We present Cluster multisatellite observations of accelerated electrons in the near-Earth magnetotail associated with substorms. We found that the hardest electron energy spectra appear in the earliest stage of substorm expansion in the near-Earth tail region and that they gradually become softer during the events. Enhancement of the high-energy electron flux occurs generally associated with the bulk acceleration of ions (fast flow) and electrons. It is also shown that the high-energy electrons sometimes show preferential perpendicular acceleration associated with the temporal enhancement of the normal component of the magnetic field, and then the anisotropic distribution quickly becomes isotropic. During the dipolarization interval, in which no convection signature is observed, perpendicular flux drops to less than the initial value, and the parallel flux is more than the perpendicular flux. The results suggest that the electron acceleration mechanism is mostly consistent with adiabatic betatron acceleration, while Fermi acceleration is not clear in the high-energy part. The effect of the pitch angle scattering is also important. The dispersive signature of the high-energy electron flux indicates fast dawnward drift loss, namely, the three-dimensional effect of the limited plasma acceleration region.
 The origin of high-energy particles with energy beyond tens of kiloelectron volts in the magnetosphere has been discussed for decades. While many of the particles come directly from the cosmic or solar radiation penetrating deep into the magnetotail and are trapped in the radiation belts, there also exist high-energy particles in the magnetotail, supposedly accelerated inside the magnetosphere.
 Substorms are known to be the significant phenomena during which drastic increase of high-energy particle flux often occur. Observations of sudden increase of the accelerated particle fluxes are usually initially observed near the midnight around the geosynchronous orbit associated with substorms. They are typically called injection [e.g., Arnoldy and Chan, 1969; Lezniak and Winckler, 1970; DeForest and McIlwain, 1971], and they may extend to the magnetotail 7–8 RE from the Earth [e.g., Lopez et al., 1990]. The injection is phenomenologically well known. Recent numerical simulations also refer to the importance of injected particles showing that some of them are further accelerated to the relativistic energy range [Summers et al., 1998; Horne et al., 2003; Katoh and Omura, 2007]. However, the effect of the radial diffusion by the ULF (Pc-5) waves mostly inside the inner magnetosphere are considered in storm-time activities [e.g., Li et al., 2005] and the signature of the azimuthal drift path in the inner magnetosphere has been studied [e.g., Ejiri, 1978; Reeves et al., 1991; Takahashi et al., 1997]. Hence, particle acceleration mechanisms and the three-dimensional particle flux transport in the magnetotail and its relation to the substorm activities are not yet clear.
 Disruption of the tail electric current [e.g., Lui et al., 1988; Lui, 1996] associated with substorms is considered to be initiated in the magnetotail 7–8 RE from the Earth, and some related acceleration mechanisms are theoretically considered [Lopez et al., 1990]. However, it is not clear if these ideas are valid, since the acceleration in the current disruption region has not yet been reported observationally. On the other hand, Li et al. [1998, 2003] reproduced injection-related signatures such as dispersionless injections and drift echoes by applying an earthward propagating electric field pulse. Smets et al.  and Wu et al.  examined observational signatures of the high-energy electron distributions and discussed the relative efficiency of the adiabatic Fermi and betatron accelerations in time and in space.
 Recent theoretical and numerical studies have revealed that kinetic processes in magnetic reconnection in the magnetotail can also generate high-energy particles. Near-Earth neutral lines are considered to form in the region 20–30 RE down tail from the Earth [Nishida and Nagayama, 1973; Hones and Schindler, 1979; Nagai et al., 1998; Baumjohann et al., 1999; Miyashita et al., 2009] associated with substorm onsets. Hoshino  suggests that relativistic electrons can be generated by a combination of the surfing acceleration in the electric field potential structure around the X line and the nonadiabatic gradient and curvature drifts in the piled-up outflow flux region. This prediction has been supported by some observations [Imada et al., 2007]. A parallel electric field in the presence of a guide magnetic field may also accelerate electrons drastically [Pritchett and Coroniti, 2004; Drake et al., 2005]. Formation of multiple flux rope structures with multiple X lines may be able to generate relativistic electrons by Fermi acceleration [Drake et al., 2006] associated with the coalescence of the island structures [Pritchett, 2008]. Chen et al.  have reported that the island structures of the flux ropes are related to the enhancement of the high-energy electron fluxes. Retinò et al.  have shown the electron acceleration signature within the thin current sheet around an X line and in a flux rope structure. However, it is considered that the accelerated high-energy electrons in this region might drift significantly dawnward [Birn et al., 1998, 2004] and will not be transported to the active region of substorms in the near-Earth magnetotail. Vogiatzis et al.  reported an increase of the high-energy electron flux in the recovery phase of a substorm in the magnetotail, and interpreted the observation by proposing that the high-energy electrons have drifted from the near-Earth duskward sector where the accelerated particles were generated by the current disruption process.
 In order to understand global acceleration mechanisms and particle flux transport in the magnetosphere, there is now a good opportunity to study the physics of this region using recent Cluster multisatellite observations around 10 RE away from the Earth in the magnetotail. In this paper, we examine acceleration signatures of electrons and their temporal and spatial evolution during two substorms obtained by satellites and ground magnetograms, and discuss the physical processes.
 In this paper, we use plasma, magnetic field and electric field data obtained by the four Cluster satellites in the magnetotail. Magnetic field data are obtained from the Fluxgate Magnetometer (FGM) [Balogh et al., 2001]. High-resolution data are sampled at 22 or 67 Hz depending on the telemetry mode. In order to adjust to the times of each electron energy sweep, these data are linearly interpolated. In the summary plots shown in this paper, they are integrated with the time resolution of about 4 s. Electric field data are obtained from the Electric Field and Wave instrument (EFW) [Gustafsson et al., 2001]. Two-dimensional high-resolution data are sampled at 25 or 450 Hz in the spin plane of the satellites (Despun System Inverted (DSI) coordinate system, which is close to the GSE coordinate system). In the summary plots, it is also averaged over the spin period of about 4 s.
 Low- to mid-energy electron data are obtained from the Plasma Electron and Current Experiment (PEACE) instruments [Johnstone et al., 1997]. They consist of two instruments, High Energy Electron Analyser (HEEA) and Low Energy Electron Analyser (LEEA) which are located on opposite sides of the spin plane of the satellites. In combination their energy range is from 0.59 eV to 26.4 keV. We use in this paper Pitch Angle Distribution (PAD) data. The original PAD data are obtained with reference to magnetic field vector sampled once per spin onboard the spacecraft, but for the analysis presented here, we recalculate accurate pitch angles for each angular bin using the high-resolution magnetic field data together with the high-quality ground calibration. The PAD comprises two data sets per sensor, each sampled in ∼125 ms collected half a spin apart when the sensor look directions contain the field direction.
 High-energy electron data are obtained from the Research with Adaptive Particle Imaging Detectors (RAPID) instrument [Wilken et al., 2001] for the energy range 37.3–406.5 keV. In this study we use omnidirectional electron data (ESPCT6) with the time resolution of one spin (∼4 s), and the three-dimensional data (E3DD or L3DD) containing eight or two energy samples, respectively.
 Ion moment data are obtained from the Comprehensive Ion Spectrometer (CIS) [Rème et al., 2001]. Proton data obtained from the Composition and Distribution Function analyser (CODIF) instrument are used for Satellite 4 (CL4) with the time resolution of 8 s (two spins). Ion data without mass spectrometry from CL1 and CL3 are obtained from the Hot Ion Analyzer (HIA) instrument with the time resolution of 12 s (three spins).
 Moment and magnetic field data are presented in the Geocentric Solar Magnetospheric (GSM) coordinate system if not specified otherwise.
3.1. Event of 27 October 2007
 First, we show an event observed in the near-Earth magnetotail at Xgsm ∼ −10 RE. We present in Figure 1 a summary plot of an event on 27 October 2007. This is a very fortunate event in which the satellites were located just around the neutral sheet at the timing of the onset. Total intensity and three components of the magnetic field, x and y components of the electric field in DSI coordinates, ion (CL1) or proton (CL4) density and temperature and the x component of the velocity are shown from top to bottom. Black, red, and blue lines show the CL1, CL2, and CL4 data, respectively. At the bottom, position data labels for CL4 is shown as well as the time (UT). The locations of the three satellites during the event (0900–0930 UT) in xz (Figure 2, top) and xy (Figure 2, bottom) planes are shown in Figure 2. All the satellites were moving from the north to the south and from dawn to dusk in the magnetotail near the neutral sheet roughly in the same xy plane within the separation of 10000 km. CL1 and CL2 were located at [−8336, −194, −306] km and [−4793, 10170, 988] km relative to CL4, respectively. Namely, CL4 was on the earthward side of CL1, while CL2 was on the dusk side of the other satellites. Note that CL3 (not shown) was very close to CL4 during the interval within 40 km separation, and did not show any typical difference from CL4 data in the macroscopic view.
 A fast earthward flow was first detected by CL1 and CL4 at 0906:30 UT, and it continued until 0909:30 UT. From the timing difference of the associated temporal enhancement of By and Bz among the satellites, we roughly estimate that the earthward propagation speed about 600 km s−1. During this interval there was a large amplitude oscillation of Bx, followed by weak Bx. After that, continuous large Bz was observed by CL4, namely, a dipolarization only in the near-Earth region. This dipolarization at CL4 location during the interval is not so steady, and becomes more dipolar later. Smaller perturbation of Vx and the magnetic field were again observed at around 0914 UT and 0921 UT. Tailward expansion of the dipolarization region and the reconfiguration of the magnetotail during the event is further described by Nakamura et al. .
 In order to examine substorm activities during the interval, Figure 3 shows ground magnetic field data for the event. Figure 3 (top) shows the AL index. Following a previous substorm event at about 0830 UT, the AL index began to decrease again at 0906 UT. A Pi2 onset was detected at CCNV (magnetic latitude = 45.35°, magnetic longitude = 304.84°) at 0905 UT (Figure 3, bottom). No clear enhancement of the positive bay was observed here, associated with the onset (Figure 3, middle). It is also found that the tail activity at 0914 and 0921 UT mentioned above was associated with small and then large evolution of the AL index, respectively. Azimuthal evolution of the substorm activity between 0904:18 and 0908:36 UT is examined using signatures of the auroral brightening observed by the Polar UVI instrument shown in Figure 4. Local time is indicated in MLT in the first panel (Figure 4, top left). Location of the Cluster satellites is indicated in the fifth panel (Figure 4, bottom left). It is clearly seen that the initial brightening was observed on the dusk side, between 20–21 MLT at 0904:55 UT, which was also duskward of the satellite locations. After that, we can also see that the second brightening initiated at 0907:22 UT is between 21–22 MLT, and this brightening included the foot point of the satellites. Prior to the initial fast flow, CL2 observed a large magnetic field oscillation at 0904 UT (no ion data for the satellite) as is shown in Figure 1. Considering the facts that CL2 was duskward of the other satellites, and that the initial auroral brightening was observed duskward of the satellite location, it is likely that only CL2 observed the initial, spatially limited magnetospheric perturbation. Then the activity extended toward the midnight as indicated by the second brightening, so that all the satellites observed magnetospheric disturbances. This result is consistent with the evolution of the geomagnetic activity shown in the work of Nakamura et al. [2009, Figures 1 and 2].
Figure 5 shows energy-time spectra between 0904 and 0918 UT, specifically earthward and tailward ion fluxes from CL1, proton fluxes from CL4, and omnidirectional electron fluxes from the both satellites. The start time of the fast flow is marked by a vertical dotted line. During the fast earthward flow interval, the CIS ion data clearly show an intensification and also an energization of the earthward fluxes. At times the earthward fluxes are beyond the highest-energy range of the instruments, delimited by dashed vertical lines. Despite the different duration of the fast flows measured at CL1 and CL4, both satellites clearly observed a continuous earthward energetic component until 0909 UT. Between 0907:50 and 0908:20 UT, average Ey at CL1 is 9.7 mV m−1, and Vx is estimated to be about 1990 km s−1 from Ey/Bz. During this time, we see that electrons are accelerated simultaneously with the ions. In particular, after 0907:30 UT at CL1 and 0908:35 UT at CL4, electrons were highly accelerated and ions were beyond the energy range. An electron distribution of the phase space density (PSD) at 0908:53 UT during the acceleration observed by CL4 is shown in Figure 6. The left half of Figure 6 (left) shows data from the HEEA instrument. LEEA data (only low-energy part) looking at the opposite direction of the HEEA is shown on the right half. Electron pitch angle data are presented in a magnetic field coordinate system, in which parallel to the local magnetic field is set to 0° (upward) and 180° for antiparallel direction. The perpendicular direction to the magnetic field is set to 90° (horizontal). In Figure 6 (right), one-dimensional cut of the PSD is presented. The cut data are obtained from both instruments (HEEA/LEEA) in the several pitch angles indicated at each line. Each label shows the instrument (HEEA/LEEA) of the data and a pitch angle. We can see that electrons are accelerated beyond the PEACE instrumental energy range. It should be also noted that the distribution is isotropic. Such isotropic acceleration indicates either the acceleration is not caused by the adiabatic acceleration such as betatron acceleration or Fermi acceleration, or the accelerated particles are immediately pitch angle scattered. Such simultaneous acceleration was observed again, between 0912:10 and 0912:45 UT and between 0914:30 and 0915:50 UT by CL1, and between 0914:45 and 0915:35 UT by CL4.
 In Figure 7, in order to investigate signatures of high-energy electron acceleration in comparison with signatures of ions, thermal electrons and magnetic field in the substorm, we show the electron flux of both high- and low-energy components together with the high-resolution magnetic field data and Vx from top to bottom. Here, omnidirectional high-energy electron fluxes and low-energy electron fluxes with pitch angles of 0°, 90°, and 180° are plotted. After the initial decrease of the high-energy flux was observed by all satellites at 0904 UT, the flux increased until 0906:40 UT at CL1, until 0905:25 UT at CL2, and until 0906:50 UT at CL4. This flux increase is observed before the appearance of a fast flow at 0907 UT with a slightly dispersed signature. The dispersion signature was less clearly observed by CL2. Then, more drastic increase of the flux is observed at 0907 UT by CL1 and CL4. At this time, the fast earthward flow accompanied by the acceleration of thermal electrons was observed by CL1, CL3, and CL4. The peaks of the high-energy flux were detected between 0908:40 and 0908:50 UT. CL2 on the dusk side the other satellites near the southern plasma sheet lobe boundary did not observe significant acceleration. After that, there were also several enhancements between 0912 and 0915 UT associated with magnetic oscillations and earthward fast flows. In these flux enhancements, thermal electrons observed by PEACE were more accelerated than the initial enhancement. At 0912 UT, only CL1 and CL2 observed the acceleration but CL4 with enhanced Bz did not observe such acceleration without any significant perturbation of the magnetic field. After these flux enhancements, the magnetic field was relatively steady and higher-energy flux (>100 keV) preferentially decreased quickly. Figure 8 shows the spectra of the omnidirectional PSD of the high-energy electrons for CL4. Relative to the initial state at 0904 UT (thin solid line with plus marks, power law spectral index γ = −3.64), the high-energy component (>100 keV) increased in the course of the fast flow and showed harder spectra (thick solid line with circles, γ = −2.97). In the late phase of the fast flow and after the fast flow, the lower-energy component (∼50 keV) gradually increased (thick dashed line with crosses). After the fast flow (thin dashed line with triangles), the high-energy flux gradually decreased, and became even less than the initial state. Thus, a significantly softer spectrum (γ = −4.45) was observed. Namely, the spectrum of the high-energy electrons varies significantly in the course of substorms.
 Next, we investigate pitch angle distributions of electrons in order to discuss generation and loss processes of electrons in the near-Earth magnetotail. In Figure 9, we display evolution of the electron PSD spectra during the temporal Bz enhancement. In Figure 9a, we show the initial omnidirectional spectrum before the enhancement at 0906:28 UT observed by CL1 as a thin solid line with plus marks. At this time, Bz was 3.17 nT. Then a fast earthward flow was detected associated with an enhancement of the magnetic field not only Bz but also By. After the passage of the enhanced magnetic field structure, Bz was enhanced to 6.78 nT at 0907:55 UT. An omnidirectional spectrum at this time is shown as a thick dashed line with open squares. The expected spectrum assuming that the betatron acceleration caused by the increase of Bz has occurred is shown as a thin dashed line with plus marks. We found that the expected and the observed spectra are similar. Thus, the initial acceleration is supposed to be adiabatic. Note that the perpendicular component at this time (a thick solid line with filled circles) is almost the same as the omnidirectional PSD although we have only two data samples at energies <100 keV. Therefore, it is not so simple as electrons have not only been accelerated by betatron and Fermi acceleration, but experienced some additional process. Hence, we argue that the perpendicularly accelerated particles can be considered to be pitch angle scattered immediately after acceleration, and the distribution becomes isotropic. Figure 9b shows a similar signature during the second enhancement observed by CL4 from 0914:01 UT to 0914:34 UT when Bz enhances from 12.61 nT to 31.73 nT. In this case, the clear enhancement of only the perpendicular flux is observed. On the other hand, the parallel component (thick solid line with filled squares) or the omnidirectional PSD increase only slightly. Since the enhanced perpendicular component is comparable to the expected spectrum assuming betatron acceleration, the acceleration mechanism of the high-energy electrons is considered to be consistent with the betatron acceleration.
 In Figure 10, parallel (0°–20°, solid lines) and perpendicular (80°–100°, dashed lines) fluxes observed by CL1 and CL4 between 0904 and 0920 UT are shown in the two available energy steps, 37.3–50.5 keV (thin lines) and 68.1–94.5 keV (thick lines) in Figures 10b and 10g. They are presented with Bx, Bz (Figures 10a and 10f), and Vx (Figures 10e and 10j). Since the parallel and antiparallel fluxes were mostly the same within the error in this interval, we present only the parallel flux in the figure. We can see that the parallel and perpendicular fluxes sometimes differ. There are several intervals in which the perpendicular flux was more than the parallel flux near the neutral sheet. Such flux anisotropy is observed both by CL1 and CL4, and is indicated in solid squares. It is seen that such preferential enhancements of the perpendicular flux are associated with the rapid and significant increase of Bz. Such Bz increase usually occurs during the earthward flow occurrence, but sometimes slightly after the end of the flow. Associated with enhancement of the perpendicular flux, the temporal increase of the total flux is also observed (Note that the short-lived increase of Bz at CL1 at 0909 UT interval is shown in Figure 7). There is an exception: CL4 observed an increase in Bz at 0909:50 UT without enhancement of electron flux or flow velocity. It is also interesting that CL4 observed an enhancement of the perpendicular flux at 0913:30 UT without any convection under the dipolarized magnetic field configuration. At this time, CL1 and CL2 were not in the dipolarized region on the tail/dawn sides of CL4. The satellites observed enhancement of both Bz and high-energy flux with minor earthward flow observed by CL1. Thus, the enhanced flux may be drifted from the nondipolarized region to the dipolarized region whose boundary may not be simply planar as is observed in the initial activity (tailward expansion of the dipolarization region is discussed also by Nakamura et al. ). The enhancement of the perpendicular flux immediately disappeared when the temporal Bz enhancement ceased. More isotropic distribution of the energetic particle flux is observed after such a short preferential perpendicular enhancement.
 On the other hand, the excess of the parallel flux appears associated with the decrease of the perpendicular flux observed between the enhancements of the flux, as is indicated by dashed boxes. Excess of the parallel flux was observed between 0918:00 and 0918:30 UT at CL1 and more clearly between 0911 and 0912 UT and 0916 and 0919 UT at CL4. This anisotropic signature is simultaneously observed with bi-streaming electrons in the low-energy range. After 0909 UT, CL4 PEACE detected a bi-streaming electron distribution, as is shown in Figure 7. Figure 11a shows such a distribution in comparison with the almost simultaneous isotropic distribution observed by CL1 (Figure 11b) in the same format as Figure 6. At CL4, parallel (LEEA measurement up to 2 keV) and antiparallel fluxes exceeded the perpendicular flux by a factor of about ten. Such a difference is also considered to be related to the spatial structure of the magnetic field. CL4 was closer to the Earth than CL1 and was in the dipolar Bz region. On the other hand, CL1 still detected small Bz which indicates the stretched magnetic field.
 It is also noted that parallel flux in the low-energy electrons (less than a few keV) is sometimes more than the perpendicular flux, even in the condition perpendicular flux in the high-energy electrons is more than the parallel flux. Figure 11c shows this type of low-energy electron distribution observed by CL1 at 0913:31 UT, when the perpendicular high-energy flux increased more than the parallel flux. One can clearly see that while the low-energy (<40,000 km s−1, ∼5 keV) flux shows an anisotropy which has more flux in the parallel direction, perpendicular flux becomes more pronounced in the energy range of >40,000 km s−1. The result suggests that betatron acceleration is not always dominant in all energy levels.
 In order to examine the evolution of the pitch angle distribution more clearly, we present time evolution of the pitch angle distribution in PSD in Figure 12. The pitch angle distributions are shown for two major enhancements of the flux observed by CL4 at 0909 UT (dashed lines) and 0914 UT (solid lines) in comparison with the initial state at 0903:00 UT (solid line). In Figure 12, we show pitch angle distributions of the electron PSD obtained from RAPID/L3DD. The energy range is between 68.1 and 94.5 keV and the pitch angle data is averaged in each 20° bin. It is observed that the perpendicular flux was more enhanced in the initial stages of the flux increase (pancake-type), but it rapidly disappeared within 1 min, even as the overall fluxes increased, and became a parallel-dominated (cigar-type) distribution. In other words, the decrease of the perpendicular flux is more significant than that of the parallel flux. The perpendicular flux became even less than the initial state at the lowest cases.
 We also menton the electron flux anisotropy in the azimuthal direction. In Figure 10, the flux ratio of duskward to dawnward electrons (±45° in ygsm and ±60° in zgsm) averaged over 20 s are also plotted. The flux ratio in 37.3–50.5 keV and 68.1–94.5 keV energy steps are shown in Figures 10c and 10d for CL1 and in Figures 10h and 10i for CL4, respectively. It is mostly close to 1.0, but a slight excess of the dawnward component was observed associated with the enhancement of the flux indicated by error bars. It could be interpreted as either the existence of the dawnward drift by the magnetic field gradient and the curvature drift which are directed dawnward in the magnetotail configuration, or the radial density gradient in which more electrons exist on the earthward side of the satellites.
3.2. Event of 3 September 2006
 Here we show another event on 3 September 2006, which was observed at Xgsm = −15 RE. Figure 13 shows a summary plot of the event between 2140 and 2220 UT. The intensity and the three components of the magnetic field at all four satellites, high-energy and low-energy electron flux, ion flux at CL1, ion density, temperature, Vx and Vy,gse at CL1 and CL3 are shown from top to bottom. Note that CL3 failed to get part of the ion data from high-elevation (polar) directions during the interval. However, we can still work with the spin plane fast flow velocity, namely, Vx and Vy. As is shown in the ground magnetogram data Figure 14a, Pi2 onset was observed at 2148 UT and 2154 UT indicated by two vertical lines. Then a positive bay with the intensity of ∼5 nT evolved until 2220 UT. During the event, satellites were located in the postmidnight region of the near-Earth magnetotail Figure 15. CL3 was at [−15.2, −2.7, 0.7] RE in GSM at 2200 UT. Relative to CL3, CL1 was on the dawnward side, CL2 was on the northern side, and CL4 was on the earthward as well as on the northern side of CL3 by about 1 RE, forming a tetrahedron configuration. As is seen from Bx values, CL3 was the only satellite which located on the southern plasma sheet, and CL1 was the closest to the neutral sheet while CL2 was the furthest. A fast earthward flow accompanied by highly accelerated ions and electrons was observed at 2155 UT. This fast flow would be associated with the second Pi2. A decrease of the Bx difference (ΔBx) among the satellites and the enhancement of Bz more than 10 nT are typical signatures of the dipolarization just after the fast flow.
 While the thermal components of the electrons (PEACE) did not show any significant variation until 2155 UT, the more energetic electrons (RAPID) first showed a drop of the flux at 2146 UT. Then the dispersive increase of the flux up to the highest energy channel was observed at 2148 UT when the first Pi2 was observed. After that, the repeated dispersive increase was observed at 2150 UT and 2153 UT. This signature is similar to the previous event when the auroral brightening was observed on the dusk side of the satellite location. Figure 14b displays H and D components at three midlatitude ground magnetograms around the midnight: Borok (Geomagnetic Longitude = 38.23°), Uppsala (17.35°), and Black Forest (8.32°). Foot points of the Cluster satellites were slightly on the dawn side of Borok. Uppsala and Black Forest were well on the dusk side of the foot points. All three stations observed a clear positive bay. On the other hand, D components show different polarities, namely, negative at Borok, initially negative and changed to positive at Uppsala, initially weak variation and then positive at Black Forest. The results suggest that the auroral activity (current wedge) was initiated at around Black Forest, and then extended dawnward. Therefore, the initial activity is considered to have been well on the dusk side of the Cluster location.
 After 2156 UT, the increase of the high-energy electron flux became dispersionless associated with the observation of local fast flows and magnetic disturbances. The largest flux of the energetic electrons was observed at 2158 UT, which is at around the end time of the fast earthward flow. After that, while the lower-energy flux (<50 keV) remained almost constant for more than 20 minutes, the higher-energy flux (>100 keV) decreased quickly. This signature is also the same as that observed in the previous event. The time evolution of the electron energy spectra shown in Figure 16 is in the same format as Figure 8. Data points at 2208:30 UT are slightly shifted to the left for the better visibility of error bars which are also added to 2146:00 UT data. Relative to the initial state (γ = −4.15, if the data are fitted to the power law spectra) it is clear that the increase of the higher-energy (>100 keV) component was significant in the initial phase of the increase and showed the hard spectrum (γ = −2.35). The PSD reached a maximum at around the end of the fast flow, and subsequently changed to softer spectrum (γ = −5.90). We also show in Figure 17 the flux anisotropy during the interval in the same format as Figure 10. In this event, we use E3DD data set. For CL4, the lowest-energy channel data (37.3–50.5 keV) are not shown in the parallel/perpendicular flux because of the nonnegligible noise level. Again, one can see the temporal increase of the perpendicular flux associated with the rapid increase of Bz at 2156 UT and 2159 UT by CL1 and 2158 UT by CL4. The excess of the perpendicular flux lasted for one to a few minutes, otherwise, rather isotropic. It is also clearly seen that the perpendicular flux dropped during the quasi-steady dipolarization intervals (2204–2212 UT for CL1 and 2207–2212 UT for CL4) after taking several times from the enhancement of Bz. Dawn/dusk asymmetry was small, but dawnward flux sometimes exceeded the duskward flux associated with the enhancement of the flux, as in the previous event.
Figure 18 shows the high-resolution magnetic field (67 Hz) and the electric field (450 Hz) between 2155 and 2200 UT. Four colors in the magnetic fields (total intensity and three components) and Vx show the same satellites as those in the summary plot (Figure 13). Ex and Ey in the spin plane (DSI coordinate system) from CL1 and CL3 are shown separately. Red lines show the 450 Hz high-resolution data and the black lines show the spin-averaged (4 s) electric fields. At the bottom, wave intensity, degree of polarity (DOP) and ellipticity of the wave activity derived from the CL1 EFW data are also displayed. Here we mention the significant bipolar Bz which was observed at 2156 UT during the fast earthward flow detection. This bipolar Bz signature was observed from CL2, CL3, CL1, and then CL4. Thus, the structure is clearly found to propagate earthward with the speed of about 400 km s−1 derived from the timing difference. The structure became more significant with larger amplitude and shorter interval as the structure approached the Earth. The bipolar structure is asymmetric. Bz became only a few nT in the negative part, on the other hand, it became more than 20 nT in the positive part. The four Cluster satellites observed large electric field just at the enhancement of the Bz. Such large enhancement of the electric field reached 70 mV m−1. The wave activity consisted of mainly lower hybrid waves (<10 Hz) and some whistler waves (DOP ∼1, ellipticity ∼1, namely, 100–200 Hz activity with clear right-handed circular polarization).
 The origin of the high-energy particles in the magnetotail has been long discussed for decades. In this paper we examined two events which were observed at Xgsm ∼ −10 RE and Xgsm ∼ −15 RE. The former event has been supposed to be slightly outside or just around the region of the injection boundary [e.g., Mauk and Meng, 1983], while the latter event is supposed to be between the near-Earth neutral line region and the initial dipolarization (current disruption) region [Miyashita et al., 2009]. We summarize our observational results in Figure 19. Initially, the magnetotail is stretched (black lines). Associated with initial activities on the dusk side (local time with dashed lines), dispersive flux enhancement of electrons (magenta dashed arrow) are observed. In the −10 RE event, magnetic oscillations are also detected by the most dusk-side satellite (red one) associated with the dispersive flux enhancement. Then earthward propagation of fast flows associated with temporal enhancement of Bz (gray filled arrow) was observed associated with the second activities dawnward of the initial activities (local time of the magnetic field lines shown in solid lines). During the interval, both ions and electrons are highly accelerated in the region in a gray solid oval. Enhancement of the high-energy flux is dispersionless and the spectra are very hard with temporal preferential enhancement of the perpendicular flux. After the fast flow, the bulk plasma is more thermalized than the initial state, and no convection signature was observed. The quasi-steady dipolar magnetic field (brown lines) is observed more clearly by a satellite nearer to the Earth (blue one), and the dipolarization region expands tailward. The highest-energy electron flux decreases (magenta solid arrow), but the flux of lower-energy electrons remains high. During the interval, the perpendicular flux of ∼100 keV electrons decreases rapidly and parallel flux becomes relatively larger than the perpendicular one. Most of these signatures are similarly observed both in the first −10 RE event and in the second −15 RE event, but more significantly observed at the most earthward satellite of CL4 in the first event.
 Two kinds of adiabatic acceleration may be associated with the global reconfiguration of the magnetotail: betatron acceleration with the conservation of the first adiabatic invariant for the perpendicular flux and Fermi acceleration with the conservation of the second adiabatic invariant for the parallel flux. Both our events exhibit a temporal increase of the perpendicular flux followed by a drop of the perpendicular flux to a level less than the parallel flux. As the perpendicular flux increase is well correlated with the enhancement of Bz as is shown in Figure 10 and Figure 17, electron acceleration in the near-Earth magnetotail is considered to be basically associated with the betatron acceleration. Effect of the betatron acceleration was reported previously [e.g., Kivelson et al., 1973]. However, unlike a steady dipolarization at the geostationary orbit with a steady enhancement of perpendicular electron fluxes in their result, our observation results show that significant Bz enhancements inside earthward fast flows do not last so long, less than 1 minute. In addition, we also have to consider another additional acceleration mechanism and/or a very fast pitch angle scattering mechanism. As Kivelson et al.  mentioned, the appearance of the whistler mode wave can be closely related to the scattering of electrons. Such whistler waves are clearly observed during the fast flow interval as is shown in Figure 18 and the pitch angle distribution during the corresponding interval becomes immediately isotropic (Figure 17). Such observation is also reported in the first event on 27 October 2007 (A. Retinò, personal communication, 2009). Thus, the whistler wave is one of the most important candidates. While Birn et al.  showed the pitch angle distribution of the accelerated electron in the near-Earth magnetotail in their numerical simulation, we have not yet resolved the effect of such wave activities both in time and in space. It should be studied further. The local Bz enhancement similar to the local braking region of fast flows in the near-Earth magnetotail [e.g., Shiokawa et al., 1997] discussed in this paper, is frequently observed in the outflow region of magnetic reconnection sites, where the Alfvénic outflow is decelerated outside the diffusion region by pile up of the ambient plasma and magnetic field. In this region, Hoshino  proposed the existence of nonadiabatic curvature and gradient acceleration of electrons up to the relativistic energies in his numerical simulation study. Imada et al.  and Asano et al.  also show the enhancement of the high-energy electron flux in such a region. It may be possible that electrons in the propagating leading edge of the fast flows also undergo such a suprathermal acceleration process. It is noted that the appearance of the net dawnward high-energy electron flux at the enhancement of the flux may indicate that there are more electrons whose center of the gyration locates on the earthward side of the satellites than electrons whose center of the gyration is located on the tailward side [Ohtani et al., 1992]. This would indicate that particles are accelerated locally, not convected from the tail.
 On the other hand, it is not easy to confirm that the excess of the parallel electron flux is caused by Fermi adiabatic acceleration. Theoretically, the acceleration mechanism should work as a result of the significant shortening of the magnetic flux tube associated with earthward fast flows. It can be also understood in terms of the dipolarization process in the near-tail region Xgsm > −10 RE, as has been discussed by Hada et al. . In studies of ions, such parallel acceleration has been further discussed as “bounce phase–bunched” ions [e.g., Delcourt et al., 1990], although the acceleration is not caused by the transport of the midtail flux tube. However, during the interval when the drop of the perpendicular flux was observed, Ey was close to zero, and the formation of the cigar-type distribution was not associated with local convection. Considering the result that the drop of the perpendicular flux causes the excess of the parallel flux, rather than an increase of the parallel flux, it is more likely that the perpendicular high-energy flux escapes from the region where the measurements were made. The observation of a dispersive signature associated with the dusk-side substorm activity also suggests that nonnegligible dawnward drift motion is occurring in the regions shown in our study, and supports the idea that the perpendicular flux drifts as sketched in Figure 19.
 Bulk ions and electrons are thermalized during the same interval, electrons measured in the lower-energy channel flux of the high-energy particle detector may similarly be affected, such that the spectra becomes softer. However, we do not want to deny the overall possibility of the Fermi adiabatic acceleration. As the parallel velocity is proportional to the inverse of the flux tube length, only ∼3 RE earthward motion to the 10 RE region can produce an acceleration comparable to betatron acceleration by an enhancement of Bz by factor 10. The Fermi acceleration would occur at the same time as the betatron acceleration during earthward fast flows, but may be less effective. Birn et al.  showed in their numerical simulation that both betatron and Fermi accelerations are effective and betatron acceleration is dominant for higher-energy electrons. It is interesting that only CL4, which is closer to the Earth, observed the additional parallel electron flux in the lower-energy range, sometimes simultaneously with the perpendicular acceleration in the high-energy part. It may be necessary to consider another source, such as an ionospheric origin.
 Evolution of the pitch angle distributions from perpendicular (pancake-type) to parallel (cigar-type) has been reported previously, by Smets et al.  and Wu et al. . Considering these two pitch angle distributions as the betatron and Fermi acceleration, Smets et al.  interpreted the variation of the 1 keV flux pitch angle distribution in terms of a spatial structure observed due to the satellite motion from midtail to the region nearer to the Earth in the longer interval (3 hours) than in our results. Wu et al.  interpreted the same type of the pitch angle evolution as a temporal variation lasting about 20 min. In their interpretation, earthward moving flux tubes nearer to the Earth, namely, observed in the initial phase, are more affected by the conservation of the first adiabatic invariant with larger increase of Bz. On the other hand, flux tubes from the midtail region which arrive at the observation point later are more affected by the conservation of the second adiabatic invariant due to the shrinking length of the flux tubes themselves. Thus, the effect of the betatron acceleration is predominantly observed in the initial phase and the effect of the Fermi adiabatic acceleration is more clearly observed later in this two-dimensional picture. On the other hand, our results suggest that the excess of the parallel flux is not necessarily associated with the Fermi acceleration as we saw no convection (no finite Ey) during the excess of the parallel flux. And it is also clear with two similar observations at the different radial distances that the transient signature is not caused by the spatial difference. Here, we should consider a three-dimensional picture including the effect of the azimuthal drift. Although the exact drift paths of the particles in the dynamic magnetosphere are complicated and have been analyzed only under certain assumptions [e.g., Hamlin et al., 1961; Ejiri, 1978; Reeves et al., 1991; Anderson et al., 1997; Takahashi et al., 1997], it is clear that the highly energetic particles in the near-Earth magnetotail drift significantly. In the three-dimensional picture, the high-energy particles will drift azimuthally and will be lost at the magnetopause before being trapped as ring current particles. In the 3 September 2006 event, the initial dispersive signatures of electrons between 2148 and 2150 UT occur within about 10 s between 44.9 keV (center energy of the lowest energy channel) and 305.3 keV (that of the highest energy channel) with the Bz gradient of the bipolar structure. Hence we can estimate the magnetic field gradient between CL2 and CL4 near the neutral sheet to be 0.012 nT km−1. As Reeves et al.  discussed, the magnetic field gradient drift is only weakly correlated with the location of the particle along the bounce motion paths, and we can derive the drift length to be about 1.1 × 104 km for particles with enhanced perpendicular energy. Even if one considers the largest effect of the electric field drift as large as 1000 km s−1 (factor 10 larger than the normal speed), the azimuthal distance from the source region to the satellite location can be roughly estimated to correspond to the distance to the dusk-side auroral brightening. It is already known that the auroral brightenings are well correlated with the occurrence of fast plasma flows in the magnetotail [e.g., Ieda et al., 2001; Nakamura et al., 2001], and we expect the existence of the temporal enhancement of Bz and associated electron acceleration at the same local time. On the other hand, the curvature drift becomes larger only near the neutral sheet, and the averaged drift velocity of particles with smaller pitch angles along the longer bounce path becomes smaller, which may explain the preferential drift loss of particles with ∼90° pitch angles. Under the dipolar magnetic field configuration, the curvature radius even in the neutral sheet becomes larger, and hence the azimuthal drift velocity should become much smaller.
 While we argue that the existence of betatron acceleration is likely, in the above discussion, we have one more important result about the variation of the spectral indices during the activity. In the previous statistical results mostly observed in the midtail region, Christon et al.  concluded that electrons show somewhat harder spectra during geomagnetically active intervals than in the quiet time, while Åsnes et al.  reported that the spectra are independent of the geomagnetic activity (Kp indices). On the other hand, Vogiatzis et al.  have reported a midtail (Xgsm ∼ −19 RE) observation of high-energy electrons, and they found that the spectra become softer during the substorm recovery phase with a more dipolar magnetic field relative to the preonset signature. Our result shows that the evolution of the spectral signature is not simple. It becomes first harder and then softer in the course of a substorm, and the statistical results reported in the papers mentioned above may depend on the ratio of the data set from the initial (early expansion) and the later (recovery) phases as well as nonsubstorm intervals. Since simple adiabatic acceleration alone cannot change the spectral index, the result also means that an additional acceleration and/or thermalization process must be considered.
 We should also consider the context of large-scale substorm phenomena, such as magnetic reconnection and current disruption. Recent numerical simulations have proposed several interesting mechanisms to create nonthermal high-energy electrons associated with magnetic reconnection caused by interacting multiple magnetic island structures [Drake et al., 2006] and the parallel electric field under the existence of the off-the-plane magnetic field [Pritchett, 2006]. However, it is not likely that the energetic particles are directly transported to the near-Earth tail without any drift loss signature as is estimated above. It is also important that the observed flux in the 27 October 2007 event is a factor 10 larger than that observed downstream of the reconnection region reported by Asano et al. , which indicates the undersupply from the tail to account for the observed flux in the near-Earth magnetotail. On the other hand, Li et al.  have demonstrated using their numerical simulation that an earthward propagating electromagnetic pulse which may be possibly created by the earthward jet from the X line, can reproduce all the major signatures of the particle injection very well, suggesting the indirect effect of the magnetic reconnection far away and on the earthward side of the X line itself. Therefore, it is more probable that the particles are accelerated locally in the near-Earth tail region, even though the fast earthward flow originated by the near-Earth neutral line is effective.
 It is also proposed that appearance of near-Earth electric field and the magnetic field turbulence are the origin of the principal acceleration. Vogiatzis et al.  reported a substorm event observed at Xgsm = −19 RE, and interpreted the appearance of the high-energy perpendicular flux as drifting particles from the dusk flank, assuming the particle acceleration was caused by the local current disruption mechanism [e.g., Lopez et al., 1990]. Birn et al.  showed in their numerical simulation that electrons on the dusk side can be further accelerated by the temporarily enhanced electric field in the near-Earth magnetotail. This type of the high-energy particle acceleration mechanism may be one of the important candidates. However, it has not yet been clarified. The relation to the well established phenomenological signatures of the injection boundary model [e.g., Arnoldy and Chan, 1969; Russel and McPherron, 1973; McIlwain, 1974; Moore et al., 1981; Mauk and Meng, 1983] is not clear, either. In addition to the large-scale current disruption model scenario, Retinò et al.  discussed the existence of effective particle acceleration due to a turbulent magnetic field in the magnetosheath context. As is shown in our results, an intense electric field was observed in association with the arrival of the fast earthward flows and enhancement of Bz. The electric field by 15–20 mV m−1 during the flow and impulsive enhancement up to 70 mV m−1 may accelerate electrons with nonadiabatic motion much more effectively. Thus, effectiveness of such nonadiabatic process for the electron acceleration just in the thin layers of the flow braking region with dipolar magnetic field may be important (A. Retinò, personal communication, 2009).
 Finally, we discuss the relation of the high-energy flux to midenergy components. Christon et al.  reported that the ion and electron temperatures vary highly coherently not only in quiet times but also in active times. Our result clearly shows that the enhancement of the electron acceleration (<30 keV) is well correlated to the ion acceleration during fast plasma flows. On the other hand, timings of the flux enhancement of the suprathermal electrons (>30 keV) are not exactly simultaneous. As is clearly observed, some of the suprathermal electrons quickly drift dawnward from another area and so observations of such electrons do not always correspond to a local signature. Otherwise, a close relationship between the enhancement of the high-energy electron flux and magnetic disturbance indicates the existence of a local acceleration process in addition to the adiabatic accelerations. Although we could not examine high-energy ion acceleration in this paper, it is also important to understand both ion and electron acceleration mechanisms within the same frame of magnetospheric dynamics. This topic should be examined in the future.
 Using Cluster multisatellite observations, accelerated electrons in the near-Earth magnetotail associated with substorms are examined. It is found that the hardest energy spectrum is observed in the earliest stage, and that the spectrum becomes softer later on in the events. The peak values of the high-energy flux are observed at around the end of the local fast earthward flow observations. The high-energy electrons sometimes show preferential perpendicular acceleration associated with the temporal enhancement of the normal component of the magnetic field. The anisotropic distribution quickly becomes isotropic. During the acceleration interval, large-amplitude electric field and whistler wave activity were observed. During the following steady dipolarization interval without any convection signature, on the other hand, perpendicular flux drops to less than the initial value, and the parallel flux exceeds. Such anisotropic distributions of electrons are consistent with adiabatic betatron acceleration, but the effect of the pitch angle scattering is supposed to be important. The effect of Fermi acceleration is not so clearly identified in the high-energy electrons. Furthermore, variation of the energy spectrum indicates that the acceleration mechanism is not only betatron acceleration but also nonadiabatic effect should be considered. Dispersive signature associated with dusk-side activities indicates dawnward drift loss.
 We thank H.-U. Eichelberger for processing Cluster FGM data. Part of the data set was obtained from the Cluster Active Archive. AL index and Urumqi geomagnetic data were obtained from the World Data Center for Geomagnetism, Kyoto, Japan. We also acknowledge NASA contract NAS5-02099 and V. Angelopoulos for use of data from the THEMIS mission, specifically, S. Mende and C. T. Russell for use of the GMAG/CCNV data. G. Parks is acknowledged for use of Polar UVI data, and we also thank K. Liou for processing Polar UVI data. This work is supported by a Grant-in-Aid for Scientific Research for JSPS Fellows 18.5427.
 Amitava Bhattacharjee thanks Joachim Birn and another reviewer for their assistance in evaluating this manuscript.