Storm-time energy and momentum inputs (particle precipitation heating and Joule heating, ion drag) at high latitudes change thermospheric temperature and winds, create upward and equatorward winds, and drive a largescale circulation in the thermosphere (Rishbeth, 1975). There have been large equatorward winds associated with storms. Rees (1995) observed a nightside thermospheric equatorward wind up to 500 m/s during the November 8-9, 1991 storm over London (geographic coordinates: 51 • N, 1 • E or magnetic coordinates: 47 • N, 80 • E). Simulations also show that the storm-time nightside equatorward wind (200-300 km) at magnetic latitudes of 50 • -60 • reached ~450 m/s (Wang et al., 2008). FPI (Fabry-Perot interferometer) observations at a mid-latitude station (Fritz Peak Observatory, Colorado, 39.9 • N, 105.5 • W) showed that the nightside equatorward wind reached a maximum of 640 m/s during a Kp = 9 storm (Hernandez and Roble, 1976). These strong storm-time neutral winds lead to a significant redistribution of energy and momentum in the thermosphere and global changes in composition and variations in temperature, such as O/N 2 depletion (Burns et al., 1989;Strickland et al., 1995Strickland et al., , 2001;;Prölss, G. W., 1997;Zhang et al., 2004), NO and temperature enhancement (Richards, 2004;Zhang et al., 2014Zhang et al., , 2019;;Makela et al., 2014;Zhang et al., 2019). The storm-time variation in thermospheric temperature and composition extend from high to mid and low latitudes due to equatorward meridional wind.
On the other hand, energy and momentum inputs in the high latitude thermosphere are localized in magnetic latitude and local time during substorms. This affects the high-latitude ionosphere-thermosphere system by altering local neutral wind and composition, but to a much smaller degree and magnitude. Based on an analysis of CHAMP neutral wind observations at around 400 km altitude, Ritter et al. (2010) found that the neutral wind system was only weakly perturbed (<20 m/s) by substorms. FPI observations at Tromsø (Xu et al., 2019) also showed a mild neutral wind perturbation (<49 m/s) at local substorm onsets. Simulations revealed that the equatorward neutral wind around an altitude of 260 km peaked at only ~60 m/s around a latitude of 60 • during a substorm (Fujiwara et al., 1996).
The large different neutral wind responses during storms and substorms can be explained by the difference in the heating and ion drag magnitudes and heating areas in the high latitude thermosphere. Here we report a unique substorm event that caused a storm-like neutral wind perturbations at the Athabasca Observatory (latitude: 54.6 • . longitude: -113.6 • ). The observatory's magnetic latitude is 61.5 • .
The Special Sensor Ultraviolet Scanning Imager (SSUSI) (Paxton et al., 1992(Paxton et al., , 1993(Paxton et al., , 2002) ) (165-180 nm) (Paxton et al., 2018(Paxton et al., , 2021(Paxton et al., , 2017)). Sources of these emissions are due to either energetic particle (e.g. in the auroral region), photoelectrons produced by solar EUV radiation excitation, solar resonant scattering (for Lyman α, O 130.4 nm), and chemical processes (O + and electron radiative recombination and mutual neutralization for O 130.4 or 135.6 nm emissions). The N 2 LBHS or LBHL emissions are the Lyman-Birge-Hopfield vibrational bands of N 2 excited by secondary electrons created by either particle precipitation and/or photoelectrons created by solar EUV radiation (Paxton et al., 2021).
Fig. 1 shows six F18 SSUSI N 2 LBHS auroral images observed in the southern hemisphere on May 29, 2023 in the magnetic latitude (Mlat) and magnetic local time (MLT) coordinates. Dayglow due to solar EUV radiation has been removed in these images. The auroral emission around 03:35 UT (Fig. 1a) was weak and dominated by a thin poleward edge intensification. There was a poleward bulge around the midnight. By 05:18 UT (Fig. 1b), the bulge remained and the poleward edge of the oval in the pre-midnight sector moved from a Mlat below 70 • to above 70 • . About 100 min later, the bulge was intensified with an intensity around 1 kR and moved to the pre-midnight sector (Fig. 1c). The auroral oval intensity also increased at this time. Around 08:43 UT (Fig. 1d), the bulge changed its form but intensity of the bulge remained at around 1 kR. The bulge moved to 21:00 MLT and reached 75 • Mlat by 10:23 UT (Fig. 1e). The bulge disappeared around 12:04 UT (Fig. 1f) and a new bulge developed at the post-midnight sector. These features are typical for a substorm (Akasofu, 1964). Due to a limited ime resolution and spatial coverage of SSUSI data, it is difficult to estimate the true lifetime of the substorm. However, it is clear that the substorm lasted at least for ~7 h from 03:35 UT to 10:23 UT. Despite being a typical auroral substorm, there was little geomagnetic disturbance signature in nominal geomagnetic indices: Kp ≤ 0.667 between 00:00 and 12:00 UT (https:// www-app3.gfz-potsdam.de/kp_index/Kp_ap_since_1932.txt) with low AE < 150 nT (see an AE plot in Fig. 4d) and moderate THEMIS AE ≤ ~250 nT (Russell et al., 2008) (Fig. 4d). However, this event led to a significantly strong and storm-like neutral wind perturbation at a sub-auroral latitude and thermospheric composition changes extending to mid/low latitudes (~30 • ).
A new ground-based FPI was installed at the Athabasca University Geophysical Observatory (AUGO) (54.6 • N and -114.6 • E) in May 2023. The FPI measures the Doppler shift of thermospheric 630 nm emissions and provides neutral wind velocity estimations (Wu et al., 2004(Wu et al., , 2014)). Fig. 2a shows the meridional wind from the northward (blue line) and southward (red line) beams of the FPI between 05:00 and 10:00 UT on May 29, 2023. These meridional winds are all equatorward (negative values) and range from ~-50 m/s to ~ -450 m/s. These meridional winds show a similar pattern in the northward and southward beams. However, the magnitude of the blue line is larger than that of the red line between 07:00 and 09:00 UT. This is likely due to the fact that the northward beam measured the wind at a location closer to the auroral oval than the southward beam. Note that the mean measurement errors in the blue and red lines are 0.80 m/s and 2.65 m/s, respectively, which are much smaller than the absolute speeds of the meridional winds. This indicates that the wind measurements are reliable and the wind variations are significant. These meridional neutral winds (up to -450 m/s) are similar to the storm-time ones at sub-auroral latitudes reported by Wu et al. (2022) (e.g. ~-200 m/s at Boulder, CO, -400 m/s at Eureka, Canada, and ~-200 m/s at Resolute Bay, Canada).
The zonal wind from the Athabasca FPI is plotted in Fig. 2b. The zonal wind from the eastward beam (blue line) showed a negative value (westward) between ~05:00 and ~07:00UT. This westward wind turned into an eastward wind after 07:00 UT. The zonal wind from the westward beam (red line) shows a similar pattern. However, the eastward turning occurred about 1 h later around ~08:00 UT. This is likely due to the fact that eastward and westward beams were measuring wind at two different longitudes (local times) separated by ~5• (~550 km) at ~250 km of the airglow emission layer. The eastward beam was detecting the wind at a location toward the dawn sector while the westward beam measured wind at a location toward the dusk sector. The speed of the zonal wind is up to 180 m/s. The average measurement errors in the zonal wind (eastward and westward beams) are ~1.86 m/s and 1.47 m/s, respectively. These measurement errors are also much lower than the estimated zonal wind speeds, indicating a reliable measurement.
To investigate the neutral wind response to the auroral substorm, the FPI wind vector from the northern and eastward beams was overplotted in three SSUSI auroral images (Fig. 3). Note that the SSUSI data in the northern hemisphere did not cover the nightside auroral oval on May 29, 2023. The SSUSI auroral images in Figs. 1 and 3 were observed in the southern hemisphere. A statistical study shows that the differences in the average MLT and Mlat of substorm onset locations between the two hemispheres are less than 1:00 h and 1 • respectively, despite some IMF B y dependence (Wang et al., 2007). This indicates that substorm location is roughly conjugate in the two hemispheres. The substorm auroral images in the southern hemisphere could be used to represent the substorm location and activity in the northern hemisphere. Around 05:18 UT (Fig. 3a), the FPI wind was southward and weak (around 50 m/s). This was associated with the initial growth of the substorm. Note that the auroral images in Fig. 3 are plotted in a log scale to show the weak auroral activity. By 07:01 UT, the substorm intensified and this was associated with an increase of the southward wind to ~250 m/s (Fig. 3b). The wind direction is mostly southward at 05:18 and 07:01 UT (c.f. Fig. 2) as the substorm location is directly north of the FPI location. With the continuous growth of the substorm, the wind reached ~450 m/s around 08:43 UT (Fig. 3c). The wind direction is mainly southward but with a clear eastward component (c.f. Fig. 2b). This is likely because the substorm was still occurring in the pre-midnight MLT but the observatory rotated into MLT morning sector. Drivers for the changes in both meridional and zonal winds will be discussed in the next section.
Fig. 4 shows the solar wind and interplanetary magnetic field conditions during the substorm on May 29, 2023. The data were obtained from the NASA SPDF database and adjusted to the dayside magnetopause. The IMF B z (red line in Fig. 4a) was mostly positive or northward between 03:00 and 12:00 UT, except for a southward turning between 05:18 and 07:01 UT. This southward turning was associated with the auroral substorm development. On the other hand, IMF B x and B y were mostly positive and negative, respectively, and they dominated the IMF. The solar wind speed (Vx) was around -400 km/s, a very typical value (Fig. 4b). The solar wind dynamic pressure was around 0.6-1.0 nPa, which is relatively low (Fig. 4c). The geomagnetic Dst index (black line in Fig. 4d) shows a very smooth behavior with small values; it changed from ~-10 to ~1.0 nT, indicating no active storms during the period. The AE index from the Kyoto data center (also referred as Kyoto AE) was also low and changed slowly (below 150 nT, red line in Fig. 4d). This smooth varying and low AE index doesn't indicate that a substorm occurred as a typical AE substorm signature is a rapid increase (peaked around 440 nT) followed by a rapid decrease based on a super-epoch analysis (Caan et al., 1978). It remains to be addressed why the auroral substorm on May 29, 2023 is not registered in the AE index. Super-MAG auroral electrojet (SME) index (Gjerloev, 2009(Gjerloev, , 2012) ) is similar to the Kyoto AE index. SME (not shown here), derived from more ground magnetic data, also shows no clear substorm signature except for weak peaks of ~150 nT around 10:00 UT and ~175 nT around 12:00 UT (https://supermag.jhuapl.edu). On the other hand, THEMIS AE index (blue line in Fig. 4d), derived from ground magnetometer data obtained over 22 stations in Canada and US (Russell et al., 2008;Pu et al., 2010) are higher than the Kyoto AE between 09:45 and 12:00. THEMIS AE shows three weak peaks, ~186 nT, ~230 nT, and ~253 nT at ~10:00, ~11:00, and 11:50 UT, respectively. These peaks are the UTs when the auroral bulge moved to 21:00 MLT and reached 75 • MLAT by ~10:23 UT (shown in Fig. 1e) and when a new bulge developed at the post-midnight MLT sector at ~12:04 UT (as shown in Fig. 1e). These observations are consistent with a weak (or low intensity) substorm (AE<300 nT) based on the classification of Speiser et al. (1996) and Immel et al. (1997). These peaks are the UTs when the auroral bulge moved to 21:00 MLT and reached 75 • MLAT by ~10:23 UT (shown in Fig. 1e) and when a new bulge developed at the post-midnight MLT sector at ~12:04 UT (as shown in Fig. 1e). Nevertheless, the SSUSI auroral observations unambiguously demonstrate a clear auroral substorm which lasted for at least ~7 h from ~03:35 to 12:04 UT or later (Fig. 1).
The large storm-like equatorward (southward in the northern hemisphere) wind (Fig. 2) suggests strong heating (particle precipitation and Joule heating) during the substorm. The height integrated Joule heating rate (Q J ) can be represented by Σp * (E + Vn × B) 2 , where Σp, E and Vn are height integrated Pedersen conductance, electric field, and neutral wind velocity, respectively, under an assumption that the Joule heating is deposited in a limited altitude range and E, B, Vn are constant over that range (Baker et al., 2004). The total heating rate Q is the sum of Q J and Q p , where Q p is the energy flux of precipitating particles. Based on an analysis of SuperDARN plasma convection data, TIMED/GUVI (Paxton et al., 1999;Christensen et al., 2003) auroral data over 60 orbits, and the HWM93 neutral wind model, Baker et al. (2004) calculated Q J , Q p on the nightside under quiet to moderate geomagnetic conditions. They found that heat sources (Joule and particle) tend to be complementary because the strength of the electric field is often anti-correlated with auroral brightness or particle precipitation. Joule heating is dominant on the dusk side and in the poleward portion of the auroral oval. The contribution of neutral wind to the Joule heating could be positive or negative, but its magnitude is noticeably lower than Q J or Q p (Baker et al., 2004). For the substorm event on May 29, 2023, we assume that the neutral wind impact on the total heating can be neglected. In this case,
Fig. 5 shows plasma convection maps from SuperDARN (Greenwald et al., 1995) in the northern hemisphere at 05:19, 07:01 and 08:45 UT on May 29, 2023. The red circles are used to mark the substorm locations seen by DMSP/SSUSI (Fig. 3). Note that the coincidence between the strong plasma convection and the bright auroral image, two signatures of a substorm, also validates the mapping of southern auroral images to the northern hemisphere. The SuperDARN data show a strong plasma convection speed or E field in the pre-midnight sector (21:00-24:00 ML) around 70 • Mlat. This is collocated with the auroral substorm or high Q p . Note that high Q p not only enhances the Q, but also Q J as Q p is proportional to Σ p . This indicates a highly enhanced Q in the localized substorm area. Such a strong and localized heating is likely the driver for the observed storm-like neutral wind disturbance observed at Athabasca which is located at ~60 • Mlat, southward of but not far away from the substorm location. As Athabasca is located in the subauroral region and away from the strong plasma convection during the substorm, we do not expect ion drag plays an important role in the wind perturbations seen by the FPI at Athabasca.
The high total heating rate at the substorm location explains the observed high meridional neutral wind. This high or storm-like neutral wind is expected to bring the O/N 2 depleted air form the sub-storm region to mid and low latitudes (Burns et al., 1989;Strickland et al., 1995). Indeed, TIMED/GUVI did observe an O/N 2 depletion that extended to mid/low latitudes. Fig. 6a shows the thermospheric O/N 2 column density ratio derived from TIMED/GUVI data (Zhang et al., 2004) over three days (May 28-30, 2023) in the afternoon sector (~16:18LT). A strong O/N 2 depletion was detected over the North Fig. 5. SuperDarn ionospheric convection maps in the northern hemisphere at (a) 05:18-05:20 UT, (b) 07:00-07:02 UT and (c) 08:44-08:46 UT on May 29, 2023 from https://superdarn.ca/. The red oval is used to mark the sub-storm locations observed by F18 SSUSI (Fig. 3). The edge of the maps is at 60 • Mlat. The Athabasca observatory (61.5 • Mlat) is at a location slightly above the edge on the nightside. The vectors represent the plasma drift direction and speed. The shaded contours are for electric potential derived from SuperDARN. America and Pacific Ocean around the end of May 28 and beginning of May 29 (red arrow in Fig. 6a). This O/N 2 depletion was due to a weak geomagnetic storm with a minimum Dst of -32 nT at 14:00UT, May 28 (Dst was not shown). On May 29, the O/N 2 depletion over the North American sector and the East Pacific Ocean (see purple arrow) recovered from the storm O/N 2 depletion. However, there was a localized O/N 2 depletion (marked by an orange arrow in Fig. 6a) in the west Pacific Ocean (east of Japan) on May 30. Note that O/N 2 over the same region on May 29 was undisturbed (black arrow in Fig. 6a). Therefore, the O/N 2 depletion in the Pacific Ocean, east of Japan, on May 30 was not due to the storm residual O/N 2 depletion, but a new heating or the substorm on May 29 as the strong equatorward meridian wind brought depleted O/N 2 from the substorm region to the sub-auroral latitudes. It is also worth noting that the Dst index started a rapid decrease around 8:00UT on May 28 (not shown). Assuming the storm-time heating is primarily in the midnight sector (00:00 MLT) and considering the corotation of the O/N 2 depletion (Prölss, 1980), the storm O/N 2 depletion is expected to be observed about 16~17 h later at the local time of the GUVI observation (~16:18 LT). The lowest O/N 2 due to the storm occurred around 1:00 UT on May 29 (see the red arrow in Fig. 6b and the O/N 2 line plot at a latitude of 50 • , red line). The time difference between 08:00 UT on May 28 and 01:00 UT on May 29 is 17 h, consistent with the corotation theory. On the other hand, the peak equatorward meridian wind due to the substorm occurred around 08:43 UT (Fig. 2a) and peak location of the substorm was at ~21:30 LT (Fig. 1d). Following the same corotation theory, the substorm induced O/N 2 depletion is expected to be observed around 1:30 + 16:18 = 17:48 later after 08:43 UT (May 29) = 2:31 UT (May 30) at the 16:18 LT. The orange arrow in Fig. 6b shows the lowest O/N 2 at 50 • latitude due to the substorm occurred around 04:20 UT on May 30. The difference (2:31 UT versus 04:20 UT) is about 1:49 which is relatively small. Therefore, the corotation theory also roughly explains the time delay between the observed neutral wind on nightside and O/N 2 at 16:18 LT.
It is well known that thermospheric neutral wind is mostly affected by three drivers: the pressure gradient force, ion drag, and the Coriolis force. The intense localized heating around the sub-storm region can explain the strong equatorward meridional wind (Fig. 2a). This is mostly driven by the pressure gradient force as the heating causes local temperature increase that produces an equatorward pressure gradient force. The ion drag effect is negligible as the ion drift speed is very small at the location of the Athabasca observatory (61.5 • Mlat), outside of the polar plasma convection pattern (Fig. 5). The zonal Coriolis force remained westward due to the sustained equatorward meridional wind during the substorm (Fig. 2a). On the other hand, the zonal wind changed from ~-100 m/s (westward) to ~180 m/s (eastward) at 07:10 UT (see the blue line in Fig. 2b). Because of a low ion drift speed at the observatory through the life cycle of the substorm, the ion drag effect on the zonal wind can also be neglected. The Coriolis force is westward in the region with a strong equatorward meridional wind. This could explain the westward wind between 05:18 and 07:10 UT (Fig. 2b, blue line). The substorm lasted for a few hours. The direction of pressure gradient force changes with the relative position between the observatory and substorm location. Fig. 3a shows that the observatory was located at ~21:00 MLT at 05:18 UT and the center of the substorm was at ~22:30 MLT on the east side of the observatory. Therefore, the pressure gradient force is westward. This works together with the westward Coriolis force to enhance the westward zonal wind. The observatory location in the Mlat-MLT coordinates moved from the midnight to the dawn sector, for example, around 22:45 MLT at 07:01 UT (Figs. 3b) and 00:30 MLT at 08:43 UT (Fig. 3c). This indicates that the wind observation at the observatory occurred eastward of the substorm location at 08:43 UT. The pressure gradient force is thus eastward. Note the Coriolis force remained westward during the substorm with an equatorward meridional wind. The observed eastward wind indicates that the eastward pressure gradient force is larger than the Coriolis force in magnitude, resulting in a net eastward zonal wind at the location of the observatory. Here the competition between the pressure gradient force and Coriolis force determines the zonal wind direction.
It is also interesting to note that the westward beam (red line in Fig. 2b) and the eastward beam (blue line in Fig. 2b) were measuring the zonal wind at two different locations or longitudes. The FPI beams were pointed to the sky with an elevation angle of 45 • . Assuming the peak 630.0 nm emission occurs at 250 km altitude, the measurement locations of the westward and eastward beams are separated by about 500 km or about 0.55 h in MLT. This separation indicates different relative measurement locations to the substorm region and thus different pressure gradient force directions, leading to a time delay in the eastward turning of the zonal wind of the westward beam. Physics-based modeling is required to elucidate the details of the time delay. On the other hand, the northward and southward beams are also separated by about 500 km in their measurement locations. The meridional wind from the northward beam was detected at a location closer to the substorm than that of the southward beam. This is likely the key reason for the difference in the meridional winds (Fig. 2a) where the northward beam detected a stronger meridional wind.
The long lasting auroral substorm (Fig. 1) was associated with a southward turning in IMF B z , a strong IMF B y and a larger IMF B x . The solar wind dynamic pressure was low (1 nPa) with a typical solar wind speed of ~400 km/s. This suggests that IMF plays an important role in transferring solar wind energy to the magnetosphere and/or release of the stored magnetospheric energy. Previous modeling and observations showed that IMF with a dominant B y enhances the ion convection pattern, Joule heating rate, neutral winds, and consequently, a bigger O/ N 2 depletion (30% with a strong B y versus ~8% with a B y ~0 nT) (Cai et al., 2022). Furthermore, IMF B x , B y and B z also show a periodic variation during the substorm period (Fig. 4a). It is well known that a periodic fluctuation in solar wind (such as Alfven waves) can pump energy into the magnetosphere and cause a long-duration continuous magnetic disturbance (e.g. an enhanced AE index) (Tsurutani and Gonzalez, 1987) and leads to substorms (Tsurutani et al., 1990). Simulations showed that periodic variations in IMF Bz could change thermospheric meridional wind (Zhang et al., 2019). A more recent study by Zhang et al. (2021) also indicated that periodic variations in the IMF were associated with a 20-day storm-like O/N 2 depletion. However, the periodic variations in IMF did not lead to periodic variations in SymH (Symmetric geomagnetic field disturbance in H component, parallel to the Earth's dipole magnetic axis), suggesting that the impact was mostly at high latitude regions. These previous studies and this study provide a new insight into the auroral substorm and its impacts on thermospheric wind and composition.
Fig. 4d indicates a weak/no substorm signature in THEMIS/Kyoto AE during the auroral substorm observed by SSSUSI. These observations suggest that the substorm electrojet was weak. SSUSI auroral data explain the weak electrojet during the substorm. SSUSI N 2 LBHS and LBHL auroral radiances have been used to derive mean energy (Eo) and energy flux (Q) using the method described by Zhang and Paxton (2008). F18 SSUSI Eo and Q products during the substorm are shown in Fig. 7. At 05:18 UT, the initial substorm growth phase, the Q and Eo are around or above 3 erg/s/cm 2 and 5 keV, respectively. However, the Eo started to decrease around 07:01 UT and was ~2 keV at 10:23 UT, the time of the substorm peak. The Eo derived from the LBHS/LBHL ratio has a 2 keV minimum value. The actual Eo could be lower than 2 keV. Fig. 3 of Rees (1989) shows that the peak ionization altitudes for 5 keV and 2 keV precipitating electrons are ~115 km and ~130 km, respectively. The MSIS2.0 model shows that the neutral number density near the FPI location around 10:00 UT on May 29, 2023 (55 • latitude, 270 • longitude) decreased from 2.14x10 12 cm -3 at 110 km to 1.3x10 11 cm -3 at 130 km, a factor of 16 drop (https://kauai.ccmc.gsfc.nasa.gov/instant run/nrlmsis/). This indicates that ion-neutral collision frequency also drops by a similar magnitude. The relatively low ion-neutral collision frequency at 130 km suggests that the ions and electrons mostly drift together, thus lead to no or very weak electrojets or signature in ground magnetic field data.
Why is THEMIS AE higher than Kyoto AE during the auroral substorm? The Kyoto AE index is derived from the magnetic field data observed at 12 ground stations at roughly fixed latitudes and evenly spaced longitudes in the northern hemisphere. However, the THEMIS AE are obtained from the ground magnetic field data at 22 stations in a limited longitude but a broader latitude range in Canada and US. The Substorm occurred in Canada. Therefore, most of the THEMIS stations are close to the substorm weak electrojet. On the other hand, most of the Kyoto AE stations are far away from the weak electrojet. The AE index is the average of magnetic field deviations from the stations. This explains why THEMIS AE (with a weak substorm signature) is higher than Kyoto AE (without substorm signature).
The relatively low energy of precipitating electrons in the substorm region also impacts the neutral wind response to the Joule heating. As discussed in the previous paragraph, the peak ionization altitude is around ≥130 km for precipitating electrons with energy ~2 keV or less. Therefore, the peak particle and Joule heating is also around 130 km or higher. Since the neutral density at 130 km is about 16 times lower than that at 110 km, the heating neutral heating capacity is relatively low at 130 km. Any enhancement in the particle and Joule heating could create a strong wind. This is another likely reason for the large neural wind during the substorm.
F18 DMSP SSUSI, a FUV spectrograph imager, detected a long lasting (~7 h) auroral substorm on May 29, 2023. TIMED/GUVI observed a strong O/N 2 depletion that extended to mid/low latitudes (~30 • ) in the northern hemisphere on the same day. Ground FPI observations from the Athabasca Observatory, at a sub-auroral latitude, detected strong and storm-like equatorward meridional winds in the thermosphere (up to ~450 m/s). The FPI also observed strong zonal winds (up to 180 m/s). The zonal wind direction changed from westward to eastward during the substorm. Such a direction change is due to the competition between the pressure gradient and Coriolis forces. While the Coriolis force is always westward with an equatorward meridional wind in the Northern Hemisphere, the pressure gradient force direction changed from westward to eastward due to changes in the relative locations of the observatory and substorm. Coincident SuperDARN measurements showed strong plasma convection in the substorm location. These SuperDARN and SSUSI measurements indicate a strong local heating (Joule and particle precipitation) at the substorm location. This strong and localized heating caused the storm-like thermospheric response in neutral wind and composition. A dominant IMF B y and periodic variation in the IMF likely provide a favorable upstream condition for continuous energy input from the solar wind to the magnetosphere and/or release of stored magnetospheric energy to drive the long duration substorm and consequently wind and composition changes in the thermosphere. A physicsbased simulation is needed for a better understanding of the thermospheric response to this substorm. Y. Zhang et al.