The Sun-synchronous, low-altitude ( ∼ 850 km), polar orbiting National Oceanic and Atmospheric Administration (NOAA) Polar Orbiting Environmental Satellites (POES) and European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) MetOp spacecrafts have since 1978 carried the Medium Energy Proton and Electron Detectors (MEPED) instrument. Each satellite orbits the Earth 14–15 times each day. During the latest decades, a constellation of up to six operating satellites has allowed for a more global magnetic local time coverage.

The MEPED instrument includes two directional electron telescopes and two directional proton telescopes, as well as an omni-directional detector for energetic protons (16–140 MeV) (Evans and Greer, 2004). The field of view of both the 0 ° and 90 ° telescopes is 30 ° full width. The nominal energy limits of the electron telescopes are listed as > 30, > 100, and > 300 keV with upper energy cutoff of 2,500 keV. In operation, the limits will, however, depend on the incoming energy spectrum (Yando et al., 2011; Ødegaard et al., 2017). Ødegaard et al. (2017) utilize the geometric factors given in Yando et al. (2011) to determine new optimized effective integral energy limits > 43, > 114, and > 292 keV and associated geometric factors based on a wide range of realistic power law and exponential spectra. False counts due to proton contamination in the electron detectors are corrected by the simultaneously detected proton fluxes. The proton spectrum is first adjusted for degradation due to radiation damage by applying correction factors derived by Sandanger et al. (2015) and Ødegaard et al. (2016). Subsequently, the proton flux in the energy ranges known to impact the respective electron channels (Evans and Greer, 2004) are then subtracted from the electron flux observations. In the absence of protons in the fifth proton channel, the presence of a relativistic electron count rate in the sixth proton channel is registered as > 756 keV electron fluxes (Nesse Tyssøy et al., 2016; Ødegaard et al., 2017).

The 0 ° and 90 ° telescopes observe different parts of the electron pitch angle distribution. In the main precipitation regions, the 0 ° telescope detects electrons at pitch angles within the bounce loss cone (BLC), while the 90 ° telescope detects electrons at pitch angles both within and outside the BLC (Rodger et al., 2010; Nesse Tyssøy et al., 2016). Hence, the partial pitch angle coverage makes observations from either telescope a coarse estimate for the precipitating electron fluxes, where the 0 ° and 90 ° will under- and overestimate the flux of precipitating electrons, respectively (Nesse Tyssøy et al., 2016). A more realistic estimate can be achieved by combining the observed fluxes from both telescopes together with theoretical electron pitch angle distributions calculated based on wave-particle interactions in the inner magnetosphere. As described in detail in Nesse Tyssøy et al. (2016), the Fokker-Planck equation for particle diffusion (Kennel and Petschek, 1966; Theodoridis and Paolini, 1967) is solved for a wide range of diffusion coefficients. The solutions are then transformed to the satellite altitude and organized in a lookup table. The ratio between the fluxes detected by the 0 ° and 90 ° detector is used to identify the pitch angle distribution that best fits the observation taking into account the viewing directions of the telescopes relative to the magnetic field. Finally, the size of the BLC is calculated from the magnetic field strength provided by the International Geomagnetic Reference Field (IGRF) in order to estimate the precipitating fluxes.

In this study, the BLC fluxes from the four electron channels using the new optimized effective integral limits > 43, > 114, > 292, and > 756 keV, denoted as E1, E2, E3, and P6, are used. It applies all available MEPED data from the MEPED telescopes over a full solar cycle from 2004–2014 from both hemispheres. Daily averages are first created for the MLT sectors 0–6, 6–12, 12–18, and 18–24 MLT. The global daily flux is then estimated by averaging the electron fluxes across the MLT sectors over the corrected geomagnetic (CGM) latitude bands: 55°–70°N in the NH and 55°–70°S in the SH. This ensures equal representation of the MLT sectors independent of satellite MLT sampling. Note, however, that this method might mask the energy dependent precipitation characteristics associated with L-shell and MLT.

This study uses the event selection denoted “absolute events” in Salice et al. (2024). The global daily > 43 keV (top panel) and > 292 keV (bottom panel) electron flux data from 2004 to 2014 is shown in Figure 1. The black lines represent the global flux values. The electron fluxes with solar proton events (SPEs) exceeding 200 protons cm − 2 s − 1 sr − 1 have been excluded from the flux data from the SPE onset to 20 days after due to proton contamination dominating the counts in the MEPED electron detector and making the proton correction uncertain (Nesse Tyssøy et al., 2016). These excluded segments are shown in light grey (peaks without event identifications). The circles in the upper panel denote the > 43 keV electron peaks above the 90th percentile ( ∼ 1.4 × 1 0 5 cm − 2 s − 1 sr − 1 ) . To ensure isolation, the > 43 keV flux had to fall below its median value ( ∼ 2.2 × 1 0 4 cm − 2 s − 1 sr − 1 ) between flux peaks. Moreover, events with data gaps during the five first days succeeding the peak flux were ignored. Based on these criteria, 164 distinct peaks were selected. The associated flux peaks in the > 292 keV channel were identified as the maximum flux in a 4-day interval starting from the onset of the > 43 keV flux peaks. The third lowest and highest > 292 keV flux peaks are classified as E1 and E3 events, respectively. The dividing thresholds between the events in the > 292 keV flux are marked in the bottom panel: the 33.3rd percentile (blue line) at ∼ 2.8 × 1 0 3 cm − 2 s − 1 sr − 1 and the 66.6th percentile (red line) at ∼ 4.2 × 1 0 3 cm − 2 s − 1 sr − 1 . This results in 55 E1 and 55 E3 events, shown as blue and red circles in Figure 1, respectively. Onset in this paper, as in Salice et al. (2024), is the peak of the > 43 keV flux.

The solar wind flow is categorized into slow solar wind speed, high-speed streams (HSSs), and coronal mass ejections (CMEs) based on their solar origins and near-Earth solar wind parameters. Based on Richardson and Cane (2012) the following definitions are applied:

• HSS: Both high-speed streams and the associated co-rotating interaction regions (CIR) are labeled HSS. HSSs originate from coronal holes and hold solar wind speeds of v ≳ 450 km/s. CIRs are compressed regions between the fast streams and the preceding slower, cooler, and denser solar wind.

Identification and timing of substorms is based on the criteria given in Newell and Gjerloev (2011). A list covering the full solar cycle is available as an open source list at the https://supermag.jhuapl.edu/substorms. This study uses the daily sum of substorm onsets.

(Salice et al., 2024) hypothesized that high and sustained solar wind speed in the recovery phase of a storm increases the substorm onset rate, which ensures sufficient electron acceleration and subsequent scattering into the loss cone from both the ring current and radiation belts. This study test this hypothesis by examining solar wind parameters and geomagnetic indices for the events associated with Dst and Kp extremes:

• The E1 event with the strongest Dst deflection

Moreover, it investigates the predictability of E1 and E3 events combining the Dst and Kp indices. The resulting ambiguous events are finally evaluated with respect to solar wind parameters and substorm onset rates.

Table 1 summarizes the Dst and Kp extremes for the previous case studies. The associated probability of being an E3 event given by (Salice et al., 2024) accounting for the respective solar wind structure is listed in parentheses. It demonstrates that a strong Kp maximum might be accompanied by weak Dst deflection and vice versa. Using the probability assessments given in Salice et al. (2024) independently, might therefore label the same storm as either E1 or E3 event dependent on which index is applied. The left plot in Figure 7 shows the minimum Dst and maximum Kp values associated with the 110 E1 and E3 events. If ignoring the outlier presented by the E3 event with the weakest Dst deflection, 40 E1, and 25 E3 events are correctly classified by the Dst limits ≥ −26 nT and ≤ −48 nT. The Kp*10 boundaries of ≤ 33 and ≥ 40 correctly classifies 37 E1 and 31 E3 events. Note that the given boundaries are slightly different from the ones listed in Salice et al. (2024) as the data is not binned. Interestingly, the Kp-index partly classifies a different subset of E1 and E3 events compared to the Dst-index. This makes a combined determination of the limits highly effective, leaving only 9 E1 and 19 E3 events ambiguous.

As highlighted in Salice et al. (2024), if the solar wind driver is known, the Kp and Dst boundaries can be made more specific, further increasing the prediction factor. This is illustrated in the middle and right panels in Figure 7. For HSS-driven storms, the boundaries determining an E1 event stay the same, while the boundaries determining an E3 event change to Dst ≤ −35 nT and Kp*10 ≥ 37. This leaves only three ambiguous HSS E1 events and five ambiguous HSS E3 events, as listed in Table 2. For CME-driven storms, the Dst boundary determining an E1 event changes to Dst ≥ −33 nT, while those for Kp and E3 events stay the same. This renders only four ambiguous CME E1 events and five ambiguous CME E3 events, as listed in Table 3.

The 3-h magnetic activity index, Kp, is known to correlate with the strength of the magnetospheric convection electric field (Thomsen, 2004). It is proposed by Thomsen (2004) that the physical reason for this is that Kp is sensitive to the latitudinal movement of the auroral currents due to its measurements at sub-auroral latitudes. These currents map to the plasmasheet, whose motion is determined by the strength of the convection field. Moreover, the Kp index has previously been used to model ULF Pc5 pulsation power (Ozeke et al., 2014). Hence, if the daily averaged Kp values are accompanied by strong solar wind speed in an HSS structure, it is likely to be associated with Pc5 acceleration. The Kp index will, however, also be affected by other sources, such as an intensification of other ionospheric and magnetospheric currents. This implies that it will also be correlated to substorm generation.

Salice et al. (2024) finds Kp as one of the best parameters independent of the solar structure for predicting with more than a 95% probability an E1 or E3 event, accounting for 56% of events. The CME-driven E1 event with the strongest daily Kp*10 maximum (40), shown in Figure 4, was associated with solar wind speed less than 500 km/s, which, together with a northward turning Bz component on the zero epoch days, limits the Pc5 pulsation power and electron acceleration. In this case, a fairly high Kp level does not support inward radial acceleration due to Pc5 pulsation in generating high-energy tail acceleration.

Despite having the weakest Kp*10 response of 33, the CME- and HSS-driven storms shown in Figures 5, 6 evolve into E3 events. The long-lasting but weak Kp enhancements in panel G are, in both cases, accompanied by strong solar wind speed in panel F and a negative Bz component in panel E. The Dst deflection and elevated substorm rates in panels F and H appear moderate but long-lasting. Both events do, however, display a gradual build-up/broad peak in the lower energy channels in panels A and B. This build-up suggests that sustained acceleration processes over multiple days might be the key to explaining high-energy tail precipitation as these fluxes, on average, reach their peak one to 2 days after the > 43 keV does as illustrated in panel C. If this is the case, ULF Pc5 acceleration driven by high speed and negative Bz, and accompanied by sustained moderate substorm activity might be responsible for storms developing into E3 events.

The Dst index is assumed to reflect the strength of the magnetospheric ring current. Studies have, however, pointed out its limitation in climatological studies and quantitative modeling as its baseline changes over time (e.g., Olsen et al., 2005; Lühr and Maus, 2010; Babu et al., 2022; Babu et al., 2023). The outlier E3 event with the weakest Dst deflection, shown in Figure 3, might be subject to this bias. It occurs in a period of very strong geomagnetic activity close to solar maximum. Olsen et al. (2014) developed the RC index from stations at mid and low latitudes. RC was originally designed to account for the magnetospheric contributions during geomagnetic quiet conditions and used for deriving models of the Earth’s main field. However, Lühr et al. (2017) demonstrated that RC is a good alternative to Dst during more disturbed conditions as well. For example, the E3 event with the weakest daily Dst of −16 nT occurred on 01-March-2004 in an active period after the solar maximum. The corresponding RC value is −23 nT. In comparison, the E1 event on 20-Jul-2011 has a minimum Dst of −13 nT in a moderate period after the solar minima. The corresponding RC value is −8 nT. Hence, the Dst comparison is partly hampered when comparing events from different parts of the solar cycle as baseline removal damps the Dst value in periods of high geomagnetic activity and vice versa.

The long measurement record of the Dst index going back to 1957 is of high value in respect to a potential EEP parametrization in, e.g., chemistry climate models. Moreover, a rough indication of the strength of the ring current is still powerful when assessing if an event is associated with high-energy tail precipitation. The negative Dst deflection manifests the presence of trapped ions and electrons of a few to 100 s of keV in the inner magnetosphere. The subsequent decay of the ring current is a potential source of high-energy tail precipitation, particularly if the recovery phase is accompanied by chorus waves scattering the respective electrons into the loss cone. Salice et al. (2024) suggested this as the explanation for the high predictability of Dst in classifying E1 and E3 events.

Moreover, Salice et al. (2024) hypothesized that strong solar wind speed in the recovery phase of a deep Dst through would increase the probability of an E3 event, as there is a close link between substorm probabilities and solar wind speed (Newell et al., 2016). The case studies show, however, that there is far from a one-to-one scaling between the respective parameters. For example, a substorm onset rate of 12 was associated with a solar wind maxima of less than 400 km/s on 01-Nov-2011, as seen in Figure 2, and a substorm onset rate of 10 was linked to a solar wind speed greater than 623 km/s on 23-Jun-2013, as seen in Figure 5. Hence, actual substorm onset rates are needed to assess the proposed hypothesis.

Interestingly, the outlier event shown in Figure 3 partly invalidates the suggested hypothesis as it lacks a deep Dst minimum. Still, both the persistent high values of Kp and solar wind speed support ULF Pc5 acceleration and sustained substorm activity, which is likely to generate chorus wave acceleration and scattering. Moreover, the elevate prestorm > 292 keV fluxes support that a significant high-energy tail electron population already exists in the radiation belt.

The E1 and E3 events are identified from the 90th percentile of the > 43 kev flux over a full solar cycle. Hence, the selection criteria in itself guarantees the existence of source and/or seed electrons, providing favorable conditions for relativistic electron enhancement in the outer radiation belt (Jaynes et al., 2015). Salice et al. (2023) showed a strong correlation of ∼ 0.9 between the > 43 and > 292 keV flux peaks, although for a specific > 43 keV peak, the corresponding > 292 keV peak could vary by an order of magnitude.

The E1 event with the strongest > 43 keV flux strength is shown in Figure 8. It is one of three HSS E1 events that is ambiguous after applying Salice et al. (2024)’s Dst and Kp criteria. Hence, both the ring current increase and general geomagnetic activity are at a moderate level. The daily average solar wind speed reaches close to 700 km/s and remains elevated in the recovery phase of a relatively deep Dst through. Nevertheless, the substorm onset rates are below 3 throughout the recovery phase of the storm, and both the AE and Kp, as well as the > 43 keV flux strength, drop by epoch day one. Hence, the storm provides transient seed electrons by ULF PC5 and/or chorus waves acceleration, but the intermittent substorm activity likely limits the chorus wave generation and subsequent pitch angle scattering needed to cause loss to the atmosphere.

In contrast, the E3 event with the weakest > 43 keV flux strength (not shown) is deemed an E3 event by Salice et al. (2024) because of its strong Dst deflection. Moreover, the overall geomagnetic activity, substorm activity, and the seed and source electrons are elevated over several days. As such, it is not the > 43 keV peak flux level that determines the > 292 keV peak flux level. Several of the E3 case studies displayed here have a broader, longer-lasting flux response in the lower energy channels compared to the E1 events, suggesting sustained acceleration sources. This feature was also pointed out in the SEA analysis by Salice et al. (2024). It also falls in line with both Miyoshi et al. (2013) and Tyssøy et al. (2021), emphasizing that the high-energy tail response is a result of the accumulated response to sustained acceleration processes.

Miyoshi et al. (2013) performed a statistical analysis of HSS/CIR events and found that the events with a dominant southward Bz component considering the Russel-McFerron effect are associated with > 2.5 MeV electron flux enhancement of the outer radiation belt. The acceleration source responsible for the elevated flux levels is substorm generated chorus waves, which also implies pitch angle scattering and loss to the atmosphere. The SEA in Salice et al. (2024) shows that both E1 and E3 events have predominantly negative Bz values near the zero epoch day. The negative Bz deflection is, however, more prominent for the E3 events compared to the E1 events. Of the 55 E3 events, only one is associated with a positive daily average Bz component. Of the 55 E1 events, 7 are associated with a positive daily average Bz component. Although not statistically significant, it appears as a daily positive Bz reduces the probability of an E3 event. It should, however, be noted that the main requirements for large enhancement of the electrons and the whistler mode waves in the outer radiation belt during the recovery phase are fast solar wind speed and an IMF-Bz fluctuating around zero or more directed predominantly southward (Iles et al., 2002). Therefore, if Bz is averaged over a day, it is difficult to distinguish between the events with an oscillating Bz component and those that are persistently positive or negative.

The importance of a negative Bz component is likely twofold. Miyoshi et al. (2013) suggested that the IMF-Bz dependence of the flux enhancement of relativistic electrons may be attributed to sustained substorm activities. This is largely in line with the case studies shown here. A negative Bz is likely to accommodate moderate substorm activity compared to a positive Bz. This dependence is best demonstrated by focusing on how the substorm onset rates drop when the Bz turns positive, as shown in Figure 4 or 5. Moreover, Miyoshi et al. (2013) points out that during the HSS events with negative Bz, the shrinkage of the plasmapause implies a decrease in thermal plasma density, which allows for more efficient flux acceleration through substorm-generated chorus waves. Chorus waves are also responsible for scattering the high energy tail into the loss cone. Hence, Bz affects both the substorm onset rates as well as their efficiency in accelerating and scattering the electrons.

For cases where both the Dst and Kp indices are insufficient in classifying an event as E1 or E3, a negative Bz on epoch day zero or increase the likelihood of an E1 event. The northward turning of Bz effectively slows down the energy transfer from the solar wind to the magnetosphere. This reduces the substorm activity and makes the remaining substorm onsets less efficient as the plasmapause will expand and reduce the area of chorus wave growth (Miyoshi et al., 2013). As a result, applying the concurrent criteria of Dst, Kp, and Bz correctly classifies 108 of the 110 E1 events but will wrongly classify one of the E3 events as an E1 event.

In general, it is easier to exclude an E3 event than to determine one. For the ambiguous E3 events, the key appears to be enough time to nurture continuous growth. Hence, sustained moderate substorm activity and ULF Pc5 pulsation over multiple days can achieve the same high-energy tail precipitating electron fluxes as events with much stronger Dst and Kp deflections.

Time to nurture flux growth over multiple days is in line with Tyssøy et al. (2021), who developed an AE-based MEE proxy accounting for the AE activity over multiple days. The AE-based proxies were shown to account for at least 70% of the observed MEE precipitation variance at all energies. Salice et al. (2024) tested the AE-based model on the 110 E1 and E3 events and found that it captured the general features of the SEA flux analysis but failed to identify the individual E3 events. Assuming that AE is representative of the substorm onset rate, a higher substorm onset rate in the recovery phase alone does not appear to be exclusively able to explain the high-energy tail of MEE precipitation found for E3 events. This notion is in line with the findings by Miyoshi et al. (2020), which demonstrate that the high energy tail precipitation is determined by the upper limit of the cyclotron resonance energy of the chorus waves, rather than the presence or absence of the chorus wave excitation. The higher energy electrons resonate with the high latitude propagation of chorus waves (Miyoshi et al., 2010; Miyoshi et al., 2015). None of the standard geomagnetic indices are able to reflect this important nuance.