Data from three Defense Meteorological Satellite Program (DMSP) satellites (i.e., F16, F17, and F18) are used in this study. They flew in sun-synchronous orbits at an altitude of ∼840 km with an inclination angle of ∼98.8° (Rich and Hairston, 1994) primarily in the dawn–dusk direction during the October 8–9, 2012, geomagnetic storm. The cross-track ion drift (V_y) is measured by using the onboard Special Sensor for Ions, Electrons, and Scintillation (SSIES), which has a temporal resolution of 1 s. In this study, the data with the quality flag of 1 (i.e., most reliable) are used. A linear baseline correction is applied to the original V_y data to remove some co-rotation effects and to ensure V_y is zero at 45° |magnetic latitude| (|MLAT|). Afterward, a 13-point sliding window is applied to the corrected data in order to reduce very-high-frequency fluctuations and extract the large-scale ion convection at high-latitude regions.

The Gravity Field and Steady-State Ocean Circulation Explorer (GOCE) satellite was launched on 17 March 2009. The satellite was a polar-orbiting satellite (inclination angle: 96.5°) and flew in a near-circular orbit with an altitude from 250 km to 280 km (Bruinsma et al., 2014). The solar local times (SLTs) of the ascending and descending nodes of the GOCE were about 19 and 07 h, respectively. The neutral mass density (hereafter, neutral density for simplicity) was measured by using an onboard accelerometer with a temporal resolution of 10 s (Doornbos et al., 2014). To reduce the variation caused by the satellite orbit altitude change, the neutral density data are normalized to a constant altitude of 270 km using the NRLMSIS 2.0 model (Emmert et al., 2021). More details of the density normalization technique can be found in Bruinsma et al. (2006).

The AMPERE high-latitude FAC densities are derived from the horizontal magnetic field perturbations measured by 66 Iridium satellites which are distributed along six longitudinally equally spaced orbital planes (Anderson et al., 2002). Each Iridium satellite flies in a near-polar orbit at an altitude of 780 km with an orbital period of 104 min. The radial current components are calculated by fitting the magnetic perturbations measured within a 10-min time window using spherical cap harmonic basis functions (Waters et al., 2001; Green et al., 2006; Waters et al., 2020). Specifically, the patterns are fitted with a longitude order of 5 and latitude order of 20 (i.e., 3°) between colatitude 0° and 60° in the Altitude-Adjusted Corrected Geomagnetic (AACGM) coordinates (Baker and Wing, 1989). The temporal resolution of the FAC data is up to 2 min, and the spatial resolution of the FAC data is 1° in MLAT and 1 h in the magnetic local time (MLT) (Anderson et al., 2014). In this study, AMPERE FAC densities with a magnitude below 0.2 μ A/m² mainly distributed in the polar cap region and below 50° |MLAT| are removed, defined by being smaller than the noise level. In addition, due to differences in the geomagnetic field magnitude and the curvature of the magnetic field, the FAC pattern is mapped from 780 km to the APEX reference height of 110 km based on a factor of 1.343. Using a bilinear interpolation, the 10-min resolution AMPERE FAC patterns are spatially interpolated to the electrodynamic solver’s grids and then temporally interpolated into a 2 s cadence at which the simulation is performed using linear interpolation.

To determine how well the model output represents the ionospheric ion convection, a data–model comparison of the cross-track ion drift V_y has been conducted along DMSP trajectories during the focus period. As shown in Figure 3, six examples from DMSP F16-18 and the corresponding GITM runs are compared with the top and bottom panels representing the Northern and Southern hemispheres, respectively. For each plot, the black dots show the observed cross-track ion drift, and the blue and red curves represent the empirical model-based run 1 (W05) and realistic pattern-driven run 2 (FAC-driven and AMIE electron precipitation), respectively. In general, the cross-track V_y from both GITM runs is consistent with the DMSP results. When comparing the two sets of runs, the FAC-driven V_y from run 2 can better capture the realistic ion drifts in high latitudes than the W05-based results from run 1, especially inside the auroral zones.

To investigate how GITM simulations capture the high-latitude ion drifts V_y from a statistical perspective, Figure 4 compares the simulated and measured cross-track ion drift at the region poleward of 45° |MLAT| from all DMSP trajectories during the focused period for runs 1 and 2. Linear fitting is performed on the data, and the slope (k) and y-intercept (b) are calculated. In addition, the Pearson correlation coefficient (r) and the root mean square error (RMSE) are also calculated. Basically, the cross-track ion drifts V_y from both GITM simulations have smaller magnitudes than the observed cross-track V_y since the slope of the best-fit line of the fitted line is much smaller than 1. This underestimation in the simulation results could be due to the underestimation of high-latitude forcing such as the FAC magnitude. This is a known property of AMPERE FAC data in the dusk-side R2 current. However, the slope of the best-fit line is slightly higher in run 2 than in run 1. Meanwhile, compared with run 1, run 2 shows a greater correlation (e.g., for DMSP F16 run 2 = 0.81 vs. run 1 = 0.68) and a smaller RMSE between the simulated and observed ion drifts (336.20 compared to 426.33) than run 1. Overall, Figure 4 indicates that the FAC-driven GITM simulation can better reproduce the high-latitude ion drift measured by DMSP satellites than the W05.

The data–model comparisons of high-latitude ion convection given in Figure 3 and Figure 4 show a general consistency between the measured and simulated ion drift on the dawn and dusk sides, especially for run 2. Moreover, we would like to see how GITM simulations perform in response to the ion convection equatorial boundary. Figure 5 shows a scatterplot of the ion convection equatorial boundary in the GLAT for both hemispheres measured by DMSP F16 versus those from runs 1 and 2 during the focused storm period. In general, the simulations reproduce the DMSP-measured equatorward boundary latitudes well, with a better representation in run 2. Specifically, the correlation coefficients to DMSP F16 are 0.81 in the NH (Figure 5A) and 0.91 in the SH (Figure 5C) for run 1 and 0.92 in the NH (Figure 5B) and 0.95 in the SH (Figure 5D) for run 2. Moreover, as shown in Figures 5A, C, W05 has a clear underestimation of the convection expansion below 60° for the NH and at all the latitudes for the SH. This underestimation of the ion convection equatorial boundary may have some effects on other electrodynamical processes such as the ion-neutral interactions in that region. Overall, both runs show good agreement with the DMSP-observed convection equatorial boundary. The comparison between runs 1 and 2 is to give the community a rough idea about how well the FAC-driven technique could improve high-latitude electric field specification which is also a justification for using the FAC-driven technique in this study. For later scientific studies in Section 4.2 and Section 4.3, we would like to focus on the GITM run 2 since it has advantages over run 1 as discussed. A summary of the statistical parameters from the data–model comparisons of the two runs can be found in Table 2.

Figure 6 shows the temporal variations of the cross-polar-cap potential (CPCP, top panel) and hemispheric power (HP, bottom panel) during October 8–9, 2012, from run 1 (left) and run 2 (right). The vertical dashed line marks the time when IMF B_y is reversed (00:20 UT on October 9) during the storm main phase we focused on in this study. All the red and blue lines correspond to the NH and SH, respectively. Following Hong et al. (2021), the asymmetry index (AI) is used to quantify the temporal changes of IHA for a given quantity: A I = 2 × Y N H − Y S H Y N H + Y S H × 100 % , (3) where Y_NH and Y_SH stand for the quantity in the NH and SH, respectively.

As shown in Figure 6A, CPCPs from run 1 are almost the same in the NH and SH during the focused period and the CPCP stays around 145 kV. As for run 2, significant asymmetries between the NH and SH can be seen in Figure 6B even for the recovery phase (e.g., after 12 UT on October 9). An evident asymmetry index (AI up to −50%, see Figures 6C, D) occurs in the CPCP during the focused period, especially before IMF B_y is reversed. In addition, the CPCP also exhibits more pronounced temporal variations than run 1. Similarly, as shown in the bottom two panels of Figures 6E, F, the hemispheric power in run 2 exhibits more dynamic variations and greater IHA than in run 1. The AIs can reach 75% and −22% in run 2 (Figure 6H) and run 1 (Figure 6G), respectively. Different from the CPCP, the HP tends to have greater AI in run 2 when the IMF B_y is negative (after 00:20 UT on October 9). Additionally, the HP in the NH is generally larger than that in the SH in run 2 while the opposite is true in run 1. This is probably due to the fact that the empirical model is mainly dependent on the given IMF, solar wind, and input seasonal conditions. Another reason is that, as described in Section 2.2, the AMIE patterns have better data coverage from the NH than that in the SH.

To our knowledge, the IHAs in high-latitude forcing could be due to the combined action of the season-related dipole tilt and the non-zero B_y component, which contribute to the asymmetric interaction in the dayside reconnection (Reistad et al., 2021). The well-known Dungey cycle suggests that the distribution of the open magnetic field flux in the two hemispheres should be the same. However, observations showed that the open magnetic flux in the polar cap is often distributed differently between the Northern and Southern hemispheres. Global auroral images from Polar and IMAGE satellites show different polar cap shapes, indicating hemispherically asymmetric distribution of open flux (Laundal and Østgaard, 2009). This is particularly true when there is an east–west component of the IMF (IMF B_y) (Tenfjord et al., 2015). The lobe reconnection due to IMF B_y causes the magnetic flux to build up asymmetrically (Lu et al., 1994). For example, the responding speed to the solar wind conditions in the two hemispheres can be different, which can result in different temporal variations of the magnetic flux between the two hemispheres (Milan et al., 2020). Therefore, this phase mismatch may result in some instantaneous asymmetry of the CPCP while it should be reduced significantly when the response in both hemispheres is fully developed. In addition to the reconnection for the IMF B_y, the field line potential drops across the hemispheres, and the asymmetries in the current systems produced by differing ionospheric conductivities and neutral dynamos in two polar regions contribute to the possible asymmetry in the CPCP as well. Data–model comparisons are still absolutely crucial to validate the simulation results. While the GITM simulations are able to capture the fundamental features of the ion drifts at high latitudes and the convection boundaries at mid-latitudes, the potential strong asymmetry of the CPCP presented in our simulation has yet to be corroborated with realistic observations. We will make careful comparisons with SuperDARN and other measurements in the follow-on study.

An enhanced geomagnetic storm can lead to large perturbations in the high-latitude ion convection and then extend to lower latitudes. From Section 4.1.2, we have demonstrated that the ion convection equatorial boundary from DMSP observations and GITM simulations shows a fairly good agreement on both the dawn and dusk sides. To have a detailed picture of the UT variations during this storm, Figures 7A, B directly show the temporal variations of DMSP F16-measured ion convection boundary in GLAT on dawn and dusk sides during the focused IMF B_y reversed period (18 UT on October 8–10 UT on October 9). Over this period, the NH dawn-side convection boundary (red dashed line) only undergoes a small variation (<5°) around 63° N GLAT, while the dusk-side convection boundary can expand to as far as 45° N (red solid line). As for the SH, both the dawn and dusk sides expand to lower GLATs during this interval of time, with the observed lowest boundary at 45° S and 40° S, respectively. One of the most interesting features is that these two sides reach their lowest |GLAT| under opposite IMF B_y conditions: the dawn side first expands under positive B_y and then contracts during negative B_y (blue dashed line), while the dusk side shows the opposite (blue solid line). This result could be explained by the anti-correlated responses between the dawn and dusk sides to B_y (e.g., Cowley, 1981; Reiff and Burch, 1985; Kabin et al., 2003; Tenfjord et al., 2015), the SH dawn-side convection cell is much stronger than the dusk-side cell under positive B_y, while the dusk side is more favorable under negative B_y in the SH.

In terms of the ion convection boundary comparisons between the two hemispheres, the temporal expansion in the NH dusk side (7A) follows a similar tendency to changes in the SH dawn side (7B), especially under positive IMF B_y. This could be the result of reversed B_y responses in the two hemispheres (Cousins and Shepherd, 2010), a stronger convection cell on average expands to lower latitudes. On the other hand, in contrast to the expansion in the SH dusk side (7b), the NH dawn-side convection expansion (7a) seems to be less important, implying a relatively stable convection cell on the dawn side. Moreover, this response is much stronger in the SH than in the NH, suggesting a stronger B_y dependence of the two cells in the SH (Burch et al., 1985). A possible explanation to the dawn–dusk asymmetry and IHA in the expansion is that the ion convection is subject to asymmetries in FACs and ionospheric conductivities. However, these line plots represent a limited view of ion convection pattern variation at two fixed LTs. The aforementioned hemispheric asymmetries between the NH and SH on both dusk and dawn sides will be further discussed in the following section.

To better understand the spatiotemporal variations of the ion convection patterns, Figures 7C–F present polar-view snapshots of the 2-D convection patterns from realistic pattern-driven GITM run 2 in both hemispheres under positive B_y at 22 UT October 8 (T1, left) and negative B_y at 04 UT October 9 (T2, right) from run 2. It can be seen in Figures 7E, F that the SH ion convection strengths and their expansions have obvious responses to IMF B_y. From T1 IMF B_y positive to T2 B_y negative, the dawn-side cell decreased from a maxima value of 42 kV (e) to 33 kV (f) associated with the dawn-side boundary contracted from ∼45^oS (b and e, blue triangle) to 67^oS (b and f, blue triangle). Similarly, for the dusk side, the cell with a minimum value −43 kV (e) with respect to −69 kV (f) leads to an expanded dusk-side boundary to much lower at ∼ 42^oS (b and f, blue circle). However, as shown in Figures 7C, D, a stronger dusk-cell with minima value −51 kV (d) corresponds to a shrinking convection boundary (a and d, red circle). Furthermore, the dawn-cell increased from 29 kV (c) to 40 kV (d) with subtle changes in the boundary location. Again, the NH displays a clear stretch of the dawn-side cell in the noon–midnight direction instead of an equatorward expansion, which could explain the static phenomena along the DMSP F16 trajectory on the NH dawn side as shown in Figure 7A (red dashed line). This result suggests that it might be insufficient to determine the storm period ion convection expansion based on only the dawn and dusk LTs. Overall, the ion convection pattern simulated by the GITM for these two specific periods is in general agreement with the observed convection boundary expansions. Moreover, a stronger convection cell does not necessarily associate with a lower equatorial boundary.

In addition, we also conducted data–model comparisons of the equatorial boundary expansion at specific times. As shown in both the linear and polar plots, the red and blue triangles refer to the dawn side for the NH and SH, respectively. Similarly, the red and blue circles refer to the NH and SH but for the dusk side. The snapshot times T1 and T2 represent the corresponding positive B_y case at ∼ 22 UT on October 8 and negative B_y case ∼04 UT on October 9, respectively, which are marked in the left line plots. The subscripts N and S refer to the NH and SH, respectively. The boundary expansion features in the 2D contour from FAC-driven GITM run 2 are in a good agreement with the line plots from the DMSP F16 observations. For example, the blue triangle T_2S (Figure 7B, dawn side) almost overlaps the expansion of the dawn-side potential cell (Figure 7F, blue color) at around 67^oS GLAT.

To identify the possible mechanisms for the observed IHAs in the ion convection patterns, as well as their equatorial boundaries, a separation of the FAC and auroral electron precipitation (associated with the ionospheric conductance) is needed. In reality, the high-latitude FAC and auroral precipitation are connected and respond simultaneously to the changes in geophysical conditions, so it is impossible to completely separate the effects caused by these two variables. However, numerical simulations could help solve this issue by replacing the real forcing with symmetric patterns. We first examine the GITM run 2 (simulation setups in Table 1), which the high-latitude forcing is based on the FAC-driven potential and AMIE electron precipitation patterns. In GITM run 3, the AMIE electron precipitation patterns in the SH are replaced by the mirror precipitation patterns in the magnetic coordinate from the NH accordingly. Therefore, the differences in the SH between run 2 and run 3 illustrate the contributions from the auroral precipitation. Figure 8 shows the Southern hemispheric FAC-driven convection patterns and corresponding auroral energy flux from GITM run 2, run 3, and their differences at 04 UT on October 9 as an example. Specifically, due to the asymmetric electron precipitation, the auroral electron energy flux between the NH and SH is significantly different in both distributions and magnitudes: the SH has a continuous and complete aurora arc structure (Figure 8D) while the NH is more like a discrete aurora (Figure 8E). As shown in Figure 8F, the maxima difference in energy flux can vary from −45% to 55%. This result could be explained by the larger hemispheric power captured at that specific time (see Figure 6F); even the hemispheric power in the NH can be overwhelmingly superior to that in the SH during the focused period. However, the SH and NH electron energy flux-based ion convection (Figures 8A, B) illustrates quite similar distributions, as well as the maxima and minima values for the dawn (33 vs. 34 kV) and dusk (−69 vs. −67 kV) cells. Compared to Figure 8C, the differences caused by the asymmetric auroral precipitation can be ignored at this specific time. Moreover, by comparing the convection contour lines, one can find that the distributions, especially the equatorial boundaries of the convection, are more likely determined by the FAC patterns (R2 FAC edge) while its strength may also depend on the magnitude of the ionospheric conductance-associated precipitation. With this discussion, the effects from the FAC and auroral precipitation can be roughly isolated. It should be noted that this conclusion depends critically on the assumption of the FAC-driven procedure. Similar plots for all the UT times during the focus period indicate the same conclusion as the example time shown in Figure 8.