Journal of Geophysical Research: Space Physics

Global thermospheric neutral density and wind response to the severe 2003 geomagnetic storms from CHAMP accelerometer data



[1] Measurements of atmospheric density near 410 km from the STAR accelerometer on the CHAMP satellite are used to illustrate the spatial-temporal dependence of the thermospheric response to the severe solar storms occurring during 29 October to 1 November 2003. This interval includes periods of elevated magnetic activity with KP values of 5–9, as well as undisturbed intervals that serve to define quiet time baseline densities. Measurements are available from −87° to +87° latitude during both day and night at local times near 1300 and 0100 hours, respectively. During times of maximum geomagnetic activity for this study, density measurements exhibit enhancements of 200–300%. Northern Hemisphere daytime responses are much larger than in the Southern Hemisphere; the origins of this effect are unknown. Nighttime density disturbances more readily propagate to equatorial latitudes, possibly facilitated by the predominant equatorward flow in both hemispheres due to the diurnal tides driven by in situ EUV heating. The CHAMP density measurements are compared with density predictions from the NRL-MSISe00 empirical density model and demonstrate some model shortcomings. Measurements of cross-track accelerations provide the opportunity to estimate zonal winds from the equator to about ±60° latitude, transitioning to a measure of purely meridional winds at the turning point of the orbit near ±87° latitude. A periodic variation in cross-track winds with an apparent period of 24 hours appears at high latitudes and exhibits similar amplitudes and temporal-latitudinal structures to the empirical HWM-93 wind model when projected into the cross-track direction. This periodicity is due to the displacement of geomagnetic and geographic coordinates. At low latitudes, CHAMP and HWM-93 both yield westward winds of order 100 ms−1 during midday under quiet magnetic conditions; however, during severely disturbed periods the HWM-93 winds generally show a greater westward intensification (to 250 ms−1) than the CHAMP measurements. At night, CHAMP winds are near zero under quiet conditions whereas HWM-93 indicates eastward winds of order 50–100 ms−1. Under disturbed conditions the CHAMP winds shift to westward values of order 200 to 250 ms−1, while HMW-93 values do not exceed about 50 ms−1 in the westward direction. The physical origins of the observed effects are difficult to isolate, and unequivocal interpretation will require sophisticated numerical modeling taking into account self-consistent interactions between the neutral winds, drifts, and ionization densities.

1. Introduction

[2] During the period spanning from late October to early November of 2003, an active region on the Sun (Active Region 486) produced flares and coronal mass ejections (CMEs) of intensities rarely seen before. Two major geomagnetic storms were caused by CMEs arriving at the Earth on 29 and 30 October. At 0611 UT on 29 October, the CME associated with an X17 flare impacted the Earth's magnetic field. The geomagnetic aP index reached 400 for a short period of time after the impact. A northward interplanetary magnetic field (IMF) Bz component was responsible for a slight lull in geomagnetic activity from 0900 to 1800 UT on 29 October. Near the end of this period, a strong southward IMF Bz component was observed, finishing out the day with severe geomagnetic activity. The second severe storm occurred at 1600 UT on 30 October when the CME associated with an X10 flare impacted the Earth. During this storm, a southward IMF Bz component assured a severe level of geomagnetic activity, during which time the planetary aP index remained saturated at 400 for 6 hours. Early on 31 October, activity began to decline, facilitated by a northward IMF Bz component. Both CMEs had a transit time of around 19 hours, making them among the fastest recorded. In addition, this time span exhibited the highest geomagnetic disturbances of the present solar cycle.

[3] The relationship between geomagnetic disturbances and thermospheric composition, density, and winds has been studied in depth for decades [Matuura, 1972; Prölss, 1980; Breig, 1987; Crowley, 1991]. The first discoveries of this connection resulted from discrepancies between predicted and observed satellite ephemerides during periods of geomagnetic activity. Early models for predicting thermospheric density include the Jacchia total density models [Jacchia, 1970] derived from satellite drag data. Zonal winds were first studied by analyzing their average effect on satellite orbits which was termed “superrotation” of the atmosphere [King-Hele, 1964] and was estimated using measurements of satellite inclination. More recent studies of thermospheric conditions have utilized ground-based incoherent scatter radar measurements of temperature and in situ satellite measurements of total density and composition. The consolidation of these data is represented statistically in the MSIS series of models [i.e., Hedin et al., 1977; Hedin, 1983, 1987], and by the recent update to MSISe90 [Hedin, 1991], NRL-MSISe00 [Picone et al., 2002]. In a similar fashion, a global empirical model of thermosphere winds (HWM-93) was developed [Hedin et al., 1996] that addresses a range of magnetic and solar conditions.

[4] The morphology of a geomagnetic storm is reviewed in detail by Prölss [1980] and is related to changes in composition, electric fields, neutral flow, and dynamo effects all caused by an increase in energy input to the high-latitude thermosphere-ionosphere system. In the present paper, two elements indicative of these upper-atmospheric disturbances are studied, namely in situ measurements of thermospheric total density and wind speed derived from accelerometer measurements. Accelerometer measurements have been used in the recent past to estimate both total density [Berger and Barlier, 1981; Forbes et al., 1996; Bruinsma et al., 2004] and wind speed [Marcos and Forbes, 1985; Forbes et al., 1993] for quiet and active conditions. Total density measurements represent one measure of the global thermospheric response to geomagnetic storms.

[5] The German satellite CHAMP (Challenging Minisatellite Payload) was launched into a circular orbit on 15 July 2000. The project plan includes a 5-year mission duration to study the gravity and magnetic fields of the Earth, with a secondary goal of studying the upper atmosphere [Reigber et al., 2002b]. One of the instruments on CHAMP is a sensitive triaxial accelerometer that is capable of providing estimates of total mass density and cross-track winds. With a near-polar inclination, the satellite provides near-global coverage at an approximate altitude of 410 km within two local time sectors at most latitudes. Figure 1 illustrates the sampling of CHAMP in local time and latitude for the period of 27 October to 2 November 2003, and Figure 2 shows the angle between the cross-track axis and geographic east for this time period. CHAMP's high orbit inclination thus allows for the study of zonal winds near equatorial regions and meridional winds near the turning point of the orbit (near ±87°) using the cross-track accelerometer axis. The combination of these capabilities provides an unprecedented opportunity to view the global density and wind responses to the unusually severe solar and geomagnetic disturbances discussed previously. The primary objectives of this paper are to describe the global density and wind responses to these storms and to provide some interpretation. In addition, we provide comparisons with current empirical models of density and winds in order to identify their capabilities and shortcomings.

Figure 1.

Latitude versus local time for CHAMP during 27 October to 2 November 2003.

Figure 2.

The angle between the cross-track axis and geographic east for CHAMP during 27 October to 2 November 2003.

[6] In the following section, we discuss the density retrieval procedure, necessary force models, and semiempirical models used for comparison. In section 3.1, we analyze density response to elevated levels of geomagnetic activity at all latitudes and examine departures from the NRL-MSISe00 density model. In section 3.2, we analyze cross-track winds with an emphasis on low-latitude response to elevated levels of geomagnetic activity. Finally, a summary of our results and conclusions are given in section 4.

2. Data and Models

[7] CHAMP data was provided for our use by the CHAMP Information System and Data Center (ISDC). Acceleration, attitude, and orbit ephemeris data files were used in the calculations pertaining to this study. The acceleration and attitude data is provided on a 10-s interval, which is processed from the original 1 Hz data to remove spikes caused by spacecraft maneuvers. Both the in-track axis, used for density calculations, and the cross-track axis, used for wind calculations, are thought to be accurate to 3 × 10−9 ms−2 [Reigber et al., 2002a]. The radial axis is not used in these studies because it is less sensitive by a factor of ten and has had repeated trouble since launch.

[8] In order to have the most accurate density calculations, a bias and scale factor must be applied to the raw acceleration data. For this study, a least squares method was used to estimate the bias factors for the in-track accelerometer axes. With this method, the bias factors are estimated by using the STAR accelerometer measurements as an orbit determination force model while letting the estimated orbit converge on the Rapid Science Orbit ephemeris provided by the CHAMP ISDC. The in-track a priori bias factor used for this time period was −2.925 × 10−6 ms−2. The estimated values had a mean of −2.92534 × 10−6 ms−2 and rms of 1.71 × 10−8 ms−2. We considered biases to be inaccurate when they differed from the mean by more than 4.5 × 10−8 ms−2. When the numbers lie outside of this range, linear interpolation is used. An error of 1% in the scale factor translates to less than 1% of error in total density, while an error of 1% in the bias factor translates to an error of 2.5% in total density. For the cross-track axis, the bias factor was calculated assuming that on average for this time period, the measured cross-track acceleration agrees with the modeled accelerations which assume a corotating atmosphere. For this time period, the cross-track bias factor was determined to be 1.145978 × 10−7 ms−2. An error of 1% in the scale or bias factor translates to an error of about 5–10 ms−1 in wind speed.

[9] On board the CHAMP satellite is the STAR accelerometer, which measures the sum of all forces on the satellite's surface. This measured quantity is comprised mostly of the force imparted to the satellite by atmospheric drag, with lesser constituents such as solar and Earth radiation pressure also contributing. All other nongravitational forces on the satellite are neglected in this study. Subtracting the modeled effects of radiation pressure, we are left with the acceleration due to atmospheric drag. From here we arrive at a solution for atmospheric density which is a modification of the traditional drag equation taking into account a 13-plate macromodel of the satellite:

equation image

where equation imageDRAG is the acceleration caused by drag, Ai is the plate area, CDi is the coefficient of drag for the plate, m is the satellite mass, equation imagei is the unit plate normal, ρ is the atmospheric density, and equation imagerel is the satellite velocity relative to a corotating atmosphere. Typical in-track magnitudes for this term vary from 2 × 10−7 to 12 × 10−7 ms−2 over the course of one orbit. This technique requires models to account for the unwanted forces that are measured along with drag by the STAR accelerometer. For each of these models, orbit ephemeris files and quaternion files provided by the CHAMP ISDC are used to orient the satellite with respect to inertial space, terrestrial coordinates, and the Sun.

[10] The solar radiation pressure model also employs a 13-plate macromodel of the satellite. When the satellite is sunlit, attitude quaternions are used to calculate an angle between the satellite-Sun vector and the normal vector for each plate that is facing the Sun. The following equation can then be used to sum the entire force of the satellite due to solar radiation [Luthcke et al., 1997]:

equation image

where equation imageSR is the acceleration caused by solar radiation pressure, Ai is the plate area, c is the speed of light, crd,i is the coefficient of diffusive reflectivity, crs,i is the coefficient of specular reflectivity, m is the satellite mass, equation imagei is the unit plate normal, ϕinc,i is the angle of incidence of the Sun with respect to the plate, R is the flux originating from the Sun, and equation image is the unit satellite-Sun vector. The magnitude of solar flux is also multiplied by a ratio to account for shadowing when the satellite is in the umbra or penumbra shadow regions. The modifying ratio is equivalent to the percentage of the sun visible by the satellite. Finding the Sun-Earth vector and the appropriate flux acting on the satellite requires up-to-date JPL solar and planetary ephemerides (version DE-405). Typical in-track magnitudes for this term are on the order of 3 × 10−8 ms−2 during the time period studied in this work.

[11] The albedo calculations require a latitudinally varying model for short-wave radiation (in terms of albedo) and long-wave radiation (in terms of emissivity) coming from the terrestrial sphere. The effect of the Earth is summed by using discrete elements according to Knocke et al. [1988]. The following equation can be used to sum the effect of each Earth element on each satellite plate:

equation image

where equation imageALB is the acceleration caused by Earth radiation pressure, Ai is the plate area, c is the speed of light, crd,i is the coefficient of diffusive reflectivity, crs,i is the coefficient of specular reflectivity, m is the satellite mass, equation imagei is the unit plate normal, ϕinc,ij is the angle of incidence of the source with respect to the plate, Rj is the flux originating from the source, and equation imagej is the unit satellite-source vector. Both the solar radiation pressure model and the albedo/infrared models were adapted from the Geodyn II orbit determination software package (NASA Goddard Space Flight Center). Typical in-track magnitudes for this term are on the order of 5 × 10−10 and 2 × 10−10 ms−2 for albedo and infrared radiation pressure, respectively.

[12] The final unknown in this process is the coefficient of drag for each plate of the CHAMP macromodel. Usually, these coefficients are solved for in terms of a density model using orbit determination software. However, for this application, we desire that the coefficient of drag not have the same bias that the corresponding thermospheric density model possesses. Therefore a physics-based model for approximating the coefficient of drag for a flat plate was used. This method is outlined by Cook [1965] and leads to the following formula:

equation image

where CDi is the coefficient of drag, αi is the accommodation coefficient, Ta is the temperature of the atmosphere, Tw,i is the temperature of the plate, and equation imagei is the angle of incident gas flow with respect to the plate. Further, readmittance of atmospheric molecules is assumed to be mainly diffusive because the temperature of the satellite is fairly cool (assumed to be 273.0 K). For macromodel plates that are normal to the flow of atmospheric gases, this assumption leads to an underestimation of CDi of about 2.5% for every 100.0 K temperature increase in Tw,i. This assumption also leads to the following approximation for the accommodation coefficient that appears in (4):

equation image

where μi is the ratio of mass of the incident gas atom to the mass of the surface atom. The force caused by impacting and readmitting atoms and molecules can be represented by the drag equation (1) in terms of the coefficient of drag, CDi. In this study, the coefficient of lift is assumed to be small enough to ignore. At its maximum, the force caused by lift is an order of magnitude less than that caused by solar radiation. Estimating neutral density requires modeling surface forces other than drag, estimating instrument bias and scale factors, and calculating the coefficient of drag. After these steps are completed, (1) is solved for density using the in-track axis of the accelerometer denoted by equation imageSTAR · equation image, giving:

equation image

where ρSTAR is the atmospheric density calculated from CHAMP/STAR, and equation image is the unit vector of the in-track axis. Normally, neutral wind velocity is considered to be the largest source of error in this calculation. From calculations using a simulated horizontal wind vector, we see that zonal winds can skew density calculations by 5% for every 100 ms−1. Meridional winds can skew density calculations by 2.5% for every 100 ms−1. Under quiet geomagnetic conditions, the error caused by horizontal winds rarely exceeds 15% near the poles and 8% away from the poles. However, under active conditions, the error can exceed 20% in the close vicinity of the poles.

[13] It is also possible to obtain an estimate of the neutral wind vector component in the direction of the cross-track axis of the STAR accelerometer. This calculation requires the same steps as the density calculations. The following equation is now used as the drag equation:

equation image

where equation imageDRAG is the acceleration caused by drag, Ai is the plate area, CDi is the coefficient of drag for the plate, m is the satellite mass, equation imagei is the unit plate normal, ρSTAR is the atmospheric density calculated from CHAMP/STAR, and equation imagerel is the satellite velocity with respect to a corotating atmosphere. Here \vec{w} has been added as a vector representing wind in relation to an atmosphere that corotates with the Earth. Only the cross-track axis is studied here for the following two reasons. Neutral density and wind speed cannot be separated from each other in calculations using the in-track accelerometer axis, and the radial accelerometer axis does not provide measurements of sufficient accuracy. Thus when solving equation (7) for wind speed, components of equation image in the in-track and radial directions are assumed to be zero. The density term in equation (7) is now the neutral density derived from the in-track accelerometer axis.

[14] The NRL-MSISe00 thermospheric density model [Picone et al., 2002] is used for comparisons to the CHAMP/STAR density calculations. This model is an extension of the MSIS-83, -86, and -e90 versions [Hedin, 1983, 1987, 1991]. All of these models are error weight nonlinear least squares fits of atmospheric composition data measured by mass spectrometers, and neutral temperature profiles measured by incoherent scatter radars, to empirical formulas. The NRL model also includes satellite drag data and calculations of anomalous oxygen in the atmosphere. The required inputs for the model are the satellite position, day of the year, solar local time, F10.7 solar flux from the previous day, mean F10.7 solar flux from 81 days centered on the current day, and a 57-hour history of the aP index.

[15] The HWM-93 thermospheric wind model [Hedin et al., 1996], is used for comparisons to the CHAMP/STAR cross-track wind calculations. The HWM is a error weight nonlinear least squares fit to a truncated set of spherical harmonics. It is primarily based on wind data obtained from the AE-E and DE-2 satellites. Both satellites had highly eccentric orbits. AE-E had an inclination of 20° and only analyzed cross-track drift speeds, making zonal wind studies difficult and high-latitude studies impossible. DE-2 had a near-Polar orbit making it a better candidate for comparison with CHAMP cross-track winds [Killeen and Roble, 1988]. However, the DE-2 mission time span was too short to capture any long-term variability. Inputs to the model include the satellite position, day of the year, solar local time, F10.7 solar flux from the previous day, mean F10.7 solar flux from 81 days centered on the current day, current 3-hour aP index, and daily AP index. Lower-latitude winds are thought to be reproduced well by the model, while the structure of higher-latitude winds is somewhat lost by the simplicity of the spherical harmonic basis function.

3. Results

3.1. Density Response

[16] The daytime and nighttime global density responses at 410 km are illustrated in Figure 3. NRL-MSISe00 model response is also shown, sampled on the orbit of CHAMP, to draw comparisons and reveal any shortcomings of the model. For each case, density response is shown in terms of latitude and time, with geomagnetic aP and northern Polar Cap indices (Danish Meteorological Institute, indicated. Near the equator, local solar time sampling is approximately 1320 and 0120 hours, respectively. When CHAMP is at maximum latitude, local time is approximately 0720 near the North Pole and 1920 near the South Pole. However, local time sampling does not change substantially from the equator until CHAMP is above ±80° latitude. This complexity, caused by the fact that CHAMP is not in a perfectly polar orbit, is illustrated in Figure 1.

Figure 3.

Latitude versus time, daytime response of CHAMP total mass densities at 410 km (top left), the corresponding NRL-MSISe00 model densities sampled along the CHAMP orbit (top right), nighttime response of CHAMP (bottom left), and corresponding NRL-MSISe00 model densities (bottom right). Also shown are the Northern Polar Cap Index (top solid line in each panel), and 3-hourly Ap index (bottom solid line in each panel).

[17] During this period of elevated activity, there are two time intervals in which the aP index becomes saturated at 400, first for 3 hours early on 29 October (day 302) and then again for 6 hours at the end of 31 October (day 304). In between these two intervals, there is an active period of about 9 hours where aP reaches 300. During the first surge of geomagnetic activity, severe elevated levels of density can be seen confined to certain latitudes. On the dayside, elevated density levels of order 300% (the highest of this time period) can be seen in high northern latitudes. However, this density enhancement seems to die out within an hour and a half, indicating that the first spike in geomagnetic activity does not last long enough to have a global effect on density. On the nightside, density enhancements last longer and can even be seen to travel toward the equator with time. During the second spike in which aP reaches 300, density is more noticeably affected. On the dayside, a major density disturbance of amplitude 200–250% spans from 85° latitude to the equator and lasts for upward of 9 hours. Much more activity can be seen in the southern hemisphere as well. Dayside density enhancements exhibit much less dispersion to lower latitudes in comparison to the nightside density disturbances.

[18] Several of the above response characteristics are reminiscent of those revealed in accelerometer measurements near 200 km, interpreted by Forbes et al. [1996] in the context of the thermosphere modeling results of Fuller-Rowell et al. [1996]. For instance the nighttime density disturbances appear to more readily propagate to equatorial latitudes at night, possibly facilitated by the predominant equatorward flow in both hemispheres at nighttime due to the diurnal tides driven by in situ EUV heating [Forbes and Garrett, 1976]. During daytime, the flow is predominantly poleward, inhibiting equatorward extension of the density disturbances in both hemispheres. During both day and night, horizontal advection and subsidence heating may be playing a role in heating the low-latitude thermosphere. It is not understood why the daytime response in the summer hemisphere is less intense than that in the Northern Hemisphere. Although this period is between equinox and solstice, there is probably also a solstice-like net (diurnal mean) meridional flow from the South Hemisphere to North Hemisphere which modulates the above dependence of the density response at lower latitudes on time of day. This could also be caused by NRL-MSISe00 model inadequacy when attempting to normalize total density measurements to a height of 410 km. During this time period perigee of CHAMP is close to the North Pole, which might explain an increased response in the Northern Hemisphere.

[19] The nighttime response can also be viewed in the context of traveling atmospheric disturbances (TADs). If we assume that a nighttime TAD has a velocity of 750 ms−1, successive orbits would observe movement between 32 and 38° latitude, depending on whether the disturbance is southbound or northbound. While there are many disturbances in this time period that have the potential of being a TAD, the large amount of latitudinal displacement between successive CHAMP orbits makes it difficult to identify a TAD with any certainty.

[20] The geomagnetic storm response using the NRL-MSISe00 model is driven by two parameters: a 57-hour history of the aP index and F10.7 solar flux measurements, adjusted to 1 AU. During geomagnetically quiet periods, these two indices do fairly well at representing global density variations with a slight mean difference between modeled values and measured values. This can be seen by comparing prestorm modeled and measured density on 27 and 28 October (day 300 and 301) in Figure 3. However, when this model is applied to storm time conditions, two problems arise: (1) Planetary indices do not contain enough information to empirically emulate small-scale density structure, and (2) the aP index has a saturation point at 400 which precludes the model density from exhibiting the same extrema as the measured densities. Another shortcoming of the NRL-MSISe00 model is that it uses statistically based empirical fits to represent density and hence tends to smear out density response features both spatially and temporally. The effects of these issues can be seen in Figure 3.

[21] During times of elevated magnetic activity, daytime density measurements are higher than modeled density by as much as a factor of 2. While modeled density seems to be symmetric about the equator for day and nighttime density, this is clearly not consistent with measured values, especially on the dayside.

[22] Note that just after 1200 UT on day 301, there is an enhancement in density of order 100% extending between −40 and +60° on the dayside but not on the nightside. Given that this is a period of very quiet geomagnetic activity, we believe that this is the response to a large solar EUV flare that occurred near 1100 UT on day 301, as observed by the SEE instrument on the TIMED spacecraft (T. Woods, private communication, 2003). A recent study of these EUV events can be found in a paper by Tsurutani et al. [2005]. The possibility of an EUV flare response being evident in CHAMP density data on day 301, and again during a major solar EUV flare on day 308, is currently undergoing further investigation and will be reported on separately.

[23] Equatorial density responses for the October through November time series can be seen in Figure 4, for daytime and nighttime orbits. Both plots reveal a 3- to 6-hour delay between the onset of geomagnetic activity and equatorial density response. However, the amplitude of equatorial response to the first impulse of geomagnetic activity differs between dayside and nightside. During the night, disturbances are free to travel equatorward, while the dayside equatorial response is small. To an unknown extent, the equatorial density enhancement may also be due to subsidence heating, facilitated by the nighttime equatorward flow connected with diurnal EUV heating.

Figure 4.

Daytime (top) and nighttime (bottom) equatorial response of total density from CHAMP/STAR and NRL-MSISe00. Also shown is the 3-hour Ap index.

3.2. Cross-Track Winds

[24] Studies of thermospheric winds are somewhat sparse, and in many cases, the methods used limit their comparability. For this reason, it is difficult to validate the measurements from CHAMP/STAR. Our estimates indicate that CHAMP/STAR wind measurements are accurate to about 60–100 ms−1. These estimates take into account model error, bias factor error, accelerometer precision, and the error in the total density using a “propagation of uncertainty” statistical technique. Figure 5 illustrates the behavior of the measured winds in the cross-track direction for day and night orbits during days 300–306 (27 October to 2 November). Owing to the orbit constraints of CHAMP, the cross-track axis measures winds predominantly in the east-west direction at middle to low latitudes (refer to Figure 2). However, above ±80° latitude, the cross-track axis is oriented predominantly in the meridional direction, becoming fully north-south at the point of highest latitude for the orbit (±87°). At the equator, red indicates eastward winds for both day and night. Above ±80°, red indicates northward winds for daytime orbits (top) and southward winds for nighttime orbits (bottom). The HWM-93 wind model has also been sampled on CHAMP's orbit and the vector winds projected into the cross-track direction to best illustrate similarities and differences. The HWM-93 results are shown in the right panels of Figure 5.

Figure 5.

Daytime response of CHAMP and HWM-93 cross-track wind speeds near 410 km (top). At middle and low latitudes, red indicates eastward winds and near the poles red indicates northward winds. Nighttime response of CHAMP and HWM-93 cross-track wind speed near 410 km (bottom). At middle and low latitudes, red indicates eastward winds and near the poles red indicates southward winds.

[25] One of the most noticeable traits exhibited by the thermospheric wind measurements is the combined effect of the longitude and universal time sampling of CHAMP, consisting of a combination of the so-called “longitude-UT” effect and the displacement between geographic and geomagnetic coordinates. These effects can be seen in the polar regions of both dayside and nightside orbits as an apparent 24-hour periodicity, which is also reflected in the HWM-93 model results. Owing to the complexity of the cross-track sampling in the polar regions, it is not obvious how to connect our results with any specific neutral circulation pattern (i.e., two-cell), and we will not discuss the polar region wind results further. We simply note that the overall magnitudes and patterns inferred from the cross-track accelerations follow similar patterns to that of HMW-93, hence serving as a crude validation of the accelerometer methodology for deriving neutral winds.

[26] Near the equator (see Figure 6), the high inclination of the CHAMP orbit permits a good measure of the zonal winds by the cross-track accelerometer. Near 1320 LT zonal winds are consistently westward, on the order of −100 to −150 ms−1 during times of low activity. The top panel of Figure 6 shows a westward wind of −300 ms−1 during the period of highest activity at the beginning of day 304. HWM-93 is consistent with CHAMP/STAR observations during the quiet daytime. During active periods, however, the model overestimates the observed wind enhancements by about 50%, except during the aforementioned excursion to 300 ms−1 where the HWM-93 model yields 250 ms−1.

Figure 6.

Daytime (top) and nighttime (bottom) zonal wind speed at the geographic equator from CHAMP/STAR and HWM-93. Also shown is the 3-hour Ap index.

[27] The bottom panel of Figure 6 compares nighttime winds from CHAMP/STAR and HWM93. When geomagnetic activity is low, the measured nightside winds are on average only slightly positive (∼25 ms−1), whereas HWM-93 yields eastward winds of order 50–75 ms−1. This difference is within the estimated 60–100 ms−1 errors for our wind measurements. While the dayside winds increase in magnitude due to the increase in geomagnetic activity (cf. top panel of Figure 6), nightside equatorial winds are seen to reverse direction, reaching speeds as high as 275 ms−1. This shift in direction can also be seen to a lesser extent (0 to −50 ms−1) in the HWM-93 model.

[28] The normal undisturbed behavior of the neutral wind at equatorial latitudes according to the WATS instrument on Dynamics Explorer [Wharton et al., 1984; Wu et al., 1994] is westward during the day and eastward at night, with maximum wind speeds of order 100 ms−1. The majority of data from DE/WATS is between 300–400 km. This behavior is also reflected in HWM-93, which is based in part on DE wind measurements. However, analyses of equatorial DE/WATS winds by Wu et al. [1994] and equatorial UARS/WINDII winds near 250 km by Fejer et al. [2000] reveal rather small differences between active and quiet geomagnetic conditions. On the other hand, it is important to keep in mind that the present events are unusually large in magnitude and that reversals of nighttime zonal winds have been observed under extremely active magnetic conditions [Meriwether et al., 1986; Burnside et al., 1991].

[29] The largest differences between CHAMP equatorial wind measurements and those consistent with HWM-93, are the unusually large westward accelerations at night occurring in conjunction with large geomagnetic disturbances. Changes in the zonal neutral wind speed are governed by the zonal momentum equation:

equation image

where Un is the zonal neutral velocity, Ui is the zonal ion velocity, and νin is the collision frequency of an ion with neutral gas [see Herrero et al., 1985]. Further, it is known that in the equatorial region zonal winds tend to produce zonal ion drifts of similar magnitude and direction [Rishbeth, 1971; Fejer et al., 1985; Crain et al., 1993; Richmond et al., 1992]. Kamide and Matsushita [1981] speculated that during geomagnetically active periods, a low-latitude upward electric field is caused by penetration of high-latitude electric fields in the premidnight local time region. The effect of this is a westward ion drift and hence would lead to an increased westward neutral wind acceleration according to (8). This phenomenon has also been observed in the postmidnight local time region at Arecibo [Burnside et al., 1991], but in this case it was deduced that an increase in ion-drag was responsible for a reversal of the normal eastward zonal wind. For the measurements presented here, it is not possible to unequivocally determine if the observed westward surges at night are due to changes in ion drifts or ion densities or both. Examination of GPS TEC measurements (International GPS Service, at low latitudes sampled near the CHAMP orbit do reveal, however, enhancements in the anomaly peaks during midday, and decreases in equatorial-region TEC values at night (see Figure 7). Both suggest equatorial penetration of electric fields and a reduction in plasma collisions with neutrals. It is thus clear that the CHAMP measurements are revealing the neutral wind responses to significant plasma-neutral electrodynamic interaction events that suggest further experimental and modeling studies.

Figure 7.

Total electron content (TEC) sampled for daytime CHAMP orbits (top) and nighttime CHAMP orbits (bottom).

4. Summary and Conclusions

[30] We have shown that during times of extreme geomagnetic activity, thermosphere densities near 410 km exhibit enhancements of 200–300%. On the dayside, these enhancements are confined mostly to the Northern Hemisphere between the equator and 80° latitude. If not caused by the height sampling of CHAMP, we cannot explain the origin of this latitudinal asymmetry in the density response. During nighttime, density enhancements spread much more readily to equatorial latitudes, and we suggest that this is facilitated by the nighttime equatorward flow that accompanies the diurnal variation in EUV heating. Traveling atmospheric disturbances would be expected to exist during nighttime conditions based upon analyses of CHAMP density data during the 15–24 April 2002 storm study period (J. M. Forbes et al., manuscript in preparation, 2004). However, owing to the sampling of CHAMP and the complexity of the present storm interval, it is difficult to unequivocally establish their presence or analyze their movement.

[31] An apparent 24-hour periodicity appears both in the high-latitude wind measurements and in HWM-93. We interpret this as some combination of the so-called “longitude-UT effect” and the displacement between geomagnetic and geographic coordinates (i.e., the convection-driven neutral winds tend to be ordered in the geomagnetic frame) and moreover as a validation of the CHAMP winds. Unfortunately, although at 87°N and 87°S the CHAMP cross-track winds are in the meridional direction, these measurements occur at dawn and dusk and do not sense the significant antisunward flow that would be expected from the convection driven circulation at these latitudes under very disturbed conditions. Similarly, at ±60° latitude, CHAMP is sampling near midday and midnight, far from the dawn and dusk periods where significant convection-driven zonal winds might be expected to exist. Therefore and given the difficult geometry of the cross-track winds at high latitudes and the associated ambiguities and uncertainties, we have not attempted an extensive analysis of the high-latitude wind patterns. As a follow-up to this study, we intend to obtain a better global view by comparing similar measurements from the GRACE satellites which sampled local times near 0400 and 1600 during this time period.

[32] Near the equator, measured zonal winds reveal westward surges of order 100–200 ms−1 in conjunction with an increase in magnetic activity during both day and night. The observed effect is shorter-lived (but of roughly similar magnitude) than a similar feature contained in HWM-93 under daytime conditions but exceeds the magnitude of the nighttime HMW-93 response (∼100 ms−1) by about a factor of two. These data-model differences are not surprising considering the large magnitudes of the geomagnetic disturbances; however, it is also possible that this difference is caused by an error in the cross-track bias factor. It is not possible from our measurements to determine whether changes in plasma densities and/or zonal plasma drifts driven by penetrating electric fields are the root cause of the observed effects. We hope that our experimental results will provide useful comparisons for three-dimensional general circulation models of the thermosphere and ionosphere that include self-consistent neutral and plasma dynamics.

[33] For those interested in obtaining the CHAMP/STAR winds and densities for further interpretation and collaborative studies, please visit our web site (∼suttonek).


[34] The authors acknowledge the GeoForschungsZentrum Potsdam (, who provided data from the CHAMP satellite used for this study. The authors also thank Sean Bruinsma for helpful discussions. This work was supported under grant ATM-0208482 from the National Science Foundation as part of the National Space Weather Program.

[35] Arthur Richmond thanks Sean Bruinsma and Douglas Drob for their assistance in evaluating this paper.