The retrieval process for ACE-FTS v.5.2 wind profiles has previously been explained in-depth by , so only a short explanation is provided here. For each limb measurement, the atmosphere is partitioned into 4 km regions above 45 km. Below that, increasingly smaller regions are used that decrease to 2 km in size. This is done to account for refraction distorting the disk of the Sun during a measurement. A forward model calculation is then used to derive a spectrum at a tangent height near the center of each partition. This forward model is based on pressure, temperature, and volume mixing ratio (VMR) profiles taken from sr10063 (sr for sunrise and 10063 for the orbit number since launch). The observed spectra and sr10063 are cross correlated to derive wind speeds from the Doppler shift in the spectra. The wind speeds are then interpolated onto a 1 km grid using a cubic spline. Finally, the wind profiles are calibrated using line-of-sight winds from the Canadian Meteorological Center (CMC) between 19 and 24 km altitude. The final product is line-of-sight wind speeds between 18 and 140 km at a given time, location, and heading. The heading is the angle between the “look-direction” of the instrument and geodetic north, which is calculated by the Systems Tool Kit software package .

The models and instruments we compare with provide vector wind speeds, not line-of-sight wind speeds. We are able to convert the vector winds into line-of-sight winds using 1Line-of-Sight Wind Speed=U⋅cos⁡θ-90°+V⋅cos⁡(θ), where U and V are the zonal (west-east) and meridional (south-north) components, respectively, from the given model or instrument, and θ is the heading of the ACE-FTS instrument.

Version 5.3 explicitly retrieves altitude profiles for line-of-sight wind speed simultaneously with retrieving the VMR profiles of molecules contributing to the spectrum. The retrievals follow the usual procedure described previously for VMRs , but with a piecewise linear interpolation for wind speed onto the 1 km altitude grid employed for forward model calculations. Deriving wind speed using a retrieval, rather than using the wavenumber offset of a given measurement as an estimate of the Doppler shift at that measurement's tangent height (as was done in previous processing versions), should better capture the altitude variation of wind speed when there is significant structure in the profile. With the retrieval approach, the calculated spectrum employed in the analysis is also more representative of the measured spectrum compared to the “generic” calculation from a single occultation (sr10063) in a given altitude “bin” that was employed in the v.5.2 analysis.

The altitude range was divided into 5 overlapping segments. The microwindows for a given altitude segment targeted single bands or sets of closely spaced bands (including overlapping hot bands) for a particular isotopologue to maximize internal consistency. It was found that observed shifts for lines from different isotopologues or lines from molecules without distinct bands (like H₂O) were not always consistent. The microwindows employed in the different altitude segments are provided in Table A1 of the Appendix. All measurements that fall within the wavenumber, altitude, and minimum transmittance ranges are employed in the retrieval. The filter for minimum transmittance is used to avoid lines that are saturated or where the signal is too close to the noise, either of which could obscure Doppler shift information in the measurement. The upper altitude limit of a particular segment is determined by the maximum tangent height within the altitude range that has at least three lines satisfying the filter conditions, to avoid having an abnormally large noise feature being included in the analysis when the signal falls below the nominal noise level.

Wind retrieval results for the lowest altitude segment (termed “segment 1” in Table ) exhibit excessively large variability, possibly a consequence of pressure shifts and deviations of line shapes from the assumed Voigt profile becoming more prominent at higher pressures. Therefore, in this segment line-of-sight wind speed is determined using the average wavenumber shifts of the lines, similar to the approach employed in previous processing versions, rather than using the retrieved results. Attempts to push the analysis lower in altitude often yielded relatively large gradients below 18 km, which are not expected in that altitude region, suggesting that systematic errors from pressure-related issues (such as pressure shifts and line shapes) become excessive when pressure gets too high. Therefore, the lower altitude limit of the analysis remained at 18 km.

Wind profiles in the lowest two altitude segments have persistent biases compared to the results in the three highest segments, as shown in Fig. . Thus, fixed shifts are applied to the wind profiles in the lowest two segments: -35 m s⁻¹ in segment 1 and +35 m s⁻¹ in segment 2. In overlap regions between segments, results are calculated from an unweighted average of the two profiles. When no measurement occurs in the overlap region between segments, there is a small altitude gap in the reported wind profile. As with previous processing versions, the final wind profile is shifted such that the results between 18 and 24 km match the expectations from an analysis run of the Canadian weather model at Environment and Climate Change Canada . This is needed to account for the motion of the satellite relative to the atmosphere. A rough calibration of the ACE-FTS wavenumber scale is accomplished using high altitude CO₂ lines, but because this process uses sampled peaks (which may not be sampled at line centers), the calibration could be off by a fraction of the width of the instrumental line shape (0.02 cm⁻¹). The accuracy of this calibration will vary from occultation to occultation, but the wind calibration shifts can be as high as a few hundred metres per second to compensate for the resulting offsets.

The MIGHTI instrument uses a Michelson interferometer and utilizes the Doppler Asymmetric Spatial Heterodyne (DASH) technique to determine shifts in emission lines . The emission lines observed are the red (O(¹D-³P) 630.0 nm) and green (O(¹S-¹D) 557.7 nm) lines of the oxygen atom. We only compare with winds derived from the green line data products, as the red line is less frequent below 150 km. MIGHTI uses two sensors that record measurements 5 to 8 min apart. These measurements can be used to determine horizontal vector winds due to being orthogonal to each other, with MIGHTI-A observing at 45° and MIGHTI-B at 135° relative to the satellite's motion.

A detailed comparison with ACE v.5.2 line-of-sight winds with MIGHTI vector winds is available in . We provide a similar comparison of ACE v.5.3 against v05 MIGHTI using 207 of the same sunrise and 215 sunset coincidences between 2019 and 2022 used in the previous paper. Those coincidences are measurements within 2.5° latitude and 5° longitude in location and 2 h in time. In the case that there were multiple MIGHTI measurements that fit this criteria for one ACE occultation, the spatially closest measurement was used.

In Fig. , we show a comparison between an individual ACE sunrise and sunset occultation and the coincident MIGHTI measurement. It is important to note that the heading angle differs in v.5.2 and v.5.3, meaning that the converted MIGHTI winds would be different for each. However, the resulting speeds are very similar, so we display the line-of-sight winds converted using v.5.3 heading angles. There is similar agreement for both v.5.2 and v.5.3 with MIGHTI when looking at the sunrise coincidence. The sunset coincidence shows slightly better agreement using v.5.3.

In an effort to remove outliers and find trends in the data, we use the same restrictions as Data Set 2 from . Data Set 2 used only the coincidences where the average difference between ACE and MIGHTI wind speeds was less than 60 m s⁻¹, leaving us with 184 sunrises and 196 sunsets. The outliers removed are typically associated with a limited altitude window for the MIGHTI measurement. The average of these MIGHTI and ACE measurements are presented in Fig. . Looking at the sunrises, there is similar profile agreement at all altitudes for both v.5.2 and v.5.3. The sunset comparison shows good agreement for both versions from 90 to 110 km and then deviates.

We are also able to derive a sunrise-sunset bias with this comparison. To do this, we shift the MIGHTI altitude to the nearest ACE altitude (maximum of ±0.5 km) and find the difference between ACE and MIGHTI wind speeds. The differences are then averaged at each altitude over all occultations in Data Set 2 and the results are shown in Fig. a for v.5.2 and (c) for v.5.3. We then subtract the sunrise and sunset averages from the total average to find the bias. The average bias at each altitude is displayed in Fig. b for v.5.2 and (d) for v.5.3. We find that the sunrise (sunset) bias for v.5.2 is generally within ±5 m s⁻¹ of the previously found -15 m s⁻¹ (+15 m s⁻¹) up to 120 km. The bias for v.5.3 is within ±2 m s⁻¹ of the -15 m s⁻¹ (+15 m s⁻¹) bias up through the same altitude. The sunrise-sunset bias is shown in pink in Fig. .

The Institute of Geology and Geophysics of the Chinese Academy of Science has five meteor radar stations in China providing horizontal wind profiles: Mohe at 52.5° N, 122.3° E; Beijing at 40.3° N, 116.2° E; Wuhan at 30.5° N, 114.6° E; Ledong at 18.7° N, 109.2° E; and Sanya at 18.3° N, 109.6° E. The stations use VHF radar and five receivers to track the motion of meteoric trails. The measurements are reported hourly and provide an all-sky coverage . A detailed comparison of meteor radar data with ACE v.5.2 is provided in . A comparison with ACE v.5.3 is provided here, using a similar process. The spatial range for a coincidence is the same as used with MIGHTI: 2.5° latitude and 5° longitude in location and 2 h in time.

An individual comparison between an ACE sunrise and sunset occultation with the corresponding meteor radar measurements is shown in Fig. . These occultations were chosen to be shown as the coincident meteor radar measurement covers a large altitude range, whereas some only provide data in a small altitude window. We again only show the meteor radar winds converted using the v.5.3 heading angles. There is excellent profile agreement for both the sunrise and sunset comparisons.

Unlike the comparison with MIGHTI, the number of coincidences here is small, so the entirety of the coincident data set is considered. This means, unlike in MIGHTI where we removed outliers created by measurements with small altitude windows, outliers persist towards the edges of the meteor radar altitudes. This is best seen at 100 km in Fig. b, where the difference in meteor radar versus ACE is nearly 40 m s⁻¹ at the top of the window. There were 26 sunrise and 12 sunset coincidences. Of those, 18 were from the Mohe station, 12 from Beijing, and 4 from Wuhan and Ledong each. The averages of these coincidences are shown in Fig. . For the sunrises, we again see well matching profiles. The sunsets we see general agreement but with ACE showing more prominent features.

Similar to Fig.  for MIGHTI, we display the average sunrise-sunset bias when comparing with meteor radar in Fig. . Focusing on the center of the altitude window, the sunrise (sunset) bias for ACE v.5.2 is within ±2 of -9 m s⁻¹ (±3 of +18 m s⁻¹). For v.5.3, the bias is the same but with less variability, within ±1 m s⁻¹ (±2 m s⁻¹).

MERRA-2 provides instantaneous 3-dimensional 3-hourly horizontal vector winds . The model uses cubed-sphere horizontal discretizations, giving an approximate resolution of 0.5° × 0.625° and 72 hybrid-eta (altitude) levels that are best constrained up to ∼ 40 km, but extend up to ∼ 70 km . Because the model is a reanalysis, we are able to compare individual ACE occultations as well as an average of measurements with the corresponding MERRA-2 data. To do this, we find the nearest MERRA-2 time to the ACE occultation. Then, we bilinearly interpolate the nearest coordinates of MERRA-2 vector winds to the given latitude and longitude of the ACE occultation. The MERRA-2 vector winds are then converted to line-of-sight winds. We again only display the MERRA-2 profiles that are calculated using the v.5.3 heading angles.

A sample of individual sunrise and sunset occultation comparisons at northern, tropical, and southern latitudes can be seen in Fig. . These occultations were chosen as they show particularly strong altitude profile agreement, however, other occultations show similar results. We see a decrease in bias for both sunrise and sunset occultations in v.5.3 compared with v.5.2 as well as better matching profile. These improvements are best seen below ∼ 40 km, where MERRA-2 is better constrained. Near 52 km in Fig. a, there is a sharp eastward increase in ACE v.5.3 wind speed. This shift occurs at the boundary of segments 2 and 3 and is likely an issue with the retrieval at the top of segment 2 or bottom of segment 3 (see Fig. ) for this particular occultation.

All ACE line-of-sight wind speed measurements in 2019 and their corresponding MERRA-2 wind speeds are averaged and compared in Fig. . Note that there are ∼ 10 m s⁻¹ jumps in MERRA-2 wind speeds, best seen between 40 and 60 km. These are not real and are due to each MERRA-2 profile having data available at varying altitudes. However, the profile produced is real. There is a large profile disagreement between 40 and 60 km for ACE v.5.2 sunrises that now shows good agreement in v.5.3. Comparing the sunsets, we see similar profile agreement but a decrease in bias, particularly at lower altitudes.

Similar to MIGHTI and meteor radar, we are able to derive the sunrise-sunset bias, displayed in Fig. . We find that MERRA-2 and ACE differ by less than 10 m s⁻¹ up to 50 km with v.5.3, which is about half of the difference with v.5.2. We also find that the sunrise (sunset) bias has strongly improved. It is now within +5 m s⁻¹ (-5 m s⁻¹) up through 50 km, but largely within +3 m s⁻¹ (-3 m s⁻¹). Above 50 km the bias does grow to about +15 m s⁻¹ (-15 m s⁻¹), as seen in v.5.2. However, this higher altitude region is not well constrained within the MERRA-2 reanalysis.

HWM14 outputs for all of 2019 on a 5° × 5° grid with 15 minute time increments using observational geomagnetic ap indexes were retrieved from the Community Coordinated Modeling Center (CCMC) based out of NASA Goddard Space Flight Center (GSFC). The data ranges from the surface up to 1000 km in 10 km intervals.

Since HWM14 is a climatology, we only compare the average of many ACE and corresponding HWM14 wind speeds. We again compare all ACE wind speed measurements available from 2019. We first separate the ACE occultations into northern (latitude > 45° N), tropical (latitude between 30° S and 30° N), and southern (latitude < 45° S) regions based on latitude. It is not reasonable for the climatology to predict wind patterns within the winter polar vortex, so measurements during those periods are removed. Similar to MERRA-2, we then find the nearest HWM14 time to a given ACE occultation and bilinearly interpolate the vector winds of the nearby HWM14 coordinates to the given ACE coordinates. The HWM14 vector winds are then converted to line-of-sight winds.

This average of ACE v.5.2 and v.5.3 wind speeds from 2019 and their corresponding HWM14 wind speeds are shown in Fig. . For v.5.2 (v.5.3), 1037 (1032), 410 (418), and 945 (945) sunrise occultations were used for northern, tropical, and southern latitudes. For sunsets, 1300 (1294), 396 (402), and 924 (923) were used, respectively. The displayed HWM14 wind speeds are calculated using the v.5.3 heading angles. We see general profile agreement for both processing versions, with v.5.3 being slightly improved. Both versions of ACE processing show more prominent features than those found in HWM14, best seen at the ∼ 90 and ∼ 60 km features in Fig. a and e, respectively. The sunrise-sunset bias relative to HWM14 is shown in Fig. , similar to Fig.  relative to MERRA-2. We see a similar shape to that of MERRA-2, but the bias in v.5.2 appears less significant than in v.5.3. At low altitudes, the bias begins at less than 5 m s⁻¹, but increases to 17 m s⁻¹ at 60 km for v.5.2 and 20 m s⁻¹ at 70 km for v.5.3. In then decreases through 90 km before rising to it's maximum of 22 and 37 m s⁻¹ at 100 km for v.5.2 and v.5.3, respectively.

Specified Dynamics (SD) WACCM-X Version 2.2 provides global vector winds in 3 hour intervals on a 1.9 ° × 2.5 ° grid. This SD version of WACCM-X uses observational data to produce wind speeds closer to the actual atmospheric state. WACCM-X vector winds were compared with ACE-FTS line-of-sight winds for all of 2019 in a similar fashion as HWM14. The resulting line-of-sight winds are shown in Fig. . There is notably less agreement between both ACE versions and the WACCM-X profiles compared with HWM14. Both ACE sunrise and sunset averages for v.5.2 and v.5.3 in the northern and southern latitude region show additional features not found in the WACCM-X model. The model and ACE data show the best profile agreement in the tropical region. Comparing the sunrise-sunset biases, shown in Fig. b and d for v.5.2 and v.5.3, respectively, we see a similar shape as before when compared with HWM14. The sunrise (sunset) bias varies up through 80 km, but is less than -8 m s⁻¹ (+7 m s⁻¹). It then increases drastically to -34 m s⁻¹ (+32 m s⁻¹) at 103 km before dropped to near zero at 120 km.