BARREL Observations of Microburst Events With a Slowly‐Varying Component

Electron microburst precipitation has been shown to have significant potential for depletion of the outer radiation belt. We present observations from the Balloon Array for Radiation belt Relativistic Electron Losses (BARREL) of six (five unique and one dual‐balloon observation) microburst events, each containing minutes to hours of persistent microbursts. We find that each event included a slowly‐varying smooth precipitation component underlying the bursty component. The smooth component has not yet been fully characterized in the literature; we have written a program to identify microburst events and quantify the relative contributions of each component in the BARREL data. In all six events analyzed, the smooth component contributed more to the total X‐ray counts measured, indicating that the smooth component could contribute significantly to radiation belt loss.


Introduction
Earth's radiation belts are highly variable, with an ever-changing particle content (e.g., Ripoll et al., 2020;Thorne, 2010).Acceleration and injections increase the population of trapped electrons while various loss processes compete to decrease the population (e.g., Millan & Thorne, 2007).Wave-particle interactions can contribute to both acceleration and loss.Characterizing wave-particle interactions and electron losses in the radiation belts has been the subject of increasing focus, both from the perspective of preserving human interests in space as well as utilizing the radiation belts as a large-scale laboratory to probe the fundamental physics of energetic particle dynamics.
The nature of the wave-particle interaction and pitch-angle scattering mechanism that generates microbursts has remained an open question despite having been studied since the 1960s.The need for and effectiveness of nonlinear interactions is under current investigation (e.g., Chen et al., 2020;Santolík et al., 2010).Particle flux estimates for microburst events have been successfully reproduced with quasilinear theory, but nonlinear interactions are required to reproduce the bursty structure (e.g., Mozer et al., 2018).The presence of a more slowlyvarying precipitation component accompanying microbursts has been mentioned in the literature (e.g., Lampton, 1967;Reinard et al., 1997), but to the authors' knowledge, has not been studied in detail.A two-component behavior for microburst precipitation was also noted briefly in B. R. Anderson et al. (2017), and the Reimei satellite observed pulsating aurora simultaneously with energetic electron precipitation that showed two components (Miyoshi et al., 2015).The connection between the bursty and smooth components, and the connection with pulsating aurora are not currently understood.
We report observations from the Balloon Array for Radiation belt Relativistic Electron Losses (BARREL) mission of six microburst events (five unique, one observed simultaneously on two different balloons) that show a two-component precipitation structure.That is, between individual microbursts (bursty component), the observed count rate does not return to background levels and is more slowly-varying (smooth component).Balloon-based observations are ideal for quantifying the relative contributions of the two components to particle loss since the balloon detects only particles that are precipitating; most satellite observations observe a mix of precipitating and trapped particles because of their large field of view (Tu et al., 2010).Moreover, the balloon platform is essentially stationary on the timescale of a single microburst, or even a microburst event (defined here as a 10s-100s of minutes interval throughout which microbursts persist), enabling better separation of temporal and spatial variations over microburst timescales.We quantify the relative loss contribution of the smooth and bursty components for six microburst events observed by BARREL.This smooth component has not been accounted for in previous wave-particle interaction predictions and flux estimates from microburst events.All six events examined show that the smooth component makes up a significant fraction of the observed X-rays from microburst events.This has implications for inclusion of chorus-driven loss in radiation belt models and may be important for understanding the details of the wave-particle interactions that cause microbursts.

Observations and Analysis
BARREL observed microburst events that have two main features: a bursty precipitation component and an underlying smooth precipitation component.Figure 1 shows one such microburst event observed on 13 August 2015 (the spatial extent and duration of this event was studied in detail in B. R. Anderson et al. (2017)).Persisting for several hours, the event contains thousands of individual short, intense bursts of precipitation (bursty component points, shown in cyan).Between these bursty points (which appear to be resolved temporally), and even before they start and after they desist, the count rate does not remain at or return to background levels.BARREL electronics have a processing time of ∼7 μs so this cannot be due to an instrumental effect or scattered photons.Rather, the individual microbursts are superimposed over a more slowly-varying "smooth" precipitation component (magenta points).The two precipitation components are distinguishable from background both visually and quantitatively.We have developed an automated program that goes through the BARREL campaign 3 data and finds microburst events, classifies the data points for each of the two precipitation components, then separates and integrates the components to estimate their relative contribution to the measured bremsstrahlung, as described in the sections below.

BARREL Microburst Observations
The BARREL mission was designed to study electron precipitation by measuring bremsstrahlung X-rays produced by electron collisions with neutrals in the atmosphere during precipitation (see e.g., Sample et al., 2020) (Millan et al., 2013).From 2013 to 2020, 58 balloons, each equipped with a sodium-iodide scintillator, were launched during seven campaigns (1, 2, 6, 7 in Antarctica, 3-5 in Sweden).BARREL collected data at multiple energy and time resolutions, including 50 ms from 25 keV to 1.5 MeV, covering the microburst energy range at sufficient cadence to resolve individual microbursts (Woodger et al., 2015).The BARREL data set consists of thousands of hours of observations.In order to sift through the large volume of data, we wrote an automated program in Python to identify microburst events (minutes-long to hours-long time periods during which microburst precipitation was observed) that applies thresholds for dispersion and power spectral density (PSD) to the data, following the process described below.This method was developed using the campaign 3 data which consists of 7 payloads flown in Sweden during 2015.Results discussed in this paper are from campaign 3 observations only.

Microburst Event Identification
Microburst events, often lasting up to several hours, are found in the data by applying thresholds for dispersion, PSD, and microburst occurrence rate.Figure 2 shows an example of the process for finding a microburst event on 13 August 2015.For each payload, data (summed from the first three energy channels, ∼25-180 keV) are first split into 10-minute time windows.The dispersion statistic (which is the variance to mean ratio weighted by the size of the window, see Equation 1) is then calculated for each window, with N as the number of points in the window, σ 2 w as the window variance, and μ w as the window mean.
We expect the background count rate to be dominated by X-ray counts produced from cosmic rays.These are expected to have Poisson distributed count rate measurements; in a window that consists entirely of background observations, the dispersion statistic should be close to that of a Poisson distribution.A Poisson distribution has a variance equal to the mean, thus σ p ≈ 1 (for large N).Data that are more varied than a Poisson distribution (expected when microbursts are present) have σ p > 1.To differentiate between background and potential microbursts, the windows are classified to be "background" (having only static precipitation with σ p < 1.2) or "replete" (having changing background or other phenomena-driven precipitation with σ p ≥ 1.2) (Payne et al., 2017).Figure 2a shows background (black) and replete (red) intervals meeting these criteria.
The replete windows are further examined to identify those likely to contain microburst precipitation.As microbursts have a period of 100-500 ms, the windows that contain microbursts are expected to show more temporal variation at frequencies from 2 to 10 Hz compared to the windows with only background or other non-microburst precipitation.The PSD is calculated for each time window and integrated from 2 to 10 Hz.A 5% increase in the integrated PSD as compared to the background windows was used as a threshold to identify candidate microburst windows.This parameter was empirically motivated based on manual inspection of the selected windows (with the intention of setting it as low as possible to not eliminate any microbursts).The leading and trailing non-flagged windows adjacent to candidate microburst windows were also flagged to avoid potentially missing burst points because of the window placement.These windows are shown in red in Figure 2b.
After identifying the candidate microburst windows, these windows are inspected more closely for microburst content to determine which should be kept for analysis.To do this, a modification of the methods in O'Brien (2003), Shumko, Blum, and Crew (2021), and Douma et al. ( 2017) is used to identify individual microbursts in the data.For each of the candidate microburst windows, a 10 s rolling median, m b , of the bottom 50% (in magnitude) of the rolling data is taken (green line in Figure 3a).The "spike factor," N m , which is somewhat analogous to the number of standard deviations above the rolling median, is then calculated for each point.
here, n c is the number of counts from ∼25-180 keV in each 50 ms bin.Data points with N m ≥ 8 are considered to be microburst points.Microburst candidate windows that have an average occurrence rate of one microburst point per minute (10 total minimum in the window) are kept.To form the microburst events, windows that have sufficient microburst occurrences, and the windows in between them, are grouped together.Windows that have more than 2 hr between them were considered separate events.The start and end of the event is then truncated to the first and last microbursts that occur in the grouped data.Events were inspected manually for verification.An example microburst event selection is shown in red in Figure 2c.This program was run on BARREL Campaign 3 observations and six microburst events were found (with five unique events and one dual-balloon event observation).All six events occurred on the dayside (with MLT: 7.6-13.5,L-Shell: 5.4-6.7 calculated from the T89 magnetic field model with a Kp of 2 (Tsyganenko, 1989)), which is consistent with expected locations from previous microburst observations (e.g., Douma et al., 2017;Shumko, Blum, & Crew, 2021;Tsurutani et al., 2013).

Background Subtraction
As the first step to analyzing the microburst events, we must remove the background X-ray counts from the count rate data.For this analysis, background counts are defined operationally as described below.It is expected that measured background counts are primarily those from cosmic rays.In the energy range of this work, the cosmic ray background count rate depends mostly on the altitude of the balloon and the energy measured, with higher energy cosmic rays penetrating farther down into the atmosphere, hence also the overall total number of ambient X-rays also increasing with decreasing altitude.
Background definition and subtraction is done separately for each payload using the background windows that were identified with σ p < 1.2 for the given flight.Background windows with an average count rate of 75 counts/ 50 ms or less were selected (to exclude data that were flagged as background but were actually just relatively constant precipitation over the window interval).A linear fit to the number of counts per second per keV versus balloon altitude is created for each of the three energy channels using all of the data points in the selected background windows.For each microburst event, background counts at each point for each energy channel are determined using the altitude at that point along with these fit functions.The background counts are then subtracted from the total counts at each point for each channel.For data points during ephemeris data dropouts, altitudes were interpolated from balloon location prior to and following dropout.Figure 1 shows one example of a background-subtracted microburst event occurring on 13 August 2015.

Bursty and Smooth Component Identification
All six of the background subtracted microburst events examined show a persistent underlying precipitation structure (magenta points in Figure 3) concurrent with the bursty component (cyan points in Figure 3).Specifically, the X-ray count rate does not return to background levels between individual microbursts, indicating a more slowly varying precipitation component (here called the smooth component).To separate the smooth and bursty precipitation components, we first classified all of the data points in the microburst event using Equation 2. The previously identified microburst points (with N m ≥ 8) are categorized as part of the bursty component, points that have N m ≤ 3 are part of the smooth component (Figure 3a), and points that fall between 3 < N m < 8 are not classified to either component.For the example shown in Figure 3, of the over 275,000 total data points in the microburst event, about 13% of the data points were assigned to the burst component, 70% to the smooth component, and 17% left unclassified.With this identification, the smooth component is clearly visible in between the individual bursts and does not appear to be solely due to an overlap (or lack of resolution) of the microbursts, though some overlap in microbursts can be seen, for example, in Figure 3c near 08:50:12UTC.

Loss Contribution Estimates for Smooth and Bursty Components
After identifying the two precipitation components for the events in Campaign 3, we examined their relative contributions to the measured bremsstrahlung.Points attributed to the smooth component were isolated and subtracted from the total X-ray counts at each point.To create a continuous smooth component count rate, a linear interpolation was done to connect the nearest (temporally) smooth component values in the summed 25-180 keV data (see Figure 3b).It should be noted that this gives X-ray counts associated with all unclassified points to the burst component, which places an upper limit on the contribution from the bursty component (and gives a minimum for the counts from the smooth component).This was done in order to determine whether the smooth component, which has not been previously considered, contributes significantly.This also ensured that small bursts that were otherwise too weak to meet the threshold criteria were not missed in the bursty component count rate estimate.
It should also be pointed out again that microburst events were defined as the time from the first to last identified microburst burst points only.The smooth component generally appears to begin sooner and/or persist longer than the bursty component in the microburst events, so choosing these endpoints would again ensure that we are finding an upper limit for the bursty contribution.Table 1 shows the integrated X-ray count contribution as a percentage for the two components for all events.In each of the events examined, the smooth component contributes a majority percentage of the total X-ray counts, despite having maximized the opportunity for the bursty component to contribute by including all of the unclassified points into the bursty component and the choice of event endpoints.As a check on the methodology, the rolling median from Equation 2 was used instead of a linear interpolation to fit a continuous count rate for the smooth component, and the background count rates were varied by a constant ± ̅̅̅̅̅ ̅ N b √ counts/50 ms (where N b is the number of background counts subtracted from the data).These changes resulted in an average component variation of ±4 percentage points on the relative contributions.The events that peaked near mid-morning have a larger bursty contribution (see events 1, 5, and 6 in Table 1).

Conclusion
In conclusion, BARREL observed six microburst events in August 2015 that exhibit a slowly-varying smooth precipitation component in addition to the much shorter-duration microbursts that have been the primary focus of previous studies.That is, even after subtracting out the background from these events, there is a clear excess in the count rate data that underlies the individual microbursts (i.e., the bursty component).For the events examined, these two components did not always appear to have a direct correlation in intensity, and the smooth component was observed to often start before and/or stop later than the bursty component.This could be phenomenological either as a temporal effect, if the components do not have the same build-up and wind-down timescales, or a spatial effect, if the two components do not have the same spatial scale and the balloon drifted in and out of the component precipitation regions.
An automated detection algorithm was developed to identify the microburst events and to identify individual microbursts within each event.The six events identified by the algorithm had durations ranging from about 1-4 hr, and each contained many individual microbursts.The smooth and bursty components were separated, and a These events appear to be the same event observed simultaneously on two separate balloons located approximately 65 km apart (based on event time, location, and precipitation profiles).
their relative contributions to the total X-ray count rate produced by precipitating electrons were determined.In all six events, the smooth component contributed a majority fraction of the total observed X-ray count rate.This component has not been included in previous microburst flux estimates or theoretical predictions, due in part to the large field of view of LEO satellite measurements conflating trapped and precipitating particles.Our results show that this component should be taken into consideration for further investigation; in particular, the results suggest that the smooth precipitation component could contribute significantly to radiation belt loss and subsequent energy deposition into the atmosphere.

Discussion and Future Work
Questions still remain about the process that scatters electrons into Earth's atmosphere during a microburst event, both for the purpose of understanding the large-scale impact of microburst loss contributions as well as the small-scale wave-particle dynamics that cause them.The results presented here motivate further theoretical work that can account for both the smooth and bursty components, and explain their relative strength.
In particular, the role of non-linear interactions to explain microburst events has remained an open question in the field.Microburst flux estimates consistent with observations have required non-linear driving to reproduce (e.g., Osmane et al., 2016).However, time-averaged flux estimates of microburst events have been accurately reproduced using quasilinear theory (though non-linear interactions were needed to produce the characteristic microburst temporal profiles) (e.g., Santolík et al., 2010).Since the present work suggests that the smooth component could dominate the loss rate, a time-averaged flux profile of the two components in combination would represent primarily the smooth component, which occurs over a much longer timescale than the microbursts.A quasi-linear diffusive process may be sufficient to explain and model the smooth component and thus capture the bulk of the loss.The need for including nonlinear interactions to model microburst events may depend on the timescale being modeled and the relative intensity of the two components.Simulations have also shown that non-resonant wave interactions can create significant amounts of precipitation in addition to what is produced from resonant interactions during a precipitation event (e.g., Denton et al., 2019).Additionally, the structure of wave fields at the time of microburst events (i.e., the presence of chorus risers simultaneous with weaker amplitude waves, frequency and spacing properties, etc.) could be studied to shed light on the two components.Understanding the nature of both the smooth and bursty components could provide insights into the need for nonlinear interactions to describe loss from microburst events on a largescale as well as illuminate some of the micro-physics of the wave-particle interaction and microburst scattering mechanism.
Understanding, with a goal of eventually predicting, expected radiation belt electron loss is crucial for characterizing the space environment around Earth and improving our ability to protect the satellites on which our technological society depends.To fully quantify the relative electron loss rates of the two components observed during microburst events, future work will carry out the spectral inversion needed to convert from measured X-ray counts to electron fluxes.Future work will also look at the entirety of the contribution from the smooth component instead of just the portion of the events when frequent and strong microbursts were occurring.Additionally, the burst and smooth components contain further temporal structures ranging in scale from seconds to minutes, which will be investigated.The program that we have written to find microburst events can be applied to the entirety of the BARREL data to conduct a statistical analysis of the two-component microburst events.In particular, we plan to investigate whether the relative strength of the two components has a dependence on local time, geomagnetic activity or hemisphere (due to drift loss cone effects).This work also motivates future observations that can measure microbursts with greater spatial resolution to investigate both the relative spatial scales of the two components as well as the extent to which the smooth component is comprised of microburst overlap within the field of view as compared to additional more diffuse precipitation.Understanding the relationship between the characteristics and drivers of the smooth and bursty components will help illuminate their significance as a source of radiation belt loss.

Figure 1 .
Figure 1.Microburst event observed on 13 August 2015.The two component structure can be observed with the bursty structure (cyan) lying on top of the more slowly-varying smooth component (magenta).The inlay shows a zoom in of just over 2 min where the microbursts (cyan bursty component) are more clearly visible on top of the smooth component (magenta).

Figure 2 .
Figure 2. Sample microburst event selection process from 13 August 2015.Data shown are summed X-ray counts from 25 to 180 keV.Panel (a) shows the identified replete (σ p ≥ 1.2) windows in red and background windows (σ p < 1.2) in gray, (b) shows the microburst candidate (increased power spectral density) and adjacent windows in red, and (c) shows the final microburst event selection in red.

Figure 3 .
Figure 3. Component Identification and separation for microburst event on 13 August 2015.(a) Classification of data points for each component.The rolling median from Equation 2 is shown in green.Burst points (having N m ≥ 8) are shown in cyan and smooth points (N m ≤ 3) are shown in magenta.Data are not yet background subtracted.(b) Component assignment and separation.Background count values are shown in green.Smooth component with linear interpolations shown in magenta.Burst component with all non-labeled points from (a) shown in cyan.Data are not yet background subtracted.(c) Finalized components.Data are background subtracted.Magenta line is the smooth component and cyan line is burst component.

Table 1 X
-Ray Count Percentages for Precipitation Components in Balloon Array for Radiation Belt Relativistic Electron Losses Campaign 3 Microburst Events