 December 2019 to 26 January 2020) in the rest of the paper. The winter of December 2021 to January 2022 was also a non-SSW winter. However, the solar and geomagnetic conditions during this time were greater and more variable (Section 2.3). Since the thermospheric temperatures and winds are closely related to this activity, we only consider the comparison to the non-SSW winter of 6 December 2019 to 26 January 2020. While between ∼90–109 km in the atmosphere both daytime and nighttime ICON wind data are available, only daytime winds are available above ∼109 km up to about 210 km. Therefore, we use only daytime ICON winds from 90–200 km to produce a uniform analysis of the mean wind variations.

Figure 2 shows the geomagnetic and solar conditions in terms of the (a) daily mean A _p index (b) F _10.7 cm solar radio flux, respectively, during the non-SSW (December 2019 - January 2020), the SSW (December 2020 - January 2021), and non-SSW (December 2021 - January 2022) northern winters. The data are from the Geomagnetic Observatory Niemegk, GFZ German Research Centre for Geosciences, Potsdam, Germany (Matzka et al., 2021).

While the space weather conditions are overall relatively quiescent during the winters of 2019–20 and 2020–21, the solar activity during the SSW winter is slightly higher than during the non-SSW winter of 2019–20, since the former corresponds to the ascending phase of the solar activity (Solar Cycle 25), while the latter is the solar minimum (end of Solar Cycle 24).

The observed solar activity during the non-SSW winter of 2021–22 exhibited a significant increase in intensity and variability compared to the previous two winters. This period was characterized by elevated levels of the F _10.7 cm solar radio flux. Given the direct relationship between solar radiation levels and thermospheric heating, such a substantial rise in solar activity significantly influences the energy balance within Earth’s upper atmosphere. Concurrently, the geomagnetic activity during this period demonstrated increased variability. The increased solar irradiance acts as a dominant external forcing mechanism, which could potentially overshadow the more subtle dynamical and thermal responses of the thermosphere to SSWs. In light of these considerations the 2021/2022 non-SSW winter was excluded from the analysis.

Although the magnetic activity is comparable in both the SSW winter and the winter of 2019–20, it exhibits day-to-day variability with occasionally higher activity during the SSW winter than the non-SSW winter, reaching A _p ∼ 12 (K _p ∼ 3 −). In order to reduce the impact of these elevated space weather conditions on our analysis of temperature variations, we have excluded geomagnetically disturbed days in temperature analysis. Accordingly, in the analysis of GOLD temperature (Figure 11) we have removed 5 days in the non-SSW winter and 11 days in the SSW winter, corresponding to days with A _p > 7 (K _p > 2). The grey shading marks the range of the data used for ICON and GOLD analysis (6 December - 26 January in both winters). Before 6 December 2019, ICON horizontal winds are not available and in all our analysis, we have excluded days before 6 December to be consistent in our comparison of the different winters.

ICON’s scientific operation started in December 2019, observing the low-to middle-latitude thermosphere between 10^◦S and ∼40^◦N (Immel et al., 2018) and lasted till 25 November 2022 when contact with the spacecraft was lost. Data are available after 6 December 2019, therefore, for both winters, we use the 52-day data from 6 December to 26 January centered around 1 January 2021 (the SSW onset). Here we use the wind velocity observations performed by the Doppler shift measurements of the green line (λ = 557.7 nm). ICON version 5 green line daytime horizontal winds have been binned daily from 90 to 200 km altitude, including data with solar zenith angles less than 80^◦ (i.e., χ < 80^◦) to exclude solar terminator effects. Also, bins with less than 50 data points have been excluded to avoid low statistical significance. We concentrated our analysis on the Northern Hemisphere low-to middle-latitudes (0^◦–40^◦N).

Recently, validation of the ICON/MIGHTI green line winds have been successfully performed (Makela et al., 2021) and the mean horizontal winds and the associated circulation patterns in the Northern Hemisphere during solstice were characterized (Yiğit et al., 2022).

The associated latitude-local time coverage by ICON at ∼100 km is shown in Figure 3 as well as the latitude-longitude coverage at around 151 km in Figure 4. ICON’s latitude-local time coverage varies from month to month. Therefore, it would be inconsistent to compare the monthly mean fields. It is for this reason we have analyzed the altitude-time variations considering only daytime measurements. For the daytime winds, nearly all longitudes except a narrow band at 300^◦ longitude from 0–10^◦N is not covered during the analyzed months. However, the latitude-longitude coverage between the non-SSW and the SSW winters are very similar. The typical uncertainties in MIGHTI green line wind measurements are 8.7–10 m s⁻¹ (Englert et al., 2017).

For each day, ICON wind files include around 2,100–2,200 profiles, measured every 22–80 s. This gives more than 110,000 profiles over 52 days for daytime and nighttime together. Since we included only daytime winds with solar zenith angles χ < 80, we included approximately half of the number of total profiles, i.e., 52,000 profiles for the 52-day period studied in a given winter. For the daytime data included here, the profiles are available typically every 27–37 s. In order to improve the data quality in our analysis we have excluded data with the quality flag less (called “wq”) than 0.5, i.e., wq < 0.5, and winds speeds with magnitudes greater than 350 m/s.

In the analysis of the longitudinally averaged daytime zonal and meridional winds in Results (Figures 6, 7), the ICON winds are averaged within each latitude sector, i.e., 0–20^◦ and 20–40^◦N, in 5 km × 1 day altitude-time bins, by including daytime data with χ < 80^◦ and all longitudes. Since southern latitudes were not included, our analysis exclude the South Atlantic Anomaly.

GOLD observes the Far Ultraviolet (FUV) spectrum of Earth’s atmosphere at geostationary orbit, from 0610 to 0040 Universal Time (UT) every day, providing, among others, daytime thermospheric temperatures near 150 km at low- and midlatitudes (0^◦ to ±60^◦), depending on the solar zenith angle.

The temporal resolution of the GOLD temperatures is about 30 min. The longitudes covered by GOLD range from −100^◦ to 10^◦. Similar to ICON, we have used 52 days of GOLD data centered around 1 January. Thermospheric temperatures are retrieved from the daytime disk scan measurements. The effective disk neutral temperatures (T _disk ) are derived from the N ₂ Lyman-Birge-Hopfield (LBH) emission profile at a height of approximately 150 km. The T _disk data product is created from spatial-spectral image cubes from the disk scans (Level 1C data). These pixels are binned 2^◦ × 2^◦ spatially, resulting in a data product that has a spatial resolution (nadir) of 250 km × 250 km, with a precision of ±55 km (Eastes et al., 2020). We use version 4 of the level 2 T _disk data product. The longitude-latitude grids are fixed for all scans. The T _disk data file has flags for data quality issues at the file and pixel levels. Observations with dqi > 0 at both levels are not considered in our analysis. To increase signal-to-noise ratio (SNR), observations of χ > 65^◦ have been removed. Random errors in the 2 × 2 binned data vary with SNR of the N ₂ LBH emission and ranges from 20 (for high SNR) to 90 K (for low SNR). Figure 5 shows the spatiotemporal coverage for 2 representative days, 31 December 2019 and 31 December 2020, during the non-SSW winter and SSW winter, respectively. The coverage represents the good quality retained T _disk temperature profiles. In panels (d) and (i), an uneven latitude versus SZA (χ) distribution is seen. Higher latitudes are covered at higher χ′s, which has implications for the observed temperatures. Otherwise, there is an even χ coverage across all longitudes. Two successive years were chosen to demonstrate the negligible amount of variability between the 2 years in terms of the orbital coverage.

In Figure 11G in the main text, a linear model was fitted to the data for each of the latitude bands in order to better visualize the variability. The standard deviation of the residuals were then used to generate the error bars in the plot, rather than the standard deviation of the raw temperature data. Therefore, the error bars in the plot should be interpreted as indicating the variability around the fitted linear trend, rather than the absolute variability in temperature.

In order to assess changes in the thermospheric winds induced by the major SSW, we consider ICON/MIGHTI measurements for two periods with a common spatiotemporal coverage. Figures 6, 7 present the daily evolution of the daytime longitudinally averaged (i.e., zonal mean) zonal and meridional winds, respectively, during the major SSW (December 2020—January 2021) and non-SSW winters (December 2019—January 2020) at two representative latitude bands: at low-latitude (0–20^◦N, left column) and low-to midlatitude (20^◦ - 40^◦N, right column) regions. Altitudes and days, for which observations are not available, are shown in gray shading.

Even without an SSW, the observed thermospheric horizontal winds exhibit a significant degree of day-to-day variability. This could be related to a combination of physical processes, such as a) changes in the dynamics of internal atmospheric waves, b) variability of the solar and geomagnetic activity, and c) orbital effects, e.g., ICON’s orbit precession toward earlier local times by about 29.8 min every day (Figure 3). Under the non-SSW conditions (during the non-SSW winter and before the onset of the warming), the daytime zonal winds exhibit an alternating with altitude pattern at low- and low-to midlatitudes: typically eastward in the upper mesosphere, westward in the lower thermosphere and eastward again above ∼ 120 km. Above ∼160 km, the westward flow dominates, in general. The daytime meridional winds without an SSW are overall northward (representing the poleward, summer-to-winter circulation) in the upper mesosphere, southward (winter-to-summer transport) in the lower thermosphere, and poleward again above ∼130 km ( Figures 6A,B; Figure 7A,B ).

In the winter hemisphere, westward propagating GWs, owing to their faster intrinsic horizontal phase speeds c ̂ i , have favorable upward propagation conditions through the mean eastward stratomesospheric jets, maintained by the Coriolis torque. Saturation of the surviving harmonics produces momentum flux divergences (i.e., GW drag) in the MLT that reverses the zonal winds from eastward to westward and the meridional winds from northward to southward around 90–110 km (Garcia and Solomon, 1985; Yiğit et al., 2009; 2021).

Secondary or higher-order waves can be generated in the middle atmosphere by the dissipation of the primary waves (Holton and Alexander, 1999). Secondary waves could thus contribute to the momentum balance in the MLT and above, as suggested by previous theoretical modeling studies (Vadas et al., 2018), however, based on first-principle theoretical arguments, the relative contribution of secondary waves to the momentum balance is generally expected to be much smaller than those of primary waves (Medvedev et al., 2023). Nevertheless, a number of observational studies consider secondary waves as important contributor to the momentum fluxes in the MLT region. Vincent et al. (2013) studied gravity wave absolute momentum fluxes with meteor radars and stated that they could not distinguish primary from the secondary waves, but nevertheless discussed the potential importance of secondary waves. Secondary waves were also considered as a complicating factor in that study. More recently, de Wit et al. (2017) argued that the large eastward MLT GW momentum fluxes observed near the Southern Andes could not be explained by the direct upward propagating orographic waves and thus were likely caused by secondary gravity waves. Kogure et al. (2020) used two satellites to study mountain wave breaking over the Andes and their connection to secondary waves. They concluded that ring-like concentric waves above the wave breaking region were associated with secondary wave generation. The momentum forcing is also supplemented by upward propagating diurnal and semidiurnal tides at low- and middle-latitudes, respectively (Jones et al., 2019; Miyoshi and Yiğit, 2019; Griffith et al., 2021; Conte et al., 2024).

Changes in the stratosphere during SSWs can be communicated to upper layers of the atmosphere via changes in the large-scale structure and in upward propagation and dissipation conditions of internal waves. The degree of vertical coupling during SSWs can be in general different than during quiescent times. After the onset of the warming in January 2021 (Figures 6C,D, 7C,D), westward (negative) and northward (positive) winds relatively strengthen depending on the altitude and day, especially above 140 km. The dotted color boxes are thought to aid with the comparison of the wind changes due to the SSW. There is an indication that the thermospheric winds begin to change before the start of the SSW, which can probably be related to the fact that the stratospheric mean zonal wind decrease precedes the polar temperature rise by several days (Figure 1B). This phenomenon is known to modulate upward gravity wave propagation (Yiğit and Medvedev, 2012; Miyoshi et al., 2015).

Figure 8 analyzes SSW-induced changes in horizontal winds by subtracting the 7-day averaged winds before the onset of the SSW (i.e., 25–31 December 2020) from the rest of the wind observations in the 2020/2021 SSW winter. Difference winds are shown for the zonal (Δu) and meridional (Δv) components. Overall, post-SSW winds are more westward and northward relative to the pre-SSW period. Figure 9 plots the averaged horizontal circulation in terms of vectors during the pre-SSW and post-SSW periods, focusing on an altitude interval of 140–160 km and a time window of 1 week with respect to the SSW onset in each case. This representation shows more clearly the westward turning of the horizontal circulation during the SSW, especially at low-latitudes.

Diurnal and semidiurnal tides with long vertical wavelengths can reach upper mesospheric and thermospheric altitudes and play an important role in the momentum budget of the low-to midlatitude thermosphere (Jones et al., 2019; Miyoshi and Yiğit, 2019; Forbes et al., 2022). Given the different general mechanism and spatiotemporal characteristics of the different tides, the response of the upper atmosphere across the tidal spectrum is expected to vary. In other words, the different parts of the tidal spectrum can be modulated by the SSW to different amounts. Figure 10 presents the latitude-time distributions of the semidiurnal tidal amplitudes for the SW2, SW1, and SW3 (left panels) during the SSW year, as well as for the diurnal components, DW1, D0, and DE3 (right panels), as retrieved from ICON zonal winds at around 100 km. It is seen that the SW2 is the strongest component compared to the others and it is enhanced at low-latitudes after the onset of the major warming. Previous studies suggested that the SW2 is strong after many split-type SSWs (Ma et al., 2023), which could be also related to the stationary planetary wave-2 (SPW2) induced by splitting vortices after the associated SSWs (Ma et al., 2020).

Zhang et al. (2022) studied the semidiurnal tides using Hough mode extension of the zonal and meridional winds and emphasized the importance of semidiurnal tides in the variations of equatorial plasma drifts and neutral winds during the SSW. Overall, our results are in good agreement with previous findings of the enhanced semidiurnal migrating tides in the low- and midlatitude lower and upper thermosphere during SSWs (Goncharenko et al., 2010b; Liu et al., 2013; Oberheide, 2022).

Figure 11 presents the day-to-day evolution of the daytime neutral temperatures near 150 km as measured by GOLD, and averaged longitudinally and over the same two representative latitude bands discussed above. Based on GOLD’s coverage, only longitudes between 100^◦W and 10^◦E contributed to the zonal mean. The upper two rows (Figures 11A–D) show the temperature variations as a function of solar zenith angle χ. Note that the two latitude bands have different χ coverage. The observations for 25^◦ < χ < 65^◦ contributed to the low-latitude 0–20^◦N band, with a larger portion of measurements centered around 65^◦. The low-to midlatitude (20^◦ - 40^◦N) band includes observations for χ between 45^◦ and 65^◦, with a larger portion taken around χ = 55^◦. Rows three and four (Figures 11E,F) display another aspect of temperature variations: the latitude-time cross-sections at 150 km during the non-SSW and SSW winters, respectively. It is seen that, at all latitudes and solar zenith angles, thermospheric temperatures drop during the SSW. The cooling trend begins shortly before the SSW onset and lasts for about 15 days. The thermospheric cooling is more clearly seen in Figure 11G, which presents the day-to-day variations of the average temperature in the corresponding latitude bands. The error bars indicate the variability around a fitted linear trend (Section 2.5). Starting a few days before the onset of the SSW, the thermospheric temperature decreases by about 50 K, from ∼730 K–680 K, after which it returns back to ∼720 K over about 10 days. Such cooling trend is untypical in the low-latitude thermosphere in the absence of SSWs, as a comparison with the non-SSW winter shows. It is also seen that the thermosphere is much colder during the non-SSW winter, because it coincided with the solar minimum.

Observations presented above demonstrate a global response of the low-to middle-latitude thermosphere to the major SSW event. Generally, winds and neutral temperature are affected by a number of physical processes pertaining to external (space weather, or coupling from above) and internal forcing (coupling from below) (Yiğit et al., 2016). SSWs, which originate in the troposphere and lower stratosphere, are significant disturbances of the latter region. They quickly modulate the vertical propagation of internal atmospheric gravity waves and solar tides, which have a broad spectrum and can reach thermospheric altitudes with significant amplitudes. A number of observational and modeling studies found that the thermospheric GW activity decreases after a major warming is fully developed (Miyoshi et al., 2015; Nayak and Yiğit, 2019). On the other hand, the amplitude of the migrating Sun-synchronous semidiurnal tide increases during SSWs in the low- and midlatitude lower and upper thermosphere (Goncharenko et al., 2010b; Liu et al., 2013; Oberheide, 2022). More recently, Gasperini et al. (2023) analyzed ICON green line winds in the MLT, showing that the SW2 zonal wind amplitudes are amplified following the 2020/2021 major SSW, which we were able to confirm in our analysis (Section 3.2). These two changes can be related, because GWs are known to attenuate the semidiurnal tide in the thermosphere (Miyoshi and Yiğit, 2019). That is, a decrease in gravity wave activity could provide favorable conditions for SW2 tidal propagation and growth. Semidiurnal tidal sources can also be modulated owing to a redistribution of the stratospheric ozone (Pancheva et al., 2003). Thus, the modified wave forcing can directly affect the residual circulation in the thermosphere (Koval et al., 2021). Coupled changes in gravity wave and tidal activity are the probable cause for the observed wind variations in thermospheric altitudes. Systematic modeling studies are required for isolating the effects of gravity waves and semidiurnal tides (and their possible interactions) during major stratospheric warmings.

SSW-induced thermal and dynamical changes are intimately connected. In addition to direct wave forcing, thermodynamical changes can be caused by modification of the large-scale flow. Vertical motions and the associated adiabatic heating/cooling are connected to divergence and convergence of horizontal winds via continuity (Rishbeth et al., 1969). Previous research demonstrated significant low-latitude thermospheric cooling during major SSWs. During the January 2009 SSW, Liu et al. (2011) discovered large density drops in GRACE and CHAMP observations above 325 km, which corresponded to a ∼50 K cooling. Using simulations with a whole atmosphere model, Liu et al. (2013) reported a net cooling of the thermosphere above 100 km during the 2009 major SSW. Although the spatiotemporal coverage between previous studies and our analysis is somewhat different, our results are qualitatively in agreement with previous findings and contribute toward filling the gap in vertical coupling between the troposphere and the upper atmosphere. A net upwelling and enhanced poleward flow initiated by SSW-induced changes can account for the observed cooling in the low-latitude thermosphere around 150 km. However, demonstration of this physical mechanism would require three-dimensional global modeling effort, such as the one conducted by Miyoshi et al. (2015). Overall, our findings are complementary to previous studies that showed upper thermospheric cooling associated with upwelling at low-latitudes using CHAMP and GRACE satellites (Liu et al., 2013; 2011). It was later studied in more detail using whole atmosphere models, which demonstrated that the cooling extends to middle thermospheric altitudes (Liu et al., 2014; Miyoshi et al., 2015).

Finally, a subtle decrease of solar activity (from 85 to 75 × 10⁻²² W m⁻² s⁻¹) over the SSW period (Figure 2) can contribute to some extent to the observed 50 K temperature drop around 150 km. Tests with the NRLMSIS empirical model (Picone et al., 2002) however suggest that a reduction of the solar activity by ten F _10.7 radio flux units changes temperature by only 5–10 K around 150 km altitude (not shown). Obviously, more accurate and self-consistent estimates can be obtained using models extending from the lower to the upper atmosphere or even better whole atmosphere general circulation models.