Interhemispheric differences of mesosphere–lower thermosphere winds and tides investigated from three whole-atmosphere models and meteor radar observations

. Long-term and continuous observations of mesospheric–lower thermospheric winds are rare, but they are important to investigate climatological changes at these altitudes on timescales of several years, covering a solar cycle and longer. Such long time series are a natu-ral heritage of the mesosphere–lower thermosphere climate, and they are valuable to compare climate models or long-term runs of general circulation models (GCMs). Here we present a climatological comparison of wind observations from six meteor radars at two conjugate latitudes to validate the corresponding mean winds and atmospheric diurnal and semidiurnal tides from three GCMs, namely the Ground-to-Topside Model of Atmosphere and Ionosphere for Aeron-omy (GAIA), the Whole Atmosphere Community Climate Model Extension (Speciﬁed Dynamics) (WACCM-X(SD)), and the Upper Atmosphere ICOsahedral Non-hydrostatic (UA-ICON) model. Our results indicate that there are interhemispheric differences in the seasonal characteristics of the diurnal and semidiurnal tide. There are also some differences in the mean wind climatologies of the models and the observations. Our results indicate that GAIA shows reasonable agreement with the meteor radar observations during the winter season, whereas WACCM-X(SD) shows better agreement with the radars for the hemispheric zonal summer wind reversal, which is more consistent with the meteor radar observations. The free-running UA-ICON tends to show similar winds and tides compared to WACCM-X(SD).

in hourly values. GAIA has been demonstrated to be particularly good at capturing comprehensive coupling processes between 90 the lower and upper atmosphere at different temporal and spatial scales, e.g, the wave-4 structure, the thermosphere cooling during stratosphere sudden warmings (SSWs) (Liu et al., 2009a;Liu et al., 2014).
This study uses the same 21-year long reanalysis data-driven simulation results as that used for ENSO study in Liu et al. (2017). Briefly, A nudging technique is used to constrain the model output (e.g. pressure, temperature, wind, etc.) below 30 km altitude to the reanalysis data JRA-25/55 by Japan Meteorological Agency with a 1.25 • × 1.25 • spatial resolution and a 95 6-hour temporal resolution (Onogi and et al., 2007;Kobayashi et al., 2015). Due to the update of JRA-25 to JRA-55 in 2014, the simulation uses JRA-55 for 2014-2016 and JRA-25 before that. The F10.7 index as a proxy for the EUV input was set to observed values, while a fixed cross polar cap potential of 30 kV and a quiet particle precipitation condition were held throughout the simulation period to exclude any geomagnetic activity effect.
using a 300 km radius, which is a bit more than the actual beam width used in the wind retrieval of about 220 km radius, but ensures that at least 5 grid points are available from each model and for each site. We extracted for each meteor radar location the geopotential height, the zonal and meridional wind, temperature as well as pressure for all grid points that fall within the above mentioned area around the meteor radars. These reduced data sets are now further analyzed to simulate meteor radar observation. Therefore, we converted the geopotential heights (Φ) into geometric altitudes (h) for each extracted profile using the expression by taking into consideration of variable gravity; 145 h(lat, lon) = Φ(lat, lon)/ 1 − Φ(lat, lon) R Earth (lat, lon) . (1) Here R Earth (lat, long) corresponds to the Earth radius at a given latitude and longitude. In a next step, the converted height vectors for each profile are interpolated to a reference altitude vector, which has a vertical resolution of 2 km between 16-150 km, 5 km vertical resolution at altitudes from 155-200 km and 10 km vertical resolution between 210-300 km and a 20 km vertical resolution at altitudes above 320 km to account for the decreased model resolutions due to the pressure level spacing 150 in the models. Finally, we compute the median and variance for all profiles and obtain a mean zonal and meridional wind and temperature corresponding to the observation volume of each meteor radar. Furthermore, we derive the variance of these parameters, which provides a proxy for the statistical uncertainties similar to the meteor radar observations. The final result of our data reduction are time series of zonal and meridional winds with a temporal resolution of 1 hour for GAIA and UA-ICON and 3 hours for WACCM-X(SD) respectively, and a 2 km vertical resolution at the mesosphere/lower thermosphere, which is 155 identical to the meteor radar observations. MR winds are obtained using the retrieval algorithm described in Stober et al. (2018), which is basically a further evolution of Hocking et al. (2001) and Holdsworth et al. (2004). The retrieval includes a full Earth geometry based on the WGS84 reference ellipsoid, full non-linear error propagation and a spatio-temporal Laplace filter as Tikhonov regularization constraint.
Furthermore, the wind retrieval does not require w = 0 and explicitly fits for the vertical component, which are considered as 160 remaining wind bias due to the lack of independent validation sources. However, these vertical winds have proven to provide a good quality control and show a Gaussian distribution with a width of w ± 0.25 to ±0.35 m/s around the zero wind line.
The benefit of this retrieval is that we obtain for all systems a harmonized wind time series based on the same quality control criteria.
Atmospheric mean winds and tides are analyzed using the ASF, which is described in more detail in Baumgarten and Stober 165 (2019) and Stober et al. (2020a) and was already applied in several studies Pokhotelov et al., 2018;Wilhelm et al., 2019;Stober et al., 2020b) to decompose MR winds in daily mean winds, diurnal and semidiurnal tides for the zonal and meridional components, respectively. The technique is implemented based on least square fits with full error propagation, which permits to apply the algorithm to unevenly sampled data with data gaps. Similar to wavelets the window length is adapted for each of the fitted wave periods (Torrence and Compo, 1998). Furthermore, we minimize the impact of inertia-scale 170 gravity waves on the tidal analysis by applying a vertical regularization to the tidal phases. In Stober et al. (2020b) shows an example comparing the ASF2D with classical harmonic analysis for different window lengths. Due to the intermittent and non-stationary wave field generated by gravity waves and tides, long window lengths tend to produce artefacts and leak energy between the different wave scales. Furthermore, the meteor radar sampling is irregular in time, which additionally introduces spectral leakages that are significantly reduced by the ASF2D technqiue.

175
The GAIA, WACCM-X(SD), UA-ICON and MR time series are analyzed with the same ASF2D algorithm to ensure the best possible comparability and to minimize differences, which might be introduced when different analysis procedures are applied.
Thus, we obtain harmonized time series for daily mean winds, diurnal and semidiurnal tidal amplitudes and phases as well as a gravity wave spectral residuum for each data set. The model data is available with different temporal resolutions, which refers to the cadence of the data output of the meteorological fields rather than the actually numerical temporal step size (e.g.

180
WACCM-X(SD) is solved with 5 minutes resolution). Hence, the model data for each temporal step represents the numerical solution for this output period and not a temporal average as in the observation. Furthermore, we performed an additional test to ensure that the coarser temporal resolution of WACCM-X(SD) of 3 hours has no impact on the harmonized time series.
Thus, we used an earlier version WACCM-X (v.1.9) run with hourly data output for cross-validation and found no resolution dependent issues.

185
Vertical wavelengths of the diurnal and semidiurnal tides are also estimated from the vertical phase profiles. Therefore, we estimated the instantaneous vertical linear slope of the unwrapped phases in the altitude range between 80-100 km. The vertical wavelength is then estimated from the altitude difference between the −π and π phase transitions. This method allows to derive vertical wavelengths that are much longer than the actual width of the meteor layer. Such long vertical wavelengths correspond to evanescent tidal modes. However, we did not define a certain threshold, but consider vertical wavelengths that are much  The seasonal morphology in WACCM-X(SD) shows substantial differences from the MR measurements during the hemispheric winter months: the zonal wind changes from eastward to westward between 70-80 km and remains westward at most MLT heights, whereas observational data shows no reversal and are eastward in this region. The summer wind reversal from westward to eastward winds can be found in the model as well as in the observations. As such, the general seasonal morphology  Furthermore, it is noticeable that UA-ICON tends to capture the seasonal asymmetry in the zonal winds and shows a gradual decrease of the summer wind reversal altitudes as it is seen in the observations.
Meridional winds are compared in Figure 2  GAIA exhibits a very similar seasonal characteristic for both stations. However, the northward winds during the northern hemispheric winter have an increased magnitude compared to the observations at SOD and are less strong above DAV. Nevertheless, GAIA is capable of capturing the main seasonal features in the meridional component for both locations.
Comparing meridional winds in WACCM-X(SD) with the observations reveals distinct differences between the conjugate lati-230 tudes. At the southern hemisphere above DAV the seasonal morphology is well-reproduced in WACCM-X(SD) and shows the northward winds during the northern hemispheric winter and southward winds during May to August. This is not the case for SOD, where WACCM-X(SD) shows an entirely different seasonal meridional wind throughout the year, which is most of the time southward at the altitude range between 80-100 km. Furthermore, the model exhibits a wind reversal from southward to northward during the summer months May to September above 100 km, which is not indicated in the observations. Meridional 235 winds in UA-ICON show again a seasonal characteristic similar to WACCM-X(SD). However, the magnitude of the meridional winds appears to be in better agreement with the observations. In particular, the meridional winds at DAV look similar compared to the observations.

Diurnal tides
At polar latitudes diurnal tides gain only moderate amplitudes, although being still visible throughout the year, which is also predicted from the Laplace tidal equation and the corresponding Hough modes (Andrews et al., 1987;Wang et al., 2016). The amplitudes reach their largest values during the hemispheric summer months. Furthermore, the zonal and meridional amplitudes show consistent seasonal patterns. Typically, at middle and high latitudes the meridional amplitudes exceed the zonal amplitudes during the summer months as documented before (Portnyagin et al., 2004;Jacobi, 2012;She et al., 2016;Wilhelm et al., 2017;Baumgarten and Stober, 2019;Pancheva et al., 2020). Figure 3 presents the comparison between SOD and DAV 245 with a similar arrangement of the panels as for the mean winds. The MR observations reveal a characteristic vertical structure of the diurnal tidal amplitude for the hemispheric summer months, which shows a first enhancement at altitudes below 80 km and a second maximum above 95-100 km. Furthermore, there is a second hemispheric winter maximum apparent at DAV during June and July at altitudes below 80 km, which is less obvious at SOD.
GAIA and UA-ICON capture most of the seasonal characteristic of the diurnal tidal amplitudes above 90 km, but shows no 250 tidal enhancements below 80 km. The meridional amplitudes are also more amplified compared to the zonal component at both hemispheres, which is consistent to the observations. However, the vertical structure of diurnal amplitudes is less visible relative to the observations. This is also the case for WACCM-X(SD). WACCM-X(SD) also shows the diurnal tidal enhancement above DAV during the hemispheric winter below 80 km, which is not found in GAIA and UA-ICON. However, in WACCM-X(SD) this lower amplitude maximum is stronger than amplitudes above 100 km during the same period, which appears to be 255 reversed compared to the observations. In general the amplitudes of the diurnal tide appear to be almost equal in the zonal and meridional component for DAV, which is also reflected by all three models. Only at SOD the observations indicate enhanced amplitudes in the meridional component, which seems to be less pronounced in all three model outputs. One noticeable aspect of UA-ICON is the diurnal tidal seasonal climatology, which indicates much lower amplitudes during hemispheric winter compared to the observations and the other two models.

260
Comparing the diurnal tidal phase between SOD and DAV MR observations we found remarkable differences in the seasonal pattern between both hemispheres (see Fig. 4). The zonal and meridional diurnal phase at DAV remains more or less stable throughout the year indicating only little annual variation with longer vertical wavelengths in the meridional component, whereas at SOD a pronounced semiannual structure is observed with distinguished phase changes in April/May and October.

265
WACCM-X(SD) and GAIA present a very similar seasonal diurnal tidal phase characteristic for both components, which deviates by several hours relative to the MR measurements at both locations. In this respect, there is a less good agreement of the vertical diurnal tidal phase structure for both models in comparison to the observations. For the DAV location WACCM-X(SD) meridional phase even shows a jump above 100 km, which is not seen in GAIA. UA-ICON diurnal tidal phases exhibit an offset compared to the observations, but the vertical structure appears to be similar to GAIA and WACCM-X(SD).

270
Vertical wavelengths for the diurnal tide are presented in Figure 5. All three models tend to overestimate the vertical wavelength compared to the MR observations, which could be already seen in the phase plots. These very long vertical wavelengths reaching more than 1000 km n all models during certain periods practically indicate that the tidal modes are evanescent and not vertically propagating.

Semidiurnal tides 275
Semidiurnal tides are the dominating tides at mid-and high-latitudes at the MLT and reveal a characteristic seasonal structure (Portnyagin et al., 2004;Wilhelm et al., 2019;Pancheva et al., 2020;van Caspel et al., 2020), but also show significant interhemispheric differences. Figure  similar seasonal characteristic as WACCM-X(SD) and the observations. The general seasonality of the semidiurnal tidal amplitudes is fairly consistent between the models, but exhibits deviations from the observations, which are more significant in the southern hemisphere.
In Fig. 6 we present the semidiurnal tidal phases for SOD and DAV. As already shown for the amplitudes, there are significant interhemispheric differences at polar latitudes. The SOD MR observes a quasi-biannual tidal phase characteristic, whereas at 295 DAV station the annual pattern is more evident. In the northern hemisphere the semidiurnal tidal phase is basically drifting with season and shows phase variations of several hours through the course of the year. Southern polar latitudes exhibit a much smaller phase variability and more constant seasonal variation of the phase, which is still significant and exceeds several hours, but the changes are not as rapid as in the northern hemisphere.
As expected from the diurnal tidal phase analysis, the models have difficulties to match the phase variability and seasonal 300 characteristics especially in the southern hemisphere. GAIA appears to have a slightly better agreement with the observations above DAV at altitudes between 80-100 km in both components. WACCM-X(SD) indicates a better performance for the northern hemisphere and, in particular, captures the phase variability during the fall transition in both components, which is only weakly present in the GAIA data. Surprisingly (taking into account that the model is free-running), UA-ICON semidiurnal tidal phases are reasonable well-reproducing the observations for the northern hemisphere and partly also in the southern hemi-  altitudes corresponding to westward zonal wind. In the southern hemisphere, we observe a smoother vertical structure of the meridional wind, which even reverses to northward above 84 km in July and August.
As already found for the polar latitudes, the zonal wind climatologies are only to a certain extent reproduced in the models.
GAIA shows a better agreement during the winter months, whereas WACCM-X(SD) and UA-ICON exhibit a better agreement of the summer zonal wind reversal compared to the MR measurements. All three models tend to overestimate the strength and UA-ICON exhibit more remarkable differences relative to GAIA and the observations. In the northern hemisphere the meridional wind is southward at altitudes between 90 and 100 km throughout the year. Both models apparently only reproduce 360 the seasonality, that is found in the observations, below 80 km altitude. In the southern hemisphere WACCM-X(SD) and UA-ICON show a better agreement during the hemispheric summer months (Dec-Feb), but tend to overestimate the wind reversal from southward to northward as well as the seasonal duration compared to the TDF observations.

Diurnal tides
At mid-latitudes diurnal tidal amplitudes have a similar seasonal characteristic as at the polar latitudes. During the hemispheric 365 summer season the amplitudes are largest and remain rather small throughout the other months. Figure 10    Diurnal tidal phases are presented in Figure 11. Some of the features found at the polar latitudes are again visible at the mid-latitudes. The seasonal characteristic of the diurnal tidal phases indicates some interhemispheric differences, which were already indicated for SOD and DAV. In the northern hemisphere the tidal phase shows a more pronounced biannual mode at altitudes above 84 km, whereas in the southern hemisphere an annual structure is more evident. All three models show 375 dissimilarities in the seasonal vertical phase characteristic. Above 95 km GAIA and WACCM-X(SD) systematically deviates from the observations. UA-ICON exhibits an offset in the diurnal tidal phases compared to the observations as well as GAIA and WACCM-X(SD). In general, below 95 km, there is a much better agreement between the observations and models. Figure 12 shows the vertical wavelength for the diurnal tide at the mid-latitude locations above COL and TDF. All three models and the observations look dissimilar. Interestingly, also an inter-model comparison does not indicate a substantial 380 agreement for the vertical wavelengths between the different models. However, contrary to the polar latitudes UA-ICON now tends to overestimate the vertical wavelength a bit more compared to GAIA and WACCM-X(SD), which was opposite at polar latitudes.

Semidiurnal tides
At mid-latitudes the semidiurnal tide is the dominating atmospheric wave and gains amplitudes of about 50 m/s on an average 385 and occasionally up to 70 m/s above 90 km altitude. The seasonal characteristic is presented in Figure 13 and exhibits an inherent interhemispheric difference. The seasonal amplitude behaviour at COL looks rather similar to that at SOD, but reflects the much higher amplitudes during the winter months from November to February in the northern hemisphere. As expected (e.g., Jacobi, 2012), the zonal and meridional amplitudes are almost the same. Apparently, TDF in the southern hemisphere indicates a different seasonal amplitude characteristic. TDF exhibits still the largest amplitudes during the winter months from 390 April to October, but with much weaker amplitudes compared to COL. Furthermore, at TDF the zonal component shows larger amplitudes than the meridional semidiurnal tide.
The GAIA model shows only a weak seasonality of the semidiurnal tidal amplitudes and even larger amplitudes above 100 km than the MR measurements. GAIA exhibits almost no interhemispheric differences of the tidal amplitudes. Only in the northern hemisphere, GAIA indicates a semidiurnal tidal enhancement for the winter months, as it is found in the observations.

395
WACCM-X(SD) and UA-ICON indicate a rather good representation of the semidiurnal tide in the northern hemisphere for the location of COL. The seasonality of the amplitudes are well-captured and exhibit a remarkably good agreement compared to the observations. In the southern hemisphere the semidiurnal tides are less well represented in WACCM-X(SD). The model shows a hemispheric summer tidal enhancement at altitudes above 90 km, which is missing in the observations. Furthermore, the amplitudes appear to be increased relative to the observations at TDF. For the hemispheric winter months above TDF, 400 WACCM-X(SD) shows increased tidal amplitudes relative to the observations, but captures the general hemispheric winter characteristic from May to July/August. Interestingly, UA-ICON indicates the best agreement with the observation for the seasonality of the semidiurnal tidal amplitudes on both hemispheres and even reproduces the interhemispheric differences quite well.
Semidiurnal tidal phases for the mid-latitude conjugate comparison are shown in Figure 14. The seasonal phase characteristic is 405 rather similar compared to the polar latitudes. The measurements as well as the models indicate a significant interhemispheric difference that was already depicted in the amplitudes. On the northern hemisphere, we find a biannual seasonal phase characteristic in the observations that is well-reproduced in the WACCM-X(SD) and UA-ICON data. GAIA also shows a reasonable agreement, but does not reflect the quick phase change during the northern hemispheric fall transition in September/October.
In the southern hemisphere the observations at TDF show a more smooth seasonal phase characteristic that appears to be only 410 partially reproduced by the three models, which show distinguishable phase differences compared to the measurements.
Vertical wavelengths are shown in Figure 15.  Finally, we investigate the KIR high-latitude meteor radar data and the CMOR observations in Canada, which is located at the lowest latitude used in this study. The KIR meteor radar is included as sanity check for the robustness of the meteor radar observations as it is located in the proximity of SOD. In addition, these two stations provide a good comparison to show how the seasonal characteristic of mean winds, diurnal and semidiurnal tides change with latitude. The data are displayed in The mean wind and tidal climatologies at SOD and KIR are almost identical and, thus, show a similar agreement with GAIA, WACCM-X(SD) and UA-ICON. Comparing KIR and CMOR provides a more direct assessment of latitudinal differences.
During the northern hemispheric winter CMOR observes an eastward zonal wind, which reaches higher magnitudes compared to KIR, but also indicates a wind reversal above approximately 100 km to a weak westward wind. The summer wind reversal from westward to eastward winds occurs at an almost 5-8 km lower altitude relative to KIR. GAIA, WACCM-X(SD) and Both models are nudged to the reanalysis data. GAIA is nudged to the JRA-25 and from 2015 on to JRA-55 reanalysis data (Kobayashi et al., 2015) up to 30 km, whereas WACCM-X(SD) is driven by MERRA (Rienecker et al., 2011) up to 50 km. 490 Harada et al. (2016) performed a cross-comparison among various reanalysis data sets to investigate potential differences and found that JRA-25, JRA-55 and MERRA already showed some differences in storm tracks and mean winds and temperatures.
In particular, at the upper stratosphere the reanalysis data sets and observation from MLS indicated some differences among each other. Due to the nudging of GAIA and WACCM-X(SD), these differences enter also into the model fields and partly explain differences among the models as well as concerning the MR observations. The nudging height may also play some role 495 since the systematic differences have been found in the upper stratosphere (Sakazaki et al., 2018).
Furthermore, the reanalysis data used for the nudging of GAIA and WACCM-X(SD) seems to be less relevant for the representation of the diurnal and semidiurnal migrating tide climatologies. Ortland (2017) estimated the tidal forcing for the DW1 and SW2 due to absorption of solar radiation from ozone and water vapor. They found that the tidal correspondence between the Tide Mean Assimilation Model (TMAT) and observations at the MLT strongly depend on the forcing at the troposphere and 500 stratosphere for both migrating tides. Thus, differences in the tides between the models are most likely the result of different implementations of the radiative transfer and distribution of tropospheric/stratospheric water vapor and ozone causing differences in the radiative forcing. This is further supported by the free running UA-ICON, which shows very good agreement to the tides produced in WACCM-X(SD), but is not driven by any reanalysis data.
good agreement with the semidiurnal tidal phases at mid-and polar latitudes capturing many of the seasonal characteristics that are found in the MR observations. It is also obvious that GAIA tends to better agree on the southern hemisphere and WACCM-X(SD) performs a bit better on the northern hemisphere. Only the diurnal tidal phases at polar latitudes are dissimilar in both models compared to the MR observations. The vertical wavelength of the diurnal tides in GAIA and WACCM-X(SD) suggest almost evanescent diurnal tidal modes, whereas the observations indicate much shorter wavelengths and a vertically propagat-510 ing diurnal tide. Both models reduce the longitudinal grid resolution closer to the pole to avoid the singularity. This seems to favour a damped vertical propagation of the diurnal tide, but has to be investigated in more detail.
Non-migrating tides are also worth to discuss. In fact, our MR observations are local and, thus, the observed diurnal and semidiurnal tides are a superposition of the migrating and non-migrating tidal modes. At the lower latitudes, the generation of non-migrating tides is well-understood due to the latent heat release in the tropics Forbes, 2002, 2003;Oberheide 515 et al., 2011). At mid-and polar latitudes non-migrating tides can be generated by various processes such as latent heat release, nonlinear interactions with stationary planetary waves (Yamashita et al., 2002;Smith et al., 2007;Murphy et al., 2009;Miyoshi et al., 2017) and other tidal modes, variations in the mean background wind and temperature field, and gravity wave breaking or dissipation regions . Furthermore, there were some studies investigating SSW as a potential cause to excite the westward propagating semidiurnal tides with wavenumbers 1 and 3 (Du et al., 2007;Liu et al., 2010b;Stober et al., 2020a). showed that secondary or non-primary wave generation provides an essential contribution to the MLT wind forcing above GW hot spots such as the southern Andes and the Antarctic Peninsula . A first observational evidence was obtained at McMurdo investigating lidar observations . More recently a detailed study of the MLT dynamics for the year 2019 using six meteor radars from Tierra del Fuego, South Georgia, Rothera, King Sejong Station, Davis and McMurdo indicated a strong impact of non-primary waves above the Andes and Antarctic Peninsula 545 on the daily mean zonal and meridional winds and momentum fluxes (Stober et al., 2021). Including non-primary waves into the GW parameterizations for the hemispheric winter could add eastward momentum to the MLT, which is apparently too weak in WACCM-X(SD) and UA-ICON. The strong winter stratospheric eastward polar vortex efficiently removes all eastward GW by critical level filtering and, thus, only westward GW propagate to the mesosphere and can deposit their momentum resulting in an westward forcing and westward mean winds. Considering non-primary waves in the parameterization could essentially 550 balance the total forcing at the MLT and may help to get the mean winds to a better agreement with the observations.

Conclusions
In this study we compared GAIA, UA-ICON and WACCM-X(SD) predictions with local meteor radar observations applying a unified diagnostic to decompose the wind field into mean winds, as well as diurnal and semidiurnal tidal amplitudes and phases 555 in the MLT. Therefor we present observations from six meteor radars and derived climatologies from the continuous observations for the above mentioned meteorological parameters, which are cross-compared to nudged model simulations from GAIA and WACCM-X(SD) for the same periods as the measurements are available from each radar. In addition, a 21 year UA-ICON free-running GCM run was employed for comparison.
Although all models utilize similar gravity wave parameterizations schemes, but different implementations, the zonal and 560 meridional winds exhibit seasonal and interhemispheric differences between GAIA, WACCM-X(SD), UA-ICON and the MR observations. It is obvious that GAIA shows a better agreement of mean winds during the winter season in both hemispheres compared to the meteor radar, whereas WACCM-X(SD) and UA-ICON indicates a better agreement of the zonal wind reversal from westward to eastward in both hemispheres with the observations. However, one has to note that GAIA seems to produce a too weak and high up hemispheric summer zonal wind reversal. WACCM-X(SD) and UA-ICON tend to show westward winds 565 during the winter season pointing towards too much westward wave drag at these altitudes. Furthermore, meridional winds appear to be in remarkable agreement between GAIA and the MR observations at mid-and polar latitudes, while WACCM-X(SD) and UA-ICON indicates more dissimilarities in the meridional winds relative to the MR observations. UA-ICON, as a free-running model, nevertheless shows remarkably good representation of MLT wind fields compared to the two nudged models.

570
The ASF decomposition of the time series from the models and the meteor radar observations ensures a harmonized tidal comparison of the amplitude and phases. Atmospheric tides provide an essential source of variability for the coupling of the middle atmosphere to the ionosphere. Daily tidal amplitudes and phase are obtained from the ASF and vector averaged, which reduces the contamination of the amplitude and phase due to the tidal intermittency caused by non-linear wave-wave interactions, changes in the mean winds or source variabilities. There is a good agreement of the GCMs for the diurnal tide amplitude 575 and phase for the latitudes investigated here in. Diurnal tides indicate only weak interhemispheric differences and reach the largest amplitudes above 95-100 km during the hemispheric summer months. The seasonality of the diurnal tidal amplitudes is well-reproduced by GAIA, UA-ICON and WACCM-X(SD). However, diurnal tidal phases show some differences between the observations and the GCMs. GAIA, UA-ICON and WACCM-X(SD) tend to exhibit much longer vertical wavelengths compared to the MR measurements.

580
Semidiurnal tides are the dominating tidal mode at mid-and high-latitudes through the course of the year and MLT altitudes.
One of the main results of this comparison are the distinct differences of this tide between both hemispheres. It appears that for conjugate latitudes the semidiurnal tide reaches higher amplitudes in the northern hemisphere at mid-and polar-latitudes. In particular, the amplitude and enhancement and phase variability in September on the northern hemisphere is not found at the southern latitudes during the transition from the hemispheric summer to the winter circulation. More detailed investigations are 585 required to distinguish potential reasons, which are likely caused by a complex chain of interactions due to the differences in the land-sea distribution, GW sources and planetary waves between both hemispheres that alter the polar vortices and, thus, the ozone transport into the polar cap, which again provides a feedback on the excitation of tides. GAIA, UA-ICON and WACCM-X(SD) indicate a reasonable agreement of the semidiurnal tidal amplitude and phase. There is a tendency that WACCM-X(SD) and UA-ICON has a better agreement with the MR observations on the northern hemisphere, whereas GAIA seems to agree 590 better on the southern hemisphere. However, all GCMs have a tendency to overestimate the summer hemisphere semidiurnal tidal amplitudes above 100 km.
The climatological comparisons of mean winds and diurnal and semidiurnal tides underline the value of continuous observations in the MLT to evaluate/assess GCMs. GAIA, WACCM-X(SD), and UA-ICON are state of the art models, coupling the middle atmosphere with the upper atmosphere to study the forcing from below of the thermosphere/ionosphere system and a 595 potential feedback to the middle atmosphere. Therefore, we assessed the climatological state of the mean winds and the tidal activity at the MLT. We identified systematic dissimilarities in the mean zonal and meridional winds and in the seasonal characteristic of tidal amplitudes and phases. However, there was a remarkable agreement in both hemispheres of the semidiurnal tide between the observations and the free-running UA-ICON, which further underlines that the climatological behavior at the MLT seems to be not driven/improved by the nudging of GCMs to reanalysis data. 600 of the results. KB provided an essential contribution developing the ASF method. All authors contributed to the editing and writing of the manuscript. AK and ML provided SGO meteor radar data. EB and JK contributed the Esrange meteor radar observations. DJ shared the TDF meteor radar measurements and DM supported this work by providing DAV data. PB provided the CMOR radar data.
Competing interests. The authors declare that there are no competing interests.