During the past decades, radars have been used to investigate mesospheric phenomena, e.g., polar mesospheric echoes (, , ) and atmospheric dynamics (e.g. , , ). Mesospheric radars are distributed over the whole globe and depending on their antenna arrays, frequencies, locations, and transmitting power, they provide valuable information about winds at different vertical and spatial scales. One of the main advantages of radar systems, compared to other remote sensing techniques for the mesosphere and lower thermosphere (MLT), is that they provide continuous measurements independent of weather conditions. A crucial aspect of the measured winds is the reliability of each technique. This requires a proper understanding of the underlying scattering processes and possible instrumental effects, analysis related simplifications, and assumptions that could introduce biases or systematic errors in the derived winds .

In this study we analyze data from the Saura Medium Frequency (MF) Radar and the Andenes Meteor Radar (MR), which are both located on the island of Andøya in northern Norway (69∘ N, 16∘ E). The comparison is done based on data collected between 2004 and 2014. This study pursues two primary goals. First we want to quantify potential biases between the two techniques, and secondly we intend to merge both wind fields in order to compile a consistent and homogenous wind from approximately 60 to 110 km of altitude. Further we examine whether it is possible to fill gaps in the time series and generate a long continuous time series, ideally throughout the whole comparison period, that is suitable to study atmospheric patterns with periods between months and years. Similar comparisons have already been performed within the past few decades by , , and with smaller data sets and different locations. Their results showed particularly good agreement between MF and MR for altitudes between 75 and 85 km, but they indicate larger discrepancies at higher altitudes in the obtained winds.

Former studies very often used MF winds obtained by full correlation analysis (FCA). Most MFs employ a wide beam and use only three receiving antennas. However, due to the large observation volume there were issues with the analysis that can mainly be attributed to the underlying assumption that the FCA technique requires homogeneous volume-filled backscattering within the beam volume . A technical upgrade in 2003 for the Saura MF radar allows us to operate the MF in a Doppler beam swinging (DBS) mode and to derive winds from multiple oblique beams . The main difference of Saura, compared to many other MF radars, is its large antenna array (Mills Cross), which permits a rather narrow beam. The benefit of the MF DBS mode is that a wind and tidal comparison with MR can now be done based on the same wind fit routine. We operated the MF in DBS mode for this comparison.

The article is structured as follows. First we describe the radar systems used in this study. We explain the method to determine the wind and tides for both systems in Sect. 3, and in Sect. 4 we compare the winds and tides obtained from both systems. The discussion and conclusions are found in Sects. 5 and 6, respectively.

In this study we present observations from two different radars located at Andenes (69.3∘ N, 16∘ E). The systems are the Andenes Meteor Radar, which measures radial velocities from meteor trails, and the Saura MF radar, which obtains measurements from refraction index variations due to dynamic processes (e.g., gravity waves) and D-layer ionization and associated irregularities . The technical details for both systems are summarized in Table .

In 2001 the Andenes Meteor Radar started its continuous operation up to the present. The radar has been updated several times and the peak transmitting power has increased from 6 to 30 kW. The radar consists of one circular polarized transmitting 3-element Yagi antenna and five circular polarized receiving 2-element Yagi antennas. The receiver antenna array is arranged as five-antenna Jones configuration . The system operates at 32.5 MHz. Most specular meteors are detected at an approximate altitude of 90 km (, ). At this altitude, the observed measurement volume has a diameter of ∼ 400 km. A detailed description of the so-called All-Sky meteor radar is found in .

The Saura HF radar, historically called “MF”, is located 15 km south of the Andenes MR location and operates on a frequency of 3.17 MHz. Although the frequency used is in the HF band, Saura was designed and built as an MF radar (). In 2002 the Saura MF started its observations. The transmitting and receiving antenna is formed by 29 crossed half-wave dipole antennas in a Mills Cross arrangement. In addition to differential absorption and phase measurements of electron density, the system is able to provide winds based on atmospheric irregularities . The altitudinal coverage ranges from 50 to 100 km. Since 2003, the radar has allowed for Doppler beam swinging (DBS) experiments, which were done for this study, and spaced antenna applications. In DBS mode, off-zenith beams towards N, S, E, W and NW, NE, SW, SE at zenith angles between 6.8∘ and 7.3∘ were used. More technical information for the Saura MF radar can be found in , .

We compared wind measurements obtained from the MR and MF instruments using a DBS retrieval technique. In the case of the MR, horizontal winds are preserved using a modified All-Sky-fit Doppler approach (, , ), for which an ensemble of at least five randomly distributed meteors in a given time and altitude bin are used to estimate the 3-D wind. In the case of the Saura MF, winds are derived by combining the radial velocity measurements from four oblique and one vertical beam. In both cases, the wind vector (u, v, w) is obtained from the following set of equations: vi(θ,ϕ)rad=ucos⁡ϕisin⁡θi+vsin⁡ϕisin⁡θi+wcos⁡θi, where u, v, and w are the zonal, meridional, and vertical wind components, θi and ϕi are the zenith and the azimuth angle, respectively, and vradi is the radial velocity for each measurement. Hourly winds are obtained by binning the data in height and time. We use a 2 h sliding window centered at the reference time. A similar procedure is used for the altitude bins by applying a 3 km window shifted by 2 km and centered at a reference altitude.

The wind is computed considering the statistical uncertainties in each radial velocity measurement by applying an additional Gaussian weighting depending on its time of occurrence with respect to the reference time and for the altitude coordinate. The Gaussian-shaped window is used to provide an additional weighting of the individual meteors within a time and altitude bin. The regularization already estimates a temporal or vertical shear. This shear is used to penalize the impact of each measured radial velocity depending on its temporal or spatial offset from the reference grid. Further, this shear provides an estimate of the shear error for each time and altitude. This error is added to the statistical uncertainty in the radial velocity measurement. More information about the applied regularization can be found in .

The accuracy of the wind is obtained from the fitting procedure to estimate the wind by taking into account the number of measurements per bin and the statistical uncertainties in the measurements in the error covariance matrix. This leads to uncertainties of approximately 2–6 m s-1 for the hourly MR winds with larger errors at the edges of the observation range. This can be seen in Fig. , which shows the MR uncertainties based on the hourly zonal wind values for June 2011. The same structure occurs for the meridional wind component (not shown here). A more detailed description of the fitting routine can be found in .

Furthermore, we apply two additional assumptions to simplify the retrieval method. These assumptions are (1) zero acceleration within the altitude and time bin and (2) zero vertical velocity, which leads to a smoother wind field solution. The assumptions are used to constrain our wind retrieval by applying Tikhonov regularization (, ). Considering the rather large observation volume of a meteor radar, which has a diameter of approximately 400 km at 90 km of altitude, it is not advisable to fit for the vertical velocity directly. A wind field spanning such a volume likely also shows patterns of horizontal divergence or convergence. Using a simple gradient expansion of the wind field ( or ) shows that the horizontal divergence and the vertical wind are linked. Thus, applying the standard 3-D wind fit as it is typically applied for MST radars is not applicable to MRs.

The Saura MF winds are obtained from the radial velocity measurements. The determination of winds for the medium-frequency radar is done with the same wind fit routine based on the capability of using our five-beam experiments with one vertical and four oblique. In DBS mode, off-zenith beams towards N, S, E, W and NW, NE, SW, SE at zenith angles between 6.8 and 7.3∘ were formed for the measurements. The radial velocities of the MF radar are estimated using the momentum method , and there is no available information about the statistical uncertainties in the radial velocity measurements . Hence, we are not able to conduct a full error propagation to estimate the statistical uncertainties in the observed winds.

Theoretically, it would be possible to compare the radial velocities of meteors, which occur directly in the Saura beam, while Saura is measuring the radial Doppler. This would be more direct than using meteor wind fits. However, the systems are located only 20 km apart, and the Saura beam only points at approximately 6.8∘ off-zenith, so the number of meteors would not be sufficient to provide statistically significant winds by the meteor radar. already showed that the number of detected meteors in zenith over and within an MF beam is strikingly marginal. Only by interpolating over a longer time bin would it be possible to provide winds by the meteor system, but these winds are not meaningful for our study.

The tidal components are obtained by estimating the diurnal, semidiurnal, and terdiurnal tidal oscillation components and applying an adaptive spectral filter (). Therefore we decompose the time series as follows: u,v=u0,v0+∑n=13ansin⁡(2π/Tn⋅t)+bncos⁡(2π/Tn⋅t).

Here u0 and v0 are the mean zonal and meridional wind. Tn takes values of 24, 12, and 8 for the three tidal components and an and bn are coefficients of the tidal amplitude and the associated phases for each wind component. In this study we focused on the diurnal and semidiurnal tidal amplitudes and phases using a 5-day mean centered at the respective day to suppress smaller-scale variations in the amplitudes. We apply the same procedure to both data sets.

The aim of this study is the comparison of the obtained winds and tides from two different radar systems. On the one side we have the meteor radar and on the other side the medium-frequency radar; the two systems are able to measure wind in an overlapping observation altitude between 78 and 100 km. We compared data gathered between 2004 and 2014. Based on this comparison, we will present in the following section a method to combine both radar winds into a consistent and homogenous data set.

As a first step we analyzed the provided zonal and meridional wind components based on a running mean composite. The zonal wind structure in Fig.  exhibits the expected behavior with eastward-directed winds during the winter and a westward-dominated wind during the summer. By visual comparison of the annual climatology, which is computed from a 5-day running mean centered to the respective day, we obtain good agreement between the two systems below the altitude of 92 km for the zonal wind component, except during the wind transition period in spring. The presence of planetary waves in the wintertime can be seen in both radars. The main difference in the zonal wind component occurs above 92 km during the summer and shows for the meteor radar a strong eastward-directed wind pattern with mean values above 40 m s-1, which do not occur at the same amplitude for the medium-frequency radar with mean values around 10 m s-1.

The meridional wind component shows, as expected, lower amplitudes in both systems. Maximal wind values of ±20 m s-1 can be seen in both systems. Below 92 km the wind pattern of both radars looks similar, according to Fig. , expect during the transition time in spring and autumn when the systems sometimes show opposite wind directions.

Moreover, we compare the wind directions based on hourly data. According to Fig. , below 92 km the directions are mainly identical and above 92 km discrepancies occur, which is shown in a displacement to the line of equality. This phenomenon occurs for the whole year and increases with increasing altitudes.

Especially for lower wind values, these wind differences can be attributed to the different observation volumes of the two systems with ∼ 400 km of diameter for the MR compared to ∼ 30 km of diameter for the MF at 90 km of altitude. In addition to that volume effect, found a systematic bias (20 %) in the meridional wind component by the MF radar for altitudes below 90 km. They explained that the difference between the two systems occurs because the radars do not measure at the same altitude. The echoes of the MF radar measure in lower altitudes as expected. The reason for this is the group delay due to background ionization. Another reason could be that the signal the MF radar receives is purely due to the neutral wind . note that for the zonal wind component the difference is altitude dependent. They show good agreement during summer but with lower MF values (20–50 %) during the winter. further support these results. In agreement with their findings, we discovered an underestimation for the meridional wind below 92 km of approximately 10–60 %, which varies with season and altitude. The zonal component also shows lower wind values for the Saura MF in nearly every season and at every altitude. A reason to use the measured winds with Saura MF carefully above 92 km is due to E-region total reflection and the group retardation near midday . A study by , in which the Andenes MR was compared with the Navy Global Environmental Model (NAVGEM), a global spectral forecast model with a data assimilation algorithm, showed good agreement between their model and the MR measurements for the overlapping altitude range of each mean wind component. Furthermore a comparison between the VHF radar system MAARSY (Middle Atmosphere Alomar Radar System) and the MR at Andenes was done with correlations for the zonal (0.78) and meridional (0.79) wind components that supports the quality of the MR winds .

Differences in the wind measurements between the MR and the MF radars occur for two reasons. First, under the assumption of a stratified mesosphere, which means that the mesosphere is homogenously filled with electron irregularities for every layer, the measured center of scatters in a tilted Saura beam (Fig. ) is not necessarily in the middle of the beam volume (theta nominal). The measured center of backscatter is weighted by the electron density (theta eff) within the beam. In most cases the scattering center is weighted to lower zenith angles and therefore higher altitudes. This effect also plays an important role during strong electron events in the D and E regions. With higher altitudes this effect will increase due to broadening of the beam and can explain the differences between the MR radar and MF radar below 92 km in the wind amplitude and partially in the wind direction.

A second effect that influences the wind-derived measurements of the Saura radar is the scattering process. According to Singer (2003) and Singer (2007), sometimes there are differences of up to 10 dB between the main lobe of a vertical pointed narrow Doppler beam and the appropriate side lobes. However, the dynamic range of scattering in the D layer is about 35 dB. This side lobe contamination may affect the Doppler measurement itself, as the spectra contain multiple or show smeared and asymmetric peaks, which are difficult to take into account with the momentum method. More information about the momentum method can be found in . Thus the analysis is not able to derive a reliable radial velocity for these altitudes. Another relevant effect above 90 km is the pointing of the beam itself, the range, and the Doppler measurement. The electron density reaches values of 1011e-m-3 that require a more complex analysis, including height retardation, similar to ionosondes. Further, the beam pointing is no longer given by the pointing geometry alone, but becomes more and more affected by wave refraction. Considering electron densities close to 1011e-m-3 shows that the refractive index deviates significantly from n=1 at medium and high frequencies. Assuming a typical electron density profile, the refractive index n is between 0.4 and 0.8 for altitudes above 92 km and the Saura frequency. This implies that the range and the Doppler measurements need to be corrected for the group- and phase-velocity effects.

It would be possible to improve the measurement of Saura by applying interferometry or imaging to account for the angular or range distribution of velocities; imaging Doppler interferometry (IDI; more information can be found in, e.g., ) could be a suitable approach for more reliable radial velocity measurements in the D layer, but in retrospect this adjustment cannot be done for the existing data. However, within this study we compare the winds of the MF and MR to find a statistical correction factor to construct a merged MR–MF time series for the complete available data set of both systems. For altitudes that are likely not affected by wave refraction, a correction factor is used to estimate an average pointing difference from the nominal beam pointing.

Based on the hourly wind values of both systems, we are able to determine a correlation for both wind components for every year by using a running mean boxcar. The shape of the correlation coefficient pattern (Fig. ) shows the expected reasonable agreement below 92 km nearly for the whole year. The lower altitudes during summer and the wind transition time during spring show decreased correlations. The values of the meridional wind component are, according to these figures, higher than for the zonal component. The reason for the larger meridional values is an artifact due to the running boxcar method. According to Table  the values for the meridional wind are lower but the general pattern of the meridional figures is trustworthy. estimated a correlation (R) between the Esrange MR and the Tromsø MF radars at a height of 88 km for winter (zonal) with 0.71, winter (meridional) with 0.75, summer (zonal) with 0.8, and summer (meridional) with 0.83. Our R2 shows similar values for all cases (Table ). The difference between and our findings can be explained by different lengths of the observation time and further by our hourly data compared to get the same shape for the histogram as the 2-hourly mean data of .

The amplitude pattern of the tidal components (Fig. ) is similar to the wind components with strong differences in the amplitude above 92 km for both components. The tidal amplitude of the MF radar does not grow with higher altitude due to no increasing winds with height. The semidiurnal tide is the main dominant tidal component in both systems, which fits with the results of . showed, however, increasing tidal amplitudes with increasing heights for several locations. The amplitudes in tidal modes vary with height, season, component, and location. Further studies investigated differences in the tidal components between MF and MR at Tromsø with the result that the surface topography influences the deposition of momentum flux, and therefore a tidal acceleration can be expected to vary with altitude over the observed area and with season (, , ). In our case the surface topography does not play such an important role because the two systems are only 20 km apart. By taking wind values within a 1 h bin, the influence of different measurement volumes of both systems and a strong change in the wind to the previous and next time bin can almost be neglected. Along with the altitudinal differences in the tidal amplitude, a distinct difference also occurs between the semidiurnal and diurnal components; the correlation between the two systems for the 12 h tide is higher than for the 24 h tide with maximal values for the zonal case of ∼ 0.6 compared to ∼ 0.4 (not shown here). These values decrease with increasing altitudes. In addition to the amplitudes of the tidal components we compared the phases with a similar result. Below 88 km for the diurnal and 94 km for the semidiurnal component, the phases of both systems are in good agreement within the uncertainties, but the difference increases with increasing heights up to a time delay of 12 h for the diurnal phases.

Due to the mentioned reasons and under the assumption that the derived tidal components of the MR are correct, we recommend that the tidal components based on the MF should not be used for tidal studies above 92 km.

On the basis of these findings we propose MR wind measurements as primary wind for the overlapping coverage and filling existing gaps with weighted MF winds with respect to each wind component, altitude, and seasonal appearance. By comparing the amplitude wind differences between the two systems, we estimated theoretical correction factors (CFs) that fit, in most cases, to a polynomial function of second degree. Table  shows the parameters for the polynomials for each season and wind component. By using these parameters we show in Fig.  a theoretical profile of the CF down to 60 km for both wind components. The points between 78 and 100 km show estimated CFs based on hourly wind amplitudes without any smoothing but with respect to seasons. The estimation of these CFs is done by comparing these data for every altitude and according to their shape and multiplying the MF radar data with an associated CF (Fig. ). Figure  shows good agreement between the polynomial function and the estimated correction factors for all seasons. The highest theoretical CF below the overlapping area can be found during the summer. It should also be briefly mentioned that below and above the overlapping area these functions need further investigation because polynomials with a higher degree also fit within the overlapping area; beyond the overlapping area, where the CF is extrapolated, strong differences appear.

Figure  shows vertical profiles of the observed difference factors and our empirical correction factors. The green line shows the correction factors based on dividing the peak of MR by the peak of MF, and the red line and red dots are CFs and correction functions of Fig. . Both curves show a similar pattern. The black curves are the seasonal wind variances for MR (solid) and MF (dashed), which increase in the case of the MR with increasing heights. The variance curve of the MR fits well in shape with the two CF lines. This illustrates that one reason for the shape of the correction function is the growing wind values of the MR compared to the MF.

With the use of CF we show in Fig.  two examples for a combined zonal wind data set of both systems with altitude coverage between 60 and 100 km. They are based on a 5-day running mean for the year 2006 (Fig. a) and as a composite for the years 2004–2014 (Fig. b). Below 78 km the combined data set of the MF system is shown without the correction factor and above 92 km with only the meteor radar data. In the area in between we take mainly the MR data and fill gaps with the Saura radar data winds by applying the correction factors in Table . In general this leads to good results for both figures. Only the transition between 78 km and the altitudes below, at which the MF winds without any weighted function are connected, tend to still contain a small offset between the MF and MR winds.

Based on the correction factor we estimate an average pointing difference from the nominal beam pointing of the Saura beam, which can be seen in Fig. . Theoretically, by using a modified value of the Saura zenith pointing angle in the analysis, the MF and MR winds would agree better.

In this study we have compared the winds of the Saura MF and Andenes MR by applying a DBS wind analysis for both systems. Data from 11 years have been studied with the objective of obtaining a vertical wind profile between approximately 60 and 110 km. To acheive this, biases between the two systems were determined, mainly to remove system-specific differences, such as scattering processes, technical setup, and frequencies. Inside the overlapping altitude in the range from 78 to 100 km, the highest agreement (zonal and meridional R2=0.78/0.70) of the two wind components is between 78 and 94 km, except during the transition time (spring) and during the summer below 82 km. Above 92 km the correlation decreases with increasing altitude. It is clear that there is no perfect correlation between the two radars, especially on shorter timescales, which is due to fundamental differences in the systems, e.g., different measurement volume, frequency, and scattering processes. We compared the derived tidal components between the two radars. The amplitudes and the phases of the diurnal and semidiurnal tides from both sets of measurements are in agreement below 90 km. With increasing altitudes above 90 km, the mean semidiurnal phases are in agreement within ±6 h and the diurnal phases are within ±12 h. The use of tidal phases at these heights should therefore be taken with caution. The best agreement occurs during the winter period below 90 km for the semidiurnal tide. With increasing altitudes the agreement decreases because the phase and the amplitude of the MF-based tides remain almost constant with increasing altitude. This is not supported by the MR observations, which show a clear phase propagation with altitude and increasing altitudes. Based on our findings we provide a correction function for every season to minimize differences in wind amplitudes between the two systems. These correction functions fit to a polynomial function of second order, but should only be used for the altitudes at which both systems are able to obtain winds. Extrapolating the correction function beyond the overlapping area can cause problems and needs further investigation. By combining the MR and the weighted MF data set, we are able to construct a continuous data set with altitude coverage from 60 to 110 km over 11 years, which can be used for further studies.