Validation of MIGHTI/ICON Atmospheric Wind Observations over China Region Based on Meteor Radar and Horizontal Wind Model (HWM14)

: The Michelson Interferometer for Global High-resolution Thermospheric Imaging (MIGHTI) on board the ICON satellite provides effective measurement of horizontal winds in the mesosphere and lower thermosphere (MLT) region. In order to verify the measurement accuracy of the horizontal wind, this study uses the measurements of the meteor radar in Wuhan and the simulation results of a horizontal wind ﬁeld model (HWM14) to compare and analyze the measurement results of MIGHTI/ICON in the whole year of 2020. The comparative analysis indicated that two datasets from MIGHTI/ICON and meteor radar are strongly correlated (r = 0.65, 0.76) with an RMS difference of 39.21 m/s (30.31 m/s). The consistency for meridional wind from MIGHTI/ICON, meteor radar, and HWM14 is worse than that of zonal wind. The accuracy of horizontal wind observations is inﬂuenced by altitude, diurnal, and seasonal patterns.


Introduction
Atmospheric winds play an important role in determining the state and evolution of the atmosphere and ionosphere by participating in various chemical changes and physical processes in the mesosphere and lower thermosphere (MLT) region [1]. Atmospheric winds can push plasma drift by wind-driven electromotive forces and control the exchange of energy between atmospheric waves from the lower atmosphere and upper atmosphere in the MLT region [2]. Thus, an accurate measurement of atmospheric winds in the MLT region is important to facilitate our better understanding of the near-Earth space environment. Currently, atmospheric winds have been generally measured by ground-based instruments, including meteor radar [3][4][5][6], lidar [7,8], medium-frequency radar [9][10][11], passive optical instrumentation, etc.
Among the ground-based instruments, meteor radar has the advantage of 24 h continuous observation with height coverage from 70 to 110 km, which provides favorable conditions for analyzing the characteristic parameters of long-period atmospheric waves, such as tidal waves and planetary waves in the MLT region [8,12]. However, poor spatial coverage for ground-based instruments results in an inability to provide global distribution of atmospheric winds. Space-borne instruments can overcome these limitations and provide a global measurement of atmospheric winds. For example, the High-Resolution Doppler Imager (HRDI) and the Wind Imaging Interferometer (WINDII) onboard the UARS satellite first measure atmospheric winds within the height ranges of 50-115 km and 90-180 km by observing the Doppler shift characteristics of O2 airglow emission and O airglow emission, respectively [13]. The High-Resolution Doppler Imager on the Thermosphere Ionosphere Mesosphere Energetics and Dynamics Doppler Interferometer (TIDI), launched in 2001, has been monitoring MLT winds for two decades and has collected the most comprehensive global information on neutral winds [14]. The wind observations have been cross-validated by using ground-based instruments and there is no statistically significant difference for horizontal winds [13,14], which indicates that space-borne instruments can also provide an accurate wind measurement.
A new global atmospheric wind dataset can be now obtained from the Ionospheric Connection Explorer (ICON) launched in 2019. The Michelson Interferometer for Global High-resolution Thermospheric Imaging (MIGHTI) onboard the ICON can view airglow emissions from OI 557.7 and 630.0 nm to determine Doppler wind [15][16][17]. A series of cross-comparisons of ICON wind observations with ground-based and space-borne instruments have been carried out by recent research [14,18,19]. Makela et al. presented MIGHTI/ICON and ground-based Fabry-Perot interferometer measurements of the nocturnal thermospheric neutral wind and showed consistency within 10 m/s [18]. Harding et al. showed a strong correlation (r = 0.82) between ICON-MIGHTI and specular meteor radars for coincident wind measurements with an average error of 4.5 m/s [19]. The agreement of MIGHTI/ICON and TIMED/TIDI atmospheric wind measurements in the MLT region has also been shown in Dhadly et al. [14]. However, no study has yet considered the consistency of meteor radar and MIGHTI/ICON measurements of neutral wind in the MLT region on the basis of a full year of detection data. Here, we focus on the cross-validation of MIGHTI/ICON and meteor radar horizontal winds around Wuhan in 2020. The simulated results of Horizontal Wind Model 14 (HWM14) are also used to find the inconsistencies between MIGHTI/ICON measurements and HHW14 simulations.
This work is organized as follows. In Section 2, we introduce MIGHTI/ICON, meteor radar, and HWM14 in turn. In Section 3, we analyze the results at the example and statistical levels, respectively. We present the discussion process and conclusion in Section 4.

Dataset
ICON is the newest member of NASA's Heliophysics satellite constellation with a 27-degree inclination orbit at about 600 km, which is tasked with better understanding the relationship between Earth's atmosphere and the space environment [20]. To determine the height distribution of atmospheric winds and temperatures in the Earth's upper atmosphere, MIGHTI is carried on board the ICON satellite [15]. MIGHTI consists of two sensors with orthogonal fields of view, MIGHTI-A with a 45 • azimuthal line of sight to the spacecraft velocity vector and MIGHTI-B with a 135 • azimuthal line of sight to the spacecraft velocity vector. In this view, MIGHTI makes two vertical line-of-sight wind speed measurements of the same physical space while the spacecraft passes [17]. Doppler shifts of two airglow emission lines of oxygen, one green (557.7 nm) and one red (630.0 nm), are measured to retrieve the horizontal winds [16,21]. The details of MIGHTI/ICON are discussed by Englert et al. [15]. MIGHTI measurement requirements for a single field of view of wind are provided in Table 1 In this study, the cardinal vector wind data from MIGHTI in 2020 are used in cross-comparison analysis. In this work, we used meteor radar wind measurements in 2020 at Wuhan (30 • 40 N, 114 • 30 E). Horizontal winds (zonal wind and meridional wind) from 70 to 110 km were available from meteor radar with 2 km height resolution and 1 h time resolution.
HWM14 is the latest version for HWM series which has been updated in the thermosphere with new observations. Updated results provide an improved, time-dependent, observation-based, global empirical model for the general circulation of the upper atmosphere and the horizontal winds of migrating tides. In general agreement with existing theoretical knowledge of the thermospheric general circulation, additional studies have shown that the empirical wind field is consistent with the ionospheric plasma distribution and electric field pattern [22][23][24][25]. The details of HWM14 are shown in Drob et al. [23]. The simulated results of horizontal winds at Wuhan were used in this study.

Case Study
In this study, the following formulae are used to determine the coincident wind observations from MIGHTI/ICON and meteor radar: where t ICON , Lat ICON , and Lon ICON are wind observation time, latitude, and longitude of MIGHTI/ICON, respectively. t Meteor , Lat Meteor , and Lon Meteor are wind observation time, latitude, and longitude of meteor radar, respectively. To investigate the measuring accuracy of horizontal winds for MIGHTI/ICON, we used the horizontal winds obtained by meteor radar as a reference. It should be noted that the retrieved wind from meteor radar 1 h observations is a spatial average of true wind within the range of 200 × 200 km [5]. Figure 1 shows the coincident wind observations from MIGHTI/ICON and meteor radar on 7-8 October 2020. The analysis of Figure 1 shows that: firstly, since the measurement time and horizontal position of MIGHTI/ICON and meteor radar are not the same, the horizontal winds have certain deviations, the root-mean-square (RMS) differences for meridional wind and zonal wind in Figure 1a  In a qualitative sense, the horizontal wind measurements by MIGHTI/ICON and meteor radar overlap at 90-100 km. However, at 100-110 km, these differences exceed 50 m/s. The reason for this discrepancy may be that the horizontal wind measurement accuracy of meteor radar above 100 km has some errors [6]. From left side to right side, they correspond to meridional wind and zonal wind, respectively. The colored lines are MIGHTI/ICON measurements. The black lines and gray lines are meteor radar measurements during the MIGHTI/ICON observations. The error bars are the estimated errors in wind speed.

Statistical Study
To estimate the precision of horizontal winds derived by MIGHTI/ICON, using the method described above, we statistically compared the thermospheric horizontal wind measurements obtained by MIGHTI/ICON, meteor radar, and HWM14 in the whole year of 2020. Since the meteor radar detects at a resolution of 2 km over 70-110 km in altitude and the MIGHTI/ICON detects at a resolution of 2.9 km over 90-110 km, we performed three spline interpolations of the MIGHTI/ICON detection data over 90-110 km in altitude. Then, the horizontal wind data for MIGHTI/ICON, meteor radar, and HWM14 in the altitude range of 90-100 km were used for comparative analysis.
We adopted the method of statistical comparison of geophysical data using multiple instruments with different accuracies proposed by Hocking et al. [26]. We denote the MIGHTI/ICON data sets as { } and the meteor radar data sets as { } from two instruments that measure the same physical quantity but have different sensitivities to its variabilities. The variability that both instruments can detect is denoted by and for

Statistical Study
To estimate the precision of horizontal winds derived by MIGHTI/ICON, using the method described above, we statistically compared the thermospheric horizontal wind measurements obtained by MIGHTI/ICON, meteor radar, and HWM14 in the whole year of 2020. Since the meteor radar detects at a resolution of 2 km over 70-110 km in altitude and the MIGHTI/ICON detects at a resolution of 2.9 km over 90-110 km, we performed three spline interpolations of the MIGHTI/ICON detection data over 90-110 km in altitude. Then, the horizontal wind data for MIGHTI/ICON, meteor radar, and HWM14 in the altitude range of 90-100 km were used for comparative analysis.
We adopted the method of statistical comparison of geophysical data using multiple instruments with different accuracies proposed by Hocking et al. [26]. We denote the MIGHTI/ICON data sets as {x i } and the meteor radar data sets as {y i } from two instruments that measure the same physical quantity but have different sensitivities to its variabilities. The variability that both instruments can detect is denoted by v i and g 0 v i for the two instruments, respectively, where g 0 is a measure of the relative amplitude of the variability in two measurements.
The calculation gives us the variance values s 2 x and s 2 y for x i and y i , and the covariance value s 2 xy . It is easy to obtain the correlation coefficient r and the functional relations between g 0 and σ x and σ y as: Table 2 shows the correlation coefficients of the meridional and zonal winds obtained from MIGHTI/ICON and meteor radar at different altitudes. The correlation coefficients decrease with height, and the measurement deviation between MIGHTI/ICON and meteor radar increases with height. Due to the poor data quality for wind measurement from the meteor radar above 100 km, we only compared coincident thermospheric neutral wind data at 94, 96, and 98 km from MIGHTI/ICON and meteor radar at Wuhan.  Figure 2 shows the functional relations described by Equations (8) and (9). They show the relationship between the three variables g 0 , σ MIGHTI/ICON , and σ meteor radar . The intersects of the curves with the y-axis are g MIGHTI/ICON and g meteor radar , with g meteor radar always large than g MIGHTI/ICON , which means means the MIGHTI/ICON dataset corresponds to a larger error. Comparing the changes in the four curves, we find that the change in g MIGHTI/ICON with σ v for the meridional winds is flatter, suggesting a greater error in the MIGHTI/ICON observations of the meridional winds. Figure 3 shows scatter plots of the horizontal wind measurements from MIGHTI/ICON, meteor radar, and HWM14. Each data point indicates that it is from the same altitude and time. It is obvious from the Figure 3a that there is a strong correlation between horizontal wind measurements from MIGHTI/ICON and meteor radar. The correlation coefficients (fitting slopes) for meridional wind and zonal wind are about 0.65 (0.87) and 0.76 (0.85), respectively. Figure 3b compares the MIGHTI/ICON wind to HWM14 wind. The poor correlation with the maximum correlation coefficient of approximately 0.57 is found in Figure 3b. The fitting slopes for meridional wind and zonal wind are, respectively, 1.40 and 1.31, suggesting that the magnitude of horizontal winds obtained from HWM14 was relatively small.       Figure 4 shows the distributions of the difference among the horizontal wind measurements from MIGHTI/ICON, meteor radar, and HWM14. The distributions of horizontal wind differences appear normal and peak around 0 m/s. The RMS differences are 39. 21 and 46.44 m/s for meridional winds and 30.31 and 37.65 m/s for zonal winds in Figure  4a,b, respectively.  Figure 5 shows the distributions of the speed ratio and direction difference among the horizontal wind measurements from MIGHTI/ICON, meteor radar, and HWM14. The average of the radar/MIGHTI wind speed ratio is 1.08 and direction difference is −1.36° in Figure 5a. Similarly, the average of the HWM14/MIGHTI wind speed ratio is 0.57 and direction difference is −3.11° in Figure 5b.  Figure 5 shows the distributions of the speed ratio and direction difference among the horizontal wind measurements from MIGHTI/ICON, meteor radar, and HWM14. The average of the radar/MIGHTI wind speed ratio is 1.08 and direction difference is −1.36 • in Figure 5a. Similarly, the average of the HWM14/MIGHTI wind speed ratio is 0.57 and direction difference is −3.11 • in Figure 5b.
Comparing the results shown above, it could be seen that the consistency of zonal wind measurements from MIGHTI/ICON and meteor radar is better than that of meridional wind measurements, which may be caused by the orbit configuration of the ICON satellite [14]. The measurement of meridional wind from MIGHTI/ICON is naturally less accurate than that of zonal wind due to the north-south motion of the ICON satellite around Wuhan. In addition, the consistency of zonal wind measurements between MIGHTI/ICON and HWM14 is also better than that of meridional wind measurements, which is agreement with the results of previous studies [25,27]. Tang et al. compared the horizontal wind measurements obtained by meteor radar and HWM14 and found that the consistency of meridional wind between model prediction and radar observation is worse than that of the zonal wind [25]. Taiwo et al. compared results from HWM14 with a Fabry-Perot interferometer (FPI) located in Sutherland, South Africa, and found a better Comparing the results shown above, it could be seen that the consistency of zonal wind measurements from MIGHTI/ICON and meteor radar is better than that of meridional wind measurements, which may be caused by the orbit configuration of the ICON satellite [14]. The measurement of meridional wind from MIGHTI/ICON is naturally less accurate than that of zonal wind due to the north-south motion of the ICON satellite around Wuhan. In addition, the consistency of zonal wind measurements between MIGHTI/ICON and HWM14 is also better than that of meridional wind measurements, which is agreement with the results of previous studies [25,27]. Tang et al. compared the horizontal wind measurements obtained by meteor radar and HWM14 and found that the consistency of meridional wind between model prediction and radar observation is worse than that of the zonal wind [25]. Taiwo et al. compared results from HWM14 with a Fabry-Perot interferometer (FPI) located in Sutherland, South Africa, and found a better agreement between FPI measured winds and HWM14 modeled winds for the meridional component compared to the zonal component [27].
In addition to the above-mentioned reasons for the satellite trajectories, another reason is that the meridional winds in the MLT region are significantly longitudinally dependent, which means that HWM14 cannot accurately predict the meridional winds based only on the limited longitude input [9,10,28,29]. The cause of this disagreement is not clear up to now. For example, Hu et al. studied MLT winds with MF radars in Wuhan In addition to the above-mentioned reasons for the satellite trajectories, another reason is that the meridional winds in the MLT region are significantly longitudinally dependent, which means that HWM14 cannot accurately predict the meridional winds based only on the limited longitude input [9,10,28,29]. The cause of this disagreement is not clear up to now. For example, Hu et al. studied MLT winds with MF radars in Wuhan (30 • N, 114 • E) and in Yamagawa (31 • N, 131 • E), and found the longitudinal variations in atmospheric gravity waves forcing may contribute to the differences in the mean winds and tides between Wuhan and Yamagawa [10]. Jacobi et al. studied mid-latitude MLT winds over Europe and Canada and found that the zonal prevailing winds in winter show partly opposite behavior at different longitudes, which may be explained by a stationary planetary wave (SPW) influence [29].

Day/Night Differences
To investigate the differences in measurement accuracy of MIGHTI/ICON during day and night, Figure 6 shows the scatter plots of the horizontal wind measurements from MIGHTI/ICON and meteor radar during day and night. It is obvious from Figure 6a that the correlation of meridional winds from MIGHTI/ICON and meteor radar is better at night than that during the day. The correlation coefficients (fitting slopes) for day and night are about 0.60 (0.51) and 0.67 (0.53), respectively. Figure 6b compares the zonal wind measurements from MIGHTI/ICON and meteor radar during the day and night. Similarly, the consistency of the data measured at night is better than that during the day, the correlation coefficients (fitting slopes) for day and night are about 0.73 (0.59) and 0.89 (0.68), respectively.

Day/Night Differences
To investigate the differences in measurement accuracy of MIGHTI/ICON during day and night, Figure 6 shows the scatter plots of the horizontal wind measurements from MIGHTI/ICON and meteor radar during day and night. It is obvious from Figure 6a that the correlation of meridional winds from MIGHTI/ICON and meteor radar is better at night than that during the day. The correlation coefficients (fitting slopes) for day and night are about 0.60 (0.51) and 0.67 (0.53), respectively. Figure 6b compares the zonal wind measurements from MIGHTI/ICON and meteor radar during the day and night. Similarly, the consistency of the data measured at night is better than that during the day, the correlation coefficients (fitting slopes) for day and night are about 0.73 (0.59) and 0.89 (0.68), respectively.  By processing the data as before, we obtained 818 sets of data measured during the daytime and 1224 sets of data measured at night, where the daytime is from 06:00 to 18:00 and the rest of the time is considered as night. Comparing the correlation coefficients of horizontal winds from MIGHTI/ICON and meteor radar observations, we find a better agreement at night than during the day, and this difference is reflected in both meridional and zonal winds. Similar results were also observed by Harding et al., who compared MIGHTI-A/B with meteor radar results in day/night mode [19]. In this regard, we speculate that the difference in the measurement patterns taken by MIGHTI during the day and night is the reason. For example, the exposure is 30 s instead of 60 s in day mode, and the aperture is stopped at 15% of the nighttime aperture [15,19].
Additionally, we calculated fitted slopes in all four cases in Figure 6 and found that the wind field measurements by MIGHTI/ICON are larger than those by meteor radar, which is reflected in both day/night mode and meridional/zonal winds. This is also effectively illustrated by the ratio of velocities plotted in Figure 5. This difference in horizontal wind measurement is shown in previous comparisons of both satellite-based soundings and ground-based observations [30,31]. Hasebe et al. compared MLT wind velocities measured by the High-Resolution Doppler Imager (HRDI) on board the Upper-Atmosphere Research Satellite (UARS) and meteor radars located at Shigaraki, Japan, and Jakarta, Indonesia [30]. The median of the speed ratio (HRDI/radar) is 1.379 in the meteor mode for the Shigaraki MU radar. Wu et al. compared the TIDI meridional winds with meteor radar measurements at Maui. The meteor data show smaller diurnal tide amplitude than that shown in the TIDI data [31].

Seasonal Differences
It is well known that the horizontal winds not only have day and night variations, but also change with the seasons [25]. Figure 7 shows the monthly mean variation in horizontal winds observed by MIGHTI/ICON and meteor radar. The monthly mean zonal wind is mainly eastward in summer and prevailing westward in winter, and the southward monthly mean meridional wind is dominant in Wuhan. Similar seasonal variations in horizontal winds obtained by Wuhan meteor radar were also shown in Zhao et al. [32]. In Figure 7, we find an excellent correlation (r = 0.96) between the zonal winds, which are generally consistent in terms of trend and amplitude, but some differences in some months in the meridional winds result in a poor correlation (r = 0.66), which could be explained by the orbit configuration of the ICON satellite [14].
Additionally, we calculated fitted slopes in all four cases in Figure 6 and found that the wind field measurements by MIGHTI/ICON are larger than those by meteor radar, which is reflected in both day/night mode and meridional/zonal winds. This is also effectively illustrated by the ratio of velocities plotted in Figure 5. This difference in horizontal wind measurement is shown in previous comparisons of both satellite-based soundings and ground-based observations [30,31]. Hasebe et al. compared MLT wind velocities measured by the High-Resolution Doppler Imager (HRDI) on board the Upper-Atmosphere Research Satellite (UARS) and meteor radars located at Shigaraki, Japan, and Jakarta, Indonesia [30]. The median of the speed ratio (HRDI/radar) is 1.379 in the meteor mode for the Shigaraki MU radar. Wu et al. compared the TIDI meridional winds with meteor radar measurements at Maui. The meteor data show smaller diurnal tide amplitude than that shown in the TIDI data [31].

Seasonal Differences
It is well known that the horizontal winds not only have day and night variations, but also change with the seasons [25]. Figure 7 shows the monthly mean variation in horizontal winds observed by MIGHTI/ICON and meteor radar. The monthly mean zonal wind is mainly eastward in summer and prevailing westward in winter, and the southward monthly mean meridional wind is dominant in Wuhan. Similar seasonal variations in horizontal winds obtained by Wuhan meteor radar were also shown in Zhao et al. [32]. In Figure 7, we find an excellent correlation (r = 0.96) between the zonal winds, which are generally consistent in terms of trend and amplitude, but some differences in some months in the meridional winds result in a poor correlation (r = 0.66), which could be explained by the orbit configuration of the ICON satellite [14].  In Figure 8, we further plotted the scatter plots of MIGHTI/ICON and meteor radar measurements of horizontal winds with seasonal variations. The MIGHTI/ICON and meteor radar measurements all show a good consistency in different seasons, and the consistency of both is better in spring and autumn than that in summer and winter. The reasons may be that the existence of strong shears of horizontal winds and intermittent gravity waves could cause violent local horizontal wind variations, particularly in the winter MLT region at mid-latitudes, which can result in the discrepancy between MIGHTI/ICON and meteor radar due to the instrument biases [32,33].  The left side is meridional wind, and the right side is zonal wind. From top to bottom: Spring (January, February, and March), summer (April, May, and June), autumn (July, August, and September), and winter (October, November, and December).