Electron precipitation characteristics during isolated, compound and multi-night substorm events

A set of 24 isolated, 46 compound and 36 multi-night substorm events from the years 2008–2013 have been analysed in this study. Isolated substorm events are defined as single expansion–recovery phase pairs, compound substorms consist of multiple phase pairs, and multi-night substorm events refer to recurring substorm activity on consecutive nights. Approximately 200 nights of substorm activity observed over the Fennoscandian Lapland have been analysed for their magnetic disturbance magnitude and the level of cosmic radio noise absorption. Substorm events were automatically detected from the local electrojet 5 index data and visually categorised. We show that isolated substorms have limited lifetimes and spatial extents, as compared to the other substorm types. The average intensity (both in absorption and ground-magnetic deflection) of compound and multi-night substorm events is similar. For multi-night substorm events, the first night is rarely associated with the strongest absorption. Instead, the high-energy electron population needed to cause the strongest absorption builds up over 1–2 additional nights of substorm activity. The 10 non-linear relationship between the absorption and the magnetic deflection at high and low activity conditions is also discussed. We further collect in-situ particle spectra for expansion and recovery phases to construct median precipitation fluxes at energies from 30 eV up to about 800 keV. In the expansion phases the bulk of the spectra shows a local maximum flux in the range of a few keV to 10 keV, while in the recovery phases higher fluxes are seen in the range of tens of keV to hundreds of keV. These findings are discussed in the light of earlier observations of substorm precipitation and their atmospheric effects. 15

The threshold value of -50 nT comes from a long-term median of negative AL index values. OMNIWeb solar wind data, which has been propagated to the magnetopause, is used to determine the IMF B Z polarity. Juusola et al. (2011) validated these substorm detection criteria against a list of substorm onsets published by Frey et al. (2004) and concluded that the agreement was good. Instead of the global electrojet index (AL), Partamies et al. (2015) used a regional electrojet index (IL version they called IL ASC ) constructed from baselined data collected at five Lapland magnetometer stations of the IMAGE network (Tanskanen, 2009) stations were co-located with the MIRACLE auroral all-sky cameras (Sangalli et al., 2011), allowing the auroral morphology to be analysed over the same area. They also concluded that the long-term median value of -50 nT was valid for this regional index as well. Furthermore, it is important to note that an earlier study by Kauristie et al. (1996) shows that a local electrojet index, IL (including the entire IMAGE magnetometer network), corresponds well to the global AL index in the magnetic 135 midnight sector . A more recent study by Tanskanen (2009) suggested that the reliable time range could be extended to 16-03 UT. Since the magnetic midnight sector is the most favourable time range for substorm activity and the Lapland latitudes are most of the time under the substorm activity (e.g. Frey et al., 2004), results from nighttime substorm studies over Fennoscandian Lapland should be globally applicable. In this study, we use the Lapland substorm phases detected by Partamies et al. (2015) as a starting point. Thus, the IL index used in this study refers to the regional Lapland index IL ASC 140 throughout the paper. A further visual selection and sub-categorisation of events will be described in section 2.4.

From cosmic noise absorption to a regional absorption index
Measurements of cosmic radio noise absorption (CNA) from a chain of riometers owned and operated by the Sodankylä Geophysical Observatory (SGO) have been used here to describe the substorm EEP impact in the D region ionosphere. Increased ionisation in the D region leads to enhanced absorption of the cosmic noise, and at D region heights this is mainly due to 145 precipitation of electrons with energies above 10 keV (e.g. Turunen et al., 2009). The SGO riometers are wide-beam instruments which listen to the cosmic noise at approximately 30 MHz. CNA is calculated as the reduction of cosmic noise with respect to the quiet background, the so-called Quiet Day Curve (QDC). For the SGO riometer data, the QDC is calculated automatically by fitting a sinusoidal curve to the data of the ten previous days. Our automatically baselined (or QDC subtracted) dataset extends from 2008 until 2013 with a 1-minute temporal resolution. For more instrument details, see for instance Heino Partamies (2020). For this study, we selected the Lapland riometer stations at Ivalo (IVA, 68.56 • N, 27.29 • E), Abisko and Sodankylä to match the magnetometer stations used for the local electrojet index. We further calculate an "absorption index" by aligning the baselined CNA data from the three stations and taking the upper envelope curve, similar to the construction of the global AU or the local IU index. Together with the regional electrojet index (IL), this absorption index allows us to capture the magnetic disturbances and the particle precipitation enhancements occurring within approximately the same geographic 155 area over the same time period. Note, that the terms "absorption" and "CNA" in this paper refer to this regional CNA index.

Space-borne particle precipitation measurements
To characterise the particle precipitation energy spectra during the substorm events, we searched for overpasses of the lowaltitude spacecraft, Defence Meteorological Satellite Program (DMSP) and Polar Orbiting Environmental Satellites (POES).
Together the spacecraft from the two satellite programs cover electron energies from 30 eV up to almost 800 keV. The upward-160 looking spectrometers (SSJ versions 4 and 5) onboard DMSP measure fluxes of downward-going electrons at 19 energy channels from 30 eV up to 30 keV (Redmon et al., 2017). POES observes particles with two different instruments, the Total Electron Detector (TED) and the Medium Energy Proton and Electron Detector (MEPED). TED measures differential electron fluxes in four energy bands (0.15-0.22, 0.69-1, 2.12-3.08 and 6.50-9.46 keV) with telescopes pointing up and at 30 • to the vertical (Evans & Greer, 2000). We used data from the 165 upward-pointing telescope only, which may lead to an underestimation of the precipitation electron fluxes.
Similarly, the MEPED instrument comprises two telescopes, one pointing upwards and another one normal to the first. The measurements consist of fluxes of four integral channels (above 43, 114, 292 and 756 keV, Ødegaard et al., 2017). For these data, we used a combination of the measurements from both telescopes to construct the bounce loss cone fluxes as described by Nesse Tyssøy et al. (2016). Their pre-processed dataset further includes corrections for proton contamination in the electron 170 channels and identification of the relativistic electrons on the proton detector.
To make the MEPED integral fluxes comparable to the TED and SSJ differential fluxes, we converted the observed integral fluxes into differential fluxes. The resulting three flux values are set to the centre energy between the integral channel cutoff energies of 78.5, 203, and 524 keV. The data and the approach are similar to that described by Tesema et al. (2020), except for the extrapolation of the MEPED spectra into lower and higher energies, which we consider unnecessary for the purpose of the 175 current study. All particle data analysed in this study are in the format of overpass-averaged spectra, also used by Tesema et al. (2020).

Event selection and substorm categories
The event categorisation was performed visually using nightly overview plots similar to Figures 1 and 2. The middle panel of each figure shows the temporal evolution of the CNA index, and the bottom panel shows the IL index. The green, red and blue 180 shadings mark the time periods of automatically detected growth, expansion, and recovery phases respectively. As the automatic substorm phase detection routine is also sensitive to small events (thresholded by the IL index long-term median value of -50 nT), a visual inspection was done to exclude events with IL minimum above -300 nT. These mild events are generally not accompanied by an appreciable CNA enhancement. We required two days of quiet time (no automatically detected substorms) prior to all of the event groups described below. This is done to make sure that the activity starts from a solidly quiet back-185 ground, which allows us to determine whether there is a cumulative ionospheric response to the energetic particle precipitation.
Isolated substorms are defined as events with a single expansion-recovery phase pair. An example of an isolated substorm is plotted in Figure 1: the CNA (middle) and IL (bottom) index evolution with colour shadings for substorm phases. The keogram magnetic disturbances and the absorption enhancements. The expansion phase contains the brightest emission, while diffuse emission is seen in the recovery phase.
Compound substorms are defined as consecutive expansion-recovery phase pairs that are not interrupted by quiet time or a growth phase. A similar definition was employed by Sandhu et al. (2019). In case of compound substorms, the substorm onset is the beginning of the first expansion phase. Later expansion phases are called intensifications. Although the threshold 195 for the IL index minimum is -300 nT, we allow an intensification in the middle of the substorm activity to be as small as -100 nT, as long as the lifetime of the expansion exceeds 20 minutes. All these threshold values are, of course, somewhat arbitrary, but they are based on our visual comparisons of the magnetic and absorption signatures from hundreds of events. An example of a compound substorm is shown in Figure 2. This event contains two expansion-recovery phase pairs, as well as a non-detected intensification (IL dip of about 200 nT at 22:04 UT). Prior to the visually approved onset (first black vertical 200 line), a minor substorm event took place. This event was excluded from further analysis due to its low intensity (IL > -300 nT) and a short growth phase between this minor event and the major onset at 19:50 UT. Another minor event that occurred the following morning, well after the main activity, was also ignored. The top panel in Figure 2 again shows the auroral evolution from the SOD camera station as a keogram. The largest expansion (the second red shading in between the black vertical lines) corresponds to both enhancement in the CNA and bright aurora, while the first auroral brightening happened during the 205 excluded event prior to the main event onset at 19:50 UT. The third auroral brightening occurred at around 23:00 UT coinciding with the second major CNA enhancement but deep into the magnetic recovery without any appreciable IL index intensification.
Multi-night substorm events are defined as substorm activity that occurs on consecutive nights. The individual nights during these events consist of either substorms with a single-phase pair or substorms with multiple intensifications. They can look like isolated or compound substorms (as described above) during any of the individual nights, but after a magnetically calm 210 daytime (which is excluded from the analysis), the activity resumes for one or more additional nights. Each night has its own substorm onset, and may include one or more instensifications. In total, the 36 multi-night events include 134 individual nights of substorm activity, most of which were linked into a series of 3-4 nights of activity, but a handful of events was found to continue over 6-7 consecutive nights.
A summary of the identified and categorised events is given in Table 1. Each event duration excludes the growth phase prior 215 to the substorm onset. Each event has been assigned an intensity value, which is the minimum IL index value rounded to the closest 100 nT. The median value of the substorm intensity in all three groups is around -500 nT (not included in the table).
However, the range is very limited for isolated (from -300 to -800 nT) and compound events (from -300 to -900 nT), while it becomes much larger for the multi-night events (from -300 to -1800 nT). It is important to note that in the group of isolated substorms, there is only one event reaching the extreme IL value of about -800 nT. Based on our set of events, a substorm 220 negative bay with IL minimum below about -600 nT is highly uncommon for isolated substorms starting from quiet conditions. The durations and Dst indices given in Table 1 are median values for each substorm type. The range of Dst index values is larger in the group of multi-night events than it is in the other two sub-categories, but as the median values suggest, the typical events are related to very similar ring current enhancements in the groups of compound and multi-night substorm events. The number  The top panel shows the evolution of the green auroral emission as a keogram (north-south slice as a function of time) from the SOD camera station aligned to match the timing of data in the two bottom panels. 9 https://doi.org/10.5194/angeo-2020-56 Preprint. Discussion started: 24 August 2020 c Author(s) 2020. CC BY 4.0 License. of phase pairs is the number of automatically detected expansion-recovery phase pairs during the event, where phases shorter 225 than 20 minutes have been ignored (as described above). In the group of multi-night events, the table shows the number of phase pairs per night. The number of phase pairs varies between 2 and 9 from one event to another in the groups of compound and multi-night substorms. The highest numbers of phase pairs are found during multi-night substorm events.

The relationship between CNA and magnetic disturbances 230
Since the correlation between magnetic disturbances and absorption has been studied before, we want to determine the extent to which our dataset follows the previously established relationship. Figure 3 is a scatter plot showing the minimum IL index and the absorption for each event. The events are colour-coded as red, blue and black for isolated, compound and multi-night events respectively. Values for multi-night events describe individual nights. Although the Pearson correlation coefficient for the full dataset here is -0.6 (with p < 0.01), a number of obvious outliers can be seen in the figure. The correlation in the 235 isolated substorm category is insignificant, while significant correlations (p < 0.01) are found in the categories of compound and multi-night events with correlation coefficients of -0.5 and -0.7, respectively.
The temporal evolution of CNA during the substorm events from 3 hours before to 7 hours after the onset is illustrated by the superposed epoch plots of CNA in Figure 4. The typical evolution of the isolated substorms shows a mild maximum (∼0.5 dB) in the median curve (blue) at the substorm onset time, which then decays during the following two hours. The two other groups 240 of more complex substorms show a slightly higher CNA enhancement in the median curve (up to ∼0.6 dB) at the onset time, which does not recover within the seven-hour time frame shown here, not even in the 25% percentile curve (bottom red curve).
Interestingly, the top percentile for both of these event groups maximises an hour after the onset, as well as 4-6 hours after the onset, which is most likely a signature of multiple substorm intensifications. Note that there is no appreciable CNA difference between the compound and multi-night events. This is probably due to the multi-night events being a mixture of nights with 245 short single phase pair substorms and those with multiple phase pairs and long durations. Figure 5 shows the evolution of the median CNA for eight different IL index intensities from IL= −300 nT to IL< −900 nT.
In general, CNA increases with increasing susbtorm intensity. For the weakest electrojet intensities (IL≥ −500 nT), the maximum CNA occurs at the zero epoch. In contrast, for events with IL≤ −900 nT the peak CNA is delayed by one hour (top https://doi.org/10.5194/angeo-2020-56 Preprint. Discussion started: 24 August 2020 c Author(s) 2020. CC BY 4.0 License.  two curves). For these stronger events the temporal evolution of the CNA is also highly variable, and they are less obviously 250 ordered by the IL intensity.
We further investigate the CNA evolution during the multi-night substorm events, in particular, how the CNA responds to the IL change during the different nights in a series substorm activity. These results (not shown) suggest that CNA often grows

Discussion
Isolated, compound, and multi-night substorm events have been analysed with respect to the magnitude of their magnetic disturbance (IL index minimum) and the related cosmic noise absorption (CNA maximum). About 100 substorm events over 290 the course of 6 years were automatically detected and visually classified. The substorm detection algorithm used in this study has provided well-grounded results in earlier large statistical analyses (Juusola et al., 2011;Partamies et al., 2013). Since the IL index threshold value in this method allows the detection of very mild events which do not produce an appreciable CNA enhancement, we have visually pruned the events to exclude all cases with magnetic deflections less than -300 nT. We have also required two days with no detected substorms prior to any of the events included in this study. This makes it less likely 295 that there would be a significant high-energy particle storage in the radiation belt region, which could be tapped by a solar wind pressure pulse or another solar wind transient, as the nominal loss time for radiation belt particles with energies of the order of 100 keV is from hours to about a day (Summers et al., 2008). Thus, after two days of no substorm injections the particle storage build-up starts from the quiet magnetospheric conditions. This criterion is different from earlier substorm studies, which mostly require a quiet time of about 3 hours prior to isolated substorms. As pointed out by Sandhu et al. (2019), a stronger solar wind 300 driving was often maintained for several days prior to the compound substorm onsets.
It is important to note that most of the studied substorm events are not storm-time substorms, as indicated by the mild median Dst values of about -20 nT. For the multi-night events, which would be the candidates for the strongest magnetic