Fabry-Perot Interferometer Observations ofThermospheric Horizontal Winds DuringMagnetospheric Substorms

L. Cai1,2 , S. Oyama1,2,3 , A. Aikio1 , H. Vanhamäki1 , and I. Virtanen1

1Ionospheric Physics Unit, University of Oulu, Oulu, Finland, 2Institute for Space-Earth Environmental Research(ISEE), Nagoya University, Nagoya, Japan, 3National Institute of Polar Research (NIPR), Tachikawa, Japan

Abstract The high-latitude ionosphere-thermosphere system is strongly affected by themagnetospheric energy input during magnetospheric substorms. In this study, we investigate the responseof the upper thermospheric winds to four substorm events by using the Fabry-Perot interferometer atTromsø, Norway, the International Monitor for Auroral Geomagnetic Effects magnetometers, the EISCATradar, and an all-sky camera. The upper thermospheric winds had distinct responses to substorm phases.During the growth phase, westward acceleration of the wind was observed in the premidnight sectorwithin the eastward electrojet region. We suggest that the westward acceleration of the neutral wind iscaused by the ion drag force associated with the large-scale westward plasma convection within theeastward electrojet. During the expansion phase, the zonal wind had a prompt response to theintensification of the westward electrojet (WEJ) overhead Tromsø. The zonal wind was acceleratedeastward, which is likely to be associated with the eastward plasma convection within the substormcurrent wedge. During the expansion and recovery phases, the meridional wind was frequently acceleratedto the southward direction, when the majority of the substorm WEJ current was located on the polewardside of Tromsø. We suggest that this meridional wind acceleration is related to a pressure gradientproduced by Joule heating within the substorm WEJ region. In addition, strong atmospheric gravity wavesduring the expansion and the recovery phases were observed.

1. IntroductionSubstorms are transient processes that involve a sequence of loading and unloading of the solar wind energyin the nightside magnetosphere. During substorms, up to 80% of the solar wind energy input is dissipated inthe high-latitude ionosphere-thermosphere system via electromagnetic energy exchange and auroral precip-itation (e.g., Østgaard et al., 2002; Tanskanen et al., 2002; Thayer & Semeter, 2004). One of the outstandingquestion is how the energy dissipation disturbs thermospheric winds during substorms. The thermosphericwind is a key parameter that significantly affects mass transportation (e.g., Lilensten & Lathuillere, 1995;Oyama et al., 2003), ionospheric currents (Deng et al., 1993; Peymirat et al., 2002), and electromagneticenergy exchange (Aikio et al., 2012; Cai et al., 2013; Thayer & Semeter, 2004).

An isolated substorm typically has three phases: the growth phase (GP), the expansion phase (EP), and therecovery phase (RP; e.g., Akasofu, 1968; McPherron, 1970). The GP usually starts with the southward turn-ing of the interplanetary magnetic field (IMF) and lasts typically 30–90 min (Li et al., 2013). During the GP,the auroral oval moves equatorward with the expansion of the polar cap, corresponding to the dayside mag-netic merging and the accumulation of the magnetic flux in the magnetotail. The onset of the EP (substormonset) is identified optically by a sudden brightening of the preexisting quiet auroral arc with quick pole-ward expansion. During the EP, the auroral forms expand eastward, westward, and poleward from the onsetlocation, in association with the intensification of the westward electrojet (WEJ) inside the substorm cur-rent wedge. During the RP, the WEJ slowly decreases. While isolated substorms follow well the three-phaseparadigm, events with multiple onsets or intensifications are also observed. Since the ionosphere and ther-mosphere are highly coupled, the thermospheric wind may have different responses to substorm phases andauroral activity.

Thermospheric dynamics at high latitudes is affected mainly by the solar extreme ultraviolet radiation,the ionospheric convection, and auroral activity. The large-scale wind circulation depends on seasons

RESEARCH ARTICLE10.1029/2018JA026241

Key Points:• Ion drag plays an important role in

acceleration of the zonal wind duringsubstorms

• Eastward acceleration of thezonal wind is associated withintensification of westward electrojetduring the expansion and recoveryphases

• Observed meridional windacceleration may be produced bypressure gradient caused by theJoule heating

Correspondence to:L. Cai,lei.cai@oulu.fi

Citation:Cai, L., Oyama, S.-I., Aikio, A.,Vanhamäki, H., & Virtanen, I.(2019). Fabry-Perot interferometerobservations of thermospherichorizontal winds duringmagnetospheric substorms. Journalof Geophysical Research: Space Physics,124, 3709–3728. https://doi.org/10.1029/2018JA026241

Received 26 OCT 2018Accepted 17 MAR 2019Accepted article online 23 APR 2019Published online 22 MAY 2019

©2019. American Geophysical Union.All Rights Reserved.

CAI ET AL. 3709

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

(Aruliah et al., 1996; Dhadly et al., 2017; Drob et al., 2008, 2015), solar wind and IMF conditions (Försteret al., 2008; Richmond et al., 2003), and geomagnetic activity (Dhadly et al., 2018; Emmert et al., 2008; Ritteret al., 2010; Xiong et al., 2015). Ritter et al. (2010) investigate substorm-related wind disturbances based onlong-term CHAMP satellite measurements. They showed that westward wind increases from 20 to 50 m/sat low and middle latitudes after onsets.

Several first principles models, such as Thermosphere-Ionosphere-Electrodynamics General CirculationModel (TIEGCM) (Richmond et al., 1992) and the global ionosphere-thermosphere model (GITM; Deng &Ridley, 2006; Ridley et al., 2006), are able to simulate the large-scale thermospheric wind dynamics dur-ing static IMF conditions (Kwak & Richmond, 2007) and during geomagnetic storms (Lu et al., 2016; Wanget al., 2008), and substorms (Richmond & Matsushita, 1975; Wang et al., 2017). Wang et al. (2017) studied theupper thermospheric wind dynamics during an idealized substorm using the GITM model. In their results,large-scale winds are dominated by ion drag on both the dayside and nightside, but heating sources such asJoule heating and auroral precipitation become important on the nightside. In another GITM simulation byLiuzzo et al. (2015), simulated mesoscale winds during a substorm are highly dependent on the models ofionospheric convection and auroral precipitation and show clear discrepancies with the local observations.The model simulations may be improved by adding additional physical processes like cusp heating (Shenget al., 2015; Wu et al., 2015).

Observations by the Fabry-Perot interferometer (FPI) and its all-sky type, scanning Doppler imager, revealthat thermospheric winds are more dynamic and responsive to the magnetospheric forcing in both verti-cal and horizontal components than modeling or statistical studies typically show (Anderson et al., 2012a,2012b, 2012c; Aruliah et al., 2005; Griffin et al., 2008). Mesoscale and small-scale winds have been foundin association with auroral structures, such as auroral arcs (Conde et al., 2001; Kosch et al., 2010), pole-ward expanding aurora after substorm onset (Oyama et al., 2017), and auroral patches inside pulsatingaurora (Oyama et al., 2010, 2016). Those findings also suggest that the thermospheric wind has more rapidresponses than theories have predicted. However, the physics behind the rapid and structured responses ofthe thermospheric wind are not currently understood.

In this paper, we utilize the FPI (630.0 nm) measurements at the EISCAT Tromsø site to investigate theupper thermospheric wind during four substorm events, which take place during non–storm time condi-tions. During a storm time, the global thermosphere may be in a disturbed state even before the substormonset and, therefore, it may be difficult to observe the changes that individual substorms produce. We usethe International Monitor for Auroral Geomagnetic Effects (IMAGE) magnetometers to identify the regionalelectrojet currents, and we have EISCAT incoherent scatter radar data for two events. In all events, we alsohave all-sky camera data. The aim of the paper is to study the F region wind response to substorm phases andthe possible forcing that affects the wind. In section 2, measurements and data analysis are described, andthe background conditions including solar wind conditions are discussed in section 3. Section 4 focuses onacceleration of the horizontal wind in individual events. Section 5 contains discussion, and section 6 givesthe summary and conclusions.

2. Measurements and Data Analysis2.1. MeasurementsUpper thermospheric winds were measured by the FPI operated at the Tromsø EISCAT radar site (geo-graphic 69.58◦N, 19.23◦E, and corrected geomagnetic (66.73◦N, 102.18◦E). The etalon of the FPI can producemore than 10 fringes, imaged by a CCD camera with 1,024 × 1,024 pixels. We have adopted 2 × 2 binning forthe CCD image to keep equivalent level of signal-to-noise ratio at a shorter exposure time, in order to achieveshorter time resolution of the wind. The Doppler shifts are derived from individual fringes, and by averagingthe results, one line-of-sight wind speed is given with the standard deviation as the error. The line-of-sightwind speeds were measured sequentially at five directions: geographic east, west, north, south, and vertical.The combination of the data from the five azimuthal directions provides a wind vector in zonal, meridional,and vertical directions, under the assumption of homogeneous conditions within the region covered by thefive beams (Shiokawa et al., 2003). Usually, the zenith angle of the nonvertical beams is set as 45◦ or 15◦.Because the auroral fine structures may affect the assumption of homogeneous conditions, only measure-ments with the zenith angle of 15◦ are used for analysis. The FPI can measure both the 557.7-nm (green line)and 630.0-nm (red line) emissions. While the emission height of the green line is typically at 100–120 km inthe E region, the height of red line is typically at 230–250 km in the F region. In this study, we focus on the

CAI ET AL. 3710

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

upper thermospheric wind derived from the red line measurements. In addition to the wind velocity, theFPI is able to estimate the neutral temperature (Nakamura et al., 2017). However, the adoption of the 2 × 2binning causes difficulty in temperature derivation. Hence, the neutral temperature is not presented in thisstudy.

The ground-based IMAGE magnetometer chain are used to estimate the two-dimensional (2-D) equivalentcurrents by means of Spherical Elementary Current Systems (Amm & Viljanen, 1999; Pulkkinen et al., 2003).The time resolution of the 2-D equivalent current map is 1 min. The equivalent east-west current along lat-itude is taken at the geographic longitude of 19.5◦E. The equivalent current represents the divergence-freepart of the real ionospheric current. Divergence-free current is approximately equal to Hall current, if theconductance gradient is small or aligned with the electric field (e.g., Vanhamäki & Amm, 2011). Based onthis assumption, we can estimate the direction of the F region electric field using the perpendicular rela-tionship between the Hall current and the convection electric field. Furthermore, since the ion velocity inthe F region is given by the E × B drift, typically the equivalent current and the F region ion velocity areantiparallel.

The EISCAT UHF incoherent scatter radar at Tromsø provides measurements of the electron density, ionand electron temperatures, and the line-of-sight ion speed. The ultrahigh frequency Common Programme2 (CP2) experiment points the antenna in three or four beam directions, including one vertical, one fieldaligned, and one or two with high-elevation angles (usually greater than 60◦). This experiment can be usedto estimate the full vector of the F region ion velocity and the perpendicular electric field to the magneticfield with errors based on Bayesian inversion method (Nygrén et al., 2011, 2012; Virtanen et al., 2014). Acombination of the EISCAT and the FPI measurements will be used to calculate the ion drag force ande-folding time as described in section 2.2.

The auroral activity is monitored by an digital all-sky cameras (ASCs) at Tromsø. The meridional keogramwill be shown, which is produced from the green channel of the raw image. The time resolution is 1 min forthe events in 2012 and before, and 30 s after 2012.

The solar wind parameters at the Earth's bow shock are provided by the OMNI database with 1-minresolution. Solar wind coupling functions are extensively used to estimate efficiency of the solarwind-magnetosphere coupling. In this study, we utilize the coupling function by Newell et al. (2007) as aproxy for the solar wind energy input. The unit of the Newell coupling function is arbitrary, so only thenumerical values are reported in this paper. The global auroral electrojet indices (AE,AU, and AL) arederived mainly from 12 geomagnetic observations spreading along the auroral zone (e.g., Mayaud, 1980).One of the observations is made by the magnetometer at Abisko in Scandinavia, which locates at a lower lat-itude than Tromsø. The local electroject indices (IE, IU, and IL) are derived from the IMAGE magnetometerchain in a similar way for AE indices (e.g., Tanskanen, 2009).

2.2. Neutral Wind AccelerationIn order to understand the dynamics of the upper thermospheric wind during substorms, the wind acceler-ation and the relevant driving forces should be quantified. The motion of the neutral gas is described by themomentum equation as shown below (Brekke, 2012):

𝜕u𝜕t

= −(u · ∇)u + 𝜈ni(vi − u) − 1𝜌∇p − 2Ω × u + 𝜇

𝜌∇2u + g , (1)

where u is the neutral wind velocity in the geographic coordinate system, 𝛺 the Earth's angular velocity,𝜌 the neutral mass density, 𝜇 the viscous coefficient, p the pressure of the neutral gas, 𝜈ni the neutral-ioncollision frequency, vi the ion velocity, and g the gravitational acceleration. In equation (1), the term onthe left-hand side is the Eulerian derivative, and the terms on the right-hand side are advection, ion drag,pressure gradient force, Coriolis force, viscous force, and gravity.

At high latitudes, the ion drag force is usually considered as the dominant force during active auroral condi-tions. Hence, one can define the e-folding time to describe the time that the neutral wind needs to acceleratefrom one steady state to another due to the change of the ion drag (e.g., Kosch et al., 2001):

𝜏ni =1𝜈ni

=𝜌n

𝜌i

1𝜈in

. (2)

For two events (E1 and E2), the ion drag 𝜈ni(vi − u) can be estimated using the simultaneous measurementsby the EISCAT radar giving the ion velocity and the electron density and the FPI measurements giving the

CAI ET AL. 3711

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Table 1Summary of Geomagnetic Activity Conditions and Data Used in the Four SubstormEvents

Events Kp Dst (nT) FPI wind EISCATE1 on 2011-12-17 [0, 0+ ] −9 6 min CP2, 12 min EFE2 on 2012-10-17 [1− , 3− ] −35 6 min NoneE3 on 2012-12-01 [1− , 2+ ] −5 6 min NoneE4 on 2016-10-23 [1+ , 2+ ] −29 3 min CP2, 6 min EF

Note. The Kp index range is given and minimum of the Dst index within 3 days beforethe onset. For the FPI and EISCAT, time resolutions to derive the full vectors are given.FPI = Fabry-Perot interferometer; EISCAT = European Incoherent Scatter ScientificAssociation; CP2 = Common Programme 2. Dates are formatted as YYYY-MM-DD.

wind velocity. The ion-neutral collision frequency is first calculated based on the equations provided bySchunk and Nagy (2009) and then the neutral-ion collision frequency is estimated using equation (2). Therequired neutral density, constituents, and temperature are taken from the NRLMSIS-00 model, and the iondensity and temperature are measured by the EISCAT radar.

In addition to ion drag, pressure gradient is another important term that affects the thermospheric winddynamics. From local measurements this term cannot be quantified. Advection depends both on wind veloc-ity and its spatial gradient. During quiet geomagnetic conditions the advection effect should be very weakdue to small wind velocity and gradient. The Coriolis force tends to move the fluid toward the right side inthe Northern Hemisphere. For a speed of 100 m/s, the acceleration due to Coriolis force is 0.0136 m/s2 at thelatitude of Tromsø. The viscous force is considered to be less important according to previous studies (Kwak& Richmond, 2007; Wang et al., 2008). We use the HWM-14 model by Drob et al. (2015) to give the quiet timebackground winds. The model results are dominated by the diurnal tide due to the solar-radiation-inducedpressure gradient (e.g., Witasse et al., 1998).

The Eulerian derivative term 𝜕u𝜕t

represents the total acceleration rate viewed at a fixed-point on the Earth,which is called the neutral wind acceleration rate hereafter. We will calculate the wind acceleration ratefrom the FPI data using a linear regression method called the Maximum Likelihood Estimation technique(York, 1968; York et al., 2004). The method gives the error estimates of the acceleration rate based on themeasurement errors. By selecting a time window, the data within the time window is fitted and the slopeof the fitted line with error estimate is taken as the wind acceleration rate. The default time window is aperiod of 20 min, which means that at least three data points will be included in the estimate. In addition, weselected several time windows, which are typically longer than 20 min, by taking into account the continuityof the wind acceleration and substorm phases. The results will be shown in section 4.

2.3. Identification of Substorm PhasesIn this paper, we study four non–storm time substorm events (E1–E4), for which the FPI measurementsfulfill the following criteria: measurements during clear-sky condition, in a scanning mode with the zenithangle of 15◦, and with exposure time shorter than 30 s. Of the four events, two have substorm onsets whenthe FPI site was located in the postmidnight sector, and the other two had onsets when the FPI site waslocated in the evening or premidnight sector. The summary of measurements and geomagnetic conditionsfor the four events is shown in Table 1. It should be noticed that EISCAT CP2 measurements were availablein E1 and E4, so that the local ion-neutral coupling can be studied.

In the four events, substorm phases are identified by taking into account the IMF conditions, the IL index,the auroral electrojets, and ASC images. Generally, the GP starts at the southward turning of the IMF. Theonset of the EP is determined by the sudden decrease of the IL index as described above. The end of the EPis the time when the WEJ has reached the most poleward extension (Rostoker et al., 1980). The RP is theperiod with decreasing intensity of the WEJ and aurora. The end of the RP is at the time when the IL returnsbackground values.

3. Background Conditions and Neutral WindsFigure 1 shows the two postmidnight events E1 (left) and E2 (right). Substorm phases are identified asdescribed in section 2.3. The phases are separated by the vertical dashed lines and marked by the color

CAI ET AL. 3712

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 1. Background conditions and overviews of the FPI measurements for Event 1 on 17 December 2011 (left) andEvent 2 on 17 October 2012 (right). The panels from top to bottom show interplanetary magnetic field (IMF) By (black)and Bz (red) components in geocentric solar magnetic coordinate system, solar wind speed, Newell coupling function,global auroral electrojet indices (AE, AL, and AU), local auroral electrojet indices (IE, IL, and IU), FPI (630.0 nm)meridional (black) and zonal (red) winds, FPI (630.0 nm) upward wind, and FPI (630.0 nm) fringe peak counts fromfive directions: north (N), east (E), south (S), west (W), and upward (Up). Substorm phases are separated by the verticaldashed lines and marked by the colored horizontal bars on the top, where GP means the growth phase, EP theexpansion phase, and RP the RP. The red dashed vertical lines indicate the substorm onsets. For this event, there aretwo main onsets. GP = growth phase; EP = expansion phase; RP = recovery phase; FPI = Fabry-Perot interferometer.

bars on the top. The red dashed lines indicates the onsets. Panels from top to bottom are the IMF By andBz components in the geocentric solar magnetic coordinate system; the solar wind speed; Newell couplingfunction; AE, AU, and AL indices; IE, IU, and IL indices; FPI horizontal winds; FPI vertical wind; and theFPI peak counts. The peak counts indicate increasing emission intensity in relation to the auroral activity.

Both events E1 and E2 occur during southward IMF conditions. In E1, the IMF turned southward at 2026UT, which is marked as the start time of the GP. The IMF has a weak By component, and the solar windspeed is slow, only 270 km/s. As a result, the values of the Newell coupling function vary between 2,000 and4,000, suggesting relatively small amount of the solar wind energy input to the magnetosphere during theGP. For this event, we have identified two onsets of the EP: one is at 2222 UT and the other at 2320 UT (seethe detailed discussion in section 4.1). Apparently, the first EP (EP1) is stronger than the second one (EP2),according to the decrease of the AL index. In E2, the IMF has a stronger Bz southward component than inE1, as well as a strong negative By component. The solar wind speed is also relatively high, increasing from440 to 530 km/s during the GP. The decrease of the AL index in E2 is almost three times higher than in E1.

The upper thermospheric winds have similar variations in E1 and E2. The eastward zonal winds decreasein the premidnight sector before the onsets. The meridional winds are weak in the premidnight sector, andthey start increasing in the southward direction prior to the onset. During the EP, both zonal and meridionalwinds increase in the eastward and southward directions respectively. The meridional winds continue toincrease after the zonal winds reach their maximum. The winds weaken during the RP.

Although there is a second onset in E1, it has no significant effect on the horizontal winds. The reasons canbe two facts: first, the second substorm in E1 is the weakest among the substorms we have investigated, and

CAI ET AL. 3713

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 2. Background conditions and overviews of the FPI measurements for Event 3 on 1 December 2012 (left panels)and Event 4 on 23 October 2016 (right panels) in the same format as Figure 1.

second, the onset region is located at higher latitudes, far away from the FPI site (see the equivalent currentmap in next section).

The horizontal winds in E2 are stronger than in E1 after the EP. The difference suggest that more magne-tospheric energy has been converted into the kinetic energy in neutrals in E2, due to the higher substormenergy dissipation as indicated in the auroral indices.

In addition to the horizontal disturbances, the vertical winds have wave-like structures in both events,varying between −25 to 25 m/s. The wave-like structures appear at the late GP when the auroral intensityincreases, and last during the expansion and RPs. The period of the structures in E1 is shorter than in E2.

Figure 2 shows the background conditions and the FPI winds in E3 (left) and in E4 (right) for the twopremidnight events. The substorm onset in E3 occurred at 2030 UT (2236 MLT) in the premidnight sector,and the two onsets in E4 at 1730 and 1830 UT, respectively, in the evening sector. Both events are moderatesubstorms with the maximal AL index less than −450 nT under the IMF Bz− and By+ conditions. The solarwind speed in E3 is slower than in E4. The Newell coupling function in E3 stays around 5,000 with littlefluctuation, suggesting a continued and stable solar wind energy input. On the contrary, the Newell couplingfunction in E4 varies between 1,000 and 8,000, due to the fluctuations in the IMF Bz and By components. Forboth events, the AL index has clear negative values during the GP. However, the IL index is much weakerthan the AL index, suggesting that the local geomagnetic activity is low, but it may be more active away fromthe Scandinavian region.

In events E1 and E3 and in the second substorm of E4, the IL index shows clearly larger absolute valuesthan the AL index. In those events, the electrojet current maxima locate at latitudes higher than the AEindex station Abisko (to be seen in section 4), and thus, the decreases in the horizontal component of thegeomagnetic field was captured by the IL index but not by the AL index.

In E3, the zonal wind flows initially in the eastward direction, and it accelerates in the westward directionfrom the beginning of the GP. Similarly, the zonal wind in E4 accelerates in the westward direction after

CAI ET AL. 3714

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 3. Event 1 on 17 December 2011. Panels from top to bottom are east-west equivalent current (JeqE ) at 19.5◦E

glon (positive: eastward electrojet, and negative: westward electrojet), meridional keogram from the green channel ofthe digital ASC at Tromsø(TRO-DC-G), FPI (630.0 nm) zonal wind uE, and 20-min zonal acceleration rate aE, FPI(630.0 nm) meridional wind uN, and 20-min meridional acceleration rate aN (sixth panel). Substorm phases areseparated by the vertical lines and marked by the colored horizontal bars on the top, in the same format as Figure 1. Inthe first panel, the translucent area is the region out of the view of the ASC at Tromsø. The horizontal dotted lines inthe first and second panels indicate the latitude of Tromsø. In the fourth and sixth panels, the red horizontal linesindicate the acceleration rates calculated in the selected time intervals as shown in Tables 2 and 3. ASC = all-skycamera. FPI = Fabry-Perot interferometer; GP = growth phase; EP = expansion phase; RP = recovery phase.

the start of the GP. The strong westward component in both events during the GP is different from the twoprevious events E1 and E2, suggesting that the wind distribution depends highly on the MLT during the GP.

During the EP, the zonal and meridional winds have similar behaviour in event E3 as in events E1 and E2.Event E4 is different. The details are discussed in the next section.

The vertical wind in event E3 has wave-like structures, which start appearing close to the maximum of thewestward wind and continue throughout the substorm at least to the end of the RP. The duration of thewave-like structures is about 5 hr. For event E4, there is indication of wave-like structures in vertical wind ina shorter time interval from EP1 to EP2. During other times of event E4, the data are noisy with large error

CAI ET AL. 3715

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 4. Examples of estimations of wind accelerations in event E1. The left panel is for the zonal wind (positivetoward east) in the growth phase during 2027–2151 UT and the right panel for the meridional wind (positive towardnorth) in the expansion phase 1 and recovery phase 1 during 2221–2327 UT. The FPI (630.0 nm) measurements aremarked by the red dots, the linear fit by a red dashed line, and the HWM-14 results by the blue dotted lines. Theregression coefficient (acceleration rate) with uncertainties from the FPI measurements and the HWM14 model resultsare shown on the top of each panel. FPI = Fabry-Perot interferometer.

bars. The wave-like structures seen in all four events may be related to atmospheric gravity waves (AGWs),which are discussed more in section 5.

4. Acceleration of Upper Thermospheric Winds During Substorms4.1. Event 1 on 17 December 2011Event 1 (E1) on 17 December 2011 is shown in Figure 3, where the panels from the top to the bottom areeast-west equivalent currents, the meridional keogram from the green channel of the digital ASC, the zonalwind, the zonal acceleration rate, the meridional wind, and the meridional acceleration rate, respectively.The acceleration rates are calculated within a 20-min time window centered at each data point.

E1 is the weakest event in terms of the IL index and the Newell coupling function. During the GP, Tromsø islocated close to the magnetic midnight, and a weak WEJ can be seen to migrate slowly in the equatorwarddirection. ASC data and the keogram show multiple arcs drifting equatorward and the most equatorward arcis always at higher latitudes than Tromsø. During substorm onset at 2222 UT, the arcs brighten, and at thesame time the WEJ increases suddenly and expands poleward and equatorward. The poleward expansionstops at 2251 UT, which we take as the end of the EP of the first substorm. The maximum of the WEJ takesplace at earlier time of 2244 UT. Hence, the time period of EP1 identified in this event is 10 min longerthan that identified by using the AL or IL indices. The center of the WEJ is located poleward of Tromsøand the equatorward boundary passes over Tromsø during EP1. A second onset appears during the RP, andthe center of the onset is at higher latitudes than the first onset. Although the WEJ expands poleward andequatorward during EP2, Tromsø is located outside of the WEJ region.

The variation of the zonal wind in E1 is associated with the intensification of the WEJ and auroral activity.Initially, the zonal wind is directed eastward with a speed of 120 m/s at 2030 UT. The eastward direction of thewind is consistent with the diurnal tide in the premidnight sector due to solar-extreme ultraviolet-inducedpressure gradient (Kohl & King, 1967). During the GP, the eastward wind decreases gradually between 2030and 2151 UT, when the auroral arcs drifts equatorward. The eastward acceleration rates during this periodare mostly negative, but show large error bars in the 20-min calculation. The eastward wind stops decreasingand keeps around zero during the late GP, which is coincident with the weakly intensifying WEJ close toTromsø. In the beginning of EP1, the zonal wind is suddenly accelerated eastward with acceleration ratesup to 0.11 m/s2 at 2238 UT. This happens at the same time as the WEJ intensifies. At the end of the EP1,the zonal wind has its maximum value of 140 m/s in the eastward direction. After EP1, the eastward winddecreases gradually with several small enhancements. One enhancement takes place in association withthe second enhancement of the WEJ during EP2. The wind acceleration shows quasi-periodic variationswith a period of about 30 min, which can be seen already in the original wind velocity data. These kind offluctuations within timescales of 20–40 min are typical of AGWs (e.g., Hickey, 2011; Innis & Conde, 2002).

The meridional wind in E1 is weak in the southward direction during the early phase of the GP. The merid-ional wind increases gradually in the southward direction during the late GP, EP1, and RP1. The maximum

CAI ET AL. 3716

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Table 2Average Eastward Acceleration Rates of the FPI (aFPI

E ) and the HWM-14 (aHWME ) Winds During Selected Time Periods, the

Ratio of Acceleration Rates, and the Corresponding Substorm Phase

Event UT (MLT) aFPIE aHWM

E aFPIE ∕aHWM

E Phase

E1 2028–2151 (2258–0021) −0.0261 ± 0.0066 −0.0051 ± 0.0005 5.1 GP2221–2251 (0051–0121) +0.1000 ± 0.0147 −0.0026 ± 0.0003 −38.5 EP12251–2327 (0121–0157) −0.0307 ± 0.0076 −0.0015 ± 0.0003 20.5 RP1

E2 2222–2346 (0052–0216) +0.0507 ± 0.0015 −0.0003 ± 0.0006 169 EP+RPE3 1800–1907 (2030–2137) −0.0437 ± 0.0064 −0.0022 ± 0.0012 19.9 GP

2028–2107 (2258–2337) +0.0400 ± 0.0067 −0.0058 ± 0.0002 −6.9 EPE4 1651–1818 (1921–2048) −0.0189 ± 0.0038 −0.0011 ± 0.0019 17.2 GP+EP1

1818–1939 (2104–2309) 0.0297 ± 0.0033 −0.0061 ± 0.0009 −4.9 EP2+RP2

Note. FPI = Fabry-Perot interferometer; EP = expansion phase; RP = recovery phase.

of the southward wind is 150 m/s at 2309 UT. After that time, the southward wind decreases gradually. Thefluctuations in the meridional wind are also seen from the meridional acceleration rate as the wave-likestructures with periods of 20–40 min.

The 20-min window used in the calculation of the acceleration rates emphasizes short-period variations inacceleration as Figure 3 shows. However, we are interested in studying also the longer time variations, forexample, over the duration of the continuous wind acceleration in the same direction. Therefore, we haveused an alternative approach of selecting an adjusted time window for estimating the acceleration ratesduring substorm phases. In event E1, the zonal wind acceleration is determined separately by this methodfor the GP, EP1, and RP1. Since the meridional wind typically shows no or very small acceleration duringthe GP, the calculation is carried out for the expansion and RPs only.

Figure 4 left panel shows the measured zonal wind with error bars in red color between 2028 and 2151 UTduring the GP. The time period is selected because the zonal wind show clear westward acceleration. Thewestward acceleration starts at the beginning of the GP and ends at 2151 UT. After 2151 UT, the weak WEJshows intensification and the keogram shows that aurora brighten in the north. These can be signaturesof a pseudobreakup (Aikio et al., 1999). The blue dotted line shows the quiet time zonal wind values givenby the HWM-14 model. A linear fit using the method described in section 2.2 has been used to get the FPIacceleration rate and its error. The HWM acceleration rate is calculated by averaging the time differentials ofthe 1-min wind velocities in the selected time interval with its standard deviation as the error. The estimatedeastward acceleration rate is −0.0261 ± 0.0066 m/s2, which is 5.1 times higher than the acceleration thatwe get from the zonal wind given by the HWM-14 model. The results of the zonal acceleration are gatheredin Table 2.

Figure 4 right panel shows the meridional wind between 2221 and 2327 UT during the EP1 and RP1. Thislonger window shows that there is a linear trend in the meridional wind velocity, in addition to the shorterperiod oscillations in velocity (likely to be AGWs). The southward wind velocity increases during the studiedinterval, and the acceleration rate is 21.3 times higher than the acceleration by the HWM-14 model. Theresults of the meridional acceleration are gathered in Table 3.

To investigate the role of the local ion drag in the wind acceleration, we show the EISCAT CP2 measure-ments, the calculated e-folding time, and the ion drag rates during E1 in Figure 5, where panels from top

Table 3Average Northward Acceleration Rates of the FPI (aFPI

N ) and the HWM-14 (aHWMN ) Winds in the Same Format as Table 2

Event UT (MLT) aFPIN aHWM

N aFPIN ∕aHWM

N Phase

E1 2221–2327 (0051–0257) −0.0277 ± 0.0035 −0.0013 ± 0.0001 21.3 EP1+RP1E2 2242–0010 (0122–0240) −0.0200 ± 0.0017 −0.0024 ± 0.0004 8.3 EP+RPE3 2028–2307 (2258–0137) −0.0129 ± 0.0007 −0.0012 ± 0.0004 10.75 EP+RPE4 1834–1939 (2104–2209) −0.0277 ± 0.0045 −0.0042 ± 0.0009 6.6 EP2+RP2

Note. FPI = Fabry-Perot interferometer; EP = expansion phase; RP = recovery phase.

CAI ET AL. 3717

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 5. European Incoherent Scatter Scientific Association measurements during Event 1 on 17 December 2011.Panels from top to bottom are electron density Ne, ion temperature Ti, ion velocity with the geographic north (black)and east (red) components, e-folding time at altitudes between 200 and 300 km, zonal ion drag rate (black) andacceleration rate of the Fabry-Perot interferometer wind (red), meridional ion drag (black), and acceleration rate (red).GP = growth phase; EP = expansion phase; RP = recovery phase.

to bottom are electron density, ion temperature, the ion velocity at 250 km, the e-folding time, the eastwardacceleration rate from the FPI (630.0 nm) calculated at 20-min resolution (red) and the ion drag rate inequation (1) (black), and the northward acceleration in the bottom panel in the same format as the in thepanel above. The electron density is lower than 2 × 1010 m−3 in both the E and F regions during the GP. Atthat time, the auroral oval is located latitudes north of Tromsø. The densities increase suddenly during EP1,in association with the intensification and expansion of aurora from the north to the south over Tromsø.The density in the F region remains high with the lower boundary shifting from low to high altitudes afterEP1. The increase of the ion temperature occurs with the electron density.

The values of the ion velocity (third panel) during the GP are mostly smaller than 300 m/s with large uncer-tainties. The uncertainties come from the low signal-to-noise ratio due to the extremely low electron density.The ion velocity increases in the eastward and northward directions during the early phase of the EP1. Themaxima of the eastward and northward ion speeds are 500 m/s at 2235 UT and 200 m/s at 2230 UT, respec-tively. The ion velocity decreases in both directions during the late phase of the EP1 and thereafter. There isno significant change of the ion velocity during the EP2, which has onset at higher latitudes than Tromsø.

The e-folding time in the fouth panel is calculated based on equation (2), which has shown to be inverselyproportional to the electron density. During the GP, the e-folding times in both E and F regions are higherthan 10 hr, mainly due to the weak electron density. The values decrease to 1–2 hr in the F region during theEP1 and increase gradually after the EP1. The altitude profiles have similar but anticorrelated distribution

CAI ET AL. 3718

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 6. Simultaneous measurements and calculation for Event 2 on 17 October 2012 in the same format as Figure 3.

as the electron density. The smaller e-folding times during EP1 indicates that the ion drag force becomessignificant in both zonal and meridional directions. The e-folding time values are long, but they refer tothe time it takes the neutrals to catch up with the ion speed. However, even during shorter time scales theneutral velocity can change significantly compared to quiet time neutral speeds, which are typically small.

The zonal FPI wind acceleration rates together with the estimated zonal ion drag rates are shown in thefifth panel. The ion drag rates are very small during the GP, but increase during the EP1. The eastward iondrag reaches a value up to 0.07 m/s2, which is 20% smaller than the corresponding maximum in the FPIacceleration rate. During the RP1, the zonal ion drag becomes very small.

The meridional FPI wind acceleration and estimated ion drag rates are shown in the sixth panel. The localion drag has little effect on the wind acceleration during the GP and after the EP1. During the EP1, themeridional ion drag force is small and in the opposite direction than the meridional acceleration observedby the FPI. Considering that the WEJ and auroral oval expand equatorward over Tromsø during the EP1, it ispossible that the Joule heating induced pressure gradient normal to the oval boundary causes the meridionalacceleration. We will discuss the role of the Joule heating in section 5.

CAI ET AL. 3719

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 7. Simultaneous measurements and calculation for Event 3 on 1 December 2012 in the same format as Figure 3.

4.2. Event 2 on 17 October 2012Figure 6 shows Event 2 on 17 October 2012 in the same format as Figure 3. During the GP, the equivalentcurrents are weak before 2220 UT, and an faint auroral arc stays overhead of Tromsø. After 2221 UT, smallintensification of the WEJ appear close to Tromsø and mostly to the south, corresponding to the breakupand intensification of the pre-existing arc. However, the breakup arc has no significant poleward expansionuntil 2244 UT. Hence, the onset of the substorm is identified at 2244 UT, when the WEJ and the discrete arcstarts to expand simultaneously poleward and equatorward. During the EP, the WEJ has several intensifica-tions, and expands gradually poleward and equatorward. The intensifications correspond to the intensifieddiscrete aurora seen from the ASC images and keogram. Although the WEJ intensifies several times after2330 UT, the poleward and equatorward boundaries of the WEJ drifts southward and northward respectively,indicating a contraction of the auroral oval. Hence, the start of the RP is marked at 2230 UT.

In E2, the zonal wind stays around zero during most time of the GP. Continuous eastward acceleration takesplace during the late GP, EP, and early RP between 2222 and 2346 UT. During this time interval, the eastwardacceleration rate (see Table 2) is +0.0507 ± 0.0015 m/s2, 169 times higher than the HWM acceleration.

CAI ET AL. 3720

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 8. Simultaneous measurements and calculation for Event 3 on 23 October 2016 in the same format as Figure 3.

The large ratio indicates that strong ionospheric forcing during the EP affects the wind acceleration. Theacceleration is likely to be associated with the intensification and expansion of the WEJ, because the iondrag effect is in the eastward direction within the WEJ, like in event E1. The maximum of the eastward windof 210 m/s is reached at 2345 UT in the early phase of RP. After that, the eastward wind decreases gradually.

The meridional wind is disturbed during the late GP, in association with the intensification of the arc. Themeridional wind increases in the southward direction during the late GP, EP and RP until 0010 UT. Duringthat period, Tromsø is always located on the equatorward part of the WEJ. The meridional acceleration rateduring EP and RP is −0.0200 ± 0.0017 m/s2, 8.3 times higher than the HWM wind acceleration (Table 3).

Both the zonal and meridional wind accelerations show indications of AGWs with periods of several tens ofminutes starting in the end of the GP, during the EP and in the RP.

CAI ET AL. 3721

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 9. European Incoherent Scatter Scientific Association measurements during Event 4 on 23 October 2016, in thesame format as Figure 5.

4.3. Event 3 on 1 December 2012Event 3 is shown in Figure 7. The auroral images are contaminated due to the moonlight, so that the keogramhas relatively high background values. However, the main auroral structures can be identified. During theGP, an intensification of the eastward electrojet (EEJ) appears with a latitudinal width of 2◦ around 70◦N.The EEJ drifts equatorward from 1840 to 1950 UT, corresponding to an equatorward drifting arc seen inthe keogram. The arc appears at 72◦ geoLat poleward of Tromsø in the beginning and drifts to equatorwardside of Tromsø after 1830 UT. The arc has several intensifications after 1930 UT, however, without signifi-cant poleward expansion. The substorm onset is seen at 2028 UT. The arc suddenly intensifies and expandspoleward and equatorward. Correspondingly, the WEJ intensifies and expands poleward and equatorward.The WEJ is structured with enhanced EEJ on both poleward and equatorward sides, which is a typical sig-nature of westward traveling surge (WTS), which passes over Tromsø as shown by the all-sky images. Thepoleward boundary expands northward until 2115 UT, which is marked as the end of the EP. During the RP,the WEJ weakens gradually.

The zonal wind is accelerated westward during the GP starting from 1720 UT, so that the wind directionchanges from eastward to westward. The westward acceleration rate increases, when the arc drifts close toTromsø. The maximum value of 0.192 m/s2 appears at 1930 UT and after that a large peak in the oppositedirection (eastward) is observed, at the time when the auroral arc crosses the zenith of Tromsø. This indicatesthat the zonal wind acceleration is also affected by the local ion drift due to the auroral arc electrodynamics(see, e.g., Aikio et al., 2002; Conde et al., 2001), and strong meridional gradient of the zonal wind may exist inthe vicinity of the arc (within a latitudinal range of 1◦). This spatial scale is close to the observation presented

CAI ET AL. 3722

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Figure 10. A schematic figure based on this study, showing the typical responses of the upper thermospheric windsduring the growth phase (a) and the expansion phase (b) of a substorm, which are superimposed on ionosphericelectrojet patterns. Typical plasma flows and the HD region are also shown in the two panels. Panel b shows thesubstorm buldge during the expansion phase. WEJ = westward electrojet; EEJ = eastward electrojet.

by Oyama et al. (2017), who showed that clear fluctuations in lower thermospheric wind were confinedwithin a region from the edge of an arc, which had a width between 53 and 203 km.

During the EP, the zonal wind is accelerated eastward within the intensified WEJ and the value is 0.0400 ±0.0067 m/s2 (Table 2), which is slightly smaller than during the EP in event E2.

The meridional wind is disturbed in the vicinity of the intensified arc during the GP, supporting the inferenceof small-scale wind near the arc. Generally, the meridional wind is accelerated in the southward directionduring the EP and RP, with a acceleration rate of −0.0129 ± 0.007 (Table 3). In E3, a clear signature of AGWsis only observed in the meridional wind during the RP.

4.4. Event 4 on 23 October 2016In E4 on 23 October 2016 (Figure 8), intensified EEJ is seen poleward of Tromsø before the GP in the eveningsector. The EEJ expands equatorward slowly during the GP, corresponding to the expansion of the oval.During EP1, intensified WEJ and EEJ are observed on the poleward and equatorward sides of Tromsø respec-tively, indicating that the substorm current wedge mainly flows poleward of Tromsø. The all-sky data (notshown) shows a WTS. During EP2, the WEJ within the substorm current wedge expands equatorward tobecome overhead of Tromsø.

The zonal wind in E4 has westward acceleration within the EEJ, and hence, the acceleration continues untilthe end of the EP1. The acceleration rate is shown in Table 2, which is 17.2 times higher than the HWMwind acceleration. After RP1, the zonal wind is accelerated in the eastward direction with fluctuations ofseveral tens of minutes. Large-amplitude fluctuations in the zonal acceleration rate, interpreted as AGWs,are seen from the end of the GP to the end of RP in Figure 8.

The meridional wind in E4 stays weak until the end of the EP2. The wind is accelerated in the southwarddirection during RP2, suggesting that the driving forces are not balanced during the RP2.

During E4, EISCAT measurements are shown in Figure 9. The electron densities are structured in the Fregion due to passing of the WTS during the EP1 and EP2, and strong E region increases in electron densityare seen. In contrast, the ion temperature is less structured. However, around the altitude of 200 km there isa Ti enhancement during the whole period shown, and this is unfortunately a problem in standard EISCATdata analysis due to an assumption of wrong ion mass (see, e.g., Blelly et al., 2010) However, this does notaffect our data interpretation. The ions flow mainly westward with a speed of about 550 m/s during the GPand EP1, consistent with the enhanced EEJ.

The e-folding time reaches a minimum at the F peak altitudes during the late growth, expansion, and earlyRPs with values of 2–4 hr. During the late GP and early EP1, the estimated zonal ion drag is comparable withthe measured acceleration of the eastward wind and the direction of acceleration is westward. During RP1,the zonal ion drag turns eastward but has very small values. However, the direction is in agreement with thelonger period estimate by the FPI from Table 2. The meridional ion drag from EISCAT data is very small.

CAI ET AL. 3723

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

5. DiscussionWe have investigated four substorm events and shown that the upper thermospheric winds have distinctresponses during substorm GP, EP, and RP. The dynamical change of the thermospheric winds were ana-lyzed by calculating the wind acceleration rates. The wind acceleration responses are associated with theevolution of the substorm electrojets and auroral activity, which is illustrated by a schematic diagram inFigure 10.

Statistical studies have shown that both the EEJ and the WEJ originate on the dayside and terminate on thenightside, and they overlap in the 18–24 MLT region (Gjerloev et al., 2003; Kamide & Kokubun, 1996). Theboundary to separate the two electrojets is called the Harang discontinuity (HD). Although the ionosphericelectrojets flow in the lower E region, the electrojets and the upper thermospheric wind responses can berelated, because both are affected by the ionospheric electric field.

During the GP (panel a in Figure 10), the zonal winds in the evening sector tend to be accelerated in thewestward direction within the EEJ, indicating strong ion drag effect. (During the GP, we do not have mea-surements from the WEJ region in the studied four events.) Due to acceleration, the neutral wind reversesfrom the initial eastward direction to the westward direction, and the westward speed increases within theEEJ when the midnight sector is approached. The westward acceleration is also found equatorward of theouter boundary of the EEJ, as seen in event E1. The acceleration is obviously stronger than the pressure gra-dient force due to solar radiation heating, which drives the diurnal tide. Equatorward of the oval, the zonalwind did not reverse in event E1, but the eastward velocity was decreased. Near the HD, the westward windreaches its maximum, and the eastward acceleration takes place poleward of HD where the WEJ appears.As a result, the westward wind decreases within the WEJ.

The westward wind and westward acceleration observed by the FPI on the dusk side during the GP maybe linked to the large-scale ionospheric dynamics, since the EEJ and WEJ arise with the large-scale iono-spheric convection flowing along the auroral oval. In a global view, the upper thermospheric winds tendto mimic the plasma convection and form a strong anticyclonic vortex on the duskside during southwardIMF conditions (e.g., Förster et al., 2008, 2011; Huang et al., 2017). In the statistical study based on CHAMPsatellite measurements by Förster et al. (2011), they found that the outer boundary of the wind vorticityon the duskside generally overlaps with the boundary between the Region 1 (R1) and Region 2 (R2) cur-rents. Along the boundary, the Hall current flows eastward, which is equivalent to the EEJ observed by theground-based magnetometers (Huang et al., 2017). The westward wind in our observations should be a partof the duskside vortex, and the westward acceleration indicates that the vortex is strengthened with con-tinued solar wind energy input during the GP. Our observations are also in agreement with the statisticalstudy by Huang et al. (2017), who showed that the westward winds are anti-correlated with the EEJ alongthe latitudes. The speed of the westward wind rely on the duration of the acceleration. The winds increasetypically in the westward direction by 100–300 m/s within the EEJ during the GP in our observations.

The close relationship between the zonal accelerations and the auroral electrojets indicates that the ion dragis an important force driving the winds during the growth and EPs. The ion drag is the dominant force forthe high-latitude wind circulation and contributes to the thermospheric vorticity (e.g., Lühr et al., 2007;Richmond et al., 2003; Thayer & Killeen, 1991). Using the GITM model to simulate the thermospheric windsduring a substorm, Wang et al. (2017) have shown that the ion drag is a dominant force to produce thewestward winds on the duskside before the substorm onset.

So, our conclusion is that during the GP the zonal accelerations of the wind (blue and red arrows inFigure 10a) are dominated by the ion drag.

During the EP (Figure 10b), sudden eastward and southward accelerations of the winds are typicallyobserved in association with the sudden intensification and expansion of the WEJ. The strong WEJ in themidnight sector is the ionospheric closure of the substorm current wedge (McPherron et al., 1973). Thesubstorm WEJ typically expands poleward, westward, and equatorward during the EP, and the HD moveswestward. Again, the zonal accelerations are consistent with the ion drag (blue and red arrows in Figure 10b).

The joint measurements by the EISCAT radar and the FPI in events E1 and E4 show that the zonal windacceleration is indeed affected by the ion drag in the growth and EPs. In event E4, westward ion drag wascalculated in the EEJ region during the growth and EPs, and in event E1 eastward ion drag was calculated in

CAI ET AL. 3724

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

within the westward substorm current region during the EP. However, the acceleration of the winds dependsnot only on the local ion drag but also on the forces distributed over a larger area. For example, advectionand pressure gradient forces may become important in certain regions as will be discussed below.

The wind speeds were increased by 100–300 m/s within 20–40 min during the EPs. Hence, the local ion dragmay not be sufficient to accelerate the winds, as is also indicated by the EISCAT measurements in eventE1. Moreover, the EISCAT measurements showed that in the meridional direction the ion drag force wasnorthward, opposite to the southward wind acceleration close to the equatorward boundary of the auroraloval during the EP. Since the intensification of the WEJ is associated with substorm energy dissipation andJoule heating (Ahn et al., 1983), the effect of Joule heating can be important for the wind acceleration.Deng et al. (2008) estimate that by taking into account the pressure gradient produced by Joule heating, theresponse time of the thermosphere can decrease to tens of minutes. Simulation by Wang et al. (2017) showsthat after the substorm onset, thermospheric winds at 400-km altitude are strengthened in the southwardand eastward directions in the midnight, which is in agreement with our measurements. They conclude thatthe driving forces are ion drag, Joule heating and auroral heating.

Hence, we suggest that Joule heating within the substorm current wedge generates a pressure gradient atthe equatorward boundary of the current region, which accelerates winds to a southward direction (greenarrows in Figure 10b).

To fully quantify the effect of Joule heating, two-dimensional distributions of the electric field and conduc-tance would be needed. However, such simultaneous measurements are very difficult to obtain. Tsuda et al.(2009) studied the acceleration mechanism of the lower thermospheric wind in the polar cap region. Theyestimated the pressure gradient due to Joule heating under the assumption of uniform conductance, andsuggested that the pressure gradient is the major forcing to accelerate the neutral wind in the E region to anextremely high speed within 1 hr.

During the RP, the winds tend to decrease in magnitude as the substorm electrojets become weaker on thedawnside, as seen in E1 and E2.

On the other hand, the winds can continue to increase in magnitude in the pre-midnight and midnightsectors during the RP, as seen in E3 and E4. In this situation, advection may become important, transferringmomentum of the neutrals from a remote site to Tromsø.

In addition to the wind responses to different substorm phases, we inferred that strong meridional gradientof the zonal wind on a spatial scale of tens of kilometers may exist in the vicinity of the growth-phase movingarc, as shown in E3. The small-scale wind structures have been identified using scanning Doppler imagers(e.g., Anderson et al., 2012c; Conde et al., 2001; Dhadly & Conde, 2017; Kosch et al., 2010; Zou et al., 2018).Conde et al. (2001) found a meridional wind shear associated with an auroral arc in the dusk sector movingwith the arc. Anderson et al. (2012c) suggested that the meridonal gradient of the zonal wind dominates thesmall-scale horizontal wind structures, and the gradient is associated with the gradient in plasma velocity.Kosch et al. (2010) studied one event in the dawn sector, where the F region wind was not affected by theappearance of the arc. These differences may be associated with the auroral arc electrodynamics in differentMLT sectors.

The wave-like structures in the acceleration rates are clear in all four events. The periods vary between 20and 40 min, which is a typical period for mesoscale AGWs (e.g. Nygrén et al., 2015; Oyama et al., 2010).Generation of the AGWs takes place primarily in the lower atmosphere, but magnetospheric energy dissipa-tion may be a source in the upper atmosphere (e.g., Ford et al., 2008). Since the appearance of the wave-likestructures in these events is typically associated with the intensifications of the substorm WEJ, we suggestthat the AGWs are likely to be generated due to substorm energy input. As a result, the source region shouldbe close to the onset location. The AGWs may propagete from the source region to Tromsø.

6. Summary and ConclusionsIn this paper, the disturbances of the upper thermospheric wind during four non–storm time substorms(E1–E4) have been investigated using simultaneous measurements by the FPI (630.0 nm) at Tromsø Norway(corrected geomagnetic latitude: 66.6◦), the collocated EISCAT radar, the IMAGE magnetometer network,and an ASC. In these events, the Tromsø site was typically located on the equatorward part of the auroral

CAI ET AL. 3725

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

oval. Tromsø was in the post-midnight sector at the substorm onsets in events E1 and E2, and in thepremidnight sector in events E3 and E4.

We summarize the main findings as follows:

• During the GP, westward acceleration of the zonal wind is observed in the evening sector within the EEJregion. The acceleration rate was typically 0.02–0.04 m/s2 and a factor of 5–20 higher than the accelerationgiven by the HWM-14 model. We suggest that the westward acceleration of the neutral wind was causedby the ion drag force associated with the large-scale westward plasma convection in the evening sectorauroral oval.

• During the GP, there are typically no significant changes in the meridional winds.• During the EP, the zonal wind had a prompt response to the intensification of the WEJ overhead Tromsø.

The zonal wind was accelerated eastward and the acceleration rate was up to 0.1 m/s2. Since the plasmaconvection can be expected to be in the eastward direction within the WEJ region, the ion drag force is likelyto play an important role in accelerating the wind. The EISCAT measurements support the interpretation.However, additional acceleration may be produced by Joule heating.

• During the expansion and RPs, the meridional wind was typically accelerated to the southward directionwhen the majority of the substorm WEJ current was located on the poleward side of Tromsø. The heatingsources within the WEJ region are likely to generate a pressure gradient in the meridional direction, whichmay accelerate the wind equatorward.

• Periodic oscillations (20–40 min) both in horizontal and vertical winds were observed mainly during theexpansion and RPs. We suggest that the generation and propagation of AGWs during substorms may causeperturbations in the horizontal and vertical winds.

• In addition to the wind responses to substorm phases, small-scale wind disturbances were found in thevicinity of the GP arcs. Those disturbances will be a subject of a separate study.

The results above indicate that the upper thermosphere is highly dynamical during substorms. In additionto the ion drag, pressure gradient caused by Joule heating is likely to play an important role in the windacceleration. To fully understand the wind dynamics during substorms, more observations are required notonly in a single location but also at multiple sites located at different latitudes relative to the WEJ and indifferent MLT sectors.

ReferencesAhn, B.-H., Akasofu, S.-I., & Kamide, Y. (1983). The Joule heat production rate and the particle energy injection rate as a function of the

geomagnetic indices AE and AL. Journal of Geophysical Research, 88(A8), 6275–6287.Aikio, A. T., Cai, L., & Nygrén, T. (2012). Statistical distribution of height-integrated energy exchange rates in the ionosphere. Journal of

Geophysical Research, 117, A10325. https://doi.org/10.1029/2012JA018078Aikio, A., Lakkala, T., Kozlovsky, A., & Williams, P. (2002). Electric fields and currents of stable drifting auroral arcs in the evening sector.

Journal of Geophysical Research, 107(A12), 1424. https://doi.org/10.1029/2001JA009172Aikio, A. T., Sergeev, V. A., Shukhtina, M. A., Vagina, L. I., Angelopoulos, V., & Reeves, G. D. (1999). Characteristics of pseudobreakups

and substorms observed in the ionosphere, at the geosynchronous orbit, and in the midtail. Journal of Geophysical Research, 104(A6),12,263–12,287.

Akasofu, S.-I. (1968). Magnetospheric substorm, Polar and magnetospheric substorms (pp. 212–253). Netherlands: Springer.Amm, O., & Viljanen, A. (1999). Ionospheric disturbance magnetic field continuation from the ground to the ionosphere using spherical

elementary current systems. Earth, Planets and Space, 51(6), 431–440. https://doi.org/10.1186/BF03352247Anderson, C., Conde, M., & McHarg, M. G. (2012a). Neutral thermospheric dynamics observed with two scanning doppler imagers: 1.

Monostatic and bistatic winds. Journal of Geophysical Research, 117, A03304. https://doi.org/10.1029/2011JA017041Anderson, C., Conde, M., & McHarg, M. G. (2012b). Neutral thermospheric dynamics observed with two scanning Doppler imagers: 2.

Vertical winds. Journal of Geophysical Research, 117, A03305. https://doi.org/10.1029/2011JA017157Anderson, C., Conde, M., & McHarg, M. G. (2012c). Neutral thermospheric dynamics observed with two scanning Doppler imagers: 3.

Horizontal wind gradients. Journal of Geophysical Research, 117, A05311. https://doi.org/10.1029/2011JA017471Aruliah, A. L., Farmer, A. D., Rees, D., & Brändström, U. (1996). The seasonal behavior of high-latitude thermospheric winds and ion

velocities observed over one solar cycle. Journal of Geophysical Research, 101(A7), 15,701–15,711.Aruliah, A. L., Griffin, E. M., Aylward, A. D., Ford, E. A. K., Kosch, M. J., Davis, C. J., et al. (2005). First direct evidence of meso-scale

variability on ion-neutral dynamics using co-located tristatic FPIs and EISCAT radar in northern Scandinavia. Annales Geophysicae, 23,147–162.

Blelly, P.-L., Alcayde, D., & van Eyken, A. P. (2010). A new analysis method for determining polar ionosphere and upper atmospherecharacteristics from ESR data: Illustration with IPY period. Journal of Geophysical Research, 115, A09322. https://doi.org/10.1029/2009JA014876

Brekke, A. (2012). Physics of the upper polar atmosphere. Verlag Berlin Heidelberg: Springer Science & Business Media.Cai, L., Aikio, A. T., & Nygrn, T. (2013). Height-dependent energy exchange rates in the high-latitude e region ionosphere. Journal of

Geophysical Research: Space Physics, 118, 7369–7383. https://doi.org/10.1002/2013JA019195Conde, M., Craven, J. D., Immel, T., Hoch, E., Stenbaek-Nielsen, H., Hallinan, T., et al. (2001). Assimilated observations of thermospheric

winds, the aurora, and ionospheric currents over Alaska. Journal of Geophysical Research, 106(A6), 10,493–10,508. https://doi.org/10.1029/2000JA000135

AcknowledgmentsWe acknowledge the EISCATAssociation for the incoherent scatterradar data used in this study. EISCATis an international associationsupported international associationsupported by research organizations inChina (CRIPR), Finland (SA), Japan(ISEE and NIPR), Norway (NFR),Sweden (VR), and the UnitedKingdom (STFC). EISCAT data areavailable at http://www.eiscat.se/madrigal/ website. Quick-looks of theoptical measurements analyzed in thisstudy can be found at http://www.soyama.org/data website. We thankthe institutes who maintain theIMAGE Magnetometer Array: TromsøGeophysical Observatory of UiT theArctic University of Norway (Norway),Finnish Meteorological Institute(Finland), Institute of GeophysicsPolish Academy of Sciences (Poland),GFZ German Research Centre forGeosciences (Germany), GeologicalSurvey of Sweden (Sweden), SwedishInstitute of Space Physics (Sweden),Sodankylä Geophysical Observatory ofthe University of Oulu (Finland), andPolar Geophysical Institute (Russia).IMAGE magnetometer data areavailable at http://space.fmi.fi/image/www/index.php?page=home website.This research has been supported by afellowship for postdoctoral researchfrom Japan Society for the Promotionof Science (JSPS), a Grant-in-Aid forScientific Research (16H06286,16K05569, and 16H02230) and SpecialFunds for Education and Research(Energy Transport Processes inGeospace) from MEXT, Japan, and theresearch grant by Academy of Finland,decision 285474.

CAI ET AL. 3726

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Deng, W., Killeen, T. L., Burns, A. G., Roble, R. G., Slavin, J. A., & Wharton, L. E. (1993). The effects of neutral inertia on ionosphericcurrents in the high-latitude thermosphere following a geomagnetic storm. Journal of Geophysical Research, 98(A5), 7775–7790. https://doi.org/10.1029/92JA02268

Deng, Y., Richmond, A. D., Ridley, A. J., & Liu, H.-L. (2008). Assessment of the non-hydrostatic effect on the upper atmosphere using ageneral circulation model (GCM). Geophysical Research Letters, 35, L01104. https://doi.org/10.1029/2007GL032182

Deng, Y., & Ridley, A. J. (2006). Dependence of neutral winds on convection E-field, solar EUV, and auroral particle precipitation at highlatitudes. Journal of Geophysical Research, 111, A09306. https://doi.org/10.1029/2005JA011368

Dhadly, M., & Conde, M. (2017). Trajectories of thermospheric air parcels owing over Alaska, reconstructed from ground-based windmeasurements. Journal of Geophysical Research: Space Physics, 122, 6635–6651. https://doi.org/10.1002/2017JA024095

Dhadly, M., Emmert, J., Drob, D., Conde, M., Doornbos, E., Shepherd, G., et al. (2017). Seasonal dependence of northern high-latitude upperthermospheric winds: A quiet time climatological study based on ground-based and space-based measurements. Journal of GeophysicalResearch: Space Physics, 122, 2619–2644. https://doi.org/10.1002/2016JA023688

Dhadly, M., Emmert, J. T., Drob, D. P., Conde, M. G., Doornbos, E., Shepherd, G. G., et al. (2018). Seasonal dependence of geomagneticactive-time northern high-latitude upper thermospheric winds. Journal of Geophysical Research: Space Physics, 123, 739–754. https://doi.org/10.1002/2017JA024715

Drob, D. P., Emmert, J. T., Crowley, G., Picone, J. M., Shepherd, G. G., Skinner, W., et al. (2008). An empirical model of the earth's horizontalwind fields: HWM07. Journal of Geophysical Research, 113, A12304. https://doi.org/10.1029/2008JA013668

Drob, D. P., Emmert, J. T., Meriwether, J. W., Makela, J. J., Doornbos, E., Conde, M., et al. (2015). An update to the horizontal wind model(HWM): The quiet time thermosphere. Earth and Space Science, 2, 301–319. https://doi.org/10.1002/2014EA000089

Emmert, J. T., Drob, D. P., Shepherd, G. G., Hernandez, G., Jarvis, M. J., Meriwether, J. W., et al. (2008). DWM07 global empiricalmodel of upper thermospheric storm-induced disturbance winds. Journal of Geophysical Research, 113, A11319. https://doi.org/10.1029/2008JA013541

Ford, E. A. K., Aruliah, A. L., Griffin, E. M., & McWhirter, I. (2008). Statistical analysis of thermospheric gravity waves from Fabry-Perotinterferometer measurements of atomic oxygen. Annales Geophysicae, 26, 29–45. European Geosciences Union.

Förster, M., Haaland, S. E., & Doornbos, E. (2011). Thermospheric vorticity at high geomagnetic latitudes from CHAMP data and its IMFdependence. Annales Geophysicae, 29(1), 181–186.

Förster, M, Rentz, S., Kohler, W., Liu, H., & Haaland, S. E. (2008). IMF dependence of high-latitude thermospheric wind pattern derivedfrom CHAMP crass-track measurements. Annales Geophysicae, 26, 1581. atmospheres, hydrospheres and space sciences.

Gjerloev, J. W., Hoffman, R. A., Tanskanen, E., Friel, M., Frank, L. A., & Sigwarth, J. B. (2003). Auroral electrojet configuration duringsubstorm growth phase. Geophysical Research Letters, 30(18), 1927. https://doi.org/10.1029/2003GL017851

Griffin, E. M., Aruliah, A. L., McWhirter, I., Yiu, H.-C. I., Charalambous, A., & McCrea, I. (2008). Upper thermospheric neutralwind and temperature measurements from an extended spatial field. Annales Geophysicae, 26(9), 2649–2655. https://doi.org/10.5194/angeo-26-2649-2008

Hickey, M. P. (2011). Atmospheric gravity waves and effects in the upper atmosphere associated with tsunamis. The TsunamiThreat-Research and Technology, InTech.

Huang, T., Hermann, L., Hui, W., & Chao, X. (2017). The relationship of highlatitude thermospheric wind with ionospheric horizontalcurrent, as observed by CHAMP satellite. Journal of Geophysical Research: Space Physics, 122, 12,378–12,392. https://doi.org/10.1002/2017JA024614

Innis, J. L., & Conde, M. (2002). Characterization of acousticgravity waves in the upper thermosphere using Dynamics Explorer 2 windand temperature spectrometer (WATS) and neutral atmosphere composition spectrometer (NACS) data. Journal of Geophysical Research,107(A12), SIA 1–1-SIA 1-22. https://doi.org/10.1029/2002JA009370

Kamide, Y., & Kokubun, S. (1996). Two-component auroral electrojet: Importance for substorm studies. Journal of Geophysical Research,101(A6), 13,027–13,046. https://doi.org/10.1029/96JA00142

Kohl, H., & King, J. W. (1967). Atmospheric winds between 100 and 700 km and their effects on the ionosphere. Journal of Atmosphericand Solar-Terrestrial Physics, 29(9), 1045–1062.

Kosch, M. J., Anderson, C., Makarevich, R. A., Carter, B. A., Fiori, R. A. D., Conde, M., et al. (2010). First e region observations of mesoscaleneutral wind interaction with auroral arcs. Journal of Geophysical Research, 115, A02303. https://doi.org/10.1029/2009JA014697

Kosch, M., Cierpka, K., Rietveld, M., Hagfors, T., & Schlegel, K. (2001). High-latitude ground-based observations of the thermosphericion-drag time constant. Geophysical Research Letters, 28(7), 1395–1398.

Kwak, Y.-S., & Richmond, A. D. (2007). An analysis of the momentum forcing in the high-latitude lower thermosphere. Journal ofGeophysical Research, 112, A01306. https://doi.org/10.1029/2006JA011910

Li, H., Wang, C., & Peng, Z. (2013). Solar wind impacts on growth phase duration and substorm intensity: A statistical approach. Journalof Geophysical Research: Space Physics, 118, 4270–4278. https://doi.org/10.1002/jgra.50399

Lilensten, J., & Lathuillere, C. (1995). The meridional thermospheric neutral wind measured by the EISCAT radar. Journal of Geomagnetismand Geoelectricity, 47(9), 911–920. https://doi.org/10.5636/jgg.47.911

Liuzzo, L. R., Ridley, A. J., Perlongo, N. J., Mitchell, E. J., Conde, M., Hampton, D. L., et al. (2015). High-latitude ionospheric drivers andtheir effects on wind patterns in the thermosphere. Journal of Geophysical Research: Space Physics, 120, 715–735. https://doi.org/10.1002/2014JA020553

Lu, G., Richmond, A. D., Lühr, H, & Paxton, L. (2016). High-latitude energy input and its impact on the thermosphere. Journal ofGeophysical Research: Space Physics, 121, 7108–7124. https://doi.org/10.1002/2015JA022294

Lühr, H., Rentz, S., Ritter, P., Liu, H., & Häusler, K. (2007). Average thermospheric wind patterns over the polar regions, as observed bychamp. Annales Geophysicae, 25, 1093–1101.

Mayaud, P. N. (1980). Derivation, meaning, and use of geomagnetic indices. Geophysical Monograph Series (Vol. 22). Washington, DC:American Geophysical Union.

McPherron, R. L. (1970). Growth phase of magnetospheric substorms. Journal of Geophysical Research, 75(28), 5592–5599.McPherron, R. L., Russell, C. T., & Aubry, M. P. (1973). Satellite studies of magnetospheric substorms on August 15, 1968: 9. Phenomeno-

logical model for substorms. Journal of Geophysical Research, 78(16), 3131–3149.Nakamura, Y., Shiokawa, K., Otsuka, Y., Oyama, S., Nozawa, S., Komolmis, T., et al. (2017). Measurement of thermospheric temper-

atures using OMTI Fabry–Perot interferometers with 70-mm Etalon. Earth, Planets and Space, 69(1), 57. https://doi.org/10.1186/s40623-017-0643-1

Newell, P. T., Sotirelis, T., Liou, K., Meng, C.-I., & Rich, F. J. (2007). A nearly universal solar wind-magnetosphere coupling function inferredfrom 10 magnetospheric state variables. Journal of Geophysical Research, 112, A01206. https://doi.org/10.1029/2006JA012015

CAI ET AL. 3727

Journal of Geophysical Research: Space Physics 10.1029/2018JA026241

Nygrén, T, Aikio, AT, Kuula, R., & Voiculescu, M. (2011). Electric fields and neutral winds from monostatic incoherent scatter measure-ments by means of stochastic inversion. Journal of Geophysical Research, 116, A05305. https://doi.org/10.1029/2010JA016347

Nygrén, T., Aikio, A., Voiculescu, M., & Cai, L. (2015). Radar observations of simultaneous traveling ionospheric disturbances andatmospheric gravity waves. Journal of Geophysical Research: Space Physics, 120, 3949–3960. https://doi.org/10.1002/2014JA020794

Nygrén, T, Aikio, AT, Voiculescu, M., & Kuula, R. (2012). Statistical evaluation of electric field and neutral wind results from beam-swingincoherent scatter measurements. Journal of Geophysical Research, 117, A04319. https://doi.org/10.1029/2011JA017307

Østgaard, N., Germany, G., Stadsnes, J., & Vondrak, R. R. (2002). Energy analysis of substorms based on remote sensing techniques, solarwind measurements, and geomagnetic indices. Journal of Geophysical Research, 107(A9), 1233. https://doi.org/10.1029/2001JA002002

Oyama, S.-i., Kubota, K., Morinaga, T., Tsuda, T. T., Kurihara, J., Larsen, M. F., et al. (2017). Simultaneous FPI and TMA measurements ofthe lower thermospheric wind in the vicinity of the poleward expanding aurora after substorm onset. Journal of Geophysical Research:Space Physics, 122, 10,864–10,875. https://doi.org/10.1002/2017JA024613

Oyama, S., Nozawa, S., Maeda, S., Murayama, Y., Fujii, R., & Shinagawa, H. (2003). Field-aligned ion motions in the polar E-F transitionregion: Mean characteristics. Journal of Geophysical Research, 108(A8), 1334. https://doi.org/10.1029/2003JA009830

Oyama, S., Shiokawa, K., Kurihara, J., Tsuda, T. T., Nozawa, S., Ogawa, Y., et al. (2010). Lower-thermospheric wind fluctuations measuredwith an FPI during pulsating aurora at Tromsø, Norway. Annales Geophysicae, 28, 1847. Copernicus GmbH.

Oyama, Shin-ichiro, Shiokawa, K., Miyoshi, Y., Hosokawa, K., Watkins, B. J., Kurihara, J., et al. (2016). Lower thermospheric wind vari-ations in auroral patches during the substorm recovery phase. Journal of Geophysical Research: Space Physics, 121, 3564–3577. https://doi.org/10.1002/2015JA022129

Peymirat, C., Richmond, A. D., & Roble, R. G. (2002). Neutral wind influence on the electrodynamic coupling between the ionosphere andthe magnetosphere. Journal of Geophysical Research, 107(A1), 1006. https://doi.org/10.1029/2001JA900106

Pulkkinen, A., Amm, O., Viljanen, A., & BEAR Working Group (2003). Ionospheric equivalent current distributions determined with themethod of spherical elementary current systems. Journal of Geophysical Research, 108(A2), 1053. https://doi.org/10.1029/2001JA900106

Richmond, A. D., Lathuillere, C., & Vennerstrøm, S. (2003). Winds in the high-latitude lower thermosphere: Dependence on theinterplanetary magnetic field. Journal of Geophysical Research, 108(A2), 1066. https://doi.org/10.1029/2002JA009493

Richmond, A. D., & Matsushita, S. (1975). Thermospheric response to a magnetic substorm. Journal of Geophysical Research, 80(19),2839–2850.

Richmond, A. D., Ridley, E. C., & Roble, R. G. (1992). A thermosphere/ionosphere general circulation model with coupled electrodynamics.Geophysical Research Letters, 19(6), 601–604.

Ridley, A. J., Deng, Y., & Toth, G. (2006). The global ionosphere–thermosphere model. Journal of Atmospheric and Solar-Terrestrial Physics,68(8), 839–864.

Ritter, P., Lühr, H., & Doornbos, E. (2010). Substorm-related thermospheric density and wind disturbances derived from CHAMPobservations. Annales Geophysicae, 28, 1207. Copernicus GmbH.

Rostoker, G., Akasofu, S.-I., Foster, J. C., Greenwald, R. A., Kamide, Y., Kawasaki, K., et al. (1980). Magnetospheric substormsdefinitionand signatures. Journal of Geophysical Research, 85(A4), 1663–1668.

Schunk, R., & Nagy, A. (2009). Ionospheres: Physics, plasma physics, and chemistry. Cambridge: Cambridge University Press.Sheng, C., Deng, Y., Wu, Q., Ridley, A., & Hggstrm, I. (2015). Thermospheric winds around the cusp region. Journal of Geophysical Research:

Space Physics, 120, 1248–1255. https://doi.org/10.1002/2014JA020028Shiokawa, K., Kadota, T., Otsuka, Y., Ogawa, T., Nakamura, T., & Fukao, S. (2003). A two-channel Fabry-Perot interferometer with

thermoelectric-cooled CCD detectors for neutral wind measurement in the upper atmosphere. Earth, Planets and Space, 55(5), 271–275.https://doi.org/10.1186/BF03351759

Tanskanen, E. I. (2009). A comprehensive high-throughput analysis of substorms observed by IMAGE magnetometer network: Years1993–2003 examined. Journal of Geophysical Research, 114, A05204. https://doi.org/10.1029/2008JA013682

Tanskanen, E., Pulkkinen, T. I., Koskinen, H. E. J., & Slavin, J. A. (2002). Substorm energy budget during low and high solar activity: 1997and 1999 compared. Journal of Geophysical Research, 107(A6), 1086. https://doi.org/10.1029/2001JA900153

Thayer, J. P., & Killeen, T. L. (1991). Vorticity and divergence in the high-latitude upper thermosphere. Geophysical Research Letters, 18(4),701–704. https://doi.org/10.1029/91GL00131

Thayer, J. P., & Semeter, J. (2004). The convergence of magnetospheric energy flux in the polar atmosphere. Journal of Atmospheric andSolar-Terrestrial Physics, 66(10), 807–824.

Tsuda, T. T., Nozawa, S., Oyama, S., Motoba, T., Ogawa, Y., Shinagawa, H., et al. (2009). Acceleration mechanism of high-speed neutralwind observed in the polar lower thermosphere. Journal of Geophysical Research, 114, A04322. https://doi.org/10.1029/2008JA013867

Vanhamäki, H., & Amm, O. (2011). Analysis of ionospheric electrodynamic parameters on mesoscales—A review of selected techniquesusing data from ground-based observation networks and satellites. Annales Geophysicae, 29, 467. Copernicus GmbH.

Virtanen, I. I., McKay-Bukowski, D., Vierinen, J., Aikio, A., Fallows, R., & Roininen, L. (2014). Plasma parameter estimation from multi-static, multibeam incoherent scatter data. Journal of Geophysical Research: Space Physics, 119, 10,528–10,543. https://doi.org/10.1002/2014JA020540

Wang, W., Burns, A. G., Wiltberger, M., Solomon, S. C., & Killeen, T. L. (2008). Altitude variations of the horizontal thermospheric windsduring geomagnetic storms. Journal of Geophysical Research, 113, A02301. https://doi.org/10.1029/2007JA012374

Wang, H., Zhang, K. D., Wan, X., & Lhr, H. (2017). Universal time variation of high-latitude thermospheric disturbance wind in responseto a substorm. Journal of Geophysical Research: Space Physics, 122, 4638–4653. https://doi.org/10.1002/2016JA023630

Witasse, O., Lilensten, J., Lathuillere, C., & Pibaret, B. (1998). Meridional thermospheric neutral wind at high latitude over a full solarcycle. Annales Geophysicae, 16, 1400–1409. Springer.

Wu, Q., Emery, B. A., Shepherd, S. G., Ruohoniemi, J. M., Frissell, N. A, & Semeter, J. (2015). High-latitude thermospheric wind observa-tions and simulations with SuperDARN data driven NCAR TIEGCM during the december 2006 magnetic storm. Journal of GeophysicalResearch: Space Physics, 120, 6021–6028. https://doi.org/10.1002/2015JA021026

Xiong, C., Lhr, H., & Fejer, B. G. (2015). Global features of the disturbance winds during storm time deduced from champ observations.Journal of Geophysical Research: Space Physics, 120, 5137–5150. https://doi.org/10.1002/2015JA021302

York, D. (1968). Least squares fitting of a straight line with correlated errors. Earth and Planetary Science Letters, 5, 320–324.York, D., Evensen, N. M., Martínez, M. L., & De Basabe Delgado, J. (2004). Unified equations for the slope, intercept, and standard errors

of the best straight line. American Journal of Physics, 72(3), 367–375.Zou, Y., Nishimura, Y., Lyons, L., Conde, M., Varney, R., Angelopoulos, V., & Mende, S. (2018). Mesoscale F region neutral winds associated

with quasi-steady and transient nightside auroral forms. Journal of Geophysical Research: Space Physics, 123, 7968–7984. https://doi.org/10.1029/2018JA025457

CAI ET AL. 3728