High Peak Current Lightning and the Production of Elves

Elves are observed as expanding rings of light in the UV and visible optical bands. They are produced when electromagnetic pulses from lightning discharges interact with the lower parts of the ionosphere. Elves are well known to be associated with high peak current lightning discharges. Here, we use data from the Modular Multi‐spectral Imaging Array (MMIA) of the Atmosphere‐Space Interactions Monitor (ASIM), and search for observations of Elves when high peak current lightning discharges are detected by Vaisala's Global Lightning Detection network GLD360. We present two groups of events; high peak current detections associated with Elves and high peak current detections not associated with Elves. To understand why some current pulses with high peak currents do not produce observable Elves, we investigate and compare the lightning activity occurring before these two types of events, in terms of both the number of lightning discharges detected by GLD360 and the peak currents of the preceding discharges. Our results, using data from GLD360, suggest that current pulses with peak currents above |120| kA tend to produce Elves nearly always, regardless of the preceding lightning activity. For current pulses with peak currents between |70| and |120| kA, the number of observed Elves might be affected by the preceding lightning activity, or is the result of the characteristics of the storm cells that produce the Elve.


Introduction
Elves are produced at altitudes of ∼80-95 km when electromagnetic pulses (EMPs) from lightning discharges encounter the lower parts of the ionosphere (Tomicic et al., 2023;van der Velde & Montanyà, 2016).They are observed from space as laterally expanding rings of light in the UV and visible optical bands, expanding to diameters of several hundred kilometers (Fukunishi et al., 1996;Inan et al., 1996Inan et al., , 1997;;van der Velde & Montanyà, 2016).Elves are associated with very high peak current lightning discharges, and although most lightning discharges are detected over land, lightning discharges associated with Elves are more frequently observed over ocean (Chen et al., 2008;Liu et al., 2010;Peterson et al., 2017).Elves were first observed in the early 1990s, accompanying cloud-to-ground (CG) lightning discharges, and they can also be associated with Narrow Bipolar Events (NBEs) and Energetic In-cloud Pulses (EIPs) associated with intracloud lightning (Neubert et al., 2020(Neubert et al., , 2021;;Østgaard et al., 2021).
Several threshold peak current values for Elve production have been suggested from both observations and modeling studies, with different experiments using different approaches and lightning detection networks.Some networks provide a peak current estimate (e.g., GLD360), whereas others give an energy estimate (WWLLN).Modeling results by Inan et al. (1996) and Marshall et al. (2015) suggest threshold peak current values for production of observable Elves to be ∼80-90 kA for CG strokes.Results from the Imager of Sprites and Upper Atmospheric Lightning (ISUAL) experiment suggested that most Elves (∼90%) are associated with peak currents above 60 kA (Chen et al., 2008).Blaes et al. (2016) studied the global occurrence rate of Elves and predicted Elve occurrence with 90% accuracy based on peak current detections by National Lightning Detection Network (NLDN).They used a logistic regression model to predict a 50% probability of Elve occurrence at peak currents of 88 kA and a 90% probability at peak currents of 106 kA.Newsome (2010) and Tomicic et al. (2023) showed that there are many high peak current (>200 kA) CG strokes that do not generate observable Elves.
Strong EMPs from lightning are believed to affect the ionization of the lower ionosphere on very short timescales, but these changes are also believed to accumulate over time (Inan et al., 1991(Inan et al., , 1996)).The EMPs increase the electron attachment to O 2 and reduce the electron density at the bottom of ionosphere (Inan et al., 1996;Shao et al., 2013), but perhaps more important is the increased ionisation a few km above, which gives a steeper density gradient and a higher reflective altitude (Gordillo-Vázquez et al., 2016;Kotovsky & Moore, 2017).Inan et al. (1996) suggested that although the decrease in the electron density caused by the EMPs is small (∼3%), this could contribute to observed "early" very low frequency (VLF) disturbances, as reported by Inan et al. (1993)."Early" events are, as outlined by Haldoupis et al. (2013), a type of lightning-induced VLF perturbations where the delay of the VLF perturbation from the lightning return stroke is <100 ms (Inan et al., 2010).A type of "early" events with recoveries over scales of many minutes, called LOREs (long-recovery early events), were found by Haldoupis et al. (2013) to be caused by EMPs emitted by especially high peak current CG lightning (>250 kA).The production of LOREs approaches unity from the rare group of lightning discharges that have peak currents above 300 kA, which are also likely to generate Elves.
In this paper, we identify and analyze two groups of events; high peak current lightning associated with Elves and high peak current lightning not associated with Elves.To identify these events, we performed a search through Modular Multi-Imaging Array (MMIA) data for observable Elves when a peak current above |40| kA is detected by GLD360 within the MMIA field of view (FOV).To understand why some high peak current lightning discharges do not generate observable Elves, we investigate the lightning activity preceding the two types of events.We explore the number of lightning discharges detected and their reported peak currents within different time intervals before the events in the two classes, to get an indication of the impact of a disturbed ionosphere on the production of Elves.We will also discuss whether the peak current values found to produce Elves are just due to the characteristics of the thundercloud cells.

Instruments and Data
For this study we used data from the Modular Multi-spectral Imaging Array (MMIA) instrument onboard the Atmosphere-Space Interactions Monitor (ASIM) on the International Space Station, as well as lightning radio atmospheric detections by the ground-based global lightning detection network GLD360.Onwards, we refer to lightning radio atmospherics as "sferics".ASIM consists of the two detector modules Modular X-and Gammaray Sensor (MXGS) and MMIA, which are described in detail in Østgaard et al. (2019) and Chanrion et al. (2019), respectively.We used ASIM data from 2019 and 2020 when the payload was in its initial configuration, mounted on the lower starboard side of the Columbus module, facing nadir.ASIM has an absolute timing uncertainty of ∼[ 10, 40] ms, due to the transfer of ISS clock to ASIM.

MMIA
The MMIA instrument consists of three photometers and two cameras.The photometers operate in three different wavelength bands; 337 nm (4 nm bandwidth), UV (nominally 180-230 nm) and 777.4 nm (5 nm bandwidth), with a 100 kHz sampling rate (Chanrion et al., 2019).The two cameras operate in the 337 and 777.4 nm bands.The photometer and camera data are organized into observations that consist of up to 8 consecutive frames.The MMIA photometers and cameras are mounted on an optical bench, which has a tilt of 5°to limit disturbances from other instruments on the ISS, into the FOV of the MMIA cameras.The photometers operating in 337 and 777.4 nm have a square 80°-diagonal FOV.The UV photometer has a circular 80°-diameter FOV.Without taking the 5°tilt into account, this maps to a square FOV on Earth surface with the shortest distance to the edges being ∼234 km from the ISS footpoint, with the corners and the UV FOV edge being at ∼345 km from the footpoint.As the tilt can cause the FOV to shift by approximately 35 km, we limit the search for high peak current sferic detections from GLD360 to an area with a radius of 200 km from the ISS footpoint, as illustrated in Figure 1.
The 777.4 nm band is dominated by emissions from the hot lightning leader, and corresponds to atomic oxygen emissions.Hence, this band is the most sensitive to emissions from hot lightning leaders.Emissions in the 337 nm band correspond to emissions in the second positive system of Nitrogen molecules (SPN2), and this band is dominated by emissions from streamer corona, but will also detect signals from the hot leader.The UV band corresponds to the Lyman-Birge-Hopfield (LBH) system of molecular nitrogen (Chanrion et al., 2019).Emissions in the UV band from lightning are to a large extent absorbed by molecular oxygen in the air, making the UV photometer particularly suitable for detections of Elves and other Transient Luminous Events (TLEs) (Chanrion et al., 2019).As the UV from lightning is mainly absorbed in the air, it is not expected to observe significant signatures in the UV from the lightning itself.However, weak UV signals that are not from Elves are often observed when a pulse from lightning is detected in 337 and 777.4 nm.By comparing intensities of the UV pulse at these events and those from lightning, and from observing UV signals when a blue event (blue jets or blue starters) is observed by the 337 nm photometer, we conclude that these signals are highly likely leakage from the blue.This means that the UV photometer does not strictly observe signals in the 180-230 nm range, but sometimes also detects signals from up to ∼300 nm.

GLD360
Lightning discharges emit radio waves in a broad range of frequencies peaking in the VLF range.Due to their low attenuation, these waves can propagate thousands of kilometers in the Earth-ionosphere waveguide.The lightning detection network used in this study, GLD360, consists of sensors operating in the range ∼500 Hz to ∼50 kHz (Said & Murphy, 2016).GLD360 uses a time of arrival technique to locate lightning discharges, using a minimum of three sensors, and provides peak current values, but does not distinguish between CG and IC events.The median location accuracy was reported by Said et al. (2013) to be 2.5 km for lightning over the United States, and the ground flash detection efficiency is 57%.The detection efficiency is higher for CG flashes (∼80%) than for IC flashes (∼45%) and the detection efficiency is lower over Africa and South America.

Methodology
To identify high peak current detections that could be associated with MMIA observations, we performed a search for sferic detections by GLD360 within 200 km from the ISS footpoint and less than 2 s from an MMIA trigger onset time.As we were looking for high peak currents we limited our search to detection of values above/below ±40 kA.This peak current value is motivated by previously suggested thresholds for Elve production and the lower values of peak current that have been associated with Elves, as discussed in Section 6.The 200 km limit was selected to ensure that we are inside the FOV of the MMIA photometers (as outlined in Section 2.1).
For candidate sferic matches, a series of consecutive GLD360 sferic detections (minimum two) were aligned to the MMIA 777.4 nm pulses to identify the exact pulse detected by this channel that is associated with the GLD360 sferic under investigation.This time alignment was performed because of the timing uncertainty of ASIM, and follows the same procedure as outlined in Lindanger et al. (2022), Bjørge-Engeland et al. (2022), and Skeie et al. (2022).This is also similar to the procedures used by Maiorana et al. (2021) and Heumesser et al. (2021).The time alignment procedure reduces the absolute timing uncertainty from [ 10,40] ms to a few ms.Only timealigned events are used in this study.

Distinguishing Between Optical Signals
UV emissions from lightning are to a large extent absorbed in the air by molecular oxygen, and hence it is not expected to observe significant signatures in the UV band from the lightning itself.However, the UV photometer often detected weak signals when signatures of a lightning pulse were observed by the other two photometers.To distinguish whether a UV signal detected around the time of a high peak current detection originated from an Elve or from the lightning itself, we investigated the pulse shapes detected by the three MMIA photometers.We used a similar approach for identifying UV signals from Elves as in Bjørge-Engeland et al. (2022).Emissions in the 337 and 777.4 nm bands from lightning are elongated and delayed due to scattering in the cloud when they are detected by the MMIA photometers, whereas the UV signal from an Elve will be nearly unscattered by the time it reaches the photometers (Chanrion et al., 2019).We therefore expect the signal from an Elve detected by the UV photometer to have shorter risetime and reach a peak before the pulses from lightning observed by the 337 and 777.4 nm photometers.UV signals associated with Elves are usually more intense than the observed signal from the lightning discharge in 337 and 777.4 nm, whereas UV signals caused by the blue leakage at 300 nm, as described in Section 2.1, tend to be much weaker than the signals observed in 337 and 777.4 nm. Figure 2 shows an example of a detection of an Elve by the UV photometer and signatures of the lightning pulse detected by the 337 and 777.4 nm photometers.The UV pulse in panel b has a much shorter risetime than the 337 and 777.4 nm pulses, and the UV pulse also reaches a peak significantly before the other two pulses.The GLD360 sferic detection associated with this event is ∼100 km from the ISS footpoint, and the reported peak current is 86 kA.The event is located over north-east India.An example of an event where a UV pulse from lightning is detected by the UV photometer following the other signals in 337 and 777.4 nm is shown in Figure 3.The GLD360 sferic detection is located ∼165 km from the ISS footpoint, with a peak current of 81 kA.The event is located over the north of Vietnam.
To distinguish the pulse shapes and identify the events where a UV signal is not associated with an Elve, the photometer signals were normalized (as in Figure 4).This enabled studying the pulse shapes rather than the pulse intensity.For the event in Figures 3 and 4, the risetime of the UV pulse (panel b) is not shorter than the risetime of the pulses detected by the 337 and 777.4 nm photometers (panels a and c).The UV pulse follows the shape of the other pulses, and does not peak before the 337 and 777.4 nm pulses, which is different from the UV pulse of an Elve shown in Figure 2. Hence, this event was categorized as a high peak current event with no associated Elve.The UV signal in Figures 3 and 4 is rather due to the leakage from the blue channel, as described in Section 2.1.
The full MMIA-trigger containing the event shown in Figure 3 is shown in Figure 5.The event shown occurs in the second frame of the trigger, with very little activity observed by the 337 and 777.4 nm photometers before the event.Figure 6 shows a projection of the MMIA FOV, as well as the 777.4 nm camera, onto a map.A projection with the 337 nm camera shows the same.The light blue circle in the camera image represents the location of the GLD detection associated with the event shown in Figure 3. Through distinguishing the optical signals associated with high peak current GLD360 detections, the events were organized into two groups: • Elve events (E): High peak current detections associated with an Elve • No-Elve events (N): High peak current detections with no associated Elve

Storm Cells and Preceding Lightning Activity
To investigate whether the lightning activity preceding the Elve and No-Elve events impacted whether an Elve was produced or not, we examined the number of sferic detections by GLD360, as well as the peak currents reported by the network.We used three search windows; 5, 20 and 40 min before the events were detected.These search windows are motivated by the results in Haldoupis et al. (2012Haldoupis et al. ( , 2013)), and will be discussed in Section 5.2.For studying the preceding lightning activity, we restricted the geographical region to ±2.5°around the location of the Elve or No-Elve event.This is illustrated in Figure 7, where the blue circle represents the FOV of the UV photometer, and the lightning symbol marks the location of a high peak current detection by GLD360.A search box of ±2.5°, which at equator corresponds to ∼280 km, around the location provided by GLD360 is used to investigate the preceding lightning activity, and is illustrated in Figure 7 by a box around the lightning symbol.This search box was selected to include an area that is large enough to account for local disturbances of the ionosphere by EMPs from previous activity.Although we do not know if previous high peak current detections were associated with Elves, all high peak current lightning discharges emit an EMP which may disturb the ionosphere.Some stroke detections by GLD360 can be from the same flash, but in this study we focus on the power of the EMP associated with individual strokes, which is what could impact the ionosphere.

Observations
Using MMIA triggers and GLD360 detections of peak currents >|40| kA from a cluster of days in February and March 2019, December 2019 and July 2020, we have identified 127 Elve events and 276 No-Elve events.The geographical distribution of the events is shown in Figure 8.In the samples, there were several intense storms with both Elve and No-Elve events occurring within a time span of several minutes.In this section, we first explore an example of a storm where Elve and No-Elve events occur interchangeably (Section 4.1), before outlining the number of events in each group of events (Section 4.2) and exploring the lightning activity preceding the Elve and No-Elve events (Section 4.3), to investigate why some current pulses with high peak currents do not produce observable Elves.

An Example of a Storm
During a storm occurring over Indonesia on 3 March 2019, GLD360 reported six detections of peak currents over |70| kA within less than a minute, occurring within 200 km from the ISS footpoint and within MMIA triggertimes.The events are summarized in Table 1, and three events are categorized as Elve events, and three as No-Elve events.All events are produced in the same storm cell in the middle of the map in Figure 9, with an Elve being produced first.
Figure 10 shows the photometer signals of the first two events in Table 1, an Elve event and a No-Elve event, respectively.The time separation between the events is approximately 32.5 ms.A zoomed-in view of the second  event is presented in Figure 11, where the 337 and 777.4 nm photometers pick up signals from the lightning discharge, and no signal is detected by the UV photometer.Figure 12 shows the photometer signals of the other four events in Table 1, and Figure 13 shows zoomed-in views of each of the four events.The first event is an Elve event, and has a very high peak current of 222 kA, and an intense UV signal is detected by the UV photometer.
The next event, occurring nearly 78 ms later, has a peak current of 80 kA and does not have an Elve associated , where the events in Table 1 were detected.The largest cell, in the middle of the map, is considered the main cell and is where the events in Table 1 occur.Included in this plot are sferic detections 40 min before the first event in Table 1.The red triangles highlight the sferic detections with peak currents above |40| kA ±5 min around the time of the first event in Table 1.
Figure 10.MMIA photometer detections for the first two events in Table 1.The first event is classified as a high peak current sferic associated with an Elve, and the second as a high peak current sferic without an associated Elve.
with it.The third event, with a peak current of 115 kA is associated with a very dim Elve, from a distance of approximately 155 km from the ISS footpoint.This event occurs approximately 54 ms after the No-Elve event.
All the four events occur within the same MMIA trigger.These strokes occur so close in time and location that they could be part of the same lightning flash.
Figure 11.MMIA photometer detections for the second event in Table 1, which has a high peak current of |87| kA, but no Elve is detected by PHOT2.
Figure 12.MMIA photometer data for the last four events in Table 1.In Figure 14 we present the peak current distribution of an hour of GLD360 sferic detections, within ±2.5°centered at the GLD360 detection location, before the first event in Table 1.The median peak current (using absolute values) is 14 kA, and the mean peak current is 29 kA.There are 11 GLD360 sferic detections with peak currents above |200| kA (by this we mean peak currents with a magnitude higher than 200 kA irrespective of its sign) in the main storm cell in the hour before the first event in Table 1, occurring from about 50 to 4 min before the first event.An event with a peak current of 358 kA occurs 10 min before the first event.

Overview of All Events
The number of events contained in the two groups "Elve events" and "No-Elve events" (as outlined in Section 3.1), is presented in Table 2, separated into intervals of peak current.The majority of the No-Elve events have peak currents below 70 kA, and only two No-Elve events have peak currents above 120 kA.In the interval 70-120 kA, the two samples have nearly the same number of events.

Lightning Activity Preceding Events
To explore whether the preceding lightning activity would affect the peak current needed for producing an Elve, we investigated the lightning activity occurring within 5, 20 and 40 min preceding the Elve and No-Elve events, as outlined in Section 3.2.The preceding lightning activity is shown in Figures 15 and 16, respectively, divided into intervals of peak current.The same plot for 40 min is shown in Figure S1 in Supporting Information S1.The intervals are defined as number of sferic detections with peak current above a certain value within the minutes before the Elve/No-Elve event.The top panel in each plot shows the number of sferic detections by GLD360 with peak currents above |300| kA within ±2.5°of the location of the Elve/No-Elve event.The three other panels in each plot show the number of preceding detections of peak currents above |200|, |100| and |50| within the time interval.The figures indicate that for peak currents above |120| kA, an Elve is nearly always produced, regardless of the preceding activity.For events with peak currents below |100| kA, but above |70| kA, the preceding lightning activity seems to impact whether an Elve is produced or not.There are more high peak current lightning  discharges preceding the No-Elve events than the Elve events on this interval.For peak currents below |70| kA, the number of Elves decreases rapidly.The activity 5, 20 and 40 min before the events in the two groups is summarized in Table 3.
In Table 3 we present differences in the lightning activity preceding events with peak currents above |70| kA in the two groups.For all three time intervals (5, 20 and 40 min) of preceding lightning activity, the mean number of sferics with peak currents above |100| kA detected before the No-Elve events is higher than that before the Elve events.The mean of the sum of the peak currents (as a proxy for total energy) before the events in the No-Elve group is also higher than that for the Elve events.Additionally, the number of sferic detections preceding the No-Elve events, within the geographical area outlined in Section 3.2, is significantly higher for the No-Elve events.

Discussion
In this study, we used sferic detections from the GLD360 lightning detection network, aligned with optical signals both in time and space, as it provides peak current estimates, has a good global coverage and a relatively high detection efficiency.However, since different lightning detection networks report peak currents and energy estimates of lightning discharges differently, proper comparisons to peak current values in other research works can be challenging, and a peak current estimate should not be considered an exact value.GLD360 sometimes reports the ionospheric reflection (often with a higher peak current) rather than the ground wave (Cummins et al., 2008;Said et al., 2013).This could lead to an overestimation of the peak current values.The lightning detection networks could also underestimate the peak current values, for example, for cases where an Elve is associated with a NBE.Peak currents from fast pulses will be underestimated due to the bandwidth used by the GLD360 network.
We have identified two groups of events; high peak current sferics associated with Elves and high peak current sferics not associated with Elves.Among the events in the two groups outlined in Section 3.1, there are only eight Elves associated with peak currents below 70 kA.In the peak current range 70-120 kA, our sample has almost the same number of Elve and No-Elve events.For peak currents above 120 kA, nearly all (∼96%) the events in our sample are Elve events.By studying the electrical activity preceding the Elve and No-Elve events, our results indicate that for Elves associated with peak currents in the range 70-120 kA, the preceding lightning activity may influence whether an Elve is produced or not.Our results may suggest that a less disturbed ionosphere enables the production of Elves from lightning pulses with lower peak currents.We will first compare our findings with previous results, and then discuss why the threshold peak current value for producing Elves changes.2014) concluded that the lowest energy associated with Elves detected by the ISUAL experiment corresponds to a peak current of ∼38 kA.Barrington-Leigh and Inan (1999) used Fly's Eye, which is a groundbased instrument, to study Elves.Using a lower peak current threshold of 38 kA, they found 50% of the NLDN detections within the FOV to be associated with Elves.Using Elve observations by the PIPER instrument, which has the advantage of continuous data acquisition rather than triggered, and peak current detections by NLDN, Blaes et al. (2016) used a linear regression model to present an Elve production probability (in their Figure 6), given as a function of the peak current, with the probability of Elve production reaching 90% for peak currents above 106 kA.We find that 96% of peak currents above 120 kA produce Elves, and for peak currents above | 70| kA we find that 62% produce observable Elves.For peak currents above 88 kA, the model used by Blaes et al. (2016) predicted an Elve probability of 50%.Barrington-Leigh and Inan (1999) identified Elves and used sferic detections by NLDN, with all their observed Elves being associated with peak currents above 56 kA.

The Lower Threshold of Peak Current to Produce Elves
The sequences of Elve and No-Elve events shown in Figures 10 and 12 show that Elve and No-Elve events can happen interchangeably within the same storm cell.Interestingly, Figure 12 shows a very high peak current Elve (222 kA) followed by a No-Elve event with a lower peak current (80 kA), before another Elve with a lower peak current (115 kA).Modeling results by Marshall et al. (2010) suggest that for CG strokes with peak currents above ∼56 kA, the EMPs are likely to cause changes in the local electron density of the ionospheric D-region.Furthermore, they suggested that intense IC and NBEs can disturb the electron density, and that the parameters of the discharge itself, as well as the ionospheric density and the Earth's magnetic field affect the interaction between the EMP and the lower ionosphere.

What Defines the Varying Threshold
As outlined by Haldoupis et al. (2013) and suggested by Inan et al. (1996), Elves that have been triggered by very intense CG discharges can be followed by long-lasting recoveries of the conductivity in the D-region of the ionosphere.The recoveries outlined by Haldoupis et al. (2012Haldoupis et al. ( , 2013) ) range from 5 to 30 min.They found that stronger events tend to be followed by stronger perturbations of the ionosphere.Motivated by the time windows explored by Haldoupis et al. (2012Haldoupis et al. ( , 2013)), we used three intervals of lightning activity before the events; 5, 20 and 40 min, and searched for sferic detections within a geographical cell of ±2.5°.The high peak current lightning discharges preceding the Elve and No-Elve events could be associated with Elves, but this is speculation as we do not have optical measurements for the preceding interval.Haldoupis et al. (2013) found that ∼35% of the CG discharges in their data set with peak currents exceeding 200 kA produced LOREs.Newsome (2010) showed that there are many CG lightning discharges with high peak currents that do not generate observable Elves.This is in agreement with our sample of events, where there is a significant number of sferic detections with peak currents above ∼70 kA that are not associated with Elves (∼40% in our sample).Our results do not indicate a hard threshold peak current for Elve production, but rather suggest that for sufficiently high peak currents an Elve is nearly always produced (96% for peak currents above |120| kA).Tomicic et al. (2023) analyzed a large set of Elves detected during a single storm over the Adriatic Sea.They used data from a TLE observer in France, and lightning data from GLD360 in combination with data from a broadband electric field sensor in France, an ELF receiver system in Poland and a VLF receiver in Slovakia.Their Figure 10 shows that all their Elve events where a LORE is also observed are produced at altitudes above 86 km.Tomicic et al. (2023) found that the majority of the GLD360 detections of peak currents above 200 kA occurring within their FOV were not associated with observable Elves, in contrast with our results.However, their study focused on CG strokes, whereas our sample contains both CG and IC discharges.From waveforms of the vertical electric field of the lightning strokes, they found no difference between those associated with Elves and those not associated with Elves.For CG strokes with peak currents above 200 kA, they found that all Elve-producing strokes were associated with peak currents above 228 kA.
Next, we will discuss two possible candidate explanations for the varying threshold for Elve production: 1. Pre-activity disturbing the lower ionosphere 2. Type of storm cell

Pre-Activity Disturbing the Lower Ionosphere
When an EMP from lightning interacts with the lower ionosphere, the free electrons present there are accelerated, and the collision rate between charged particles and neutrals increases (Newsome, 2010).If the EMP has a strength above a certain threshold, the process of electron attachment to neutrals dominates, and the electron density of the lower ionosphere decreases.When the electron density is lower, the reflection height of the ionosphere is raised, and an EMP could reach higher altitudes, where the neutral density is lower (Newsome, 2010).If a very strong EMP interacts with the ionosphere, the process of ionization of neutrals dominates over attachment processes, and the electron density of the lower ionosphere increases.When the electron density is higher, the reflection height of the ionosphere is lowered, to where there is a larger neutral density.The neutral density there is too large for the electrons to get accelerated to the energies needed for excitation, and the EMP is just attenuated (van der Velde & Montanyà, 2016).Hence, an Elve cannot be produced unless the EMP is strong enough to compensate for the attenuation.This is needed for the same reduced electric field to be obtained, which is when an Elve could be produced.
We present in Figure 17 the distribution of the maximum peak current detected by GLD360 within 5 min before Elves with peak currents above 120 kA or below 70 kA.We identified 54 Elves associated with peak currents above 120 kA, and just eight Elves associated with peak currents below 70 kA.Although there are very few Elves with peak currents below 70 kA, the distribution could indicate that a less disturbed ionosphere allows for Elve production with lower peak current lightning pulses, and that when the ionosphere is more disturbed, current pulses with higher peak currents are needed to produce Elves.The median number of sferics before the Elves with peak currents above 120 kA is 117, and the median number of sferics before those with peak currents below 70 kA is 62.
The lowering of the ionosphere in response to very strong EMPs could explain both the distribution in Figure 17 and the sequences of events in Table 1 and Figure 10, where an Elve associated with a very high peak current is observed, and tens of milliseconds later, a No-Elve event is observed.The conditions for Elve production are not present for this value of peak current when the ionosphere has been lowered to where the neutral density is too high.Each No-Elve event in Table 1 is associated with a lower peak current than the Elve produced tens of milliseconds before, which points to the EMP not being strong enough to compensate for the attenuation, and the conditions not being present for obtaining the same reduced electric field.
To study this effect further, we present in Figure 18 the distribution of Elve/No-Elve events with all preceding GLD360 detections within 5 min prior to the Elve/No-Elve events having peak currents below |300|, |200| and |  100| kA.Panel (c) shows that when the preceding lightning activity only has peak currents below |100| kA, the peak current needed for producing an Elve is also below |100| kA.The other two panels, (a) and (b), show that when the preceding lightning discharges have peak currents up to |200| and |300| kA, respectively, Elves are produced at higher peak currents.This supports the candidate explanation that larger peak current values are needed for producing an Elve when the ionosphere is disturbed, but, as will be discussed, it can also be explained differently.
In Figure 19 we show a logistic regression of the Elve/No-Elve events for panel c in Figure 16, with the y-axis showing the number of sferic detections by GLD360 with peak currents above |100| kA within 20 min prior to the Elve/No-Elve events.By fitting our data, the logistic regression estimates the probability of an event being Elve or No-Elve as a function of the peak current and the preceding lightning activity.The provided p-value measures the confidence with which we can reject the null hypothesis that the preceding lightning activity has no influence on the Elve/no-Elve features.As the p-value obtained in Figure 19 is low, it may indicate that the previous lightning activity influences whether an Elve is produced or not.Similar logistic regression plots for sferic detections above |200| kA and for 5 min of previous lightning activity are shown in Figures S3-S5 in Supporting Information S1.

Type of Storm Cell
Another possible explanation for the distributions in Figures 17 and 18 is that the Elves associated with high and low peak currents are likely to be observed over different types of storm systems/cells.The distribution in Figure 17 indicates that the Elves associated with the lowest peak currents are produced in storm cells with generally lower peak currents, whereas the Elves associated with peak currents above |120| kA can also be preceded by very high peak current lightning discharges.Figure 18 shows that low peak current events in both groups tend to be preceded by other low peak current lightning pulses, suggesting that some storm cells primarily produce lightning discharges associated with very high peak currents, whereas other cells primarily produce lower peak currents.We could speculate that an explanation for this could be the dipole moment of the cell, with a large dipole moment in a cell being associated with the higher peak current pulses.This would imply that for a large cell, most of the peak current detections should be large, which could disturb the ionosphere.This could also mean that the spatial scales are larger.However, Figure 19 and the p-value obtained more strongly supports the explanation of the pre-activity disturbing the lower ionosphere.

Summary and Conclusion
Through exploring the lightning activity preceding two groups of events (high peak current detections associated with Elves, and high peak current detections not associated with Elves), we can conclude the following: 1.For peak currents above |120| kA, Elves are produced nearly always, regardless of the preceding lightning activity.2. For peak currents in the range 70-120 kA, our results may indicate that the number of observed Elves depends on the lightning activity in the minutes and tens of minutes before.3. Considering all the Elve and No-Elve events with peak currents above 70 kA, GLD360 detected a larger number of preceding lightning discharges, a larger number of preceding peak current detections above 100 kA and a higher mean sum of detected peak currents before the No-Elve events than before the Elve events.4. We can suggest two possible explanations for the two groups of events; • A more disturbed ionosphere requires current pulses with high peak currents in order for an Elve to be produced.If a very strong EMP interacts with the ionosphere, the electron density of the lower ionosphere increases, and the reflection height of the ionosphere is lowered to a region of higher neutral density.An Elve cannot be produced at this altitude unless the EMP is sufficiently strong to compensate for the attenuation experienced by the EMP.• In a storm cell that only produces low peak current discharges, it is more likely to observe an Elve from a peak current below |70| kA.

Figure 1 .
Figure 1.Illustration of the Modular Multi-spectral Imaging Array field of view (FOV) and the search window for sferic detections with high peak currents.The red cross is the ISS footpoint and the blue cross is the center of the FOV (tilted 5°).

Figure 2 .
Figure 2. Example of the photometer signatures of an Elve in the UV photometer data (panel b), and optical signals from lightning in the other two photometers (panels a and c).

Figure 3 .
Figure 3. Example of an event where a UV pulse (panel b) is observed at the same time as optical pulses are observed by the 337 and 777.4 nm photometers (panels a and c) from lightning.

Figure 4 .
Figure 4. Normalized photometer signals for the event shown in Figure 3.

Figure 5 .
Figure 5. Full Modular Multi-spectral Imaging Array (MMIA)-trigger, containing the event shown in Figure 3 in the second frame.The red line shows the GLD360 sferic detection, after time-alignment has been performed.

Figure 6 .
Figure 6.Projection of the MMIA FOV onto a map, including a projection of the image by the 777.4 nm camera.The light blue circle is at the location of the GLD360 sferic detection associated with the event.

Figure 7 .
Figure 7. Illustration of how the geographical area for searching for preceding lightning activity was selected.The blue circle is the event search window from Figure 1, and the box is the window used to search for sferic detections by GLD360 preceding an Elve/No-Elve event.The size of the box is 560 km only at equator.

Figure 8 .
Figure8.The geographical distribution of the Elve and No-Elve events, using the locations of the sferics reported by GLD360.For both Elve and No-Elve events more than 85% of the events occurred over ocean or coastal regions (within 150 km from the shoreline).

Figure 9 .
Figure9.Map of the storm systems over Indonesia on 3 March 2019, where the events in Table1were detected.The largest cell, in the middle of the map, is considered the main cell and is where the events in Table1occur.Included in this plot are sferic detections 40 min before the first event in Table1.The red triangles highlight the sferic detections with peak currents above |40| kA ±5 min around the time of the first event in Table1.

Figure 13 .
Figure 13.Zoomed-in plots of the MMIA photometer data for the four events in Figure 12.For weak UV signals, a moving mean line was added for clarity.

Figure 14 .
Figure 14.Peak current distribution of one hour of GLD360 sferic detections from the main storm cell in Figure 9, in bin sizes of 20 kA.

Figure 15 .
Figure 15.Number of high peak current detections by GLD360 5 min before the Elve and No-Elve events, within a geographical cell of ±2.5°.(a) Number of preceding sferic detections with peak currents above 300 kA, (b) Number of preceding sferic detections with peak currents above 200 kA, (c) Number of preceding sferic detections with peak currents above 100 kA, (d), Number of preceding sferic detections with peak currents above 50 kA.

Figure 16 .
Figure 16.Number of high peak current detections by GLD360 20 min before the Elve and No-Elve events, within a geographical cell of ±2.5°.(a) Number of preceding sferic detections with peak currents above 300 kA, (b) Number of preceding sferic detections with peak currents above 200 kA, (c) Number of preceding sferic detections with peak currents above 100 kA, (d), Number of preceding sferic detections with peak currents above 50 kA.

Figures
Figures 15 and 16 show Elves with peak currents down to 45 kA, which is in agreement with earlier reported values down to ∼40 kA Elves detected by the ISUAL experiment.Using Equation 2 inHutchins et al. (2012),Chen et al. (2014) concluded that the lowest energy associated with Elves detected by the ISUAL experiment corresponds to a peak current of ∼38 kA.Barrington-Leigh and Inan (1999) used Fly's Eye, which is a groundbased instrument, to study Elves.Using a lower peak current threshold of 38 kA, they found 50% of the NLDN detections within the FOV to be associated with Elves.Using Elve observations by the PIPER instrument, which has the advantage of continuous data acquisition rather than triggered, and peak current detections by NLDN,Blaes et al. (2016) used a linear regression model to present an Elve production probability (in their Figure6), given as a function of the peak current, with the probability of Elve production reaching 90% for peak currents above 106 kA.We find that 96% of peak currents above 120 kA produce Elves, and for peak currents above | 70| kA we find that 62% produce observable Elves.For peak currents above 88 kA, the model used byBlaes et al. (2016) predicted an Elve probability of 50%.Barrington-Leigh and Inan (1999) identified Elves and used sferic detections by NLDN, with all their observed Elves being associated with peak currents above 56 kA.

Figure 17 .
Figure 17.Distribution of the highest detected peak current in the 5 min before two sub-groups of the Elve events; Elves with PeakCurrents >120 kA and <70 kA.

Figure 19 .
Figure 19.Plot showing panel (c) in Figure16with computed logistic regression.On the y-axis is the number of sferic detections by GLD360 with peak currents above |100| kA detected within the 20 min preceding the Elve/No-Elve events.The p-value is a measure of the confidence with which we can reject the null hypothesis that the preceding lightning activity has no influence on Elve/No-Elve events being produced.

Table 1
High Peak Current Detections Within 2 s of MMIA Trigger Times During the Storm

Table 2
Summary of the Two Groups of Events

Table 3
Summary of Lightning Activity Preceding the Elve and No-Elve Events >70 kA