A numerical simulation was performed to investigate quasiperiodic ionospheric oscillations that were observed with periods of 4–5 h by the ionosonde network (Okinawa, Yamagawa, Kokubunji, and Wakkanai) in Japan during the 15 December 2006 magnetic storm. This simulation used the Coupled Magnetosphere Ionosphere Thermosphere (CMIT) 2.0 model. The CMIT model reproduced the main characteristics of the observed ionospheric oscillations, although it remains a challenging task to simulate the observations in a quantitative sense. Term analysis of the ion continuity equation demonstrated that the ionospheric oscillations in this event were mainly induced by the disturbed neutral winds, which were associated with the large scale thermospheric circulation and traveling atmospheric disturbances (TADs) during the storm. The TADs simulated from the model were then compared with those observed by the GPS Earth Observation Network (GEONET) in Japan to validate the simulation results. A prominent northward propagating large-scale traveling ionospheric disturbance (LSTID) during daytime, seen by the GEONET total electron content (TEC) data, was captured by the CMIT model. Two southward LSTIDs observed by GEONET GPS network were also reproduced by the CMIT model. However, the model gave faster phase speeds for the southward propagating LSTID occurred during 0620–0800 UT and the northward propagating LSTID; furthermore, the model missed the LSTID seen in the TEC perturbation data during 0140–0220 UT. Finally, both observations and simulations showed a strong hemispheric asymmetry for the TAD propagation that occurred during 0000–0400 UT, which may be associated with the hemispheric asymmetry of the change of Joule heating at high latitude.
 The magnetospheric electric fields have significant effects on ionospheric electrodynamics at both high and middle-low latitudes [Kelley, 1989]. The penetration of these magnetospheric electric fields, which are controlled or modulated by oscillations in the solar wind, can cause quasiperiodic ionospheric disturbances [Huang et al., 2002, 2003]. A large amount of energy is also dissipated in the high latitudes during storms or substorms, which can generate traveling atmospheric disturbances (TADs) in the thermosphere [e.g., Balthazor and Moffett, 1999; Lu et al., 2001; Prölss, 1993; Richmond and Matsushita, 1975; Richmond, 1979a, 1979b; Shiokawa et al., 2007]. The generated TADs can propagate to middle-low latitudes and even into the opposite hemisphere. The neutral wind perturbations associated with TADs can move the plasma upward/downward along magnetic field lines, thus leading to ionospheric fluctuations at middle-low latitudes, which are referred as large-scale traveling ionospheric disturbances (LSTIDs, see the review papers by Hunsucker  and Hocke and Schlegel ). Therefore, the ionospheric oscillations observed at middle-low latitudes can also be caused by a series of TADs propagating from the auroral regions. Another possible source of the variations is the enhanced high-latitude momentum forces and energy during storms, which can greatly change the global neutral wind circulation and generate a neutral wind disturbance dynamo [Blanc and Richmond, 1980]. These changed background neutral winds and the dynamo electric fields can also introduce ionospheric perturbations.
 The LSTIDs generated during storms or substorms may show time delays at different stations along the propagation direction of TADs [e.g., Hocke and Schlegel, 1996; Hunsucker, 1982]. However, sometimes it is difficult to judge whether there are time delays or not in the ionospheric oscillations when stations are too close or the time resolution of the observations is low. The ionospheric perturbations caused by penetration electric fields usually occur simultaneously at different latitudes in the same longitudinal zone because of the quick penetration of magnetospheric electric fields from high latitude to middle-low latitudes. Penetration electric fields also alter the equatorial ionosphere through the fountain effect, so the response time of equatorial ionosphere to electric fields can vary with latitude [e.g., Abdu et al., 1990]. Hence it is difficult to use the limited observations available to identify the sources, which drive the ionospheric oscillations, owing to the interplay of several physical processes.
 The purpose of this paper is to interpret the quasiperiodic ionospheric oscillations with periods of 4–5 h observed by the ionosonde network in Japan on 15th of the 2006 December storm using a Coupled Magnetosphere Ionosphere Thermosphere (CMIT) 2.0 model simulation. The CMIT 2.0 model [see Lei et al., 2008; Wiltberger et al., 2004; W. Wang et al., Ionospheric electric field variations during geomagnetic storm simulated by a coupled magnetosphere ionosphere thermosphere (CMIT) model, submitted to Geophysical Research Letters, 2008] couples the National Center for Atmospheric Research Thermosphere-Ionosphere-Electrodynamics General Circulation Model (NCAR-TIEGCM) [Richmond et al., 1992] with the Lyon-Fedder-Mobarry (LFM) global magnetohydrodynamic (MHD) magnetospheric model [Lyon et al., 2004]. This coupled model can self-consistently simulate the effects of the imposed magnetospheric convection field, neutral wind dynamo and penetration electric fields, neutral composition as well as neutral winds on the ionosphere.
 In a recent paper, Lei et al.  showed that the CMIT model could capture the temporal and spatial variations of the ionospheric storm effects in the initial phase of the December 2006 geomagnetic storm seen in the GPS TEC, ionosondes, and the Millstone Hill ISR observations. The sudden storm commencement (SSC) for this storm occurred at ∼1414 UT on 14 December. As shown in Figure 1, Dst went down to ∼147 nT at 0800 UT on 15 December, and then started to recover. Bz became northward after 1800 UT on 14 December and it continued to oscillate before its rapid southward turning at 2300 UT on 14 December, indicating the beginning of the main phase of this storm. Here we use the CMIT model to distinguish the relative importance of electric fields, TADs and background neutral winds as drivers of the quasiperiodic ionospheric oscillations seen in the observations on 15th of the 2006 December storm.
 In this paper we try to address the following questions: (1) Can the CMIT model capture the observed ionospheric oscillations in this event (section 2.1)? (2) If it does, what is the relative importance of electric fields, TADs, and background neutral winds in explaining these ionospheric perturbations (sections 2.2–2.3)? (3) If the ionospheric oscillations are associated with TADs, what are the differences in the occurrence time and propagation velocity of TADs between the model and the observations (sections 3.1–3.2)? (4) How does the propagation of TADs depend on hemisphere (season) and Joule heating (section 3.3)?
2.1. Variation of NmF2 and hmF2: Data-Model Comparison
Figure 2 shows a comparison between NmF2 and hmF2 measured from the meridional ionosonde chain consisting of four stations in Japan (Okinawa (26.7°N, 128.2°E), Yamagawa (31.2°N, 130.6°E), Kokubunji (35.7°N, 139.5°E), and Wakkanai (45.4°N, 141.7°E)) and results from the CMIT simulations for 13 and 15 December 2006. It should be pointed out that 13 December was a quiet day, whereas 15 December was a storm day (Figure 1). The ionosonde hmF2 was deduced from the measured M(3000)F2, foF2 and foE [see Maruyama et al., 2004]. When foE was not available, it was calculated from the IRI model [Bilitza, 2001].
 As seen in Figure 2, on 13 December the observed NmF2 showed two peaks at Okinawa, Yamagawa, and Kokubunji, one at about 0400 UT and another at 0900 UT. The second peak was not obvious at Wakkanai. On this quiet day hmF2 had its lowest values at about 0000 UT (0900 LT) at all stations, and then it increased in the afternoon at ∼0800 UT (1700 LT) and reached a diurnal maximum around midnight. hmF2 dropped again after sunrise. On the storm day (15 December), the variations of observed NmF2 and hmF2 were much more complicated than those on 13 December. At these four stations, NmF2 increased by a factor of 3–5 around 0330 UT on 15 December compared with its quiet time value; quasiperiodic oscillations are clearly seen in both NmF2 and hmF2. Four peaks in NmF2 can be seen at about 0330, 0800, 1400, and 2300 UT, and at least five peaks in hmF2 were detected at around 0200, 0630, 1100, 1530, and 2000 UT at Okinawa, Yamagawa, and Kokubunji. Note that the occurrence time and magnitude of peaks were slightly different at each of these stations.
 Now we turn to the simulation results in Figure 2. The CMIT model reproduced the observed salient features on both the quiet and storm days. The diurnal shapes from modeled NmF2 and hmF2 were in agreement with those from the observations on 13 December. The model calculated the quasiperiodic oscillations well, in both NmF2 and hmF2 in a qualitative sense. But it is still a challenge to reproduce the observed values in quantitative sense. The model gave higher hmF2 than the data on both days except for some short periods, and it produced a smaller increase of NmF2 at around 0330 UT on 15 December. There were also time shifts of about 20 min between the modeled and observed peaks of NmF2 on 15 December, which will be discussed later.
2.2. Ionospheric Oscillations and Neutral Winds/Electric Fields
 Given that the CMIT reproduced the ionospheric quasiperiodic oscillations on 15 December observed by the Japanese ionosonde network, it is worth examining the variations of neutral winds and E × B electric field drifts, which are possible drivers of the ionospheric oscillations, as mentioned in section 1.
Figure 3 shows contours of CMIT simulated electron density, neutral meridional winds and E × B vertical drifts on 15 December. The geographic location was at 32.5°N and 135°E, which was close to the ionosonde stations used in Figure 2. As shown in Figure 3a, the quasiperiodic oscillations in electron density can be seen clearly from 200 to 500 km. However, the amplitudes of the oscillations in electron density are much smaller during nighttime than those in daytime although the oscillations of hmF2 during nighttime are also significant. There are many oscillations in the meridional neutral winds (Figure 3b). These neutral winds changed their directions from equatorward to poleward at 0200, 0600, and 1200 UT, and from poleward to equatorward at 0100, 0500, 0900, and 1900 UT. Equatorward winds move the plasma upward along magnetic field lines, whereas poleward winds push the ionosphere downward. The upward E × B vertical drifts (Figure 3c) did not show as many oscillations as seen in neutral winds, but the strong upward and downward drifts can be seen during the periods 0000–0530 UT and 1730–1900 UT, respectively. Therefore, the changes in E × B vertical drifts also contributed to the oscillations in electron density.
Figure 4 shows the latitudinal variations of the differences between storm and quiet time hmF2, meridional winds, and E × B vertical drifts at a longitude of 135°E calculated from the CMIT results. As can be seen in Figure 4a, the difference of hmF2 in the Northern Hemisphere shows many oscillations, which are consistent with the results in Figure 2. hmF2 increased simultaneously at about 0030 UT from 10°N to 50°N, whereas the enhancements of hmF2 during 0400–0700 and 0800–1000 UT depended on latitude.
Figure 4b shows that the disturbed meridional winds oscillated many times in the latitude range between 10°N and 50°N prior to 2000 UT and varied greatly with latitude. For example, at about 0000 UT, the equatorward perturbation winds propagated from high to low-middle latitudes in both the Northern and Southern Hemispheres. Interestingly, the perturbation winds in the Southern Hemisphere even penetrated into the low-middle latitudes in the Northern Hemisphere. Enhanced equatorward winds lifted the F layer, whereas increased poleward winds lowered it. Again, the disturbed meridional winds (Figure 4b) can account for most of the latitudinal variations of the hmF2 changes in Figure 4a.
 From Figures 4c, it can be seen that the upward E × B vertical drifts were enhanced during 0000–0530 UT and became downward during 1730–1900 UT at all latitudes. Comparing the variations of disturbed E × B vertical drifts with hmF2 changes demonstrates that the electric field was also important in generating the ionospheric oscillations in the Northern Hemisphere, in particular during the 0–0530 UT and 1900–2200 UT periods (Figures 3–4).
2.3. Term Analysis of the Ion Continuity Equation
 A term analysis of the ion continuity equation solved in the CMIT model has been undertaken to identify the relative importance of neutral winds, E × B drift, ambipolar diffusion, and neutral composition in generating the observed ionospheric oscillations. The term analysis processor, which was described by Lei et al. , is also given in Appendix A. Figure 5 shows, from top to bottom, the variations of electron density and terms: production-loss, trans_E × B, trans_wind, amb_diff, and op_dt at (32.5°N, 135°E) on 15 December.
 The term op_dt (Figure 5f), the rate of change of O+, stands for the net result of the terms in Figures 5b–5e. Hence the contributions of the terms production-loss, trans_E × B, trans_wind, and amb_diff to the term op_dt will help us to isolate the relative importance of these processes to the oscillations in the electron density. First we will look at the term trans_wind (Figure 5c) because the neutral winds could play an important role in generating these ionospheric oscillations, as shown in Figures 3–4. It can be seen clearly that the enhancements of op_dt below (or around) the F2 peak during 0200–0400, 0600–0800, 1200–1400, and 1600–1730 UT mostly arose from trans_wind, when the neutral winds are poleward (Figure 3); and the term trans_wind had negative values in the topside and positive ones in the bottomside ionosphere. During these periods, the F2 peak height hmF2 decreased due to the negative op_dt at the top side, whereas the F2 peak density NmF2 still increased because the poleward winds became much smaller below 250 km (Figure 3b) and resulted in a convergence of the field-aligned component of the ion motion. So, the vertical changes of neutral winds [Wang et al., 2008] are very important for the variations of ionospheric plasma. Our results explain well the approximate anticorrelation between the variations of NmF2 and hmF2 on storm day 15 December, as shown in Figure 2. These results and interpretation are consistent with the conclusions given by Lu et al. .
 Besides poleward winds, there were equatorward winds at some intervals (Figure 3). Equatorward winds can lift the ions from the bottomside to the topside ionosphere. Correspondingly, the term trans_wind was negative in the bottomside and positive in the topside ionosphere during these intervals: 0200–0300, 0600–0700, 0900–1100, and 1900–2200 UT (Figure 5d), and consequently the equatorward winds during these periods would contribute to an increase of hmF2.
 However, the term op_dt shows significantly positive values above 310 km during 0000–0300 UT and 0500–0700 UT when the term trans_wind was relatively small. This was caused by the effects of trans_E × B because the strong upward drifts during these periods transported plasma from the bottom side of the F2 peak to the topside ionosphere, leading to large enhancements of hmF2 (Figure 5c).
 It is evident that the striking enhancements of NmF2 around 0330 UT (Figure 2) were caused by a combination of the effects of the electric field and the neutral winds at all stations. Ionospheric plasma moved upward because of the stronger eastward electric field, and both NmF2 and hmF2 increased prior to ∼0200 UT. After that, the poleward winds with vertical shear led to decreases of hmF2 and a continuous increase of NmF2 until 0330 UT. This prominent peak of modeled NmF2 was 15 min later than it was in the observations at Okinawa, Yamagawa, and Kokubunji. This timing difference between the observations and simulations were caused mainly by the fact that the current CMIT 2.0 model uses the averaged solar wind speed during the event to calculate the propagation time from the L1 point where the solar wind is measured to the magnetopause (see Wang et al., submitted manuscript, 2008). When the solar wind speed was calculated using the observations on 15 December, the propagation time was shorter than the one that calculated from the averaged solar wind speed.
 The photochemical term, production-loss (Figure 5b), also contributed to the oscillations of op_dt during local daytime. For instance, NmF2 increases during 2100–2300 UT were accompanied with a rapid decrease of hmF2, both of which were associated with rapid change of production-loss. Finally, the term amb_diff, the ambipolar diffusion term, also made some contributions to the changes of electron density, especially during 0500–0800 and 0900–1100 UT (Figure 5e).
 Overall, the disturbed neutral winds were of primary importance in producing the ionospheric oscillations in this event, whereas electric fields, the ambipolar diffusion and photochemical processes only played a secondary role. In the next section, we will discuss the relationship between LSTIDs and the neutral winds perturbations associated with TADs.
3.1. Relation Between the Neutral Wind Perturbations and TADs
Figures 6 and 7show the maps of the vertical neutral wind flowing across a pressure surface at about 300 km during 0000–0430 UT and 0630–0830 UT, respectively. The vertical neutral winds across the pressure surface can track the modeled TADs better than horizontal winds because the background vertical neutral winds are much smaller than the background horizontal winds. The propagation of TAD leads to vertical neutral wind oscillations. Thus the propagation of TAD can be easily identified from the variations of vertical winds.
 A number of prominent features were seen from the maps in Figure 6. The first feature was that stronger TADs occurred in the night sectors (America and Africa) than in the day sectors (Asia and Australia). This was probably associated with less ion drag and thus smaller dissipative effects during nighttime. Another feature was that the TADs generated in the two hemispheric auroral zones propagated equatorward and then met near equator (see the maps between 0000 and 0200 UT); after that, they penetrated deep into the opposite hemispheres. The third feature was that the amplitudes of TADs were enhanced significantly from 0100 to 0330 UT, especially during the night. There are two possibilities to explain this: (1) Nonlinear wave-wave interactions could occur when the TADs propagating from high latitudes arrived coincidentally in the equatorial region. (2) Some potential energy has converted to kinetic energy during this period, allowing the amplitude of wave in the vertical winds to become larger; this can occur because wave energy is closely related to the sum of kinetic and potential energy of the atmosphere [Richmond, 1979b].
 The TADs in Figure 7 were much weaker during 0630–1100 UT than those during 0000–0430 UT. It is a surprise that the daytime TADs from 0630 to 0830 UT were stronger than those during nighttime. This feature was different from that shown in Figure 6. The nighttime TADs became dominant again during 0900–1100 UT. However, these nighttime TADs propagating from the Northern Hemisphere were stronger than those from the Southern Hemisphere.
 Next, we will discuss the relationship between the neutral wind perturbations and TADs over East Asia. As can be seen from Figures 6–7, there were four TAD events over East Asia: during 0–0430 UT the daytime TAD in the Northern Hemisphere disappeared rapidly (see maps during 0000–0130 UT), whereas the daytime TADs launched from the Southern Hemisphere penetrated to the high latitude of Northern Hemisphere and last for more than 4 h; another two equatorward TADs propagated from 0630 to 1000 UT and from 0900 to 1100 UT, respectively. Obviously, these four TADs were also seen in the neutral wind perturbations (Figure 4b). Note that the perturbation winds (blue color, Figure 4b) during 0800–1100 UT and 1100–1400 UT had a time delay from high to low-middle latitudes; however, they were not related to the TADs because we did not see the corresponding TAD signatures in the variations of vertical winds. The seen time delay in the perturbation winds during 0800–1100 UT and 1100–1400 UT is due to the cutoff of previous TADs. Therefore, the ionospheric oscillations seen in observations are not necessarily induced by the TADs and they can be caused by the disturbed winds due to the large scale thermospheric circulation during storm periods.
 GEONET is a dense and wide-area GPS network in Japan. The GPS array consists of about 1000 GPS receivers and provides GPS data at every 30 s. Time sequences of two-dimensional maps of TEC perturbations provide a powerful tool to identify LSTIDs, as Tsugawa et al. [2003, 2004, 2006] showed. The area from 124°E to 148°E longitude and from 24°N to 48°N latitude was divided into small pixels with a size of 0.15° latitude ×0.15° longitude. The TEC value for each pixel was an average of perturbations for all satellite receiver paths which crossed the pixel at 300 km altitudes. The perturbation components of TEC values from the line-of-sight TEC were derived by subtracting the large-scale trend of the TEC values that is defined by a 60-min running average.
Figure 8 shows a time sequence of two-dimensional maps of TEC perturbations over Japan during 0240–0430 UT. An east-west band structure of TEC enhancements traveled over Japan. A positive TEC perturbation band first occurred at 0240 UT over the southern coast of Japan, and a strong negative band appeared at the same time to the northern edge of this positive region. A phase front, characterized by the border between the positive and negative regions, can be clearly seen. This phase front traveled in a northward direction. It reached Hokkaido at 0320 UT and then passed over the northern coast of Japan 10 min later. The phase front traveled about 800 km within half an hour from 0300 to 0330 UT. Another phase front, which was on the southern border of the positive band over the southern coast of Japan, was seen at 0330 UT. These large-scale structures of TEC perturbations are the classic features of a LSTID.
 To investigate the temporal evolution of LSTID and to compare the observed LSTID with the simulations, we plotted the TEC perturbations in Figure 9 along the horizontal distance (HD) axis, which was perpendicular to the wavefronts of the LSTIDs. The HD axis was set along the meridional direction. This HD axis was defined from a point of reference (50°N, 136°E) and the horizontal length of the axis was 3000 km in a roughly southward direction. Tsugawa et al. [2003, 2004, 2006] gave details of this processor. The detrended TEC along the HD axis (Figure 9) shows that there were several southward LSTIDs associated with the geomagnetic storm from 0000 to 1000 UT, in addition to the clear northward LSTID shown in Figure 8.
 The CMIT simulations in Figures 6–7 indicated that three TADs passed over Japan during this period. These three TAD events corresponded to the distinct LSTIDs seen in the detrended TEC data in Figure 9: two southward LSTIDs during 0040–0130 UT (event “a”) and 0620–0800 UT (event “d”), and a northward LSTID between 0240 and 0420 UT (event “c”). The phase speeds of these three TADs estimated from the propagation of “vertical neutral winds” (Figures 6–7) were 777, 802, and −766 m/s, respectively, whereas they were about 763, 466, and −441∼−667 m/s in the TEC observations (the positive phase speed stands for the equatorward TAD/TID, while the negative phase speed is for the northward TAD/TID). One may notice that the phase speed of the observed northward LSTID varied greatly from 667 m/s during 0240–0300 UT to 441 m/s during 0300–0420 UT. However, the model simulations did not show so significant change of the phase speeds. The variations of TEC from several GPS receivers located over Japan (yssk, tskb, ccjm, and guam) and around geomagnetic conjugate latitudes over Australia (jab1, alic, and cedu) indicated that this northward propagating TAD over Japan was also modulated by the northern movement of equatorial ionization anomaly (figure not shown). Finally, we need point out that the LSTID during 0140–0220 UT (event “b”) seen in the detrended TEC data was not reproduced by the CMIT model.
 The model tends to overestimate the phase speed of TADs during 0240–0420 UT (event “c”) and during 0620–0800 UT (event “d”), which may indicate the overestimation of the high-latitude energy input in the model or differences in the propagation conditions between the model and the data. The faster phase speeds could explain why the modeled peaks of NmF2 during 0400–1600 UT appeared earlier than the observations, as illustrated in Figure 2.
Shiokawa et al.  compared the TADs calculated from the NCAR-TIEGCM, using the assimilative mapping of ionospheric electrodynamics (AMIE) inputs, with those from observations over Japan during the 31 March 2001 storm. They showed that the phase speed of a TAD from the TIEGCM was 1100 m/s, which was much faster than the corresponding observed phase speeds of 370–640 m/s. In our case, the TAD speeds calculated from the CMIT model were smaller than those reported by Shiokawa et al.  and much closer to the observed phase speeds. As shown below, the Joule heating in this event was greater than that in their event. The differences of the modeled phase speeds in these two storm events could also be associated with the local-time dependence of the TAD propagation [Shiokawa et al., 2007] because the background themosphere for the propagation condition of TAD is different at different local times: the TADs propagated to Japan in the midnight sector in the case of Shiokawa et al., whereas they traveled to Japan during daytime in the present case.
3.3. Relation Between Joule Heating and TADs
 It is interesting to note that a northward propagating LSTID was detected during daytime (event “c” in Figure 9) by the Japanese GPS network. Though Tsugawa et al.  found one northwardly propagation LSTID event during 45 months from April 1999 to December 2002, there have been quite few reports on such northward LSTIDs over Japan. Also, this northward LSTID was also captured by the CMIT model, so we know that this daytime TAD was launched from high latitudes in the Southern Hemisphere and then penetrated into middle and high latitudes of the Northern Hemisphere. As mentioned before, the daytime TAD in the Northern Hemisphere in the East Asian sector decayed rapidly from 0000 to 0130 UT (Figure 6). This strong hemispheric asymmetry in TAD propagation may be associated with the asymmetry of the energy input in the high-latitude upper atmosphere or the nature of the wave propagation from the source. Further investigation is required.
Figure 10 shows the hemispheric-integrated Joule heating calculated from the CMIT simulations for two hemispheres on 14 and 15 December. The Joule heating in both hemispheres increased significantly after the onset of the storm (at about 1414 UT on 14 December) and then decreased gradually after the midday of 15 December. The temporal variation of Joule heating showed a good correlation with the variation of solar wind (Figure 1, and Figure 1 of Lei et al. ). The Joule heating was larger in the Southern (summer) Hemisphere than in the Northern (winter) Hemisphere from 2200 UT on 14 December to 1500 UT on 15 December. The greater change of Joule heating in the Southern Hemisphere may account for the penetration of the TAD generated in the Southern Hemispheric auroral region into the middle latitudes of the Northern Hemisphere over Japan.
Lu et al.  simulated the response of the ionosphere during the geomagnetic storm event of 10 January 1997, using the TIEGCM with the high-latitude energy inputs from AMIE. Their simulation results showed that several LSTIDs propagated to the equator from the northern auroral oval and penetrated deep into the Southern Hemisphere, while no TIDs were generated in the southern auroral zone. This situation also can be found in the simulations for the November 1993 storm by Emery et al. . They explained this by the larger dissipative effects by ion drag in the sunlit summer hemisphere. However, the CMIT model did show that LSTIDs can be generated in the summer hemisphere and even penetrate into the opposite hemisphere. This probably represents a different thermospheric response in different storm cases. Another possible cause for this discrepancy is that much less data are assimilated into AMIE in the Southern Hemisphere than in the Northern Hemisphere (G. Lu, private communication, 2007).
 The strongest Joule heating around 0 UT on 15 December generated two TADs in the two hemispheres (Figure 6). However, there is not always a direct one-to-one correspondence between the generation of the TADs and the high-latitude energy input. For instance, the Joule heating was relatively small during 0600–0700 UT when a TAD was generated in the northern auroral zone (Figure 7). A secondary peak of Joule heating around 0200 UT did not produce a clear TAD signature. Therefore, the relation between the generation of the TADs and high-latitude energy input must be complex. The generation of TADs in the thermosphere is related not only to the total energy input at high latitudes but also the precondition of thermosphere [Shiokawa et al., 2007] and to the high-latitude convection pattern [Balthazor and Moffett, 1999].
 We have used the Coupled Magnetosphere Ionosphere Thermosphere (CMIT) 2.0 model to investigate quasiperiodic ionospheric oscillations that were observed by the ionosonde network in Japan during the 15 December 2006 geomagnetic storm. The main conclusions of this study are summarized as follows.
 1. The CMIT model simulations have reproduced the main characteristics of the ionospheric oscillations observed by the ionosonde stations at Okinawa, Yamagawa, Kokubunji, and Wakkanai.
 2. The term analysis showed that the disturbed neutral winds were of primary importance in producing the ionospheric oscillations in this storm event. Furthermore, neutral wind perturbations were also related to the disturbed winds due to the large-scale storm-induced thermospheric circulation in addition to the TAD-associated winds.
 3. A prominent northward propagating LSTID during daytime has been observed by the GEONET TEC data. Meanwhile, this northward LSTID between around 0240 and 0420 UT was also captured by the CMIT model. The model suggested that it was generated in the southern hemisphere, and then propagated from the high latitudes of the Southern Hemisphere into the Northern Hemisphere over Japan.
 4. Two southward LSTIDs were seen by the GEONET GPS network over Japan during 0040–0130 UT and 0620–0800 UT in addition to the northward LSTID that occurred between 0240 and 0420 UT. These waves were also reproduced by the CMIT model. However, the model missed the LSTID that was observed between 0140 and 0220 UT.
 5. The phase speed for the TAD occurred during 0040–0130 UT that was calculated by the CMIT was comparable to that from the observations. However, the model calculated faster phase speeds for the southward LSTID during 0620–0800 UT and the northward LSTID during 0240–0420 UT. This may be related with the overestimation of Joule heating in the model or differences in the propagation conditions between the model and the data.
 6. The strong hemispheric asymmetry of the TAD propagation during 0–0400 UT may be associated with the hemispheric asymmetry of the change of Joule heating at high latitude. However, the relation between the occurrence of TADs and Joule heating has been shown to be rather complicated.
 Ionospheric F region ion density changes dominated by O+ can be described by the O+ continuity equation:
where NO+ stands for the O+ concentration, βO+ is the loss coefficient, and the terms qO+ and βO+NO+ represent the rate of production and loss of O+, respectively. The last term in the right-hand side represents transport effects including the transport caused by electric fields, neutral winds, and ambipolar diffusion.
 We denote terms , qO+ and βO+NO+ as op_dt, production, and loss. On the other hand, the transport term −∇ · (NO+O+) is divided into three parts: transport by electric fields, neutral winds, and ambipolar diffusion. They are denoted as trans_E × B, trans_wind, and amb_diff, respectively. Note that the change of the sign of density gradient change the sign of the transport term, even although the velocity has not gradient. From O+ continuity equation (A1), we thus have
The terms in equation (A2) can be use to investigate the cause of the changes of ionospheric plasma density. More details of the term analysis can be found in the work of Lei et al. .
 We thank A. D. Richmond and H.-L. Liu for their insightful discussions. This material is based upon work supported in part by the Center for Integrated Space Weather Modeling (CISM) which is funded by the STC Program of the National Science Foundation under agreement ATM-0120950. The GPS data from GEONET are provided by the Geographical Survey Institute, Japan. The ionosonde data were provided from WDC, National Institute of Information and Communications Technology, Tokyo. The National Center for Atmospheric Research (NCAR) is supported by the National Science Foundation.
 Wolfgang Baumjohann thanks Ivan Kutiev and another reviewer for their assistance in evaluating this paper.