We use observations from a variety of different ground- and space-based instruments, including ionosonde, ground- and space-based Global Positioning System (GPS) receivers, magnetometers, and solar wind data from the Advanced Composition Explorer (ACE), to examine the response of the ionospheric F2-layer height during the November 2003 superstorm. We found that the topside ionosphere responded unusually to the 20 November 2003 severe storm compared to behavior observed in a number of previous storms. While ground-based GPS receivers observed a large enhancement in dayside TEC, the low-Earth orbiting (∼400 km) CHAMP satellite did not show any sign of dayside TEC enhancement. The real-time vertical density profiles, constructed from ground-based GPS TEC using a tomographic reconstruction technique, clearly revealed that the ionospheric F2-layer peak height had been depressed down to lower altitudes. Ionospheric F-layer peak height (hmF2) from the nearby ionosonde stations over Europe also showed that the dayside F2-layer peak height was below 350 km, which is below the orbiting height of CHAMP. The vertical E × B drift (estimated from ground-based magnetometer equatorial electrojet delta H) showed strong dayside downward drifts, which may be due to the ionospheric disturbance dynamo electric field produced by the large amount of energy dissipation into high-latitude regions. This storm demonstrates that data from LEO satellites varies widely among different superstorms.
 The interconnection between southward interplanetary magnetic fields (IMFs) and the Earth's magnetic field leads to a large amount of energy deposition into the high-latitude regions. Such high-latitude heating, in turn, can create strong dawn-to-dusk disturbance dynamo electric fields [Blanc and Richmond, 1980; Tsurutani et al., 2004] and meridional neutral winds [Sica and Schunk, 1990]. These dawn-to-dusk electric fields and meridional winds can impose dramatic effects on the ionospheric plasma distribution. The westward (eastward) electric field can modify ionospheric density distribution by driving the plasma vertically downward (upward) [Blanc and Richmond, 1980; Fejer et al., 1983; Fejer and Scherliess, 1995]. Similarly, the poleward (equatorward) neutral winds can cause downward (upward) ionospheric plasma drifts [Sica and Schunk, 1990]. The convection of ionospheric plasmas by zonal electric fields, through ion collisions with neutrals, leads to the production of neutral winds with velocities in excess of 1 km s−1. Joule heat dissipation within the high-latitude thermosphere, especially during a severe magnetic storm period, may drive large-scale wind surges [Buonsanto, 1999].
 The complex interaction of the solar wind, magnetosphere, thermosphere and ionosphere causes dynamical changes over a variety of spatial and temporal scales. For example, it can cause a profound change in the upper thermosphere on a global scale [e.g., Fuller-Rowell et al., 1994; Yizengaw et al., 2005b] and modify ionospheric electrodynamics in general. Studies of stormtime ionospheric electrodynamics have shown the occurrence of large electric fields and currents during and after geomagnetically disturbed periods [e.g., Fejer and Scherliess, 1995; Mannucci et al., 2005a; Foster and Rideout, 2005]. Electric field perturbations with timescales of about an hour, occurring nearly simultaneously at all latitudes, have been attributed to the prompt penetration of high-latitude electric fields to lower latitudes [Tsurutani et al., 2004; Chi et al., 2005; Mannucci et al., 2005a]. However, this electric field can be reduced quickly by magnetospheric shielding, which is expected to occur within an hour of the onset of geomagnetic activity; this has been shown both experimentally [Fejer and Scherliess, 1995] and theoretically [Spiro et al., 1988]. In addition to these sudden perturbations, long-lasting middle- and low-latitude electrodynamic disturbances often occur from a few to several hours after large enhancements occurred in the high-latitude currents [Fejer et al., 1983]. These electric field perturbations are most probably associated with ionospheric disturbance dynamo effects produced by enhanced energy dissipation into the auroral ionosphere during geomagnetically active periods [Blanc and Richmond, 1980].
 Using the ground magnetometer observations of field line resonance (FLR)-inferred magnetospheric density and ground-based GPS TEC, Chi et al.  found huge density enhancements in both the magnetosphere and ionosphere. This coincided with intervals of southward IMF and high-speed solar wind. They concluded that these huge density enhancements could be due to the dawn-to-dusk convective electric field imposed by the solar wind. However, during the 20 November 2003 severe magnetic storm, unusual ionospheric density behavior occurred. While ground-based GPS receivers observed a large enhancement in dayside TEC, consistent with Chi et al.'s  result, the topside ionospheric TEC alone (as measured from CHAMP (Challenging Multisatellite Payload)) did not show any sign of dayside TEC enhancements during the 20 November severe storm. This shows that the topside ionosphere and plasmasphere did not respond to this storm as generally inferred by earlier studies using ground- and space-based observations [e.g., Foster et al., 2002; Chi et al., 2005; Mannucci et al., 2005a]. Mannucci et al. [2005a, 2005b], using GPS data from CHAMP, has observed a dramatic TEC enhancement (∼900% increase) above CHAMP orbiting height (∼400 km) during the severe Halloween 2003 storm time. At the same time, they also observed similar ground-based TEC enhancement. However, in the case of 20 November storm the ionospheric F-layer peak height (hmF2) was also unusually depressed down to lower altitudes (below 350 km) for more than 24 hours.
 The goal of this paper is to examine this unusual topside ionospheric density response to the severe magnetic storm period on 20 November 2003. The ionospheric effect is observed at different latitude regions both in the northern and southern hemispheres. In this paper we present the temporal variation of the ionospheric F-layer peak height and the response of both the ground- and space-based GPS TEC. Data from the ground-based GPS receivers are used to examine the ionospheric plasma response. In addition, ionosonde data for the F-layer peak height are analyzed. We also applied a tomographic reconstruction technique to the ground-based GPS TEC to track the temporal variation of hmF2 of the ionosphere. The interplanetary event is analyzed using solar wind data from the Advanced Composition Explorer (ACE) spacecraft. The topside ionospheric and plasmaspheric integrated column density is also estimated using the GPS receiver on CHAMP. CHAMP is a German small satellite mission for geoscientific and atmospheric research and applications, managed by GFZ (GeoForschungsZentrum Potsdam). It is equipped with highly precise, multifunctional, and complementary payload elements such as GPS receiver, magnetometer, star sensor, laser retro reflector, and ion drift meter. CHAMP is a LEO (low-Earth orbit) satellite, orbiting at ∼400 km altitude [Mannucci et al., 2005a].
2. Method of Analysis and Data Used
 Global Positioning System (GPS) satellite signals are received by both ground- and space-based receivers. Currently, there are about 29 GPS satellites distributed in six circular orbits at an altitude of 20,200 km (∼4.2 L). Each GPS satellite broadcasts two spread spectrum L-band radio signals: f1 = 1.57542 GHz and f2 = 1.2276 GHz. The phase delays in the dual frequency electromagnetic radio signals from the GPS satellites to the receivers are directly related to the enhanced index of refraction due to the total number of electrons along the lines of sight. GPS observations from a worldwide network of ground-based GPS receivers are available to the scientific community by the international GPS service (IGS). Figure 1 presents the geographic locations of all ground-based GPS receivers used in this study. The circles indicate where the subionospheric points lie at a 25° cutoff elevation angle. The ground-based GPS data used in this study have been analyzed using the GPS TEC extraction technique, which is described by Yizengaw et al. . The integrated electron density along the lines of sight between the satellite and the receiver (slant TEC) is calibrated from instrumental biases [e.g., Mannucci et al., 1998; Yizengaw et al., 2004] and then mapped to the vertical using a single spherical shell layer approximation of the ionosphere at hsp = 400 km altitude [e.g., Jakowski et al., 1996].
 Similarly, the topside ionospheric and plasmaspheric integrated column density can be estimated using the upward looking GPS receiver antenna deployed on CHAMP. A TEC extraction technique from CHAMP is well described [Mannucci et al., 2005a].
 The tomographic reconstructions have been performed along two distinct longitudinal sectors: over the European (at ∼15°E geographic) and the American (at ∼290°E geographic) sector using ground-based GPS receivers as shown in Figure 1. The dashed broken lines indicate the tomographic meridian sectors and are identified as the median longitudes of the stations. Bias-corrected slant TECs are used as input to the inversion algorithm to obtain a two-dimensional (altitude versus local time) vertical electron density structure. The tomographic reconstruction algorithm used in this study is described by Yizengaw et al. [2005a].
 The F-layer peak height (hmF2) is also obtained from the European ionosonde stations: Rome (42.97°N, 12.2°E geographic) and Athens (38.44°N, 24.8°E geographic). The ionospheric vertical plasma drift, which has a direct link with the zonal electric field, at and in the immediate vicinity of the magnetic equator were also inferred from equatorial electrojet Delta H response using ground-based magnetometer data [e.g., Anderson et al., 2002; Yizengaw et al., 2004, 2005a]. For this study, ground-based magnetometer data from American longitude sector stations, Huancayo (12.05°S, 284.67°E geographic; 1.62°S geomagnetic) and San Juan (18.38°N, 293.88°E geographic; 28.58°N geomagnetic), are used.
 Solar activity that produced the CME shock on 18 November 2003 led to a severe geomagnetic storm on 20 November and early 21 November 2003. Figure 2 illustrates the interplanetary events of this storm caused by shock compression of the IMF Bz fields. The top six panels are interplanetary parameters taken from data measured by ACE spacecraft. From top to bottom the panels display the solar wind speed (Vsw), the proton density (Np), the proton temperature (Tp), the magnetic field magnitude (∣B∣), the magnetic field BZ component in GSM coordinates, and the solar wind ram pressure (Pram). The corresponding ring current activity with a minimum Dst value of ≈−472 nT at about 2000 UT on 20 November (black curve) and the three hourly Kp index are shown in the bottom panel of Figure 2. The convection time delay between ACE and the ground is estimated using the measured solar wind speed of ∼700 km s−1 and is equal to ∼34 min. The solar wind data has therefore been shifted by this time in Figure 2 to match the ground-based Dst and Kp index data.
 The dashed vertical line in Figure 2 indicates the sudden storm commencement (SSC) time which started at 0803 UT on 20 November. As can be identified by the abrupt increase of the solar wind speed from ∼450 km s−1 to more than 750 km s−1, the shock arrived at about 0800 UT on 20 November 2003. At the same time the other solar wind parameters also responded to the shock positively; i.e., the density rises from 5 cm−3 to over 25 cm−3, the temperature increases from 1.0 × 105 K to more than 7.0 × 105 K, and the magnitude of the magnetic field increase to ∼25 nT from about 5 nT before the shock. The shock also gives an increase in pressure from 0.5 nPa to greater than 10 nPa. The sharp increase in Pram, Vsw, and B indicates a strong dayside magnetospheric compression leading to the generation of this storm. Right after the shock arrival, IMF Bz experienced a small fluctuation and then turned to the north for about an hour. About 3 1/2 hours after the shock arrived, the Bz sharply turned south and remained south for more than 12 hours. The density and pressure show significant decreases from the time when Bz turned south until Bz reached its minimum value at about 1530 UT on 20 November. The pressure and density again simultaneously show abrupt increases (even much greater than the values that each parameter revealed right after the shock) when Bz just passed its minimum value. However, the solar wind speed continues declining when Bz passed its minimum value.
 The tomographic technique also allows us to image the temporal variation of the ionospheric density profiles. The electron density is reconstructed over an extended (24-hour) period of time. The tomographic inversion has been performed at about 15°E longitude in the Northern Hemisphere. Figure 3 shows an example that covers the diurnal behavior of the vertical electron density distribution (altitude-local time (LT = UT + 1.0) map of ionospheric density) during the superstorm period on 20 November 2003. This can be done by imaging the spatial distribution of ionospheric electron density using tomographic reconstruction technique for the whole day with an interval of 25 min. The combination of over 57 images that are reconstructed in a day provide a single temporal map of density profiles as shown in Figure 3 (top). Plots of this kind illustrate that ionospheric tomography has a potential to be used to create temporal maps of ionospheric electron density profiles in a two-dimensional plane covering substantial spatial sections of the ionosphere. The four different horizontal panels, at the top, middle, and bottom, represent different latitudinal regions, namely, from left to right, low-latitude, midlatitude, auroral, and the cusp region. These nominal regions are used to show the hmF2's response to this superstorm as observed from different latitudinal regions. As can be seen in the figure, the normal solar-produced ionization trend is evident, i.e., increase in ionization during the day and decreased ionization at night. The middle panels of Figure 3 represent the temporal variation of the F-layer peak height (hmF2) extracted from the corresponding reconstructed electron density structures shown in the top panels of Figure 3. The peak height is extracted by simply identifying the maximum (peak) density at every 25 min interval, which is the time interval used to perform each tomographic reconstruction. These extracted peak heights have been interpolated as a function of time in order to fill the gaps. These extracted peak heights show that the ionospheric F-layer has been depressed to lower altitudes than normal during the November 2003 superstorm. The F-layer peak height remained below 350 km for more than 24 hours. Especially during the dayside the hmF2 was at much lower altitude, below 250 km, near noon. At the cusp region hmF2 was even at lower altitudes, ∼220 km, around local noon as shown in the middle panels of Figure 3. Tomographic analysis has been made for the 19 November data (not shown here) and found similar hmF2 heights. However, unlike the 20 November temporal tomographic images, the 19 November tomographic images show significant density in the topside ionosphere. The bottom panels in Figure 3 present the ground-based GPS TEC on 19 November (dashed curve) and on 20 November 2003 (solid curve), the IRI-2001 model simulated TEC (dot-line curve), and TEC above 470 km height (blue triangle curve). The TEC values above 470 km are estimated from the corresponding tomographic images shown in the top panels. As can be seen in the figure, except in the auroral region, the topside ionospheric TEC contributed less than 20% to ground-based GPS TEC in the dayside. In the auroral region, however, the topside ionospheric TEC contribution slightly increased (∼33%) around local evening. In fact the GPS TEC value in the evening was even more than the GPS TEC values at noon. A similar evening time TEC enhancement is also evident at the cusp region. The GPS TEC values on 19–20 November were much less than IRI-2001 simulated TEC around local noon in the auroral region.
 A similar F-layer depression is also evident at the southern hemisphere but at a different meridian (see Figure 4). Like the hmF2 height in the northern hemisphere, it was also below 380 km throughout the day in the southern hemisphere. However, the TEC response shown in the bottom panel of Figure 4 has different behavior compared to the TEC response in the northern hemisphere mentioned above. The storm time GPS TEC values were below both IRI-2001 TEC and GPS TEC values during the quiet day. This happened mainly at the vicinity of the magnetic equator (see first and second panel from left). However, in the tropical region (third and fourth panels from left), the storm time TEC showed a dramatic enhancement around local noon but remained below IRI and quiet day GPS TEC values during nighttime. The topside ionospheric contribution to the ground-based GPS TEC is still low (<20% during the day).
Figure 5 presents the vertical TEC derived from data recorded using ground-based GPS TEC and hmF2 obtained from the nearby ionosonde stations. The geographic latitude and longitude of each ionosonde and ground-based stations are indicated at each panel in the figure. The vertical dashed line depicts the SSC time. The horizontal dashed lines shown in the first and third panels from the top indicate the 350 km altitude mark. The open and filled rectangles along the horizontal axes of the panels represent dayside and nightside of the corresponding stations. The ionosonde also revealed that hmF2 was mostly below 350 km as shown in Figure 5 (first and third panel from the top). The ground-based GPS TEC shows a huge density increase (∼67% increase), compared to the previous day TEC value, around local noon on 20 November 2003. At the time when GPS TEC showed large density increases, the F-layer peak density was below 300 km altitude (see the hmF2 values at the second and third panels of Figure 5). A transient hmF2's height rise to an altitude level higher than 450 km during local premidnight time is evident around 1730 UT (1830 LT) on 20 November. This is consistent with hmF2 height observed by tomographic reconstruction technique described above. Unfortunately, the nighttime ionosonde hmF2 data were not available on 20 November and we are unable to see how long this transient height rise lasted.
 The bottom panel of Figure 5 shows the equatorial electrojet (EEJ) current intensity over the American equatorial station Huancayo (12.05°S, 284.67°E geographic; 1.62°S geomagnetic) between 19 and 21 November 2003 (solid curve). The EEJ current intensity is obtained by subtracting the diurnal range of the horizontal component of magnetic field intensity over a non-EEJ station San Juan (18.38°N, 293.88°E geographic; 28.55°N geomagnetic) from that of Huancayo (ΔH = HHua − HSjg). The dashed curve depicts the reference day's curve of ΔH, which is taken during magnetically quiet period (28 November 2003). When the shock arrives and during the storm main phase, the European sector was in the dayside sector and the Huancayo (Hua) was in the nightside sector. The consonance between the top four panels and the bottom panel has to be viewed as a function of local time instead of universal time. Attention should be given to the local time difference between the top four panels, which are approximately in the same local time sector, and the bottom panel which is ∼6 hours behind the top four panels' local time. The ΔH curve, which is a good indictor of the vertical E × B drift [Anderson et al., 2002; Tsurutani et al., 2004; Yizengaw et al., 2005a], indicated a downward E × B drift starting near 1400 UT (0900 LT) and reached its maximum downward E × B drift value at about 1800 UT (1300 LT) on 20 November 2003. The downward E × B drift turned upward and reached to an upward drift peak at ∼2100 UT (1600 LT). As can be seen in the figure the ΔH slowly decreased and closes to the reference curve (dashed curve) after ∼2300 UT (1800 LT) on 20 November, indicating the absence of EEJ activity.
 The CHAMP satellite, equipped with an upward looking GPS receiver antenna, did not record similar TEC enhancement as the ground GPS receivers did. The vertical GPS TEC above the polar orbiting satellite [Mannucci et al., 2005b] is shown in Figure 6. The quiet day (19 November 2003) vertical TEC as a function of geomagnetic latitudes is shown in Figure 6a. Similarly, the vertical TEC recorded during severe magnetic storm period, 20 November, is shown in Figure 6b. The corresponding ground traces of CHAMP are shown in Figure 6c. Except slight transient TEC enhancements at high latitudes (compared to quiet time TEC values in Figure 6a) and large-scale fluctuation throughout the latitudes covered by CHAMP, the vertical TEC did not show any sign of a strong enhancement like the ground-based GPS TEC did. CHAMP was at a nearly constant altitude of ∼410 km, which was well above the F-layer peak density altitude.
 The initial response at high latitudes is that Joule heating raises the temperature of the thermosphere, while ion drag drives high-velocity neutral winds. The heat source drives global wind surges, from both polar regions, which propagate to low latitudes and into the opposite hemisphere. The global dynamical changes drive a disturbance dynamo often following, but sometimes competing with, the magnetospheric penetration electric fields.
 Ionospheric disturbance dynamo electric fields, driven by Joule heating in the auroral zone [e.g., Tsurutani et al., 2004], have been studied extensively and are well documented [Blanc and Richmond, 1980; Mazaudier and Venkateswaran, 1990; Fejer and Scherliess, 1995]. Blanc and Richmond  have studied the characteristics of disturbance dynamo electric field using numerical simulations. Similarly, Mazaudier and Venkateswaran  have used midlatitude plasma drift, neutral wind measurements, and equatorial magnetic field data to examine ionospheric disturbance effects following a severe magnetic storm. During severe magnetic storm periods, the dissipation of enhanced energy deposition into the high-latitude ionosphere generates a meridional circulation with equatorward winds in the F-region (above about 120 km altitude). This equatorward wind, through the action of the Coriolis force, in turn, leads to westward winds and plasma drifts. As Kelley  clearly described, the equatorward Pederson current generated by these westward ion drifts at altitudes where the cross-field conductivity is high (∼150 km) builds up polarization charges at the equator and a poleward electric field which eventually cancels the equatorward Pederson current. The resulting eastward Hall current, which maximizes at midlatitudes, sets up polarization charges at the terminators and a dusk-to-dawn electric field is generated. This electric field is eastward during night and westward during daytime according to the model assisted extensive study by Scherliess and Fejer . They have determined the average characteristics of the vertical plasma drifts driven by the ionospheric disturbance dynamo electric fields as a function of the time history of the auroral activity determined from the Auroral Electrojet (AE) index.
 The large decrease of ΔH starting about 1400 UT (0900 LT) marks the occurrence of a strong westward electric field associated with the shock that hits 6 hours earlier. The F-layer height (hmF2) decrease that occurred over Rome and Athens ionosonde stations (Figure 5) and also observed from tomographically reconstructed electron density temporal profiles (Figure 3 and 4) could be caused by this disturbance westward electric field. This daytime westward electric field could be the ionospheric disturbance dynamo electric field, which may be generated by the huge energy dissipation in the auroral zone when Bz turns southward.
 Enhancements in the auroral currents are associated with increases in the energy input into the high-latitude ionosphere through Joule heating and particle energy injection. The AE index (not shown here but can be accessed via http://swdcwww.kugi.kyoto-u.ac.jp/aedir/), which is an indicator of energy activity in the auroral zone, shows dramatic enhancements (more than 2500 nT) between 1200 UT and 1800 UT on 20 November 2003. This is the time when the E × B drift was downward (see Figure 5). Adopting the equation developed by Ahn et al. , the empirical relationship between the global energy injection rate and AE index is given by P(W) = 2.9 × 108 AE(nT). This energy generates stormtime wind and dynamo electric field disturbances [e.g., Blanc and Richmond, 1980; Fejer and Scherliess, 1995]. Blanc and Richmond , in their disturbance dynamo drift model simulation have found dayside downward drifts of 5 m s−1 for a hemispheric power input of about 1.2 × 1011 W. This corresponds to an increase in AE index by 400 nT over its quiet day value. Similarly, during 20 November 2003 storm period the energy injection rate became ∼7.2 × 1011 W (corresponding to the change in AE index by ∼2400 nT). Therefore according to model simulation results [Blanc and Richmond, 1980] and experimental observations [Fejer and Scherliess, 1995], the 7.2 × 1011 W power input could trigger a downward disturbance dynamo drift of up to 30 m s−1. This speed could produce the sharp downward drift as is inferred by the ground-based magnetometer data shown in Figure 5.
 The dayside downward drifts have a very strong effect on ionospheric plasma in the vicinity of the magnetic equator where the magnetic field becomes nearly horizontal. This is evident from our observation shown in Figure 4 bottom panel. At the equatorial region (first and second panel from left in Figure 4), the dayside plasma on 20 November was pushed down to lower altitude (where the recombination rate is high), and thus the TEC values became below the level of quiet day and IRI-2001 TEC values (see the bottom first and second panels from the left).
 Previous studies show the ground- and space-based density observations usually respond in a similar way [e.g., Chi et al., 2005; Mannucci et al., 2005a, and references there in]. Even during severe storm periods the integrated densities detected by the space-based instruments go almost to the same level as the magnitude of the integrated densities obtained by ground-based measurements. Mannucci et al. [2005a] have observed that the magnitude of topside (above 400 km altitude) TEC was greater than the ground-based GPS TEC during the most severe Halloween 2003 storm period. This is of course due the raypaths transmitted from different satellites and detected by ground-based or space-based or by both GPS receivers may traversed through different regions. The recent modeling studies by Maruyama et al.  have also conclude that the penetration of a magnetospheric origin eastward electric field has a dominant effect during daytime, causing ionospheric plasma uplift to higher altitude where the recombination rate is low. All these previous studies showed that both ground- and space-based GPS TEC responded in a comparable way to the magnetic storm effect. However, the storm time response of ground- and space-based GPS TEC appears to be different during the 20 November 2003 superstorm period.
 Comparison of temporal variation of vertical ground-based GPS TEC (Figures 5) and space-based (CHAMP) GPS TEC (Figures 6) show that when the ground-based TEC showed a significant enhancement (∼67% increase), the space-based TEC did not show as a dramatic response. This is different than the expected response according to previous studies, particularly in comparison with the topside ionospheric behavior observed during the Halloween storm [Mannucci et al., 2005a]. Similarly, the total density integrated above 470 km height (topside ionospheric TEC) shows very low TEC values compared to the ground-based GPS TEC enhancement during 20 November 2003 severe storm (see Figures 3 and 4). This indicates that the topside ionospheric plasma has been taken away, probably vertically depressed down below 470 km height. This could be, as described above, due to disturbance dynamo electric fields that may push ionospheric plasma down to lower altitudes during daytime. According to ionosonde (Figure 5) and ground-based tomographic reconstruction density profiles (Figures 3 and 4), ionospheric F-layer peak height was below about 350 km altitude in the northern hemisphere (Figure 3) and below 370 km in the south (Figure 4), which are well below the orbiting altitude of space-based GPS receiver (CHAMP) that was used for this study. This means the upward looking antenna of CHAMP did not see the huge density below its orbiting height.
 There is also another possible mechanism that may contribute to the downward depression of the ionospheric F-layer; i.e., poleward meridional winds [Buonsanto, 1995]. The meridional wind is poleward in the winter and equatorward in the summer hemisphere. From the tomographically reconstructed images of the vertical density profile the hmF2 height in the southern hemisphere (summer) is a bit higher than the hmF2 value in the north (winter). This could be due to the meridional wind direction difference in the north (poleward) and south (equatorward). The upward looking GPS receiver on CHAMP also detected a slight TEC difference between southern and northern hemisphere high-latitude regions as shown in Figure 6b. The figure also revealed TEC fluctuations, possibly indicating the presence of a meridional wind surge traveling across different latitudes. Details of this traveling surge are beyond the scope of this paper and will not be discussed further here.
 In addition to the vertical depression, mentioned above, a composition disturbance could be another possible contributor to the ground-based GPS TEC enhancement when hmF2 revealed lower height value. The poleward neutral winds, which occur equatorward of the composition disturbance zone, causes downwelling of the neutral species at low-middle latitudes, increasing the [O] density relative to [N2] and [O2] [e.g., Buonsanto, 1999]. This may lead to the ground-based GPS TEC enhancements as can be seen in Figure 5. The electron concentration is, roughly speaking, directly proportional to the [O]/[N2] ratio at the F2 layer maximum height [Yizengaw et al., 2005b].
 Ionosonde hmF2 data (Figure 5) shows a transient ionospheric F-layer peak height rise, reaching up to 450 km at ∼1800 UT (1900 LT) on 20 November 2003. This transient height rise occurred in-phase (in terms of local time) with sudden direction change of E × B drift to upward that occurred between 2100 UT (1600 LT) on 20 November as shown in Figure 5 bottom panel. The ground-based GPS TEC also responded positively to this transient hmF2 height rise, showing a TEC peak labeled by “P” as shown in second and fourth panels of Figure 5. This clearly indicates that the plasma density was driven to higher altitude regions where the recombination rate is low.
 The unusual response of the ionosphere to the severe magnetic storm event on 20–21 November was analyzed using data from ground- and space-based instruments. We have investigated that the ionospheric F-layer peak height was pushed down below ∼360 km altitude for more than 24 hours during this interplanetary event. As clearly revealed by the ground-based magnetometer delta H, there was a significantly strong downward E × B drift which may be the result of ionospheric disturbance dynamo electric fields. The AE index experienced dramatic enhancement during this severe magnetic storm period. This indicates that a large amount of energy was injected into the high-latitude region through Joule heating and particle energy injection. The enhanced energy input to the ionosphere is a possible main cause for the production of strong dawn-to-dusk electric fields.
 There is one other important point from our observation; i.e., the plasmasphere and topside ionosphere had no significant contribution to the density enhancements (∼67% increase from the TEC value before the shock arrived) observed from the ground. These density enhancements were only sensed by ground-based GPS TEC (the integrated column density from the ground to ceiling height of GPS satellite) not by space-based GPS TEC (the integrated density between CHAMP orbiting height and GPS satellite). We conclude that this unusual topside ionospheric density response to the 20 November 2003 superstorm period could be the result of ionospheric disturbance dynamo electric field generated by the large energy dissipation in the high-latitude region. The down drift of density should generally create a decrease in TEC, but here the opposite appears to be occurring. Perhaps a composition anomaly is operating the increases [O]/[N2] ratio (either by increasing [O] or decreasing [N2]) which counteracts the downdraft so that net TEC increases, not decreases. These results suggest that more intensive studies are required to understand why the topside ionosphere and plasmasphere contribution to the ground-based TEC dramatically decreased during this severe magnetic storm period.
 This research work has been financially supported by a NASA Guest Investigator grant (NNGO4GG343G), NSF grant (ATM-0348398), NSF grant (ATM-0524711) and a JPL/UCLA partnership grant. Portions of this research were carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (NASA). The authors thank the IGS for providing GPS data and the European ionosphere data service for ionosonde data provision. The ACE data are from CADWeb (http:///cdaweb.gsfc.nasa.gov-/cdaweb/istp_public/) and the ground-based magnetometer data, Dst, and Kp data are from WDC-2 at Kyoto (http://swdcdb.kugi.kyoto-u.ac.jp).
 Arthur Richmond thanks Eduardo Araujo-Pradere and Alexey Danilov for their assistance in evaluating this paper.