Observations of substorm fine structure

Particle and magnetic field measurements on the CRRES satellite were used, together with geosynchronous satellites and ground-based observations, to investigate the fine structure of a magnetospheric substorm on February 9, 1991. Using the variations in the electron fluxes, the substorm activity was divided into several intensifications lasting about 3–15 minutes each. The two main features of the data were: (1) the intensifications showed internal fine structure in the time scale of about 2 minutes or less. We call these shorter periods activations. Energetic electrons and protons at the closest geosynchronous spacecraft (1990 095) were found to have comparable activation structure. (2) The energetic (> 69 keV) proton injections were delayed with respect to electron injections, and actually coincided in time with the end of the intensifications and partial returns to locally more stretched field line configuration. We propose that the energetic protons could be able to control the dynamics of the system locally be quenching the ongoing intensification and possibly preparing the final large-scale poleward movement of the activity. It was also shown that these protons originated from the same intensification as the preceeding electrons. Therefore, the substorm instability responsible for the intensifications could introduce a negative feedback loop into the system, creating the observed fine structure with the intensification time scales.


Introduction
The development of magnetospheric substorms on large temporal and spatial scales is well established. Dis-turbed periods can be morphologically divided into intervals like the growth, expansion and recovery phases with corresponding signatures (Rostoker et al., 1980), while the physical processes involved can be divided into directly driven and loading-unloading processes (Rostoker et al., 1987). Many details, however, are still not understood. For example, the substorm de®nition by Rostoker et al. (1980) included the concept of``multiple substorm onsets'', and many observational results verify that a poorly understood ®ne structure is an inherent feature of substorms (e.g., Sergeev et al., 1996). Because of this, Sergeev et al. (1996) postulated the 1±2 min impulsive dissipation events (IDE) as the elementary building blocks of the substorm expansion.
Substorm related particle injections (Anderson, 1965;Mauk and McIlwain, 1974;McIlwain, 1974) and other intensi®cations in the equatorial magnetosphere can also have internal ®ne structure. For example, Belian et al. (1984) showed multiple peaked proton injections at the geosynchronous orbit. In a case study based on CRRES data, Grande et al. (1992) reported on two injections of 10 min duration that exhibited internal ®ne structure, ``injectionlets''. The ®ner details of the substorm onset structure have also been studied (e.g. Lui et al., 1988Lui et al., , 1992Ohtani et al., 1992;Rasinkangas et al., 1994;Maynard et al., 1996;Sergeev et al., 1998). For example, the spiky, activation time scale electric ®eld structures observed by Maynard et al. (1996) were interpreted as radial oscillations of¯uxtubes due to AlfveÂ n waves associated with the substorm current wedge formation. However, similar spikes observed deep in the inner magnetosphere by Sergeev et al. (1998) were interpreted as fast magnetosonic waves radiated from the distant current disruption region. Finally, we note that even the high-speed¯ows in the plasma sheet show ®ne structure comparable with the time scales of both intensi®cations and activations (Angelopoulos et al., 1992).
In this study, we will present a case study of a substorm observed by the CRRES satellite. We show how the activity is characterized by ®ne structure at both intensi®cation and activation time scales, and propose a scheme that could explain the former.

Data
The CRRES satellite was launched on July 25, 1990 into an orbit with a perigee height of 350 km, an apogee of 35 786 km, and an inclination of 18 . During the substorm on orbit 484, February 9, 1991, CRRES approached the midnight sector of the auroral magnetosphere at L $ 6.3, MLT $ 22.9, magnetic latitude decreasing from À4X4 to À6X7 between 17 and 18 UT. Figure 1 shows a projection of the CRRES orbit into the GSM xy-plane and the geographic footprints of the ®eld lines through the spacecraft (calculated using T89 model, up $ 3). The locations of some important event characteristics, which will be described later, are marked along the orbit.
On CRRES, the electron proton wide-angle spectrometer (EPAS, also known as MEB; Korth et al., 1992) measured electrons with energies between 21.5 and 285 keV and ions (with no mass resolution) between 37 and 3200 keV. In this study, data with 30 s (one spin) time resolution and 19 pitch angle (PA) bins have been used. The¯uxgate magnetometer provided magnetic ®eld measurements (Singer et al., 1992). The LANL synchronous orbit particle analyzer (SOPA) data from the geosynchronous orbit was also used in the study. The instrument measures electrons in the energy range 50 keV±1.5 MeV and protons in the range 50 keV±50 MeV. During the present event, 1990 095 was located near the magnetic midnight, only 40±60 minutes MLT eastward of CRRES, while 1987 097 was located in the dusk sector (about 1730 MLT). Both satellites are also marked in Fig. 1. In addition, ground-based magnetometer data from Tromsù, Apatity, Dixon, Tixie and Cape Wellen were used to investigate the spatial extent of the substorm activity; pulsation magnetometer data from Yakutsk provided timings for the activity onsets; and all-sky camera (ASC) pictures from KilpisjaÈ rvi made it possible to investigate auroral activity to the west of the CRRES meridian. The locations of these stations are plotted on Fig. 1.
Finally, the IMF and solar wind conditions were studied using the data from IMP-8, which was located at about (34, 2, A16) R e GSM during the event.

Ground-based data
The development of the substorm is outlined in Table 1, which lists the main features as seen from ground, geosynchronous orbit and CRRES. The ground-based magnetometer data (H-component) from ®ve auroral zone observatories are shown in Fig. 2. The MLT midnight times in UT hours are given for each station in the parenthesis, and marked along the curves with dots when they fall in the time period shown. The strong positive magnetic bay indicating increased eastward current started in Apatity at about 1610 UT. This growth phase was followed by a negative magnetic bay that started at about 1657 UT and contained several intensi®cations during its two-hour active phase. The substorm onset time is veri®ed also from the Yakutsk data (not shown) displaying Pi2 onset at the same time (both timings are subject to AE1 min errors). The activity region extended azimuthally from at least Tromsù to Cape Wellen with several localized substructures. More careful analysis (including also the -component not shown here) indicates that the westward electrojet stayed south of Dixon until about 1740 UT, when the poleward expansion started. During this time the substorm reached also the KilpisjaÈ rvi station, as seen from the  Table 1). The vertical lines in the ®gure will be discussed later.
3.2 Overview of the satellite data Figure 3 shows an overview of the satellite data used in this study. The uppermost panel displays IMF f z component (SE) as measured by IMP-8. Although there are serious data gaps, it seems that the IMF had a strong southward component during most of the event. This period started at about 1606 UT (not shown here), and continued until about 1735 UT. In addition, there is a sharp decrease in the strength of the southward component at about 1651 UT, i.e., just before the substorm onset, and a short excursion to zero level shortly after 1700 UT. The solar wind parameters were quite constant, the velocity being about 420 km/s (data not shown). The rest of the panels in Fig. 3 show the magnetospheric data. The second panel shows the 1987 097 protons, the next three panels the CRRES electron uxes, proton¯uxes (both at 90 PA) and magnetic components, respectively, and the lowermost two panels the 1990 095 electron and proton¯uxes. The start of the substorm growth phase can be seen as stretching of the ®eld lines at CRRES, i.e. decrease of f z starting at about 1630 UT (panel 5; note that Table 1 lists several other growth phase related signatures in the particle data, some of which are best seen in the colour-coded pitch angle distribution plots not shown here). At least four substorm related electron injections followed (I1±I4 in Table 1, marked as horizontal bars in panel 3), the ®rst one being delayed from the ®rst ground based onset signatures by a couple of minutes. The ®rst two injections related to local magnetic ®eld dipolarization and ®eld-aligned currents (f y variations in panel 5), and the last one correlate with the northward turning of the IMF f z and the ®nal poleward expansion of the activity. Because of their durations, about 3, 7, 7, and 12 min respectively, we consider these injections to be intensi-®cations as de®ned in the Introduction. The ®ve vertical dashed lines in Fig. 3 mark the (CRRES) growth phase onset and the starting times of the intensi®cations. The same information is also marked along the CRRES orbit in Fig. 1, and as vertical dashed lines in Fig. 2. See Table 1 for corresponding UT times (note that a data gap just after 1740 UT makes it impossible to estimate exactly the start time for the last intensi®cation, and that its exact duration is also dicult to de®ne due to the very small¯ux variations that follow).
Of the geosynchronous satellites, the duskside 1987 097 showed the gross features of the substorm related injections as the protons drifted from the midnight sector toward it (panel 2). On the other hand, 1990 095, located very close to CRRES, showed details that resemble those in the CRRES data (panels 6 and 7 in Fig. 3). The substorm growth phase is seen as an electron¯ux drop out that recovered during the second CRRES intensi®cation. Also the third intensi®cation is clearly seen in the electron data with some delay.
To end this overview, we note that there are several reasons to consider the two ®rst intensi®cations as separate events, although the ®rst one is so short: (1) the electron spectra and pitch angle distributions at CRRES are dierent, as will be discussed later; (2) there is an additional Pi2 enhancement related closely with I2 (Table 1, the timing has an error of about one minute); (3) the 1987 097 data shows a double peaked increase in the duskside proton¯ux a few minutes later (see the two vertical ticks in panel 2, Fig. 3); (4) the 1990 095 electrons and protons behave dierently for these two intensi®cations (panels 6 and 7, see also next subsection); and (5) the CRRES electric ®eld data (not shown, personal communication with N. Maynard) displays a strong, separate duskward spike during each of the four intensi®cations.
We will ®rst examine the ®ne structure within the intensi®cations (most clear during the second and third intensi®cation; see Fig. 3, panel 3), and then discuss a repeating pattern that can be seen in the data, i.e., the fact that, in the ®rst three intensi®cations, electron injection is followed by a proton injection that coincides in time with the end of the intensi®cation (Table 1 and panel 4 in Fig. 3).

Intensi®cation characteristics
Figures 4 and 5 display CRRES data from the intensi®cation periods with a higher time resolution than available in Fig. 3. The dierent panels contain 90 PA electron and proton¯uxes, f y -components of the magnetic ®eld, and the inclination of the ®eld vector (angle between f x and f z ). In addition, Fig. 4 contains pitch-angle distribution (PAD) for the 21.5±31.5 keV electrons. The intensi®cation onset times are again shown as vertical dashed lines, and the particle energy ranges (in keV) are listed on the right hand side of the corresponding panels. In the following we will discuss each electron intensi®cation with detail in order to show that they can be divided into shorter time periods of increased¯uxes, i.e. activations. The ®rst intensi®cation was closely related to the substorm onset (Table 1). It consisted of two activations (labelled A1 and A2 in Fig. 4, panel 1), the former being strongest below 31.5 keV, the latter at somewhat higher energies (the dierent energy spectra are illustrated by the¯ux decrease at the lowermost energy bin during A2). However, the most notable particle feature during this period is the existence of almost ®eld-aligned electrons at`31X5 keV (panel 3). In addition, the satellite observed strong signatures of ®eld-aligned currents (panel 4) and a partial dipolarization that occurred during A2 (1701 UT, panel 5). The FAC signature started about one minute before the electrons were observed.
The second intensi®cation showed the activations more clearly, being composed of three separate but similar electron activations (B1±B3 in Fig. 4). They had durations of about 1±2 min, a repetition rate of about 2± ; 5 CRRES f y and f z ; 6 1990 095 electrons (50±500 keV); and 7 1990 095 protons (50± 400 keV). The CRRES¯uxes are given in cm À2 s À1 sr À1 keV À1 , while the geosynchronous data is given in counts only. The growth phase onset and the four intensi®cations as observed by CRRES are marked as dashed vertical lines 3 min, and they were best observed at energies of 40± 81 keV. They were quite well seen also in other pitch angles (data not shown). The main magnetic ®eld dipolarization occurred with B1.
There were at least three electron activations during the third intensi®cation (C1±C3, panel 1 of Fig. 5); the last one may be composed of two separate activations. The duration and recurrence times of these activations were comparable to those of the second intensi®cation, but the particle energies were lower (`50 keV).
The 1990 095 data support these observations (Fig. 3, two lowermost panels). For example, during and after the second intensi®cation (i.e. at 1703±1717 UT), geosynchronous electron and proton¯uxes showed periodicity that resembled the CRRES activations. On the other hand, there was no one-to-one correlation between the signatures at the two satellites, and the character of the modulation was dierent: one can see some indications of even smaller ®ne structure within the 1995 095 intensity maxima, and the intensity minima were very pronounced, reaching the drop out level. During the third intensi®cation, three or four activations can be recognized in the electron¯ux variations, obviously verifying the CRRES ®nding. Even during the ®rst intensi®cation there is a decrease of proton¯ux above 75 keV (second energy channel) that coincided with A2, again verifying the existence of ®ne structure within the intensi®cation.

Proton injections
A repeating, intensi®cation related pattern can be seen in the CRRES data (Figs. 3±5 and Table 1). In the three ®rst intensi®cations, electron injection was followed by a proton injection that coincided in time with the end of the intensi®cation. We will now investigate this further by calculating the particle and magnetic ®eld energy densities at CRRES and tracing the observed protons back in time to their source regions. Figure 6 (upper panel) presents the perpendicular energy density of protons and electrons and the magnetic ®eld energy density during the event. Also here the dashed vertical lines are used to indicate the starting times of the growth phase and the intensi®cations, the bold horizontal lines showing also the durations of the latter. We can see that the electron energy density e , although peaking during the intensi®cations as it should, was of little importance for the total energy density, and that the magnetic energy density m was generally dominant. However, the proton energy density p was comparable to that of the magnetic ®eld around the ®rst three intensi®cations. To study this further, we noted that the ®rst proton energy channel (37±54 keV) behaves often dierently from the higher energy channels (see, for example, the second and third intensi®cations in Fig.  3, panel 4). Accordingly, the proton energy densities were recalculated, now for dierent energy ranges. The lower panel of Fig. 6 presents ratios of both the 37± 54 keV dierential energy density and the b 69 keV integral energy density to the total energy density p X The relative importance of the low energy protons is obvious during the substorm growth phase and during all the intensi®cations. However, intervals of proton energy density increases were created also by the b 69 keV protons, and they correlate very well with the three major high energy proton¯ux increases seen at the ends of the ®rst three intensi®cations. There is thus a clear anti-correlation between the low and high energy proton¯uxes, the former behaving more like the electron¯uxes.
This dierence between electrons and protons can be studied further. Figure 7 presents intensities from selected electron and proton energy channels as a function of f z during the growth phase (1610±1659 UT, the diamonds) and start of the active phase (1659± 1708 UT, the pluses). Before the onset, both the electron and proton¯uxes decreased generally together with the local magnetic ®eld. After the onset of the substorm, however, the proton intensity variations were , and inclination of f (angle between f x and f z , pnel 4). Flux intensities are expressed in cm À2 s À1 sr À1 keV À1 X The vertical tick mark in the lowermost panel indicate the interval of increased ®eld line stretching that coincide with the proton ux increase independent of the magnetic ®eld variations, whereas the electron¯ux variations were still ordered by f z X Because the high energy protons showed clear energy dispersion during the ®rst (Fig. 4, panel 2) and the third intensi®cation (Fig. 5, panel 2), we can also estimate when and where they were injected. Backward tracings, using the simple dipole model in Roederer (1970), showed that the protons were, in both cases, injected during the (CRRES electron) intensi®cation they relate to, i.e. about 1701 and 1719 UT (see Table 1), and about 40 min and 90 min MLT eastward, respectively, from CRRES. The triangles on the x-axis of Fig. 1 show the approximate locations of these source regions (numbers referring to the intensi®cations).
Finally, note that the proton injections seem to correlate with the intervals of increased ®eld line stretching, which are marked in the lowermost panels of Figs. 4 and 5 with vertical tick marks. For example, at the end of the ®rst intensi®cation, the dipolarization that had occurred during A2 was partly cancelled. Similar features can be seen during the second and third intensi®cations.

Discussion
In this study, we want to stress two important points seen in the case study presented: (1) ®ne structure in the activation time scale (2 min or less) can be seen in the injected electron¯uxes and (partly) in the magnetic ®eld uctuations just as it has been observed in protons (Belian et al., 1984); (2) it is possible that high energy protons produced by substorm activity are also able to aect the way in which the activity develops further, and thus create some of the observed ®ne structure (at least on the time scale of intensi®cations, i.e., several minutes).

Intensi®cations
The substorm growth phase was characterized with a prolonged period of southward IMF f z component (Fig. 3, uppermost panel). The southward turning occurred at about 1606 UT, and taking into account the $6 min time delay with the observed solar wind speed of $420 km/s, this corresponds well with the growth phase onset derived from ground, about 1610 UT. CRRES observed the corresponding changes at about 1630 UT, i.e., a little later. However, also the IMF f z reached its minimum, about A9 nT, only at about 1615 UT (this is not seen in Fig. 3). Later on, the sharp decrease in the strength of the southward component at about 1651 UT ®ts very well with the ®rst ground-based onset signatures at about 1657 UT, indicating that the substorm was triggered.
The substorm itself consisted of a series of electron intensi®cations seen both by CRRES and 1990 095. This activity was distributed over a sector several hours wide in MLT. We can use dierent data sets to investigate the details: 1. Observed dierences between particle¯uxes at CRRES and 1990 095, the latter being about 40±60 min MLT east of CRRES (Fig. 3). 2. Drift calculations on the dispersive ions at CRRES, indicating the azimuthal distance to the proton source (Figs. 4 and 5). 3. The ground based magnetometer and ASC recordings (Fig. 2, Table 1).
The ®rst intensi®cation was observed only by CRRES, and the drift calculations on the dispersive ions indicated that they originated from about 40 min MLT away from CRRES (triangle 1 in Fig. 1). Therefore the position of the eastward edge of the source was not far from 1990 095. As the second intensi®cation was seen both by CRRES and 1990 095, it extended further toward east. However, the lack of proton dispersion makes it dicult to make any estimations about this. The third intensi®cation was also observed by both satellites, and on the ground it is best seen in the Dixon data. The dispersive ion source was calculated to be about 90 min east of CRRES (triangle 3 in Fig. 1). Note that the Cape Wellen data indicates strong activity peaking about 6±7 min later Fig. 6. Upper panel. Energy densities of the magnetic ®eld m , perpendicular electrons e and protons p during the growth and active phases on February 9, 1991. Lower panel. Relative contribution of the 37±54 keV protons and the b 69 keV protons to total proton energy density eastward of CRRES. This is considered to be a separate intensi®cation occurring in the dawn sector. The fourth CRRES intensi®cation was registered close to Tromsù and KilpisjaÈ rvi, the eastward edge being near the satellite (as derived from the weak electron dispersion and absence of protons). Thus, the azimuthal extension of an intensi®cation can be rather limited or span a few hours in MLT. This is not surprising, as for example Belian et al. (1984) found longitudinal extensions of 45 (3 h). It is also obvious that the longitudinal location of the active region can change from one intensi®cation to another. Therefore, an intensi®cation can be de®ned as the longest interval of local substorm enhancement.
It has been shown that the substorm active phase often consists of two separate phases, active-convective and expansion (e.g. Lazutin et al., 1984;Mishin et al., 1992). During the active-convective phase the loading process is still strong and comparable with the unloading of energy, while during the expansion phase the unloading becomes the main process. The present event seem to consists of a active-convective phase during which intensi®cation occur more independently of each other and the possible poleward excursions are restricted and localized. Only during the fourth intensi®cation the global northward expansion commenced, as seen from the ground based data (Table 1). This view is backed up by the IMF data showing northward turning around Fig. 7. The particle intensities during the growth phase (the diamonds) and the ®rst minutes of the active phase (the pluses) as a function of f 2 . Left column, trapped electrons; right column, protons. Units are in cm À2 s À1 sr À1 keV À1 the time of the last intensi®cation (Fig. 3, uppermost  panel).
The ®rst intensi®cation was dierent from the others, and we have reasons to believe that CRRES was closest to the activity centre, i.e., the formation of the substorm current wedge, during this time. This is supported by the (nearly) ®eld-aligned 21.5±31.5 keV electrons (Fig. 4,  panel 3) as well as the ®eld-aligned keV electrons (Johnstone et al., 1996) seen during this intensi®cation. Satellite observations of the former, energetic electron uxes at substorm onset are rather rare (Nielsen et al., 1982;Kremser et al., 1982Kremser et al., , 1988, and ground-based measurements have shown that the precipitating particles are most energetic during the onset phase of the substorm (Olsson et al., 1996). However, note also that the ®rst electron signatures are somewhat delayed from the ®rst FAC and ground based signatures (Table 1), and that at about 1659±1701 UT the magnetic ®eld shows precursory eects where it¯uctuates without any lasting dipolarization eect (Fig. 4, panel 5).
One feature common to the three ®rst intensi®cations is the partial return to more stretched ®eld line con®guration at or near the end of intensi®cation. We will discuss this more later, in connection with the proton injections coinciding in time with these signatures.

Activations
The division of intensi®cations into separate activations was inferred from the CRRES electron data (panel 3 in Fig. 3, and panel 1 in Figs. 4 and 5). In addition, the 1990 095 data supported this view, although we do not expect perfect correlation between the satellites within these time scales. Note also that the fact that the variations in the geosynchronous particle¯uxes at 1703± 1717 UT were of periodic drop out type is not a problem for our scheme, since also the drop outs must be related to injections that occur somewhere close. Thus, when the two satellites were within the same intensi®cation source, they registered particle variations with comparable, but not the same activation structure. The lack of one-to-one correlation between individual activations could even be used to de®ne an upper limit for their azimuthal extension (note that the dierence in radial distance can also have an eect). The resulting one hour is comparable to the length of the individual activation arcs within a westward travelling surge (e.g., Murphree and Cogger, 1992;Nakamura et al., 1993). Since no high time resolution optical data from our event exist, we do not know how our activations/intensi®cations aect the auroral display along the CRRES ®eld line. However, the observed ®eld aligned electrons and the ®eld aligned current signatures indicate that there is a connection, at least during the ®rst intensi®cation, between the observed activity in space and conjugate auroral ionosphere. This would be natural, as Yahnin et al. (1990) have shown that there is a connection between groundbased ®ne structure signatures and high energy proton injection ®ne structure ®rst described by Belian et al. (1984).
The CRRES measurements during the ®rst intensi-®cation show also how partial magnetic ®eld dipolarizations and ®eld-aligned current structures can occur at the time scales of activations (Fig. 4, panels 4 and 5). This indicates that the dipolarization process can be step-like, each step representing a localized activation. That similar signatures are not seen elsewhere may well be due to the fact that CRRES was closest to the activity centre during this time, as argued already.
To conclude this subsection, we note that substorm intensi®cations can be comprised of shorter elements that we call activations. The time scale of individual activations are of the order of 2 min which agrees, e.g., with the impulsive dissipation event (IDE) scenario proposed for substorm development (Sergeev et al., 1996). The present data set does not allow us to speculate further what this means in terms of substorm onset theories.

Role of the protons
Our results suggest that the injected high energy protons may play an important role in the intensi®cations. According to Figs. 4 and 5 (panels 2) and Fig. 6 (lower panel), the main b69 keV proton¯ux increases occurred at the end of the three ®rst intensi®cations. In addition, they correlate with periods of local magnetic ®eld line stretching (lowermost panels of Figs. 4 and 5). We suggest that these high energy protons give dynamic input that has important consequences, that they are in fact quenching, if only locally, the ongoing intensi®cation, and are simultaneously creating favourable conditions for the fourth (last) intensi®cation to occur. Note that we still think that the lower energy protons create the high level of particle energy density and the largescale enhancement of the cross tail current, and hence the word``dynamic'' was used.
That the fourth intensi®cation is not related to such a proton signature at CRRES is easily understood by the drift direction of the protons. It is obvious that, in order to observe the present events, the satellite must be situated in the western part of azimuthally extended acceleration region. This may not be always the case. However, we would like to note that the event presented here is not exceptional, and other similar cases are currently being studied using CRRES data.
Our scheme goes as follows (we will ®rst make a more general statement and then point out, in parenthesis, how our observations support the claim).
1. Along with the formation of enhanced energetic proton¯ux during an intensi®cation, we get a particle population that carries a relatively large amount of energy density with it. (Our drift calculations indicate that the protons are produced during the same intensi®cations as the electrons preceding them. We also calculated that the energy densities of the drifting proton clouds are comparable to the magnetic ®eld energy densities. The electron energy densities are not comparable with these two values. This dierence between the particles is also seen in Fig. 7.) 2. Thus, because of the increased cross-tail current and the diamagnetic eects of the ion cloud, the local magnetic ®eld can be stretched tailward again. (Our data shows good correlation between the ion injections and periods of magnetic ®eld stretching.) 3. This process is competing with the original, still ongoing substorm instability that created the injections in the ®rst place, and resulted in ®eld line dipolarization. (Note that we assume a situation where the source region is azimuthally extended, and the produced protons drift, at least partly, over this active region before leaving it. The particle signatures presented in this study support this view.) 4. If there are dierences in the injection process for the electrons and protons, for example dierent azimuthal extents or locations, or if the higher energy ions drift dierently from other particles, we may have time delays between the¯ux increases. (The time delay is very obvious in our data, and can be explained simply by drift from an eastern part of the intensi®cation region during the ®rst and the third intensi®cation, as seen from the observed energy dispersion. The dispersionless proton injection during the second intensi®cation must be explained in some other way.) 5. While within the active region, the protons dominate both the electron population and the magnetic ®eld, with the result that the instability responsible for the intensi®cation may be aected or even killed. (In this study, the arrival of protons correlate very well with the end of intensi®cations.) 6. At the same time, the protons could be creating a favourable condition for another intensi®cation to occur at a dierent location in the direction of their drift. (The fourth intensi®cation close to KilpisjaÈ rvi could be formed this way. Protons were drifting towards 1987 097 during the whole active period, see Fig. 3, panel 2, keeping the local ®eld lines in a stretched con®guration until the last intensi®cation which started the main northward expansion of the substorm. That the intensi®cation seems to be eventually triggered by a change in the IMF is not a contradiction.) In addition to this quenching eect, note that thè 120 keV protons may also have something to do with the onset of the ®rst intensi®cation, as their¯ux intensity started to increase just after the substorm onset, i.e. about two minutes before the ®rst electrons were observed (Fig. 3, fourth panel, and Table 1).
It is possible that also the activation structures are controlled by the protons, although it is more dicult to prove from the observations. For example, CRRES proton data shows some indications of ®ne structure: see the short lived proton peak just before 1701 UT in Fig. 4, panel 2, and the two maxima in the proton uxes relating to the third intensi®cation in Fig. 5, panel 2.

Conclusion
We have shown, using multisatellite, near-geosynchronous data, that magnetospheric substorms can consist of several intensi®cations, the dynamics of which may be controlled by bursts of high energy (b69 keV) proton uxes produced during the same intensi®cations. We suggest that the substorm instability responsible for the intensi®cations introduces a negative feedback loop into the system, creating this way the observed ®ne structure at the intensi®cation time scales. In addition, each intensi®cation can be subdivided into several short-lived activations, coinciding sometimes with separate partial dipolarizations.