Wuhan MST radar: Technical features and Validation of wind observations

The Wuhan MST radar is a 53.8 MHz monostatic Doppler radar, located in Chongyang, Hubei Province, China, which has the capability to observe the dynamics of the mesosphere-stratosphere-troposphere region in the subtropical latitudes. The radar system has an antenna array composing of 576 Yagi antennas, and the maximum peak power is 172 kW. The Wuhan MST radar is efficient and cheap, which employs simplifier and more flexible architecture. It includes 24 big TR modules, and the row/column data port of each big TR module connects 24 small TR modules via the corresponding 15 row/column feeding network. Each antenna is driven by a small TR module with peak output power of 300 W. The arrangement of the antenna field, the functions of the timing signals, the structure of the TR modules, and the clutter suppression procedure are described in detail in this paper. We compared the MST radar observation results with other instruments and related models in the whole MST region for validation. Firstly, we made a comparison of the Wuhan MST radar observed horizontal winds in the troposphere and low stratosphere with the radiosonde on 22 May 2016, as well as the 20 ERA-interim data sets (2016 and 2017) in the long term. Then, we made a comparison of the observed horizontal winds in the mesosphere with the meteor radar and the HWM-07 model in the same way. In general, good agreements can be obtained, and it indicates that the Wuhan MST is an effective tool to measure the three-dimensional wind fields of the MST region.


Introduction
The mesosphere-stratosphere-troposphere (MST) radars has been used for studying the dynamics of the lower and middle 25 atmosphere up to 100 km altitude for several decades (Hocking et al., 2011), since Woodman and Guillen observed radar echoes from the stratospheric and mesospheric heights with the Jicamarca radar in 1970s (Woodman et al., 1974). In general, large antenna array is employed by these MST radars to measure the weak echoes scattered by the turbulence (Green et al., 1979). Many MST radars have been developed world-wide by different countries and groups, and the MST community plays a significant role in technique sharing. According to the antenna array shape, the existing MST radar in the world can be 30 2 divided into two types: the square array arranging the elements in a square gird and the circular array arranging the elements in a triangular grid. The MST radars using the square array mainly include the Jicamarca radar (Woodman et al., 1974), the Sousy radar (Schmidt et al., 1979), the Poker Flat radar (Balsley et al., 1980), the Esrange MST radar (Chilson et al., 1999), the Gadanki radar (Rao et al., 1995;Rao et al., 2019), the Chung-Li radar (Rottger et al., 1990;Chu et al., 2009) and the NERC MST radar (Vaughan et al., 2002;Hooper et al., 2013). The MST radars using the circular array include the MU radar 35 (Fukao et al., 1985;Kawahigashi et al., 2017), the EAR radar , the MAARSY radar (Latteck et al., 2012) and the PANSY radar (Sato et al., 2014).
The MST radar has been developing slowly in Chinese sector by reason of the high price. Until 2008, the Wuhan MST radar and Beijing MST radar began to construct with the support of Meridian Project of China (Wang, 2010). We have introduced the two MST radars of Chinese Meridian project in 2016 (Chen et al., 2016). This paper briefly introduced the 40 antenna array of the Beijing and Wuhan MST radars and their preliminary observations. The two MST radars work more than 280 days every year and their data can be freely accessed in the data center for the Meridian Project (http: //159.226.22.74/). Thus, the radar system and their data are gained extensive attention and we have received many letters inquiring about the details of the radio system, as well as the data format and reliability. Therefore, we plan to write a new article in response to the readers' and users' demand, who want to build a low-cost MST radar or apply the data of the 45 MST radars of Chinese Meridian project. The paper presents more details of the Wuhan MST radar including some optimized circuits, as well as its recorded data.
The Wuhan MST radar is located in Chongyang, Hubei Province, China (29.5°N,114.1°E). The location is far away from the bustling city, so as to better avoid interference by radio noise. Considering this is China's first attempt to develop its own MST radar, the radar station is not selected in some areas of great difficulty in construction, such as equatorial low latitudes, 50 polar regions and plateaus. Chongyang is located in the central plain of China, which is an appropriate choice. As one of few MST radars in the midlatitudes, it can be one important member of the global MST radars. In addition to the scientific research goals, the Chongyang station also serves as a students' training base for the practice of radar technologies and meteorological applications. The Wuhan MST radar was completed preliminarily in 2011. The system was upgraded in 2016, and the TR modules were updated for better stability and better detection capability. The facility costs only about 55 $1,000,000.00, which is far lower than the high cost of other MST radars. However, it provides an average power aperture product (PAP) product of 3.2×10 8 Wm 2 . Considering the balance of system performance and project implementation, simpler and more flexible architectures are applied in the system.
The first aim of the present paper is to introduce the technical features of the Wuhan MST radar. In particular, the antenna field, the timing signal, the TR module, the digital receiver and the clutter suppression will be discussed in detail. The 60 second aim of this paper is to present the the recorded data and compared with the wind fields recorded by other instruments and related models for validation. The Wuhan MST radar is arranged in a 24 × 24 matrix with a side length of 96 m, which consists of 576 Yagi antennas. Each antenna is driven by an individual small TR module (300 W). According to the antenna radiation pattern, the beam width is 3.2°. The shortest width of the subpulse is 1 μs to satisfy the requirement for a maximum range resolution of 150 m.
The radar system allows very high flexibility of waveform parameter for different detection modes (low mode, middle mode, 70 and high mode). The basic specifications of the Wuhan MST radar are list in Table 1. The hardware of the Wuhan MST radar consists mainly of five subsystems: the antenna array, the TR module, the radar controller, the digital transceiver, and the signal processor. Fig. 1 shows the schematic block diagram of the system. row of the antenna array via the row feeding network, while the column data port connects 24 small TR modules in a column via the column feeding network.

5
The radar controller consists of a master oscillator, a frequency synthesizer and a control timing generator and a main The digital transceiver consists of the DDS module, and the 24-channel receiver. The DDS module is used to generate the binary-phase-coded continuous wave for transmitting and the test signal for channel calibration. The 24-channel receiver includes the amplitude limiter, the bandpass filter, the linear amplifier, the analog-to-digital converter (ADC) and the digital 95 down converter (DDC). The amplitude and phase weight algorithm is realized in the digital beam forming (DBF) module.
The data processing implemented in the digital signal processing (DSP) chip involves pulse compression, coherent averaging, fast Fourier transform (FFT), and spectrum averaging. Then, the output of the DSP chip is transferred to the online data processor by a peripheral component interconnect (PCI) bus. The main functions of the online data processor are clutter suppression, parameter extraction, wind field retrieval and wind field displaying. Eventually, the product data of the 100 wind field in the troposphere, lower stratosphere, and mesosphere is produced. The voltage standing wave ratio (VSWR) of the antenna is less than 2.5.

Antenna field
As shown in the right part of Fig. 2, there are 144 shelters mounted at regular intervals, and each one consists of 4 small TR modules, a 4:1 row divider/combiner unit (DCU), a 4:1 column DCU, and a power supplier and a small TR controller. It 6 should be pointed out that the 36 small TR controllers are located in the shelters of odd rows and columns. The eight yellow boxes, labeled as F1-F8, represent the row feeding boxes (F1-F4) and the column feeding boxes (F5-F8). Each feeding boxes contains six 6:1 DCUs, and each one feeds four 4:1 DCUs in the shelters. The DCUs in the row/column feeding boxes are all fed by the big TR modules in the observation house, and the row or column drive state is switched by the control signal. The row/column data from the 24 big TR module feeding the 6:1 row/column DCUs is labeled as R1-R24/C1-C24.

115
As illustrated by the red box in the right part of Fig. 2, there are four shelters (S0101, S0102, S0201, S0202) and the surrounding antennas. The left part of Fig. 2 shows the inter connections of the shelters, and the lines of different shelters are red for easy review. In the shelter S0101, the row DCU is fed by R1-1, which represents the first divided signal of R1. The other 4 ports of the row DCU connect to the row data ports of small TR module 1 and 2 in S0101 and S0102 respectively.
Similarly, the column DCU is fed by C1-1, and the other 4 ports connect to small TR module 1 and 3 in S0101 and S0201 120 respectively. By analogy, the row/column DCUs of the shelter S0102, S0201, and S0202 connect to proper data ports of the small TR modules. With this system configuration, the beam can be steered to north-south direction in the row drive state and east-west direction in the column direction. The antennas are aligned in the northwest-southeast direction for symmetrical radiation pattern. The beams are usually steered to five directions (vertical, north, south, east, west) with offzenith angles of 15°.

125
The feeding network of the Wuhan MST radar uses feeding cables of equal length. In this situation, the feeding cables of different channels have stable characteristics, which need no compensation. The big TR modules, the 6:1 dividers and combiners, the 4:1 dividers and combiners , and the antennas are connected via coaxial cable (-3dB/100m). The feeding cables of above modules are 100 m, 50 m, 10 m, and 7m respectively. Therefore, the feeding line loss from the TR module in the observation house to the end antenna of the array is about 5 dB.  All timing signals for radar operation are generated by the timing generator in the radar controller. Fig. 3 presents the timing diagram of the signals at the radar controller. They are generated from the reference clock signal of CP-20M. The RESET signal is used to activate the timing generator. The excitation (EX) signal is used to generate the pulse repetition period (PRP), and the value is different according to different detection modes. The EX signal is 200 μs later than the RESET signal.

140
The sign τ is the delay of the TEST signal and the TR signal compared to the EX signal. The delay can be adjusted by software, and the rang is from half period of clock ahead to half period of clock delay. The transmitted pulse width of the TR signal is from 11 μs to 650 μs, which is related to the compressed pulse width and the coding scheme. The beam switch (BS) signal controls the switching of the vertical beam and four oblique beams, and it is synchronised with the RESET signal.
Besides the transmitting and receiving timing signals, the radar controller also generates the control signals for system 145 control, which are shown in the last two lines of Fig. 3. It should be pointed out that the control signals are valid 100 μs after the RESET signal rising edge. The different control signals are listed in are used for beam forming in the DBF module. The DDS phase weighting coefficient (DDS-P) signal causes the DDS to generate the beams of different zenith angles with a step size of 1°, and the maximum angle is 20°.

TR Module
160 Figure 4. Block diagram (a) and photograph (b) of the big TR module.
Block diagram and photograph of the big TR module is shown in Fig. 4. The big TR module amplifies the DDS output (53.8 MHz) supplied from the digital transceiver module, and feeds it to the feeding network. This module consists of a three-stage amplifier with a gain of 40.6 dB. The D2089UK and D1001UK are employed in the first and second power amplifier stage, 165 whose drain-source voltage is 3.2 V and 2.5 V respectively. They are metal gate RF silicon field effect transistors (FET) with different power output. The DDS output (10 dBm) is amplified to 30 dBm in the first stage, while the first stage output is amplified to 40.8 dBm in the second stage. The push-pull power metal oxide semiconductor (MOS) transistor BLF278 is employed in the final stage with a gain of 9.8 dB, whose drain-source voltage is 24 V. Built-in test (BIT) technique is adopted to detect VSWR and power information of the amplifier output (50.6 dBm) via a directional coupler. The 170 information is not only used to monitor status of the big TR module, but also avoid damage to the amplifier. The insertion loss of the T/R switch and R/C switch in the big TR module is about 0.3 dB. Therefore, the total output becomes 50 dBm.
The control signals transferred from the big TR controller are transformed over twisted-pair for better transmission ability.
The TR signal (differential signal) is converted into a single ended signal by the differential converter, and then it controls the R/C switch to realize transformation of row/column. The differential receiver chip DS96F173 is used as the differential 175 converter, which allows operating at high speed while minimizing power consumption. The TR signal controls the T/R switch to receive the signal from the row/column data port, or transmit the amplifier output to the row/column data port. The MOD signal is transferred to the digital module, so as to generate timing signals to control the T/R switch and R/C switch for different observation modes. The recovery time of the two switches is less than 5μs, which reaches the allowable level. The power supply provides -48 V for the two switches. It should point out that the differential converter is involved in the digital 180 module, as shown in Fig. 4(b). Block diagram and photograph of the small TR module are shown in Fig. 5. The small TR module consists of various 185 submodules: an amplifier, a T/R switch, a R/C switch, a BIT module, and a differential receiver. Each row/column signal is divided equally into 24 signals in the feeding network. Considering the attenuation of the dividers and cables (100 m cable: -3 dB; 1:6 divider: -7.78 dB; 50m cable: -1.5 dB; 1:4 divilder: -6.02 dB), the transmitting signal from the big TR module is reduced to 31.7 dBm. Then the signal is transmitted through a low power R/C switch with 0.3 dB insertion loss. A two-stage amplifier in the small TR module amplified the signal from 31.4 dBm up to 55.4 dBm, and the drain-source voltage of 190 BLF278 is 40V with a gain of 13.2 dB. Then, a band pass filter centered on 53.8 MHz is provided for spurious emission suppression. The T/R switch in the small TR module has an insertion loss of about 0.5 dB and provides an isolation of 60 dB.
Ultimately, the output of the small TR module is about 300 W, and the signal is fed to the Yagi antenna via a 7m coaxial cable. The low-noise amplifier (LNA) in the receiving channel is a 53.8 MHz tuned amplifier with a gain of 28 dB and a bandwidth of 1 MHz.

195
The two switches embedded in the small TR module allow the controller to select proper signaling pathway, and they are both controlled by two control signals from the small TR radar controller. The small TR module is the easiest damaged part in the system. Therefore, it needs to be repaired every year. As shown in Fig. 5(b), the small TR module adopts the modular design, which is convenient for maintenance.

200
The digital-up-converter (DUC) chip AD9957 is used in the DDS module, which has 1 Gsps internal clock speed with 18-bit IQ data path and 14-bit digital-to-analog converter (DAC). The 16-bit or 32-bit complementary code with different pulsewidth (1μs, 4μs and 8μs) are generated by the DDS module, as well as the test signal. The passive calibration algorithm is used for amplitude and phase calibration of the 24 channels in the receiver. The test signal can be set with different value of Doppler shift, and is divided into 24 channel signal by the divider. The intrinsic amplitude and phase differences among 205 channels can be extracted by comparing the output of each channel. Then, the amplitude and phase calibration factors are stored in the register. Through correction, the receiver has an amplitude consistency of -0.5-0.5 dB and a phase consistency of -2°-2°after calibration.
In the receiver, the signal firstly goes through the amplitude limiter for protection of the receiver, the linear amplifier with 50 dB gain, the bandpass with the center frequency of 53.

220
The clutter suppression is carried out in the online data processor. Kumar et al. (Kumar et al., 2019) identified the turbulence echo in the multipeaked VHF radar spectra during the precipitation, and here we mainly aimed at ground clutter and high frequency interference during fair weather. Ground clutters from surrounding mountains, trees, and buildings can severely degrade parameter estimates of the turbulence (Schmidt et al., 1979). It is because the weak echoes from the clear air are 225 easy to be contaminated for the lager amplitude of the ground clutters. From the perspective of radar infrastructure, the construction of fence is an effective method to isolate ground clutter, e.g. the 10 m height MU radar fence for ground clutter prevention (Rao et al., 2003). The fence is also constructed for the Wuhan MST radar, but there are still some ground clutters in the echoes. Therefore, ground clutter suppression is an essential step for signal processing. The ground clutter echoes have narrow central peak near zero frequency with small temporal changes, and are weakened with increasing altitude. 230 Fig. 6 shows the processing procedure of the Doppler spectra recorded by the east beam in the low mode. Note that the Doppler spectra at the range gates is normalized. As show in Fig. 6(a), the ground clutters severely bias the desired signals, and most turbulence echoes are submersed. As shown in Fig. 6(b), after the ground clutter suppression, the ground clutters near zero frequency are rejected effectively, and the weak signals appears at heights of 5.4-7.05 km and 9.75-12 km. The median filter is applied to remove the high frequency interference at each range gate. As shown in Fig. 6(c), the high 235 frequency interferences decrease, and the power spectra qualities are improved obviously. Data acquisition rate is one of the most important index to describe the MST radar performance, which is calculated using the relation: 100 × number of samples with valid horizontal wind data / total number of samples (Kumar et al., 2007). The 270 horizontal wind velocity is estimated by the radial velocity of the four inclined beams through the use of Doppler beam swinging (DBS) technique (Anandan et al., 2001). If any one of the inclined beams has serious interference or lower signalto-noise (SNR), which led to failure of the horizontal wind velocity inversion, then the samples are judged to be invalid. Fig.   8(a) shows the profile of total data acquisition in the low and middle mode during January 2016 to December 2017. In the low mode, the data acquisition rate remains >90% at heights of 3.5-10 km, and then decreases rapidly to 50% at height of 15 275 km. Note that the profile of low mode clearly shows a reversal at heights of 14-16 km corresponded to the tropopause (Chen et al., 2019). In the middle mode, the data acquisition rate remains >90% at heights of 10-20 km, and then decreases rapidly to 19% at height of 25 km. Therefore, the connection height of the low mode and middle mode is usually selected at height of 10 km for optimal data acquisition. In some situations which require high range resolution, the data acquisition rate (>50%) of the low mode is also available for the heights of 3.5-16 km.

280
As shown in Fig. 8(b), the data acquisition rate of the high mode is mainly concentrated at heights of 66-86 km with a maximum up to 17%, which is much lower than that of the low and middle modes. It is because the winds in the mesosphere are only available during the daytime (8 LT-16 LT) in the D region (due to insufficient D region ionization during nighttime) 14 (Rao et al., 2014). Actually, if the time range is limited in the daytime, the maximum data acquisition of the high mode is more than 50%. The analysis of the data acquisition rate indicates that the Wuhan MST radar can receive the backscattered 285 echoes from the troposphere, stratosphere and mesosphere effectively.

290
In order to verify the validity of the wind measurements in the height ranges of 3.5-25 km, simultaneous observations obtained from the radiosonde were compared with the Wuhan MST radar observations. The radiosonde launch site (30. 6°N, 114.1°E) is about 120 km away from the Wuhan MST radar, and the radiosonde was launched at 00 UT and 12 UT on 22 May 2016. It took about an hour for the balloon to rise up to 25 km, while the repetition period of the Wuhan MST radar is 295 30 min. Therefore, after the balloon was launched, the following two periods of the Wuhan MST radar data were averaged to compare with the data from the radiosonde.
Note that since the balloon was detected at regular intervals by the tracking radar, the height resolution of the radiosonde is not uniform. Fig. 9 shows the comparison of the zonal and meridional winds obtained by the Wuhan MST radar and the radiosonde launched at 00 UT and 12 UT on 22 May 2016. In these figures and throughout the paper, the positive zonal 300 component corresponds to eastward wind, while the positive meridional component corresponds to northward wind. The zonal wind profiles are in good agreement in the altitude ranges of 3.5 to 23 km. The meridional wind profiles also show 15 good agreement at most altitudes, but the winds observed by the Wuhan MST radar are weaker around the height of 14 km.
The underestimates of meridional winds could be due to the effect of aspect sensitivity (Thomas et al., 1997). The small discrepancies at some heights could be attributed to the variations of atmospheric activities at different temporal and spatial 305 scales, and the different measurement principles and errors in both instruments are also significant reasons (Belu et al., 2010;Hocking et al., 2001). atitude-longitude (Dee et al., 2011;Houchi et al., 2010). The dataset of monthly means of daily means is applied for the present study, which is produced by the average of the four main synoptic monthly means at 00, 06, 12, and 18 UTC (Berrisford et al., 2009).
It can be seen from Fig. 10(a) and Fig. 10(b) that the mean zonal wind observed by the Wuhan MST radar captures the 320 major feature of the ERA-interim, which shows a clear annual oscillation with one westward jet and one eastward jet every year. The eastward jet occurs from September to June below~20 km, and the westward jet occurs from May to October above~20 km. The observed zonal winds in the eastward jet are~10 m/s weaker than the ERA-interim reanalysis. The maximum magnitudes of the westward jet from the observation and the reanalysis are~14 m/s and~20 m/s, respectively. As 16 shown in Fig. 10(c) and Fig. 10(d), compared to the zonal winds, the meridional winds show larger discrepancies between 325 the observation and the reanalysis. There are one northward jet and one southward jet exhibited in the observed mean meridional winds every year. The northward jet occurs from November to April below~18 km, and the southward jet occurs from May to September above~18 km. In the ERA-interim, the southward jets are extended in the two years. Especially in 2017, the southward jet occurs from April to October, and extends down to the low height in April and May. The discrepancies are mainly due to the differences of the average time periods. The meridional wind changes more over time, so 330 as to show larger discrepancies in the monthly mean meridional winds. In conclusion, the Wuhan MST radar can measure the zonal and meridional winds in the troposphere and low stratosphere effectively. In order to verify the validity of the mesospheric wind measurements, simultaneous observations obtained from the meteor radar at Wuhan were compared with the Wuhan MST radar observations. The Wuhan meteor radar (30.6°N, 114.4°E) is about 120 km away from the Wuhan MST radar, which is an all-sky interferometric broadband radar system with a peak 340 power of 7.5 kW and a frequency of 38.7 MHz (Xiong et al., 2004;Zhao et al., 2005). The averaging of daytime (8 LT-16

Mesospheric observation
LT) observations was used as the daily mean wind estimation for the Wuhan MST radar, while the 24 hours average was used for the Wuhan meteor radar. Because of the effect of diurnal variations, the estimated mean winds of the Wuhan MST 17 may be biased less than 5 m/s compared with that of the Wuhan meteor radar below 85-90 km (Nakamura et al., 1996).
Considering the observation height range of the two radars, the comparison range was set at heights of 76 to 86 km. 345 Fig. 11 shows the daily mean zonal and meridional winds observed by the Wuhan MST radar and the Wuhan meteor radar on 3 individual days in January 2016. Interestingly, the measurements at height of around 81 km show better agreement than other heights. It is because the measurements of the meteor radar are more reliable above 80 km (Ratnam et al., 2001;Kumar et al., 2008), while the data acquisition rate of the Wuhan MST radar is relatively high at heights of 70-85 km in the mesosphere. The zonal and meridional winds are of concordance in the aggregate. Two reasons might result in the 350 discrepancies between the observations of the two radars. The first one is the localized gravity waves, tides or planetary waves could make the differences between them (Rao et al., 2014;Ratnam et al., 2001). The second is that the low data acquisition rate of the Wuhan MST radar in the mesosphere could lead to the fluctuations of the daily mean data, which shows the sudden changes of the MST radar measurements at some heights.  altitude, day of year, and time of day, which is the upgraded version of HWM-07 (Drob et al., 2015).
As shown in left panels of Fig. 12, Wuhan MST radar zonal winds clearly show strong seasonal variations in 2016 and 2017. In general, the trends of the observational and predicted zonal winds match well, especially the reversal from eastward 365 to westward in spring and the reversal from westward to eastward in autumn. However, HWM-14 overestimates the zonal winds in winter. Especially in Feb the magnitudes of the observational and predicted results have big differs. Many studies indicated that stronger northward and westward winds happen after the stratospheric sudden warming (SSW) events (Mbatha et al., 2010;Chau et al., 2015), and this factor is not considered in the HWM-14. The 2016 Feb SSW is a minor SSW, and the day of peak warming is on Feb 5 (Medvedeva et al., 2017). Two minor warming events happened during the winter of 370 2017, with two days of peak warming on Feb 2 and 26 (Eswaraiah et al, 2019). The SSW events happened during the observation period may influence the mean winds in the mesosphere. Hence, the stronger westward winds may result in smaller mean zonal winds during winter. Moreover, the differences of the zonal winds in summer are noticed. The first difference is the reversal height in summer, which is a useful index for the mesopause. The wind shear around 78 km is prominent during the summer from the HWM-14. Meanwhile, the reversal height observed by the Wuhan MST radar is 375 about 84-85 km, which is consistent with the result observed by the MU radar at similar altitude (Namboothiri et al., 1999).
Further study may be needed to analyze the difference. The second difference is the westward jet (the bluer region) occurred in summer. There are some differences of the westward jet between the observations and predictions in occurrence time and height, which could be due to interannual variability. As seen from right panels of Fig. 12, it appears that the observational meridional winds of 2016 and 2017 have the same trend as the predicted results. They all have one northward jet occurred 380 above~75 km in the period from August to April, but the observational results are larger than the predicted results in winter because of the stronger northward winds during the SSW events. In general, the Wuhan MST radar wind measurements of the mesosphere are in agreement with the HWM-14 predictions in trend.

Conclusion
The technical features of the Wuhan MST radar are described in this paper. We use the TR modules and digital receiver with 385 smart structure, and reasonable feeding network, to realize the beam steering for three-dimensional wind field measurements.
Wind observations of the Wuhan MST radar are compared with other instruments and related models for validation, and the results are summarized as follows: 1. Compared with the radiosonde (120 km away) and the ERA-interim, the zonal and meridional winds are in good agreement at heights of 3.5-25 km, and large discrepancies in meridional winds could be due to the temporal and spatial 390 differences.
2. The daily mean zonal and meridional winds are in good agreement at heights of 76 to 86 km with the Wuhan meteor radar (120 km away), and the measurements at height of around 81 km show better agreement than other heights.

19
3. The monthly mean zonal and meridional winds are in agreement with the HWM in trend at heights of 66 to 86 km. The amplitudes of the observational results are different from that of the predicted results in winter, and it could be due to the 395 influence of SSW events.
The comparisons indicate that the Wuhan MST radar is an effective tool to measure the three-dimensional wind fields of the MST region. These results encourage us to do more for the improvements, such as improving the data acquisition of the high mode, and correcting the nominal zenith angle for the aspect sensitivity. In the future, we will use the Wuhan MST radar to study precipitation, gravity waves, and stratosphere-troposphere exchange processes during typhoon, cold front or 400 other events, as well as the dynamics of the mesosphere.
Author contributions. LQ prepared the main part of the paper and performed the statistical analysis. GC is the project leader 405 of the Wuhan MST radar and supported the preparation of the paper. SZ supervised the paper writing. QY implemented the construction work. The measurements were led by WG and FC. WZ and HZ helped with the statistical analysis of Wuhan MST radar. MS and EL provided valuable suggestion for data processing. The data analysis was supported by XC and HS.
HZ and LZ edited the article.

410
Competing interests. The authors declare that they have no conflict of interest.