Outer radiation belt and inner magnetospheric response to sheath regions of coronal mass ejections: A statistical analysis

. The energetic electron content in the Van Allen radiation belts surrounding the Earth can vary dramatically on several timescales, and these strong electron ﬂuxes present a hazard for spacecraft traversing the belts. The belt response to solar wind driving is yet largely unpredictable and especially the direct response to speciﬁc large-scale heliospheric structures has not been considered previously. We investigate the immediate response of electron ﬂuxes in the outer belt to driving by sheath regions preceding interplanetary coronal mass ejections and the associated wave activity in the inner magnetosphere. 5 We consider events from 2012 to 2018 in the Van Allen Probes era to employ the energy and radial distance resolved electron ﬂux observations of the twin spacecraft mission. We perform a statistical study of the events using superposed epoch analysis, where the sheaths are superposed separately from the ejecta and resampled to the same average duration. Our results show that wave power of ultra-low frequency Pc5 and electromagnetic ion cyclotron waves, as measured by a geostationary GOES satellite, is higher during the sheaths than during the ejecta. However, the level of chorus wave power remains approximately 10 the same, despite on average stronger ring current enhancements during the ejecta. Electron ﬂux enhancements are common at low energies ( < 1 MeV ) throughout the outer belt ( L = 3–6), whereas depletion occurs predominantly at high energies for high L -shells ( L > 4 ). Distinctively, depletion extends to lower energies at larger distances. We suggest that this L -shell and energy dependent depletion results from magnetopause shadowing dominating the losses at large distances, while wave-particle interactions dominate closer to the Earth. We also show that non-geoeffective sheaths cause signiﬁcant changes in the outer 15 belt electron ﬂuxes. in the Energetic Particle, Composition, and 20 Thermal Plasma (ECT; al., 2013) instrument suite onboard the RBSP spacecraft. MagEIS covers electron energies from 30 keV to 1.5 MeV (source and seed populations), while the core electron population is monitored by REPT, covering energies from 1.8 to 6.3 MeV . In this study, we employed the Level 2 spin-averaged differential electron ﬂux data. The temporal resolution of these data is 11 s . We focused our study to the outer radiation belt between L = 2 . 5 and L = 6 . The lower bound was chosen to avoid proton contamination of REPT in the inner belt, and the upper bound was constrained by the Van Allen 25 Probes apogee. The L parameter (McIlwain, computed using the TS04D magnetic ﬁeld model was extracted from the magnetic ephemeris data available on the ECT website (https://rbsp-ect.lanl.gov/). The very-low frequency (VLF) wave activity in the inner magnetosphere, including chorus waves and plasmaspheric hiss, was obtained from the Electric and Magnetic Field Instrument Suite and Integrated Science (EMFISIS; Kletzing 2013) on the Van Allen Probes. The utilised data product was the Level 2 Waveform Receiver (WFR) diagonal spectral matrix containing 30 the autocorrelations of electric and magnetic ﬁeld components in 65 frequency bins. The frequency range spans from 2 Hz to 12 kHz and the spectra are available with a 6 s time cadence. The EMFISIS team also provides electron densities estimated from the upper hybrid resonance frequency as Level 4 data products (Kurth et al.,


Introduction
The Van Allen radiation belts are toroidal regions surrounding the Earth that trap charged particles in the geomagnetic field (e.g., Van Allen, 1959). Traditionally, the belts are divided into two zones of energetic populations: the relatively stable inner belt for source electrons and usually at lower L-shells for seed electrons, whereas the response of relativistic electrons is more evenly divided between enhancement, depletion and no change events (Turner et al., 2015(Turner et al., , 2019. The most important drivers of geomagnetic activity are interplanetary coronal mass ejections (ICMEs; e.g., Kilpua et al., 2017) that enable effective magnetic reconnection at the magnetopause when their magnetic field has a strong southward component. An ICME that is sufficiently faster than the preceding solar wind will create a shock in front of it, and the tur-5 bulent region between the shock front and ICME ejecta is called the sheath region. The shock, sheath and ejecta of an ICME have distinct solar wind properties and magnetospheric impact (see review by Kilpua et al., 2017): Sheaths are turbulent and compressed structures with large-amplitude magnetic field variations and high dynamic pressure, while ejecta are typically characterised by smoothly changing magnetic field direction and low dynamic pressure. The outer belt response to sheaths and ejecta separately and their combination ("full ICME") have been studied e.g., by Kilpua et al. (2015) and Turner et al. (2019). 10 They find that energetic electrons (> 1 MeV) are more likely depleted during geomagnetic storms driven by only sheaths or ejecta, while full ICME events are more likely to result in enhancement at this energy level. Kilpua et al. (2019b) performed a case study of a complex driver consisting of multiple sheaths and ejecta. They found that sheaths were associated with stronger wave activity in the inner magnetosphere than the ejecta.
However, most previous studies have considered only moderate or stronger geomagnetic storms (Dst or SYM-H minimum of 15 -50 nT or less) and computed the belt response over long time intervals, usually excluding fluxes in a 24-hour period centred around the Dst or SYM-H minimum. Our study focused on the immediate outer belt response to sheath regions. The radiation belts are an open system that particles enter via injections and exit through losses to the magnetopause and atmosphere. Thus, to account for the total energy budget in the inner magnetosphere, we need to quantify enhancement and losses on timescales shorter than 24 hours. This immediate response is fundamental to distinguish the effects of ICME sheaths and ejecta and critical 20 for enhancing our understanding of the Earth's radiation belt environment.
In this study, we consider the changes in the outer radiation belt electron fluxes by comparing the fluxes from only a few hours before and after the sheath region, as a comparison to the up to a few days intervals used in previous belt response studies. We also comprehensively analyse, for the first time, the geospace response during sheath regions and compare it to the responses during the preceding solar wind and the trailing ejecta. This analysis includes geomagnetic activity indices, subsolar 25 magnetopause and plasmapause locations, energy and L-shell dependent outer belt electron fluxes and inner magnetospheric wave activity (chorus, Pc5, EMIC and hiss). Our study includes sheaths that caused only a weak geomagnetic storm (−30 nT > SYM-H min > −50 nT) or no geomagnetic storm at all (SYM-H > −30 nT).
The paper is organised as follows. Section 2 presents the in situ data sets and the methods employed in our statistical study.
We describe an example event and detail our statistical results in Section 3. In Section 4, we conclude our study and discuss 30 future possibilities.
Additionally, observations of wave activity in the ultra-low frequency (ULF) range were taken from the GOES-15 spacecraft at geostationary orbit (L ∼ 6.6). The magnetic field data is sampled at 0.512 s by the GOES fluxgate magnetometers (Singer et al., 1996).

Superposed epoch analysis
In superposed epoch analysis, the median of a given parameter is calculated from the data of all events aligned with respect to 5 some reference time (i.e., the zero-epoch time). This technique has been used in various studies to statistically investigate for example solar wind properties, wave activity and electron fluxes (e.g., O'Brien et al., 2001;Kataoka and Miyoshi, 2006;Kilpua et al., 2013Kilpua et al., , 2015Hietala et al., 2014;Murphy et al., 2018;Turner et al., 2019). We chose the zero-epoch time at the shock and set an additional reference time at the ICME ejecta leading edge (i.e., at the end of the sheath region). The duration of sheath regions varied widely between events from 3.0 h to 22.7 h with a standard deviation of 5.3 h. Therefore, we resampled 10 the sheath regions to match the mean sheath duration of 12.0 h (Kilpua et al., 2013;Hietala et al., 2014). The resampled data was acquired with linear interpolation. The superposed epoch analysis was performed for geomagnetic indices, solar wind parameters, inner magnetospheric wave activity, and electron flux in the heart of the outer radiation belt (L = 3. . We considered wave activity in the VLF and ULF ranges in the superposed epoch analysis. Chorus waves appear outside the plasmasphere (where plasma density is approximately < 50-100 cm −3 ) in two distinct frequency bands (Burtis and Helliwell, 15 1969; Koons and Roeder, 1990): the lower band (0.1-0.5f ce ) and the upper band (0.5-0.8f ce ), where f ce is the electron cyclotron frequency. Plasmaspheric hiss waves occur inside the plasmasphere in a frequency range from about 100 Hz to 0.1f ce .
We calculated the electron cyclotron frequency f ce based on the TS04D geomagnetic field model (Tsyganenko and Sitnov, 2005). To determine whether the spacecraft was located inside or outside the plasmasphere at the time of wave measurement, we estimated the plasmapause location with the AE index based and MLT-dependent model by O'Brien and Moldwin (2003). 20 A plasmapause model was used because the density estimate data is sporadic. For the ULF waves, we calculated the wavelet spectra for each three magnetic field components measured by GOES-15 and summed them together to estimate the total wave power spectral density. We calculated the Pc5 wave power in the range from 2.5 to 10 min (2-7 mHz) and the EMIC wave power in the range from 0.2 to 10 s (0.1-5 Hz), which corresponds to the range of Pc1 and Pc2 pulsations as given by Jacobs et al. (1964). The power spectral densities were averaged in the given frequency ranges of the wave modes to obtain the wave 25 power data for the superposed epoch analysis, and resampling was performed after this averaging.
For electron flux in the superposed epoch analysis, we considered the median flux in the heart of the outer belt at L = 3.5-5.
The MagEIS and REPT electron flux measurements were binned in time (∆t = 4 h) and L-shell (∆L = 0.1) to combine the data from the two spacecraft. The 4-hour cadence was chosen to reduce the effect of the Van Allen Probes orbits in order to minimise the data gaps in the binned flux data during all events. We selected four energy channels to represent the source 30 (54 keV), seed (346 keV), core (1,079 keV) and ultrarelativistic (4.2 MeV) populations. We also calculated the mean electron flux at L = 3.5-5 with the same time and L-shell bins (not shown), and note that it is very similar to the median values.
In addition to investigating the median sheath properties of all 37 events, we divided the events based on the level of associated geomagnetic activity inferred from the SYM-H index. The SYM-H index (Iyemori, 1990;Iyemori and Rao, 1996) is derived from perturbations in the horizontal component (H) of the geomagnetic field that is affected by changes in the ring current. The SYM-H index is essentially the same as the hourly Dst index but with a higher time resolution of one minute, and it is also more sensitive to substorm activity. The strength of a geomagnetic storm is usually characterised with the minimum Dst value, where the thresholds for small, moderate and intense storms are −30 nT, −50 nT and −100 nT, respectively (Gonzalez et al., 1994). In previous studies, events have typically been divided with the threshold of −50 nT, or only these moderate or 5 larger storms are considered (e.g., O'Brien et al., 2001;Reeves et al., 2003;Kilpua et al., 2015;Lugaz et al., 2016;Turner et al., 2015Turner et al., , 2019. However, due to the relatively low number of well-defined sheath events during Van Allen Probes measurements, only nine events out of our total 37 events have a SYM-H minimum below −50 nT during the sheath region or two hours after. Therefore, we set the threshold to −30 nT to obtain a statistically adequate subset of 17 geoeffective events. The interval where we took the minimum was extended two hours after the sheath to accommodate for lag in the ring current response. Note that 10 the geomagnetic disturbance of the ICME ejecta was not considered.

Electron flux response
We binned the MagEIS and REPT spin-averaged electron flux data from both spacecraft in 0.1 L-shell bins and 1-hour time bins, differing from the superposed epoch analysis in order to have a higher time resolution. Based on the methodology of Reeves et al. (2003) and Turner et al. (2015Turner et al. ( , 2019, we define the outer belt electron response (R) as the ratio of post-event 15 flux to pre-event flux. The pre-event (post-event) flux was obtained by averaging the electron flux over a 6-hour interval before (after) the sheath region. The response parameter R was computed for each considered electron energy and L bin. The response was categorised as enhancement when the post-event flux had increased by over a factor of 2 as compared to the pre-event flux (R > 2), depletion when it had decreased by over a factor of 2 (R < 0.5) and no change when the flux level had not changed significantly (0.5 ≤ R ≤ 2). 20 In previous studies (Reeves et al., 2003;Turner et al., 2015Turner et al., , 2019, the pre-and post-event fluxes were defined as the maximum flux from > 12 h up to a few days before and after the event, since the outer radiation belt response to entire geomagnetic storms was examined. These studies also excluded the 24-hour period during the storm. In the current study, we use the mean flux values close to the sheath region as we focus on the outer belt response to the sheath region only, and all sheaths did not generate geomagnetic storms (in 20 out of 37 events the SYM-H index does not drop below −30 nT). A post- 25 sheath maximum flux value is not meaningful as it would be embedded in the CME ejecta and subject to possible fluctuations driven by the ejecta. The 6-hour averaging period aims to capture the changes generated by the sheath while excluding the main response to the ejecta, which is expected to occur later (mean duration of the ejecta was 28.4 h with a standard deviation of 11.1 h). Figure 1 shows the solar wind conditions and geomagnetic indices during 7-9 February 2014, when an interplanetary coronal mass ejection (ICME) driving a sheath region impacted the Earth. The shock (first red vertical line in Figure 1) was identified as an abrupt and simultaneous increase in the magnetic field and solar wind speed, as well as a small increase in dynamic pressure. 5 Both the sheath and ICME were relatively slow (≈ 400-450 km s −1 ). The shock was also quite weak, as the speed jump across the shock was about 100 km s −1 . The sheath was characterised by fluctuating magnetic field and variable dynamic pressure, which had high values (≈ 20 nPa) in the trailing half of the sheath. The ejecta had smoother field and clearly lower dynamic pressure. This ICME is included in the Richardson and Cane ICME list (http://www.srl.caltech.edu/ACE/ASC/DATA/level3/ icmetable2.htm; Richardson and Cane, 2010) and is reported there as a "magnetic cloud", i.e., the event shows signatures of a 10 magnetic flux rope. This is because the magnetic field components ( Figure 1b) show some organised rotation during the ICME and the north-south magnetic field component (B Z ) rotates from north to south. In the sheath, the field was predominantly northward.

Example event on 7 February 2014
The event was only mildly geoeffective. The SYM-H index dropped to −29 nT in the middle of the sheath (and briefly below −30 nT an hour after the sheath ended) and the ICME caused only a weak SYM-H storm. The substorm activity was also weak 15 (but quite continuous) as shown by the AL index.
Despite the low geoeffectiveness of both the sheath and ejecta, there were clear changes in the outer radiation belt electron fluxes at source, seed and core energies as shown in Figure 2. Before the shock arrived, the outer belt showed typical undisturbed conditions (e.g., Reeves et al., 2016) with the seed and core electrons being enhanced at the highest L-shells and the source population featuring the extension of the inner belt to L ∼ 3-3.5. The ultrarelativistic 4.2 MeV electrons in turn peaked at 20 L ∼ 4. After the shock arrival, the fluxes increased at all energies. The most distinct increase was detected at ultrarelativistic (4.2 MeV) energies. The fluxes at all energies also widened towards lower L-shells during the sheath. The flux of the source population (54 keV) continued to increase significantly throughout the sheath, and by the end of the sheath the entire outer belt was flooded by the source population. At higher energies, on the other hand, the flux was depleted near the sheath-ejecta boundary. Interestingly, ultrarelativistic electron fluxes were enhanced already in the front part of the ICME ejecta, while seed 25 and 1 MeV electron fluxes increased clearly only near the middle of the ejecta.
The wave activity in the inner magnetosphere during the event is illustrated by Figure 3, which shows the wave power spectral density of both VLF and ULF waves as measured by RBSP-B and GOES-15, respectively. Some chorus activity appeared immediately after the shock, and it was enhanced in the latter half of the sheath region. Chorus activity persisted during the ejecta. The chorus waves might have caused some acceleration, e.g., the enhancement of 1 MeV electrons during 30 the ejecta, but the waves would not have yet acted long enough to cause the enhancement of ultrarelativistic electrons during the sheath (e.g., Jaynes et al., 2015).
The ULF wave power in the Pc5 and EMIC ranges was elevated during the sheath. The widening of electron fluxes towards lower L-shells could thus be a result of inward radial transport by ULF waves (e.g., Turner et al., 2013;Jaynes et al., 2018). EMIC waves can also be responsible for the loss of relativistic electrons (Usanova et al., 2014). The subsolar magnetopause was located at about 12.7 R E before the shock arrival, according to the Shue et al. (1998) model (Figure 1e). The shock pushed the magnetopause nose inward and the it was located closest to Earth (≈ 7.5 R E ) at the end of the sheath when depletion occurred. At the ejecta leading edge, the magnetosphere started to recover and the subsolar magnetopause stayed at about 10 R E during the ejecta. Therefore, losses at the magnetopause (i.e., magnetopause shadowing) could be the main driver 5 of depletion, possibly coupled with outward transport by ULF waves (Turner et al., 2012). During the ejecta, chorus waves continued, but ULF Pc5 and in particular EMIC wave activity weakened. The enhancement of seed and core fluxes in the ejecta was thus likely associated with continued chorus activity and possibly also with inward transport by ULF Pc5 waves.

Statistics of 37 sheath events
Statistics from the superposed epoch analysis of 37 events with sheath regions are presented in Figure 4. Results are also 10 shown for 10 hours of solar wind before the shock and for one day of the ICME ejecta after the sheath (note that unlike the sheath regions, the ejecta were not resampled). The results show the general characteristics of sheath regions (e.g., Kilpua et al., 2017Kilpua et al., , 2019a: lower magnetic field magnitude than in the ejecta (but about twice as strong as in the quiet solar wind), elevated dynamic pressure (as well as solar wind density) as compared to quiet solar wind conditions and the ejecta, and contracted magnetopause nose due to the high-dynamic pressure sheath. The SYM-H index usually has a positive peak at the 15 shock (corresponding to the storm sudden commencement and initial phase/sudden impulse), and then it gradually decreases during the sheath. However, on average, the main geomagnetic storm impact is imposed by the ejecta as seen in both SYM-H and AL indices. In 17 events, the SYM-H index dropped below −30 nT (weak storm) during the sheath or two hours after, and it dropped below −50 nT (moderate storm) only in nine events. On average, only weak substorm activity is evidenced by the AL index during the sheath region. 20 Panels (g)-(l) of Figure 4 show the statistics of different wave modes in the inner magnetosphere during the selected events.
Pc5 wave power peaks in the sheath, showing a growing trend from the shock towards the end of the sheath region. The mean of median Pc5 wave power in the sheath is about 10 2 nT 2 Hz −1 as measured by GOES-15 at L ∼ 6.6, which is one order of magnitude larger than during quiet solar wind. The wave power of Pc5 waves declines in the ejecta. Also the EMIC wave power is much larger in the sheath than in the ejecta and quiet solar wind, with a median wave power of about 10 −2 nT 2 Hz −1 . 25 The median EMIC wave power quickly decreases to the pre-event level of about 10 −3 nT 2 Hz −1 in the ejecta.
The main power of chorus waves is in the lower band, where the order of magnitude during the sheath is 10 −9 nT 2 Hz −1 .
The median wave power of upper band chorus is an order of magnitude lower, but in a quarter of the cases the power can reach values comparable with lower band waves as shown by the upper quartile curve (Figure 4k). Chorus waves are excited by substorm injected electrons. However, despite the clearly stronger SYM-H and AL response during the ICME ejecta, the chorus 30 wave power is very similar in the sheath and the ejecta. The chorus wave power is also on average only about 4 times higher during the sheath than during the pre-event conditions. The chorus wave power increases gradually for a few hours before the shock arrival. This could be driven by the very weak substorm activity in front of the ICME event. Comparison of the pre-sheath to post-sheath fluxes shows that the median response of 54 keV electrons is an enhancement by a factor of 5. For the seed population, the flux is slightly enhanced at the shock but, on average, the flux remains unaffected in the sheath. However, the flux suddenly increases after the sheath ends and continues to be enhanced in the ejecta. That is, the shown in Figure 5 both as independent on magnetic local time (MLT) throughout the event and as MLT-dependent for the 5 pre-event time (−6 hours from shock), sheath region (+6 hours) and ejecta (+18 hours). One event in 2018 was excluded in the analysis due to the AE index data not being available. The MLT-dependence of the model shows that the plasmapause is closer to the Earth on the dayside and further away in the nightside during both quiet and disturbed times. In the preceding solar wind, the plasmapause is located at about 5 R E . During the sheath, the plasmasphere moves earthward, and it moves even further earthward during the ejecta. The variation is consistent with the general AL levels in the preceding solar wind, sheath 10 and ejecta (the AE index should roughly follow AL behaviour). At noon MLT, the median plasmapause location moves from about 4.4 R E during the quiet solar wind conditions to 3.6 R E in the middle of the sheath, and 6 hours after the sheath region (18 hours after the shock) the median distance has decreased to 3.3 R E .
The electron flux response of the whole outer radiation belt for a wider selection of energies than in the superposed epoch analysis is shown in Figure 6, where the response is divided to the three categories of enhancement, depletion and no change. 15 The source population at L > 3.5 is enhanced in 80% of the cases, and practically never depleted. Closer to the inner boundary  of the outer belt, no change events are very common at all energies. Electrons at seed energies are enhanced in about half of the cases throughout the belt, with a higher possibility for enhancement in the heart of the outer belt. In a small subset (< 15%) of the seed electrons, depletion occurs near L ∼ 3. Depletion is most common in the ∼ 1-3 MeV core population at high L-shells (L > 4.5). At lower L-shells, core electron flux is enhanced at most in 10% of the cases, and in a major fraction of the events (> 60%) the core electron fluxes do not change significantly below L ∼ 4.5.  Interestingly, a feature in the outer belt response is that the depletion progresses to lower energies when L increases. At L ∼ 4.5 depletion dominates only at > 2 MeV energies, while at L ∼ 6 it has reached down to seed energies at around 500 keV.
Depletion is most likely at high energies and high L-shells.

Impact of geoeffectiveness
Dividing the studied 37 sheath events based on the geomagnetic disturbance they cause, inferred from the SYM-H index during phase), and as expected, substorm activity is greater during geoeffective events as evidenced by the AL index. In other words, geoeffective events are associated with strong shocks, large dynamic pressure and large positive SYM-H peaks at the shock.
Wave power of Pc5, EMIC and chorus waves is higher by a factor of about 7 in geoeffective sheaths as compared to nongeoeffective ones. In geoeffective events, the jump in wave power at the shock is larger in all considered wave modes. For example, the median Pc5 wave power is about 50 times higher during the sheath than before the shock arrival in geoeffective 5 events, whereas in non-geoffective cases it is only about 20 times higher. Pc5 wave power also gradually decreases in geoeffective ejecta, but during less effective ejecta it remains at an approximately constant level that is lower than the median power in the sheath. The median EMIC wave power behaves similarly between the two groups of events.
While the median chorus wave power in geoeffective events increases, on average, by an order of magnitude from pre-event conditions to the sheath region, the chorus activity does not significantly differ between the sheath and ejecta, where it is about cases the upper band chorus wave power is significantly enhanced at the shock (Figure 7k). Non-geoeffective sheaths that are associated with modest substorm activity drive chorus waves only in about a quarter of the events, and the median chorus wave power remains roughly at the pre-event level throughout the ICME (10 −9 nT 2 Hz −1 for lower band and 10 −10 nT 2 Hz −1 for upper band chorus), as opposed to geoeffective events where substorm injections during the ICME excite stronger chorus 5 activity. The median wave power of plasmaspheric hiss is on average twice as high during geoeffective events than during non-geoeffective events.
The median fluxes in the heart of the outer belt experience enhancement from source to core and ultrarelativistic energies in geoeffective events. The strongest increase occurs in the seed population, whose median flux increases by a factor of about 30. During the sheath, the flux gradually increases at source and seed energies, while an initial depletion occurs at the shock 10 impact and the flux is the lowest during the sheath at MeV energies.
For non-geoeffective events, the source and seed populations are enhanced, but higher energy electrons, on the other hand, are depleted both at 1079 keV and 4.2 MeV. This differs from the geoeffective case where enhancement occurred at all four energies. The median electron flux at 54 keV increases throughout the event, but the seed population is gradually depleted during the sheath and the flux begins to increase only after a few hours in the ejecta. The 1079 keV electron flux is slightly 15 enhanced during the sheath before the depletion, while electron losses at 4.2 MeV energies take place already at the shock.
The change in median fluxes is also lower than in geoeffective events, with the largest change being an increase by a factor of 5 at source energies in non-geoeffective events.
Again, the outer belt response as a function of L-shell and a wider range of electron energies is considered and the results are shown separately for geoeffective and non-geoeffective events in Figure 9. It is immediately evident that for geoeffective 20 sheaths, enhancement events are more common at all energies and L-shells, and the source and seed populations are practically always enhanced in the heart of the outer belt (L = 3. . However, deviating from the superposed epoch analysis results, > MeV electrons experience depletion more frequently in geoeffective events throughout the outer belt. In non-geoeffective events depletion begins to dominate the core population response only at around L > 5. Virtually all non-geoeffective events result in no significant change at low L-shells (L < 4.5) at almost all energies, while flux enhancements take place only at 25 source energies and are limited to L > 4.

Discussion and Conclusions
In this paper, we studied statistically the inner magnetospheric wave activity as well as the energy and L-dependent outer radiation belt electron flux response during ICME-driven sheath regions. Our study included 37 sheaths during the Van Allen Probes era (2012-2018). 30 We found that turbulent sheath regions preceding ICMEs caused significant changes in the outer radiation belt electron fluxes. While the response was the most dramatic for geoeffective sheaths, we emphasise that these changes also occurred during the sheaths that caused only a weak geomagnetic storm or that were not geoeffective at all in terms of their SYM-H response. These results are consistent with previous findings that have reported clear responses during small geomagnetic storms (Anderson et al., 2015) and also during non-geoeffective sheaths in case studies (e.g., Alves et al., 2016;Kilpua et al., 2019b). The ejecta in our data set had larger AL and in particular larger SYM-H responses than the sheath regions.
Our analysis showed that the inner magnetospheric wave activity was clearly enhanced in the sheath when compared to the preceding solar wind; Pc5 and EMIC wave power was enhanced by one order of magnitude and chorus wave power was four 5 times higher than in the preceding solar wind. We also found that ULF Pc5 and in particular EMIC wave power were larger in the sheath than in the following ejecta. This is in agreement with a previous case study by Kilpua et al. (2019b). As discussed in Kilpua et al. (2019b), the enhancement is likely due to higher and variable dynamic pressure and more turbulent variations of the magnetic field in the sheaths than in the ejecta. We also note that in the solar wind sheaths have in general a clearly higher level of ULF Pc5 wave power than the ejecta and the preceding solar wind (Kilpua et al., 2013;Hietala et al., 2014).

10
Chorus and plasmaspheric hiss wave power had in turn more similar levels in the sheath and ejecta. In addition, high ULF Pc5 wave power in sheaths can enhance the growth rate of chorus waves (e.g., Coroniti and Kennel, 1970).
Previous studies have statistically investigated sheath response at geostationary orbit (e.g., Hietala et al., 2014;Kilpua et al., 2015) and with radially resolved Van Allen Probes data over several days time periods (e.g., Turner et al., 2015Turner et al., , 2019. In this work, we detailed more precisely the more immediate sheath response over wide L-shell and energy ranges. We found 15 that sheaths deplete relativistic MeV electrons at higher L-shells (down to about L ∼ 4.5). We further showed that the highest were enhanced in the heart of the outer belt, while more energetic seed electrons (> 500 keV) depleted in about half of the cases at high L-shells. Additionally, our example event showed that even weakly geoeffective sheaths can in some cases result 5 in clear outer radiation belt response up to ultrarelativistic energies. Since the sheaths cause enhancements in source electrons but mostly depletion of more energetic seed electrons and the core population, they cannot, statistically, produce the so-called killer electrons (> 1-2 MeV), at least not under the studied timescales.
The results described above agree on a general level with the results of ICME sheath impacts presented by Turner et al. (2019), who only considered events that caused a geomagnetic storm with a SYM-H minimum below −50 nT. However, 10 Turner et al. (2019) found that seed electrons are enhanced more often than the source population and most enhancements occur at L < 4, while our results show the opposite. They also found that depletion of MeV electrons was as likely throughout the outer belt, whereas we show that the immediate depletion is more restricted to higher L-shells. The different results between our study and the one by Turner et al. (2019) are most likely attributed to the difference in the time intervals considered in these studies. We investigated the immediate sheath response over six hours before and after the sheath, while Turner et al. (2019)    In summary, the immediate response to the sheath has a clear energy and L-shell dependence. It is well-known that the highenergy electrons cannot access low L-shells <∼ 4 except during strong magnetic storms and very strong solar wind drivers (e.g., Baker et al., 2014a;Reeves et al., 2016). At low L-shells (L < 3.5), the high percentage of no change events at lower 5 energies (< 300 keV) is a result of the unaffected inner radiation belt population. At larger energies, no change events at L < 3.5 are likely due to the typically weakly populated slot region. At high L-shells (L > 5), the electron fluxes at seed energies do not change much as substorm injections effectively replenish the population (e.g., Turner et al., 2019). One distinct feature we found was the clear energy and L-shells dependence in the losses (Figure 6). Depletion becomes more likely when energy and L-shell increase. Such dependence was not found in previous sheath response studies (e.g., Turner et al., 2019). We suggest 10 that this energy and L-shell dependent depletion can be explained by energy-dependent wave-particle interactions contributing significantly to electron losses in the heart of the outer belt, while magnetopause shadowing at large radial distances depletes all energies equally.
We also found clear differences in the wave activity and energetic electron response between geoeffective (SYM-H minimum ≤ −30 nT) and non-geoeffective sheaths: wave activity is higher during geoeffective events, and enhancement of the source and seed populations and depletion of the core population are more common. In addition, significant response takes place also at lower L-shells for all energies during geoeffective events (similiar to the results presented in Turner et al., 2019), while non-geoeffective events usually cause significant changes only at L > 4. This can be attributed to geoeffective sheaths having tendency for larger dynamic pressure, stronger ring current (SYM-H) and substorm activity (AL). Consequently, they show strong seed energy enhancement due to substorms, while MeV fluxes are depleted more often due to stronger magnetopause 5 shadowing and possible EMIC wave scattering.
The results in this paper roughly agree with the general conclusions of phase space density (PSD) studies. Using PSD analysis of the electron fluxes, Reeves et al. (2013) showed that local acceleration, i.e., energisation via wave-particle interactions (WPI), dominates in the heart of the outer belt. Turner et al. (2013 showed in statistical and case studies that outer belt enhancements during geomagnetic storms are associated with local acceleration via chorus waves. Prompt depletion is 10 consistent with magnetopause shadowing and enhanced outward radial transport, and pitch-angle scattering by EMIC waves leads to precipitation loss (e.g., Turner et al., 2013.
In this work, we detailed the immediate energy and L-shell dependent response of the outer radiation belt to ICME-driven sheath regions. Our statistical analysis showed that the inner magnetoshperic wave activity is enhanced during sheaths, including those sheaths that do not cause a geomagnetic disturbance. Similarly, non-geoeffective sheaths also cause a significant 15 response in the outer belt electron fluxes. Enhancements occur predominantly in the heart of the outer belt at source and seed energies, while the dominant response of the core population is depletion at high L-shells affecting also the seed population at the highest sampled radial distances.
Future work could make use of the PSD analysis method (e.g., Kivelson, 2001, 2004;Chen et al., 2005Chen et al., , 2007Turner et al., 2012;Shprits et al., 2017), which excludes the effects of adiabatic processes, to study sheath response in more 20 detail. With the decommissioning of the Van Allen Probes, future missions surveying the radiation belt environment through various radial distances with high energy resolution are needed for the continuous study of the near-Earth space and its response to solar wind driving. In addition to large-scale missions such as the Van Allen Probes, radiation belt missions can be realised even with cost-effective nanosatellites (Palmroth et al., 2019).
Author contributions. MK carried out the data analysis, prepared the plots and interpreted the results under the supervision of EK and AO.
AJ helped in the interpretation of the results. MK prepared the manuscript with contributions from all authors.
We acknowledge H. Spence and G. Reeves for the ECT data, B. Blake for the MagEIS data and D. Baker for REPT data. We are also thankful to the Van Allen Probes, GOES, Wind and OMNI teams for making their data publicly available.