Long-term changes of mesosphere/lower thermosphere (MLT) winds have frequently been analysed in the past, mainly based on radar observations or modelling . However, when considering long-term trends it must be kept in mind that there exist periodicities at decadal or very long timescales. One of them is the 11-year solar cycle (SC). Its influence on the MLT dynamics was investigated in the past, among others, by , , , , and .

The SC studies, however, were partly contradictory. From low-frequency (LF), closely spaced receiver (D1) measurements in northern and eastern Germany, found a strong positive correlation of both zonal (positive eastward) and meridional (positive northward) prevailing wind in winter, which was confirmed by using additional D1 data and also meteor radar (MR) data from Obninsk, Russia (55° N). Another study was carried out by , who, using data from different stations at northern middle latitudes, qualitatively confirmed the findings of , and also reported a positive SC effect during summer. However, in contrast to the earlier results found negative correlation of solar flux and zonal prevailing wind in winter as well as in summer. analysed more than 30 years of mesopause region wind data, and reported positive correlation of the zonal as well as of the meridional mean wind with the solar flux in winter, and negative correlation in summer, although their results were not statistically significant. and confirmed the negative summer mean wind correlation and a weaker positive correlation in winter. Using partial reflection (PR) measurements over Juliusruh, Germany, also found the positive/negative SC effect in winter/summer, which in winter was particularly strong near the mesospheric wind jet, but weaker near the mesopause, which confirmed the results from LF and MR observations. Observations over Castle Eaton, UK (52° N), however, revealed contrasting results .

Using the extended UK Meteorological Office stratosphere mesosphere model, reported enhanced high-latitude zonal winds during solar maximum winters. This was partly due to a latitudinal shift of the zonal wind jet, and the effect decreased above the mesosphere, which is supported by observations . also reported enhancement of the winter zonal middle atmosphere winds after geomagnetic activity. Since the latter is connected with the SC, this effect may also lead to stronger westerly winter winds in the middle atmosphere during high solar activity.

Already earlier observations until the 1990s showed that a possible SC effect is not necessarily stable with time. Analyses of the Obninsk long-term MR measurements were also presented by , who pointed out that oscillations with a period of 11 years were present in the time series of the prevailing wind. However, they also reported a long-term change at the bidecadal-year time scale they attributed to a possible effect of the 22-year solar Hale cycle. also reported a change of solar influence, but mainly seen in the semidiurnal tidal amplitude and not in the zonal prevailing wind.

In the late 1990s and after the turn of the millenium, more long-term observations became available using meteor radars at different latitudes. While essentially showed a positive SC effect on the winter mean wind at a high northern hemisphere (NH) latitude (Esrange, 68° N), reported a negative zonal wind effect of the SC over Rothera (68° S), and confirmed this with WACCM-X global circulation model results. On the other hand, their SC effect in summer was positive, while and found effects changing with height. At midlatitudes, reported a SC effect of the zonal wind during winter and summer, but at different altitudes for Europe and North America. Also satellite observations have been analysed with respect to a SC effect , revealing mostly positive SC dependence of balanced zonal winds at low and middle latitudes. WACCM-X model results by at about 90 km showed mainly a positive SC effect in both NH and southern hemisphere (SH) higher latitudes.

Generally, the results of SC analyses of winds observed since the 2000s are partly inconclusive and not always agree with those from earlier observations. This may again be interpreted as a possible influence of changes or periodicities at the interdecadal time scale. Actually, linear trends in the MLT have been found to change with time . Recently, reported a change in winter MLT zonal mean wind trend similar as in , connected with a change of the Northern Annular Mode trend and planetary wave propagation trends in the NH winter. Finally, trends may even have a different sign across longitudes , which may also affect a possible SC signature .

To conclude, it is not clear whether there is a real SC effect on MLT winds, and if so, whether and how this changes with time. To contribute to the database regarding a possible SC effect on the MLT, we here present observations from two radar sites in Germany, namely Collm 51.3° N, 13.0° E, and Juliusruh 54.6° N, 13.4° E,. Owing to the different observation methods in Collm and Juliusruh, their data refer to slightly different heights around the MLT. We analyse long-term changes, with a focus on the possible influence of the 11-year cycle. A possible effect of the 11-year SC has also been found in gravity wave (GW) proxies e.g.,, so we also compare the SC effect in winds with GW activity in the following.

At Collm, MLT winds had been measured from 1979 to 2008 by the LF D1 method using the sky wave of three commercial radio transmitters with frequencies of 177, 225, and 270 kHz. The total reflection of the LF radio wave in the lower E region has been used by this method. Therefore, the reference height was changing during the day, i.e. strongly decreasing/increasing in the morning/evening, so that mean wind values over more than half an hour could not be obtained. Therefore, monthly median half-hourly winds have been calculated by combining the three respective wind measurements in each direction using weighting factors based on the “chaotic velocity” . Note that LF wind observations at Collm have started earlier , but using single LF transmitters only, so that data from 1979 are more reliable. The half-hourly means have already been used to calculate monthly mean zonal and meridional prevailing winds and tidal parameters . The data are attributed to the mean nighttime reflection height of LF radio waves of about 90 km . LF reflection heights have been observed since 1983 (with half-hourly resolution since 1984) until 2007 by phase comparisons of the ground wave and the sky wave in a modulation frequency band near 1.8 kHz . Due to group retardation the estimated virtual heights are too large, and have been corrected here using an empirical height-dependent reduction factor based on phase comparisons of the solar semidiurnal tide derived from LF and very high frequency (VHF) MR observations . Note that the reflection heights are changing during the day, so that observations at higher altitudes around 90 km were only possible during nighttime. During daylight hours, the reflection height decreases to about 82 km, however, due to LF radio wave absorption in the ionospheric D region particularly in the summer mesosphere, the daytime measurements were only possible during winter.

Since 2004, VHF MR observations have been performed at Collm on 36.2 MHz . An upgrade of the radar was made in 2015/2016, including the replacement of the Yagi antennas by crossed Yagi antennas and peak power increase, which resulted in a more regular azimuth distribution of registered meteors, an extended height range, and a higher meteor count rate. The observations are now also synchronized with other systems in northern Germany, which allows multistatic observations . Details of the radar configuration can be found, e.g., in . The radar now delivers wind observations in the approximate height range of 75–105 km. Meteor heights are provided by an interferometer . 21-year mean monthly mean winds in the height range 82–97 km are shown in Fig. . The data used for Fig.  is based on monthly means calculated in 3 km height gates, except for the upper height gate, which is 5 km wide . Therefore, the actual height range covered is 80.5–100.5 km. At very high and low altitudes, data gaps appear especially during the late afternoon hours due to low meteor rates then, in particular, before the upgrade. Therefore, and since in the following we analyse wind mainly around 90 km, in Fig.  we only show winds in the height range where considerably large count rates were obtained.

However, in the following, we do not use the height information to analyse the mean winds at the mean height of the LF measurements. The LF and VHF winds have been combined using a seasonally dependent correction for the meridional winds as described in . Comparison of overlapping observations during 2004–2006 presented in had shown that, when neglecting height finding, zonal winds observed by LF and MR agree, while meridional winds observed by the spaced receiver method usually underestimate the real winds. The seasonal cycle of meridional winds therefore led to a difference that depends on season. The combined dataset has been used in and . Here we present an updated dataset of monthly mean zonal and meridional winds.

From 1990 to 2003 a PR radar was in operation in Juliusruh working with the FMCW (Frequency Modulated Continuous Waves) method at 3.18 MHz . This radar has been replaced in spring 2003 by a new modular transmission and reception system with distributed power and a new so-called Mills-Cross-Antenna. Now the radar is working with a peak pulse power of 64 kW. Other technical parameters are: 27 µs pulse width, single pulse or 4 bit complementary code, 18°  3 dB beam width, and sampling resolution 1 km. Here we use observations obtained at 84 km in a 2 km height window. We analyse monthly winds from 1990 to 2024, with missing data in the year 2000 as well as in May 2005 and January/February 2027. Medium Frequency/PR radar observations above 85 km tend to be affected by group velocity retardation. This is why heights there are generally called virtual heights like in the case of ionosondes. One would need to know the actual electron density profile, which is not available, to correctly calculate the offset. Therefore, 84 km is regarded the height level closest to the Collm reference height that can be considered real.

The following analysis is based on observations from 1979 to 2024 for Collm data, and 1990 to 2024 for Juliusruh winds for comparison. We use monthly means calculated from the daily PR data for Juliusruh. For Collm LF D1 observations, we calculated monthly median winds for each half-hourly time interval of a day. These, but excluding daytime LF winds to avoid effects of the low reflection heights during daylight hours, were used in a regression analysis. This regression included the mean wind and the circularly polarised component of the semidiurnal tide , to account for the quasi-regular daytime data gaps of the LF observations e.g.,: 1u=u0+a1⋅sin⁡(ωt)+a2⋅cos⁡(ωt)+ϵ,v=v0+a1⋅cos⁡(ωt)-a2⋅sin⁡(ωt)+ϵ, where u,v represent the zonal and meridional half-hourly winds, respectively, u0 and v0 are the mean horizontal winds, ω is the angular frequency of the semidiurnal tide, t is time in hours, and a1 and a2 are coefficients describing the tidal amplitude and phase, which are not used here. u0, v0, a1 and a2 are determined by minimizing ϵ. The diurnal tide is not considered in this analysis, because the regular daily data gaps would lead to large errors.

Since our MR observations do not suffer from regular daily data gaps, assuming circular polarisation is not necessary for analysing the data, and also the diurnal tide can be taken into account e.g.,their Eqs. 1 and 2: 2u=u0+∑i=12a1,u,i⋅sin⁡(ωit)+a2,u,i⋅cos⁡(ωit)+ϵ,v=v0+∑i=12a1,v,i⋅sin⁡(ωit)+a2,v,i⋅cos⁡(ωit)+ϵ, with ω1 and ω2 now as the angular frequencies of the semidiurnal and diurnal tide, respectively.

In the Collm time series a few short data gaps of up to a few days appear, but no monthly means are missing. So these gaps do not affect the analysis of trends in the time series.

The mesosphere and lower thermosphere are dynamically coupled through GWs that are filtered in the mesospheric jet and force the zonal mean wind reversal above. Therefore, modulations of the mesospheric zonal wind jet, e.g. through the 11 year SC, potentially modulate GW amplitudes at the same time scales and also modulate the GW drag in the lower thermosphere. It is therefore of interest to analyse a possible SC effect in GW records that could give hints to the forcing mechanisms of the SC effect in the MLT.

To estimate GW proxies from Collm wind observations with height finding since 1984, deviations u′, v′ of the LF half-hourly means have been determined according to and . These deviations have been calculated as the differences of subsequent half-hourly mean winds, and variances u′2+v′2 have been analysed. Using half-hourly means is equivalent to a filter passing periods of about 0.7–2 h with a maximum near 1 h. The wind differences are not included in the further analysis, if the height changes by more than Δh=2 km from one half-hourly mean to another. For the VHF observations since 2004, the same method was applied using meteors from a 3 km height gate centered at 88 km.

3-monthly mean gravity wave proxies u′2+v′2 over Collm at an altitude of 88 km for summer (June–August, JJA, red) and winter (December–February, DJF, blue), taken from , are shown in Fig.  in Appendix . has reported a strong quasi-decadal variation of GW proxies from LF observations until 2007, and this is also visible in the first part of the time series shown in Fig. . Later, as with the background wind, this signature disappears. Analysing the SC effect c according to Eq. () in Fig.  shows significant positive values for the full dataset (1984–2024), except for December, February and March when the effect is still positive, but insignificant. But this is only owing to the first part of the time series. After 2002, there is no significant SC effect visible any more (mostly open red symbols in Fig. ), except for November. Thus, as for the mean winds, this indicates a possible change after the 1990s. The seasonal cycle of the c coefficients for GW proxies in the time interval 2002–2024 is qualitatively similar to the one of the meridional mean wind (red curve in Fig. ), but the GW values are not statistically significant.

analysed hydroxyl nightglow emissions over Zvenigorod (55.7° N, 36.8° E) during 2002–2024 and found a strong positive SC for temperature oscillations with the same period range as used here. Thus, given that GW kinetic and potential energy is connected, their results are qualitatively similar to our ones for 1984–2001, however, we cannot confirm their results for the observations after the turn of the millenium. Hydroxyl emissions are observed without height finding. Often, their height and full width at half maximum are given as 87 and 8 km, respectively, but they may change due to dynamics e.g.,. Thus, it is possible that the nightglow GW analyses are contaminated by a SC in background dynamics. On the other hand, our LF height observations may also be affected by a possibly more variable lower E region ionosphere during solar maximum and the large SC effect observed by LF might not be exclusively due to neutral wind dynamics.

For comparison, we also analysed TIMED/SABER GW potential energy per volume Epot,V,. Since GWs are strongly intermittent, a sufficient number of data points is needed for obtaining robust averages (see also ). Therefore, we averaged over 30° longitude and 12° latitude over the Collm area see for years 2002–2022. However, the seasonal distribution of SC coefficients is different from the one for Collm GW proxies, or meridional mean winds (see Fig.  in Appendix ). Earlier analyses have shown qualitatively similar long-term linear trends of Epot,V and Collm variances, but this correspondence obviously only holds for the long-term linear trend and is not visible at decadal time scales. Differences may be due to (i) the different parameters analysed; (ii) the different wavelength ranges seen by radar and satellite ; (iii) the spatial averaging of the satellite observations, so that regional effects are possible. One explanation for the insignificant SABER response could be that SABER is more sensitive to longer intrinsic wave periods, i.e., slower intrinsic phase speeds their Fig. 8b. This means that SABER GW activity should be more susceptible to short-term variability of the atmospheric background by global-scale waves which could mask longer-term variations. Such short-term variability has been found, for example, for planetary waves at mid and high latitudes e.g.,, or for tides at low latitudes e.g.,.

Analyses of more than four decades (1979–2024) of Collm MLT wind measurements show long-term trends and SC effects. Similar tendencies of the seasonal distribution of the MLT wind response to solar EUV changes are visible in most months from Juliusruh PR observations 1990–2024. Together, the results suggest that solar forcing remains an important, season-dependent driver of mid-latitude MLT dynamics. After the 1990s, linear trends and the SC effect changed and strongly weakened. A possible SC effect after the year 2000 remains only in early winter, which is also seen in satellite balanced winds . The overall solar signal after the year 2000 seems to be weak and partly inconclusive. This indicates a change of MLT dynamics after the 1990s, possibly due to changes in trends of the NH circulation , the ozone turnaround in the stratosphere, or the recently weakened SCs 24 and 25. Since the SC effect, particularly in summer, decreases with altitude in the MLT it is possible that the change in mean wind and consequently in the height of GW mean wind interaction modulates the height up to which the SC effect is visible. showed varying influence of the SC on stratospheric dynamics which they attributed to varying SC strength. However, one has to take into account a long-term decrease in measuring height as well e.g.,. It is also possible that simply the weaker recent SCs are not strong enough any more to produce a significant effect. And finally, we cannot rule out artefacts because the LF reflection heights decrease with increasing solar flux due to the strengthening of the ionospheric D and E layers .

GW proxies from the Collm winds show a SC variability until the end of the 1990s, but this strongly decreases in the following decades, in most months, except for autumn. GW proxies in the 21st century still show a possible correspondence of the GW SC effect with mean wind variations in autumn, however, GW proxies from SABER potential energy do not show a clear correspondence of the SC effect with the one seen in winds. The latter may be due to the different parameters analysed, the spatial averaging of satellite observations, or the different and hardly overlapping observational filters of VHF radar and SABER observations.

Apart from a solar influence, there are other (potential) drivers of NH MLT wind variability such as the equatorial Quasi-Biennial Oscillation (QBO), El Niño-Southern Oscillation (ENSO), or North Atlantic Oscillation (NAO). These drivers generally show a clear dynamical signature in the winter NH stratosphere. Their possible contribution to MLT dynamics has been analysed, e.g., by or . While found a positive response of the QBO at 30 hPa on NH high-latitude zonal winds in the upper mesosphere, the effect in the SH as observed by was strongest in summer. In both cases the QBO and ENSO influence decreases with height. However, reanalyses and model results showed that the QBO effect on mesospheric winds changes between middle and high latitudes, and in parts of the higher midlatitudes it is therefore weak . Furthermore, the QBO effect changes sign with altitude. For example, did not find a conclusive QBO effect on MLT summer wind analyses at European mid and high latitudes, and only some indications for oscillations at QBO and ENSO periods in midlatitude mesospheric winds. Earlier analyses of Collm winds only showed a weak QBO effect on MLT winds observed by LF at about 90 km in winter. Similarly, only weak signatures of ENSO in Collm LF winds at 90 km were found and later analyses using height resolved MR observations revealed that the possible ENSO effect on the MLT is strongly height dependent, changes sign with height, and is actually weak at 90 km, which was also reproduced by numerical modelling . Also, earlier analyses of LF Collm winds have shown a positive NAO signature in zonal winds 1979–1996 , but this connection was not stable when including later years into the analysis.

We conclude that, since the signature of QBO, ENSO and NAO is strong in the stratosphere but changes sign in the upper middle atmosphere, observations at 90 km are not well suited to capture their dynamical effects in the MLT. Further analyses of winds below and above that height might help to more clearly identify the influence of QBO, ENSO and NAO on the MLT, and may also show whether these potential influences change with time as the solar forcing does.

Here, we only analysed monthly mean winds and included the effect of tides in Eqs. () and () only to avoid biases due to daily data gaps. Earlier publications from observations and tidal amplitude analyses since the 1960s relied of analyses of time series obtained by one method. They show decreasing tidal amplitude trends, but these trends were not uniform, and decadal and interdecadal variations are visible in the time series. Analyses of a possible SC effect revealed results that change with time . Moreover, at different locations the semidiurnal tidal trends differ substantially , which suggests the rather regional character of these trends, possibly due to the influence of nonmigrating tides. In principle, trend analyses of tidal amplitudes from a combination of Collm LF D1 and MR observations would be of interest. But comparison of LF D1 and MR wind analyses has shown that LF spaced receiver measurements strongly underestimate tidal amplitudes . Therefore, so far we have not yet tried to combine the LF D1 and MR tide observations into one time series.