On the solar cycle dependence of winds and planetary waves as seen from mid-latitude D1 LF mesopause region wind measurements

At the Collm Observatory of the University of Leipzig LF D1 low-frequency total reflection nighttime wind measurements have been carried out continuously for more than two decades. Using a multiple regression analysis to derive prevailing winds, tides and the quasi-2-day wave from the half-hourly mean values of the horizontal wind components, monthly mean values of mesopause wind parameters are obtained that can be analysed with respect to long-term trends and influences of solar variability. The response of the prevailing wind to the 11-year solar cycle differs throughout the year. While in winter no significant correlation between the zonal prevailing wind and solar activity is found, in spring and summer a negative correlation between the TWC can be seen from the measurements. This is connected with stronger vertical gradients of the zonal prevailing wind during solar maximum than during solar minimum. Since the amplitude of the quasi-2-day wave is dependent on the zonal mean wind vertical gradient, this is connected with a positive correlation between solar activity and quasi-two-day wave activity.


Background
One of the ®rst investigations of the solar cycle dependence on the mesopause region winds was made by Sprenger and Schminder (1969). From low-frequency, closely spaced receiver LF D1 measurements at KuÈ hlungsborn (55°N) and Collm (52°N), Germany, they found a strong positive correlation of both zonal and meridional (which is positive if directed towards the north) prevailing wind with solar activity in winter, and a decrease in the winter semidiurnal tidal amplitude with increasing solar¯ux. These results were con®rmed by Portnyagin et al. (1977) using additional D1 data and also D2 (meteor radar) data from Obninsk, Russia. However, considering the zonal prevailing wind and the semidiurnal tidal amplitude, the solar cycle dependence as obtained from the D2 measurements was much weaker than the one found using the D1 measurements. Gregory et al. (1981) analysed mesospheric winds at Saskatoon, Canada, which were measured in 1978 and 1979 during high solar activity, and found dierences compared with the CIRA 1972 model data, which were mostly obtained during years of low solar activity. They found a negative solar cycle dependence in some summer months as well as in February and March, while in December and February a positive solar cycle dependence was found, which coincided with the results of Sprenger and Schminder (1969).
Analyses of the Obninsk D2 long-term measurements were also presented by D' Yachenko et al. (1986), who pointed out that oscillations with a period of 11 years were present within the time series of the prevailing wind and the semidiurnal tidal amplitude. Another study was carried out by Dartt et al. (1983), who, using data from dierent stations at northern middle latitudes, qualitatively con®rmed the results of Sprenger and Schminder (1969) concerning the solar cycle dependence of the winter prevailing wind. Additionally, they found a negative correlation of the spring and early summer zonal prevailing wind with the solar activity, and a positive correlation of the meridional prevailing wind with the solar activity from January through July. Greisiger et al. (1987), however, in contrast to the earlier results of Sprenger and Schminder (1969) or Portnyagin et al. (1977), found negative correlation of solar¯ux and zonal prevailing wind in winter as well as in summer, although they could con®rm the negative correlation of solar¯ux and semidiurnal tide that was found by Sprenger and Schminder (1969) or Portnyagin et al. (1977). Since Greisiger et al. (1987) used D1 data from the same sites as Sprenger and Schminder (1969), this is a hint that the apparent solar cycle dependence of the prevailing wind in winter as established using the results of the earlier years was not a real solar eect, but rather a result of either measurement uncertainties or pure chance due to long-term variations in the winter mesopause region prevailing wind that are of non-solar origin. The latter explanation, however, seems to be more probable, since the apparent positive solar cycle dependence of the winter zonal prevailing wind in these years was found from winds measured at several northern mid-latitude sites at dierent longitudes (Sprenger and Schminder, 1969;Portnyagin et al., 1977;Gregory et al., 1981). Namboothiri et al. (1993a) examined Saskatoon data from the period of 1979 through 1990. They found a positive correlation of the zonal prevailing wind with solar activity in winter, and a negative one in summer, thus con®rming the results of Dartt et al. (1983), but not those of Greisiger et al. (1987) for winter. Using data from the southern hemisphere, Fraser (1990) pointed out that the results concerning a solar control of the prevailing wind and the semidiurnal tide are not conclusive.
Recently, Bremer et al. (1997) analysed more than 30 years of mesopause-region wind data. They reported positive correlation of the zonal as well as of the meridional wind with the solar¯ux in winter, and negative correlation in summer, although their results were not statistically signi®cant. It has to be kept in mind that they also included some Collm and KuÈ hlungsborn D1 data in their analysis, which had already been used in the studies of Portnyagin et al. (1977) and Greisiger et al. (1987). Using Collm data only, Jacobi et al. (1997a) found no solar cycle dependence of the semidiurnal tidal amplitude. This was already found by Fraser et al. (1989), in contradiction to the results of Sprenger and Schminder (1969) and Greisiger et al. (1987). As shown by Jacobi et al. (1997a), the time series of the semidiurnal tidal amplitude yields an oscillation with a period of more than 20 years which, however, does not coincide with any solar cycle and which is responsible for dierent results concerning an apparent solar cycle dependence detected from a time series shorter than about two decades [see also Jacobi et al. (1997b) their Fig. 7]. The positive solar cycle dependence of the winter zonal prevailing wind that was found by some authors could not be con®rmed by Jacobi et al. (1997a), but a negative solar cycle dependence of the spring and summer zonal prevailing wind was reported.
Recently, Arnold and Robinson (1998) presented model results that showed a solar cycle eect on the winter mesospheric circulation which, however, decreases rapidly above 80 km altitude. This could be one of the reasons for the dierent results that were obtained in the past, and thus one may summarise that the results concerning a possible solar cycle eect on the winter mid-latitude mesopause region are not very conclusive and it is doubtful whether a measured eect on the region above the mesosphere really is of solar origin. However, as far as the summer zonal prevailing wind is concerned, a negative solar cycle dependence was found by most authors. With respect to the meridional prevailing wind it is also doubtful whether there is an 11-year solar cycle eect on the mid-latitude mesopause dynamics. Namboothiri et al. (1993a) found only a tendency towards such an in¯uence, while the stronger winter southerly (or weaker northerly, respectively) winds that were found by Sprenger and Schminder (1969) during solar maximum could not be con®rmed by measurements in later years. So Bremer et al. (1997) did not ®nd a signi®cant solar cycle dependence of the meridional prevailing wind in winter from KuÈ hlungsborn D2 measurements, and this was con®rmed by the results of Jacobi et al. (1997a,b) who used Collm D1 measurements, but from dierent years to those used by Sprenger and Schminder (1969).
One of the most prominent patterns in summer mesopause dynamics is the quasi-2-day wave known since the early 1970s (Muller, 1972;Babadshanov et al., 1973). Measurements and theoretical investigations have shown that this phenomenon generally occurs shortly after the summer solstice or reaches its maximum amplitude then. From many of the investigations it was found that the quasi-2-day wave is a westwardpropagating wave of zonal wavenumber 3 (Muller and Nelson, 1978;Rodgers and Prata, 1981). Recently, however, Poole and Harris (1995) obtained a wavenumber of less than 3 from simultaneous measurements at two southern hemisphere sites, while Meek et al. (1996) obtained zonal wavenumbers between 3 and 4 from phase comparisons using radar wind measurements in the northern hemisphere. Long-term studies of the quasi-2-day wave from measurements at single sites have been carried out by Harris (1994) and Jacobi et al. (1997c). The period of the quasi-2-day wave is often found to be slightly longer than 48 h (e.g. Muller and Nelson, 1978), but recent results have shown that during the maximum of the wave amplitude the period is smaller than 48 h (Thayaparan et al., 1997a,b;Jacobi et al., 1997c).
Two mechanisms of the development of the wave are discussed: Salby (1981a,b) proposes that the quasi-2-day wave is a resonant ampli®cation of the antisymmetric (3,3)-normal mode. Plumb (1983), however, suggests that the wave appears due to baroclinic instability near the summer stratospheric wind jet. The climatology of the quasi-2-day wave at Collm was described in detail by Jacobi et al. (1997c), who also pointed out that an in¯uence of the 11-year solar cycle on its amplitude could be seen.
In the following, mesopause wind data measured at the Collm Observatory of the University of Leipzig are investigated with respect to a possible solar cycle dependence. The investigation will for the most part relate only to the zonal prevailing wind, being the parameter that in the past has been found to be most reliably in¯uenced by solar variability (Jacobi et al., 1997a,b), and the quasi-2-day wave which also seems to be in¯uenced by the solar cycle due to their connection to the vertical gradient of the prevailing wind.

Description of measurements and data analysis
The wind ®eld of the upper mesopause region has been continually observed by daily D1 radio wind measurements in the LF range since 1959, using the ionospherically re¯ected sky wave of three commercial radio transmitters on 177, 225 and 270 kHz. The measurements are carried out according to the closely-spaced receiver technique. A modi®ed form of the similar-fade method has been used to interpret the wind measurements automatically since 1973 (KuÈ rschner, 1975), and half-hourly mean values of the horizontal wind components are calculated for each of the three measuring paths. Monthly means of the half-hourly wind values are constructed. Since in summer during the day the absorption of the sky wave is too strong, the daily measuring period is then restricted to night and twilight.
Since the measurements are inhomogeneously distributed in time, a multiple regression analysis is used to determine the monthly prevailing wind and the semidiurnal tidal wind components using the half-hourly mean values of the measured zonal and meridional wind components. The spectral selectivity of the separation of prevailing and tidal wind is improved through ®tting the measured monthly means of the half-hourly wind values for the two horizontal wind components as a vector, assuming clockwise circularly polarized tidal wind components (KuÈ rschner, 1991): where v z and v m are the zonal and meridional monthly mean wind components, respectively, and v om and v oz are the components of the horizontal prevailing wind. The semidiurnal phase (T 2z ) and amplitude (v 2z ) can be calculated from the coecients b and c, which along with the prevailing wind values are determined by a least-squares ®t. The uncertainty of the monthly mean prevailing wind values amounts to 5m s A1 , which is mostly due to the variability of the prevailing wind on time-scales of less than 1 month. This is especially the case during spring and autumn, but also in winter due to rapid wind ®eld transitions during stratospheric warmings (e.g. KuÈ rschner, 1981a,b, 1984). In summer, the variability of the prevailing wind is smaller, but then the vertical gradients are larger (about 2 m s A1 km A1 on a long-term average), which leads to an additional statistical uncertainty due to changes of the re¯ection height. The diurnal tidal components are not taken into account here, because the daily, quasi-regularly distributed data gaps would lead to a large error in their calculation. Thus an additional error of the prevailing wind can occur when only night-time measurements are performed. At mid-latitudes the diurnal tide generally is a less dominant feature than the semidiurnal tide (e.g. Manson et al., 1989), although this is not necessarily the case in early summer. Estimates of the diurnal tide near 95 km can also be obtained from the Collm measurements [but using the original half-hourly wind values for a regression analysis with a modi®ed form of Eq. (1) with height-dependent coecients], and it was found from analyses in the past that in most of the months the diurnal tidal amplitude is much smaller than the semidiurnal one, although in spring and early summer they are of the same order of magnitude  and thus its in¯uence on the measurements should be kept in mind. To estimate the in¯uence of neglecting the diurnal tide in Eq. (1), calculations were performed using an extended form of Eq. (1) that includes the diurnal tide. These calculations led to prevailing wind values that diered from those calculated using Eq. (1) by less than 4 m s A1 in each month of the year. However, one has to keep in mind that the diurnal tide in the case of the Collm measurements may lead to a bias in the calculation of the prevailing wind values, which is dependent on the tidal amplitude and phase. As this investigation aims at the interannual variability of the prevailing wind, this would mean that in some cases one cannot clearly distinguish between long-term variations of the prevailing wind and those of the semidiurnal tidal amplitude and phase. However, as will be shown in the following, the solar cycle variability of the prevailing wind amounts to about 10 m s A1 , which is much larger than a possible in¯uence by the diurnal tide, and thus, although some small tidal eect on the results is conceivable, this does not aect the results qualitatively.
Since September 1982 the re¯ection height h has been measured on 177 kHz, using travel-time dierences of corresponding modulation bursts between the ground wave and the re¯ected sky wave in the 1.8-kHz modulation range . To avoid apparent wind variations due to re¯ection height variations in the presence of vertical gradients of the mean wind, only these times of day are used for the regression analysis given by Eq. (1), when the long-term mean re¯ection height is found to vary only slightly around the mean height of about 95 km (e.g. Jacobi et al., 1998).

Results of monthly mean winds
In Fig. 1 time series of the winter (December±February) and summer (June±August) mean zonal prevailing wind are shown. In the upper panel the 13-monthly mean sunspot number R is shown. Only those data that have been obtained automatically since 1973 are used here to avoid possible artifacts due to the analysis by hand that was performed before that year. Regarding the winter time series of v oz , some variability on a time-scale of 2±5 years can be seen. Additionally, Jacobi et al. (1996) found a weak dependence of the winter zonal prevailing wind on the equatorial QBO, but a connection between QBO, solar activity and winter mid-latitude circulation, as was found for the stratosphere (e.g. Loon, 1992, 1996), cannot be seen from the Collm measurements. This is probably due to the strongly variable reaction of the mesopause circulation to sudden stratospheric warmings. A solar eect on v oz in winter is not inferable from Fig. 1. However, in summer there is some evidence of a negative solar cycle dependence of v oz . This is indicated by the two arrows in Fig. 1 that point to relative minimum values of the smoothed summer zonal prevailing wind. The correlation coecient between R and v oz in summer (using June±August means) is r = 0.65, with 99% signi®cance (t-test).
It can be seen from Fig. 1 that the long-term variability of the wind parameters is dierent for summer and winter. To take this into account, the time series of the zonal prevailing wind v oz,i in each month of the year are investigated with respect to a long-term trend and a solar cycle dependence using a multiple regression analysis according to where yr is the time, counted in years, and R is the 13monthly smoothed sunspot number, which is strongly correlated to the 10.7 solar radio¯ux (e.g. Rishbeth and Edwards, 1989) and can thus be taken as an indicator for solar activity. The parameters v x,i , D t,i and D R,i are obtained by least-squares ®t. The latter two represent the change in the zonal prevailing wind per year (i.e. the long-term trend) and the change in v oz per sunspot number (or the solar cycle dependence, respectively), while v x is a base value that is not of interest here, since it depends on the starting date of the time series.
The results for each month are shown in Fig. 2. The long-term trend is positive in most of the months, but it is relatively weak. It is interesting that this trend is dierent to the negative one reported by Bremer et al. (1997, their Figs. 4 and 7), and this shows that there is obviously a variation in the mean wind at a time-scale of more than two decades, since the trend detected by Bremer et al. (1997) is for the most part due to the stronger westerlies in the 1960s and early 1970s.
The solar cycle dependence is shown in the lower panel of Fig. 2. In winter no signi®cant in¯uence of the solar cycle on the zonal prevailing wind is found. This coincides with numerical results of Arnold and Robinson (1998), who found a strong solar cycle dependence of the mesospheric wind, but this eect was strongly decreasing above about 80 km. From Fig. 2 some of the results from literature cannot be con®rmed (e.g. Sprenger and Schminder, 1969;Portnyagin et al., 1977), so that it may be concluded that the apparent solar cycle dependence that was found for certain periods is due to other eects of non-solar origin. Considering, for example, the results of Bremer et al. (1997, their Fig. 3) for D1 (1964±1994) and D2 (1976±1994) measurements, one can see dierent results for January and February, so that the most part of the solar cycle eect of the D1 measurements is obviously an eect contributed by the ®rst years of measurements, and so it is not surprising that in this study, where only data since 1973 are used, no solar cycle eect on v oz is found in winter.
Thus the most striking point in Fig. 2 is the negative solar cycle dependence of the zonal prevailing wind in spring and summer that has already been found by other authors (e.g. Bremer et al., 1997;Jacobi et al., 1997a,b). From their analyses, also using Collm data, the same order of magnitude was obtained for the solar eect, and this is also the case for the analysis of Saskatoon data made by Namboothiri et al. (1993a).
Considering the results from Collm, however, one has to take into account that the wind values used in Eq.
(2) here are calculated without explicitly taking the re¯ection height into account, and therefore a solar cycle dependence of the re¯ection height would lead to an apparent solar eect on the measured zonal prevailing wind. For example, Entzian (1967) showed that during solar maximum the height of a ®xed electron density is lower during solar maximum than it is during solar Coecients D t (trend) and D R (solar cycle dependence) for the zonal prevailing wind, calculated using Eq. (2) for each month. Solid symbols denote signi®cance on the 95%-level (after Jacobi et al., 1997a, updated) minimum, and Cossart (1984) and Cossart and Taubenheim (1987) showed from indirect LF phase-height measurements that there is a dierence of about 1 km in the re¯ection height between high and low solar activity. Using medium-frequency height measurements, Namboothiri et al. (1993b) found even larger dierences for virtual re¯ection heights. Since at Collm the re¯ection height has also been measured since 1983, the solar in¯uence on the re¯ection height can be estimated. Using yearly means of mean night-time re¯ection heights it is found that the mean height for years with high solar activity (R > 100) is 94.5 0.4 km, while it is 95.6 0.9 km for the years of low solar activity (R < 50). The uncertainties given here are taken from the standard deviation of the mean annual heights. Since the data-base consists only of 4 and 7 years for high and low solar activity, respectively, this result is not statistically signi®cant, but as it con®rms the results in the literature, one may conclude that a mean night-time height dierence of about 1±1.5 km is a reasonable estimate for the solar in¯uence on the re¯ection height. The largest vertical gradients of the zonal prevailing wind are found in summer, when on a long-term average values of about 2 m s A1 km A1 are found (e.g. . Thus an apparent solar cycle wind variation of up to 3 m s A1 due to height variations has to be taken into account, and has to be subtracted from the values found from the analysis according to Eq. (2), since in summer at lower heights the zonal westerly winds are smaller and thus the solar cycle dependence of v oz due to height variations is negative as well. From Figs. 1 and 2 one can see that in summer the zonal wind dierence between solar maximum and solar minimum amounts to about 8 m s A1 , so one may conclude that although the height variability may possibly contribute up to nearly 40% to a solar cycle dependence of the measured v oz , it is not responsible for the whole eect, and a real wind variation has to be present. Additionally, from analyses of the years 1983 through 1995 that included the measured re¯ection heights at Collm and therefore excluded height eects   Fig. 7), wind variations within a solar cycle were found that are of the same order of magnitude as those shown in Fig. 2, and although the analysis of trends from a data base which is not much longer than one solar cycle has to be interpreted with care, the coincidence between the results presented here and those of Jacobi et al. (1997b) gives some hint that the coecients shown in Fig. 2 are for the most part due to real wind variations.
In summer the westerly mesopause jet is situated at about 95 km height. However, the interannual variability of this jet re¯ects the behaviour of the mesosphere below as well. To visualise this, in Fig. 3 shows summer mean pro®les of the zonal prevailing wind, calculated from joint analyses (Schminder et al., 1994 of Collm D1 LF and Juliusruh medium-frequency (MF) radar data. The MF radar data were provided by the Institute of Atmospheric Physics at KuÈ hlungsborn, Germany. It can be seen that the stronger westerly winds in the mesopause region in 1995 compared to 1990 are connected with weaker easterlies in the mesosphere below. Furthermore, the interannual variations appear to be stronger in the mesosphere, which can already be seen from the results of Schminder et al. (1997b, their Fig. 2). This gives a hint of a stronger vertical zonal prevailing wind gradient during solar maximum. The interannual variability of the gradient itself can be seen from the original half-hourly Collm wind data of the years 1983±1996, using a modi®ed form of Eq. (1) with height-dependent coecients:  Fig. 4. The 13monthly smoothed sunspot number is also added. Besides a long-term trend of the vertical gradient, a solar cycle dependence is also indicated. This is well visible in Fig. 5, showing the detrended summer vertical zonal wind gradients in dependence of the sunspot number R. The correlation is signi®cant at the 95%-level (t-test).
The eect of the solar cycle on the meridional prevailing wind v om , the semidiurnal tidal amplitude v 2z , and the semidiurnal tidal phase T 2z is presented in Fig. 6, showing the coecient D R as in Eq. (2) but using monthly analyses of the respective wind parameter as indicated in the legend. A weak positive solar cycle dependence of the meridional prevailing wind in winter is visible, so that during solar maximum the southerly Fig. 3. June±August mean pro®les of mesosphere/lower thermosphere zonal prevailing winds for high (1990, R = 151) and low (1995, R = 15) solar activity. The data are taken from joint analyses of Collm LF D1 and Juliusruh MF radar data (Schminder et al., 1994 winds above 90±95 km are somewhat weaker, although this correlation is not statistically signi®cant except in March and October, when rapid changes in the mesopause region wind ®eld usually occur (e.g. . So this may hint to a possible longer duration of the winter circulation in solar maximum, although analyses of the day-to-day variability of the wind parameters during March and October (not shown here) have so far not led to clear conclusions, and the interannual variability of the spring and autumn transition should thus be left to future investigation.
The amplitude of the semidiurnal tide in the most part of the year also shows no signi®cant solar cycle eect, except in April, but during this month rapid changes of the tidal wind ®eld also occur, and so any analysis of this month using monthly means should be regarded with care. The phase of the semidiurnal tide in winter is positively correlated with the solar cycle, so that the maximum winds are found at later times in years of high solar activity. It has been found from the analysis of winter winds that stratospheric warmings have some in¯uence upon the semidiurnal tidal phase (e.g. KuÈ rschner, 1981a,b, 1990), so that, considering the in¯uence of the solar cycle upon major stratwarm events (e.g. Loon, 1992, 1996), one may speculate about a solar eect on the semidiurnal tide due to winter stratospheric variability. However, the eect of stratwarm on the semidiurnal tide is small compared to the one upon the prevailing wind (e.g. , and the in¯uence may even vary from case to case (Schminder and KuÈ rschner, 1981b). Furthermore, the large day-to-day variability of the wind ®eld in winter leads to uncertainties of the estimated tidal phases, so that it is doubtful whether monthly analyses are suitable for the investigation of the long-term variability of the tidal phase, and more detailed analyses based upon shorter periods should preferably be used for such studies. In addition, similar analyses as in Fig. 6, but using the period from 1980 through 1996 only, do not show any signi®cant solar cycle eect upon T 2z , so that the results shown in Fig. 6 are mainly due to the variability of the tides in the 1970s, and this may be of non-solar origin. However, a possible solar cycle eect on the semidiurnal tide cannot be excluded. As Arnold and Robinson (1998) showed, the mesosphere below about 80 km altitude in winter is strongly in¯uenced by solar variability, and this may lead to changes in the tidal propagation during a solar cycle.

Interannual variability of the quasi-2-day wave
The quasi-2-day wave at Collm is calculated from the original half-hourly wind and re¯ection height values using an extended form of Eq. (1), but with linearly height-dependent coecients:  where two oscillations are taken into account. The frequencies included are x 1 = 2p/12 h for the semidiurnal tide and x 2 = 2p/48 h for the quasi-2-day wave.
Since P is generally found to be not exactly 48 h and therefore is not known a priori, Jacobi et al. (1997c) calculated a regression analysis according to Eq. (3) for each period P in the range 43 h < P < 58 h and used the one that provided the best ®t between measured and calculated data. They found that during the maximum of the events the period is smaller than 48 h, while before and after that time the periods can be longer. This was also found by Thayaparan et al. (1997a,b) for the years 1993 and 1994. However, since during a 2-day wave event the period values change from large ones to smaller ones and back to long periods, an analysis that includes only the 48-h period also leads to reasonable results in the vicinity of the amplitude maximum. The amplitude components of the quasi-2-day wave were calculated according to: Each analysis was made using a 10-day data window. To give an impression of the interannual variability of the 2-day oscillation, the zonal amplitudes at 95 km, a height near the maximum measuring density, are shown in Fig. 7. It can be seen that wave activity is quite variable, but there is a tendency for enhanced activity in the years 1988±1992, thus a dependence of the 2-day amplitude on the solar activity is suggested. Therefore, in Fig. 8  is shown in dependence on the sunspot number R for the months June±August. In the lowermost panel the mean summer amplitudes are presented. It can be seen that the 2-day wave amplitude is positively correlated with the solar activity, as already suggested by Jacobi et al. (1997c) for the period 1983±1995. Obviously this correlation is due to the fact that the quasi-2-day wave is positively dependent on the mean mesospheric zonal prevailing vertical wind gradient, so that the excitation of the wave is enhanced when the mesospheric easterly jet is stronger, or, in turn, the upper mesopause westerlies are weaker. Since in summer the vertical zonal prevailing wind gradient is positively correlated with the solar activity, this leads to the positive correlation of V 48 and R. The correlation coecients are shown in the respective panels of Fig. 8. However, from Fig. 8 it can be seen that only in August is the correlation relatively strong and therefore statistically signi®cant at the 95% level (t-test, r > 0.55). This could be due to the fact that the quasi-2-day wave is a very regular phenomenon and appears nearly every summer, and additionally that the events are more or less irregularly distributed, as can be seen from Fig. 7. This means that during solar maximum only an enhanced probability for the appearance of 2-day wave events is found. Since in July this probability is large in any case, the enhancement during solar maximum is not signi®cant for July, while in August, when during solar minimum the Fig. 7. Zonal amplitude v 48,z of the 2-day wave in June±August 1983±1996. The values are calculated from running 10-day data windows with a step rate of 1 day events appear more rarely, the enhanced probability for the appearance of the events can be seen more clearly. However, as this result, like that for the vertical gradient of the zonal prevailing wind, is taken from a data base of only 14 years and bearing in mind the longterm oscillations of the semidiurnal tide and the possibility that such oscillations may also be present in the signal of the quasi-2-day wave, one has to be careful in interpreting the results presented here. In addition, the correlation between quasi-2-day wave and solar activity is statistically signi®cant only in one month, so that the results concerning the solar cycle dependence of the quasi-2-day wave should be considered with some care, although they ®t into the general picture and are therefore a further hint of solar eects on the upper middle atmosphere dynamics.

Conclusions
From the long-term time series of mesopause region winds measured at Collm, possible eects of the solar variability on upper middle atmospheric dynamical properties can be inferred. It was found that the spring and summer zonal prevailing wind is negatively dependent on the solar activity, meaning the westerly winds are weaker during solar maximum. As can be seen from Fig. 3, this relation seems to be due to a forcing from below, which is indicated by joint analyses of Juliusruh mesospheric and Collm mesopause region wind data from 1990±1996 (Schminder et al., 1994. This also suggests a possible positive correlation of the strength of the summer mesospheric easterlies with the solar cycle. Although Juliusruh MF radar data are only available for a period of less than one solar cycle, the data available do support this suggestion . Additionally, regarding the vertical zonal prevailing wind gradients from 14 years of Collm measurements, this suggestion is again supported, since the gradients are positively correlated with the solar activity. However, as found, for example, by Jacobi et al. (1997a), the mesosphere/lower thermosphere circulation is variable on partially very long time-scales and the detection of any dependence may lead to incorrect results if the time-series used is shorter than about two decades. This means, considering all the measurements regarded here, that a forcing of the solar cycle dependence of the mesopause region mean circulation from below is indeed strongly suggested, but for a direct proof still longer time series of measurements may be necessary.
The quasi-2-day wave activity, however, also ®ts into this picture. The mean summer amplitude ± which is actually a mixture of maximum amplitudes and probability of appearance ± is enhanced during solar maximum, and this can be explained by the stronger vertical wind gradients at this time. Even if the data base from which the solar activity quasi-2-day wave correlation is inferred is rather short compared to the timescales of periodicities present in the upper mesosphere/ lower thermosphere dynamics, this again indicates that the mesospheric circulation is dependent on the solar activity as described in the previous paragraph.
Thus the solar cycle dependence of the mesopause region mean circulation and quasi-2-day wave would be a result of a solar cycle dependence of at least the lower thermosphere and the mesosphere. Therefore the mesopause region measurements might be useful for monitoring the long-term variability and solar cycle dependences of the entire middle atmosphere. Fig. 8. Correlation between the monthly mean 2-day amplitude V 48 and the 13-monthly smoothed sunspot number R for the months June±August, from Collm measurements of the years 1983±1996. The respective correlation coecients r are also given