Testing loss mechanisms capable of rapidly depleting relativistic electron flux in the Earth s outer radiation belt

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 109,, doi: /2004ja010579, 2004 Testing loss mechanisms capable of rapidly depleting relativistic electron flux in the Earth s outer radiation belt J. C. Green, 1,2 T. G. Onsager, 3 T. P. O Brien, 4 and D. N. Baker 1 Received 7 May 2004; revised 27 August 2004; accepted 21 October 2004; published 14 December [1] We investigate how relativistic electrons are lost from the Earth s magnetosphere in order to better understand the dynamic variability of the radiation belts. We identify 52 events where the >2 MeV electron flux at geostationary orbit decreases rapidly and use a superposed epoch analysis of multispacecraft data to characterize the accompanying solar wind and geomagnetic conditions and examine the relevance of potential loss mechanisms. The results show that the flux decrease events follow a common sequence. The electron flux is reduced first in the dusk sector concurrent with the stretching of the magnetic field to a more tail-like configuration. The extreme stretching at dusk is caused by the formation of a partial ring current driven by changing solar wind conditions. We investigate three possible causes of the ensuing flux decrease: adiabatic electron motion in response to the changing magnetic field topology, drift out the magnetopause boundary, and precipitation into the atmosphere. The analysis reveals that the flux depletion is likely due to enhanced precipitation into the atmosphere, but the exact cause of the enhanced precipitation is still uncertain. INDEX TERMS: 2716 Magnetospheric Physics: Energetic particles, precipitating; 2720 Magnetospheric Physics: Energetic particles, trapped; 2730 Magnetospheric Physics: Magnetosphere inner; 2784 Magnetospheric Physics: Solar wind/magnetosphere interactions; KEYWORDS: relativistic electrons, radiation belts, electron losses, adiabatic motion, electron flux Citation: Green, J. C., T. G. Onsager, T. P. O Brien, and D. N. Baker (2004), Testing loss mechanisms capable of rapidly depleting relativistic electron flux in the Earth s outer radiation belt, J. Geophys. Res., 109,, doi: /2004ja Introduction [2] Since the discovery of Earth s radiation belts, researchers have sought to identify the mechanisms that dictate the wildly fluctuating and seemingly erratic electron flux levels of the outer belt. The outer belt (3 10 R E )is populated by electrons with energies >0.1 MeV, sometimes referred to as relativistic electrons because their high energies put them in the regime where relativistic corrections are no longer negligible and sometimes referred to as killer electrons owing to their ability to penetrate shielding and damage satellite electronics. [3] The capriciousness of the outer belt electrons is demonstrated by their inconsistent response to geomagnetic activity. Often only a small increase in geomagnetic activity induces large flux changes. For example, in one particular event the CRESS satellite observed the 1 MeV electron flux increase by a factor of 10, while the Dst index remained above 30 nt [Meredith et al., 2002]. In another example, on 5 March 1997, the GOES >2 MeV electron flux 1 Laboratory for Atmospheric and Space Physics, University of Colorado, Boulder, Colorado, USA. 2 Formerly at Space Environment Center, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. 3 Space Environment Center, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA. 4 Aerospace Corporation, Los Angeles, California, USA. Copyright 2004 by the American Geophysical Union /04/2004JA decreased by a factor of 100, while Dst remained above 35 nt. Even substantial geomagnetic activity may produce a variety of responses. Reeves [1998] found only a rough correlation between storm intensity, as measured by the Dst index, and maximum electron flux measured at geosynchronous. In fact, 53% of geomagnetic storms produce elevated radiation belt electron fluxes at geosynchronous altitude, while 19% produce decreased fluxes and 28% produce no change [Reeves et al., 2003]. The range of responses to geomagnetic activity suggests that acceleration and loss are both capable of producing dramatic flux changes, are enhanced during different conditions, and ultimately compete to determine final flux levels. Thus to predict electron flux variations, both processes must be understood. [4] In recent years, significant progress has been made toward developing, refining, and testing theoretical models of radiation belt electron acceleration, yet little is known about how these electrons are lost from the magnetosphere. Probable loss mechanisms have been identified, including precipitation into the atmosphere due to wave particle interactions and drift out the magnetopause boundary, yet only limited progress has been made toward testing and quantifying these loss mechanisms. [5] We investigate how relativistic electrons are lost from the magnetosphere, focusing our attention on flux decrease events like those first identified by Onsager et al. [2002]. They presented detailed observations of the local time, radial, and energy dependence of one flux decrease event that occurred on 15 April Surprisingly, the decrease 1of12

2 Figure 1. Demonstration of the automated routine used to pick flux decrease events. (a) The ratio of the >2 MeV electron flux (#/cm 2 s str) measured by the GOES 8 satellite on the current day divided by the electron flux measured on the previous day. (b) The measured >2 MeV electron flux from the current day (solid line) along with the measured >2 MeV electron flux from the previous day (dashed line). The black vertical line marks the start time of the flux decrease event identified by the automated routine. was not globally coherent but had distinct local time dependence. The decrease, as observed by geosynchronous satellites, began at 1500 MLT and then progressed to all local times after a period of 10 hours, substantially longer than the 5 10 min required for a relativistic electron to complete one drift orbit about the Earth. The long duration of the observed local time progression challenges the intuitive notion that a flux change at one local time should be apparent at all local times within one electron drift period. The flux decrease was concurrent with a stretching of the magnetic field to a tail-like configuration at dusk, suggesting that the decrease was related to adiabatic electron motion in response to the changing magnetic field topology. However, the electron flux remained low for 10 days, implying that eventually some permanent electron loss occurred. [6] In order to better understand the unusual local time dependence of the flux decrease and the permanent loss of electrons, we have identified 52 additional flux decrease events and performed a superposed epoch analysis. Section 2 describes the datasets used and the method for identifying events. Section 3 describes the statistical features of the events using a superposed epoch analysis. Section 4 discusses mechanisms that may contribute to the flux decreases including adiabatic electron motion, magnetopause encounters, and precipitation to the atmosphere. Finally, Section 5 concludes that adiabatic electron motion produces the initial local time asymmetry of the flux decrease but cannot explain the entire flux reduction. Permanent loss of electrons is likely the result of enhanced precipitation into the atmosphere but the cause of the precipitation is uncertain. 2. Data and Event Identification 2.1. Data [7] A wide variety of datasets from January 1996 to August 2002 are used in this work. The study focuses on the geosynchronous region and includes measurements from the Geostationary Operational Environmental Satellites (GOES) and Los Alamos National Laboratory (LANL) geostationary satellites. The Space Environment Monitor (SEM) instruments on board the GOES satellites provide the flux of relativistic electrons with energies >2 MeV, the flux of protons with energy between 0.8 and 4 MeV, and the three component magnetic field. The LANL Magnetospheric Plasma Analyzer (MPA) [Bame et al., 1993] and Synchronous Orbit Particle Analyzer (SOPA) instruments measure both proton and electron flux and provide the flux of protons with energies from 0.13 to 0.4 kevand electrons with energies from 1.5 to 3.5 MeV used in this study. Two GOES satellites and up to three LANL satellites simultaneously monitor the geosynchronous region at different longitudes providing local time coverage for the study. [8] To complement the geosynchronous measurements, we use data from the Polar satellite and two Highly Elliptical Orbitor (HEO) satellites, HEO1 and HEO3, which provide extensive radial coverage. The Polar High Sensitivity Telescope (HIST) [Blake et al., 1995] measures relativistic electrons with energies between 0.7 and 7.0 MeV. For a discussion of this instrument and associated problems, please see Contos [1997], Selesnick and Blake [2000], Green [2002], and Green and Kivelson [2004]. The DSU instruments on board HEO1 and HEO3 measure both electrons and protons over a range of energies and for this study provide the flux of electrons with energies >1.5 MeV and protons with energy >0.320 MeV. [9] Data from the Proton Electron Telescope (PET) [Cook et al., 1993] on board the Solar Anomalous Magnetospheric Particle Explorer (SAMPEX) [Baker et al., 1993] is used to measure precipitating electron flux. The SAMPEX satellite has a low-altitude (600 km) high-inclination (82 ) orbit with a period of 100 min. The PET instrument provides the flux of electrons with energy >400 kev. To supplement the particle measurements, solar wind data and geomagnetic indices are obtained from the omni dataset available at NSSDC OMNIweb ( Event Identification [10] We seek to identify all sharp flux decrease events in the GOES and LANL data like those identified in the 2of12

3 Figure 2. Histogram showing the local time where the flux decrease events were first observed. Onsager et al. [2002] study. To do so, we first smooth the electron flux data with a 2-hour running average. Events are picked by an automated routine that steps through each geosynchronous satellite dataset 1 day at a time, dividing the electron flux of the current day by the electron flux from the previous day. The division essentially compares flux measured at identical local times. A flux decrease event is triggered when this flux ratio drops below The start of the event is found by moving back in time until the ratio equals 0.5. Figure 1 demonstrates the identification process. Final events include only those where at least two or more GOES or LANL satellites observe a decrease within 24 hours. The automated routine picked 61 events. Eight events were discarded because they were caused by the GOES dataprocessing routine that attempts to remove contamination from solar proton events. One event was discarded because it was sufficiently close to another event to be considered a duplicate. Appendix A lists the start times of all 61 events and labels those events not used in the study as either solar proton or duplicate events. 3. Statistical Description of Events [11] The 52 flux decrease events identified in the geostationary data show similar local time dependence and magnetic field changes identified in the single event described by Onsager et al [2002]. For example, the histogram plotted in Figure 2 shows that most events began in the dusk/ premidnight sector between 1500 and 2400 MLT. The events also show a similar dependence on local time and a distinct relationship to the changing magnetic field, as demonstrated by Figure 3. This figure is a superposed epoch representation of the electron flux and magnetic field at geostationary altitude as a function of both time and magnetic local time. To create these superposed plots, we first average the data for each event in 1-hour time and local time bins. Next we align the events on their start times and find the median values in the 1-hour time and local time intervals. Figure 3a shows the median value of the >2 MeV electron flux, and Figure 3b shows the median elevation angle of the magnetic field in each bin. The elevation angle is defined as the atan(b p /B e ), where B p is the magnetic field in the poleward direction and B e is the magnetic field in the earthward direction. Thus a 90 elevation angle corresponds to a dipolar field and smaller elevation angles correspond to more stretched tail-like fields. The data are plotted from 8 hours before the start of the events to 8 hours after. As seen in Figure 3a, the electron flux decreases first at dusk between 1800 and 2400 MLT. After 8 hours, low electron flux is observed at all local times. At the same time as the flux decreases the magnetic field becomes more stretched first at dusk, as seen in Figure 3b. 4. Analysis [12] The observed location of the flux decreases and their relationship to the magnetic field shown in Figures 3a and 3b prompt two questions: What caused the unusual local time dependence of the flux decrease events and ultimately where did the electrons go? To answer these questions, we investigate how three mechanisms affect the electron flux levels: adiabatic motion, magnetopause encounters, and precipitation to the atmosphere Adiabatic Motion [13] The correlation between the flux decrease and the stretching of the magnetic field immediately suggests that adiabatic electron motion in response to the changing magnetic field topology affects the measured flux levels. Adiabatic electron motion was first used by Dessler and 3of12

4 Figure 3. Superposed epoch plot of electron flux, magnetic field elevation angle, and electron phase space density as a function of local time and time. (a) The median >2 MeV (#/cm 2 s) electron flux at geosynchronous measured by GOES and LANL satellites for all 52 decrease events in 1-hour time and local time bins. (b) The median elevation angle (degrees) of the magnetic field at geosynchronous measured by the GOES satellites in the same format. (c) The median phase space density ((MeV-s) 3 )for m = 6000 MeV/G and K =0G 1/2 km. Data are plotted from 8 hours before the start time of the events to 8 hours after. See color version of this figure at back of this issue. Karplus [1961] to describe how high-energy electrons respond to the development of the storm time ring current. The theory suggests that current enhancements produce time varying magnetic fields and induced electric fields that alter particle drift paths. If the magnetic field changes slowly compared with the electron drift, bounce, and gyromotion, then an electron will move such that all three adiabatic invariants are conserved. Under such conditions the ensuing motion can be conveniently described in terms of these adiabatic invariants. [14] Dessler and Karplus [1961] used adiabatic motion to explain how symmetric field changes such as those produced by an enhanced ring current modify measured electron flux. However, the same arguments can be applied to an asymmetrically stretched field such as seen in Figure 3b. As the magnetic field decreases, even in a limited local time region, electrons conserving the third invariant move outward in order to compensate for the reduced field and conserve magnetic flux enclosed in a drift orbit. A satellite fixed at geosynchronous altitude now measures electrons that once resided at lower L. More importantly, the geosynchronous satellite, which measures a fixed energy range, no longer measures electrons with the same m value as it did previously at the same local time. In the stretched field low B region, the satellite measures electrons with high m values, and in the high B region it measures electrons with low m values. At the energies of interest (>100 kev), the flux of electrons generally decreases with increasing m, resulting in a localized flux decrease in the low B region. This kind of effect is regularly seen in electron flux measured at geosynchronous orbit even during undisturbed periods. High flux is typically observed in the compressed dayside magnetic field region, and lower flux is observed in the stretched nightside region, as seen in Figure 3a in the 8 hours prior to the start of the events. [15] To determine whether or not the asymmetric flux decreases can be explained by adiabatic motion in response to the changing magnetic field, we invoke Liouville s theorem. The theorem states that phase space density expressed as a function of the adiabatic invariants should be unaffected by slow changes of the magnetic field such as the observed stretching. Thus transforming flux measured as a function of energy, pitch angle, and position to phase space density as a function of the adiabatic invariants removes changes related solely to the stretching of the magnetic field. [16] We transform the electron flux measured by the GOES satellites to phase space density of electrons with 4of12

5 Figure 4. Superposed epoch plot of electron flux and magnetic field as a function of local time and time. (a) The median >2 MeV (#/cm 2 s) electron flux measured by GOES and LANL satellites at geosynchronous for all 52 decrease events in 1-hour time and local time bins. (b) The median elevation angle (degrees) of the magnetic field at geosynchronous measured by the GOES satellites in the same format. Data are plotted from 4 days before the start time of the events to 4 days after. See color version of this figure at back of this issue. K =0kmG 1/2 and m = 6000 MeV/G using the method described by Onsager et al. [2004] which we only briefly describe here. The method requires several assumptions. A magnetic field model and pitch angle distribution are required to map electron flux from the satellite location, which is up to 8 off the equator, to the equator where K = 0. We assume that the GOES detectors, which have wide apertures, locally measure electrons with predominantly 90 pitch angle. To map the 90 pitch angle particles measured locally off the equator to 90 pitch angle particles at the equator with K = 0, we use the Tsyganenko 89 field model. We use the measured Kp as input to the model and assume the electron pitch angle distribution follows a sin.25 a form. In addition, an energy distribution is required to transform flux at the measured energy to flux at the energy corresponding to constant m. To do this mapping, we assume an energy distribution of the form j = C*exp( E/E 0 ), where j is the integral flux, E is energy, and E 0 = 250 kev. Errors in the phase space density estimates are introduced with each assumption, but in this case they do not significantly impact our interpretation. We use the phase space density estimates only to argue whether or not it is feasible, using those assumptions outlined above, to explain the flux depletions by invoking adiabatic motion. [17] Figure 3c shows the superposed electron phase space density with m = 6000 MeV/G and K =0G 1/2 km. The phase space density decreases shortly after the start of the events but the local time dependence that was apparent in the flux data has now been greatly reduced. The reduced correlation between the magnetic field and the electron phase space density suggests that adiabatic motion can plausibly explain the unusual local time dependence so obvious in the flux measurements following the start of the events. This conclusion is appealing, since it is problematic for any mechanism to selectively remove electrons along the dusk portion of their drift orbit while maintaining high flux at other local times for time periods considerably longer than the very short 10 min drift period. While adiabatic motion may be a satisfactory explanation for the local time dependence of the flux levels, it cannot entirely explain the flux depletions. Figure 3 shows that the electron flux and phase space density both decrease and remain low even after 8 hours. This result suggests a nonadiabatic loss of electrons but, as mentioned previously, uncertainties in the phase space density estimates exist. Different assumptions about the particle distributions or the use of another magnetic field model to calculate the phase space density could produce different estimates. The Tsyganenko 89 field model we used lacks a partial ring current and any dawn/ dusk asymmetry and may not be the most accurate representation of the field during these times. Additionally, it is unlikely that the shape of the pitch angle distribution is independent of local time as was assumed. To determine precisely how these assumptions modify the phase space density estimates is complex. Instead, we use the later time evolution of the electron flux and magnetic field (Figure 4) to confirm that, indeed, some permanent loss of electrons occurs. Figure 4 is a superposed epoch representation of the electron flux and the magnetic field at geosynchronous now spanning from 4 days before the events to 4 days after. If the flux decreases were explained completely by adiabatic motion, then as the field returned to its dipolar configuration, the electrons would move inward, and the electron flux would return to its previous level. Yet Figure 4 shows that 1 day after the events begin, the magnetic field returns to the original dipolar configuration, but the electron flux remains low even after 4 days, indicating that some electrons were permanently removed from the magnetosphere. [18] Before investigating the cause of the permanent loss of electrons, we consider the process responsible for the dusk stretching of the magnetic field and thus the local time asymmetry of the flux decrease. Figure 5 is a superposed 5of12

6 Figure 5. Superposed epoch plot of solar wind and geomagnetic conditions during the events from 5 days before to 5 days after the start of the events. (a) (d) The solar wind total magnetic field, B x, B y, and B z (nt) components in GSM coordinates. (e) The Dst index (nt), (f) the solar wind dynamic pressure (np), (g) the solar wind velocity (km/s), and (h) the AE index (nt). The vertical black line marks the start time of the events. epoch plot showing the median and quartile values of solar wind and geomagnetic indices during the events and reveals some insight into the cause of the stretched duskside magnetic field. The figure shows that the events are preceded by 1 2 days of quiet conditions, evidenced by extremely low AE and small solar wind B z, and they begin with either a pressure pulse or a southward turning of B z. Similar conditions have been associated with observations of a super dense plasma sheet at geosynchronous. In fact, six of the flux decrease events occurred during times that Thomsen et al. [2003] identified as having a super dense plasma sheet. The authors conclude that quiet conditions lead to a high-density plasma sheet which is observed at geosynchronous orbit when either a pressure pulse or enhanced convection pushes the high-density particles earthward. [19] To investigate how dense plasma sheet conditions relate to the asymmetrically stretched magnetic field, we plot the geosynchronous plasma data during the events as a superposed epoch representation in Figure 6. This figure shows the median value of the kev ion density and temperature as a function of time and local time. In the 1 2 days of quiet conditions prior to the start of the events, the density of the plasma sheet at geosynchronous decreases at all local times. At the start of the events, the density increases in the nighttime sector and the temperature increases in the dusk sector. The increased density and temperature likely represent the formation of a partial ring current. The locally enhanced current consequently produces the stretched magnetic field at dusk and ultimately the asymmetry of the flux decrease Magnetopause Encounters [20] The previous discussion of adiabatic motion leads us to conclude that some permanent electron loss occurs during the flux decrease events. We now consider whether magnetopause encounters contribute to this permanent loss. This scenario proposes that electrons drifting about the Earth encounter the magnetopause boundary and are swept away by the solar wind. However, the radial extent of the flux depletions observed by multiple satellites does not coincide with predicted magnetopause locations and suggest that this scenario is unlikely. Figures 7a and 7b show a superposed epoch plot of the >1.5 MeV electron flux from the HEO 6of12

7 Figure 6. Superposed epoch representation of plasma data measured at geosynchronous altitude by the LANL satellites. (a) The median density (cm 3 ) of ions with energy between 0.13 and 45 kev in 1-hour time and local time bins. (b) The median perpendicular temperature (kev) of the ions and (c) the median parallel temperature. See color version of this figure at back of this issue. Figure 7. Superposed epoch plot of HEO1 and HEO3 electron and proton flux as a function of L and time. (a) The median value of the HEO1 >1.5 MeV electron flux (#/cm 2 s). (b) The median value of the HEO3 >1.5 MeV electron flux (#/cm 2 s). (c) The median value of the HEO1 >0.3 MeV proton flux (#/cm 2 s). (d) The median value of the HEO3 >0.3 MeV proton flux (#/cm 2 s). The white trace in all panels shows the minimum magnetopause standoff distance calculated from the Shue et al. [1997] model. See color version of this figure at back of this issue. 7of12

8 Figure 8. Electron phase space density obtained from the Polar HIST instrument during a flux decrease event (event 9 as listed in Appendix A). (a) Electron phase space density versus L measured over 10 Polar orbits. Phase space density from each orbit is color coded in time. Blue traces represent data before the flux decrease observed at geosynchronous. (b) The ratio of the Roederer L parameter (L*) to the McIllwain L parameter. (c) The >2 MeV electron flux measured by the GOES satellite for reference. Vertical bands show times when Polar was in the radiation belts. See color version of this figure at back of this issue. satellite which samples flux as a function of L. The white trace plots the minimum value in all the events of the magnetopause standoff distance determined from the Shue et al. [1997] model. The figure shows that the predicted magnetopause location moves earthward just prior to the start of the events, but the flux decrease is observed at significantly lower L values. Polar measurements that have been transformed to electron phase space density as a function of the adiabatic invariants to eliminate changes due to adiabatic motion [Green and Kivelson, 2004] also show a discrepancy between the magnetopause location and the region of decreased flux. One event, plotted in Figure 8, shows the phase space density decrease at L* values as low as 4.5, which is very near the minimum plasmapause location, Lpp = 4.2, estimated by the Carpenter and Anderson [1992] method. (Here L* represents the L parameter as defined by Roederer [1970].) The minimum magnetopause position location predicted by the Shue et al. [1997] model for this event is 7.9. [21] The discrepancies between the location of the magnetopause and the region of decreased flux suggest that magnetopause encounters are not responsible for the loss of electrons. However, an alternative hypothesis is that the changing magnetic field topology and ensuing adiabatic motion moves electrons outward to the magnetopause, producing an observed flux decrease at regions well inward of the expected magnetopause location. To test this hypothesis, we compare the flux of energetic protons and electrons with similar energies because both types of particles should respond identically to the changing magnetic field. The motion of these high-energy particles is dominated by the gradient and curvature drift, which is proportional to the m value of the particles. Protons and electrons with the same m follow similar drift trajectories only in opposite directions. Both will experience the same outward adiabatic motion. Figures 7c and 7d are superposed epoch plots showing the median flux of high-energy proton measurements as a function of L and time from two HEO satellites. The plots show no dramatic decrease of the proton flux. Figure 9 shows the median proton flux measured by the GOES satellites as a superposed epoch and also shows no flux decrease. Thus magnetopause encounters cannot explain the permanent loss of electrons Precipitation [22] Lastly, we consider how precipitation contributes to the permanent loss of electrons during the events by examining the electron flux measurements from the SAMPEX satellite that is in a low Earth orbit. The satellite samples different electron populations as it moves in geographic latitude and longitude due to the nonuniformity of the Earth s surface magnetic field. The measurements can be 8of12

9 Figure 9. (a) The median value of the GOES.8 4 MeV proton flux in 1-hour time and local time bins. (b) The magnetic field elevation angle for reference. The data are plotted from 4 days before to 4 days after the start of events. See color version of this figure at back of this issue. divided into three populations: electrons in stably trapped orbits that will not precipitate into the atmosphere, electrons in the drift-loss cone that will eventually precipitate within one drift orbit, and electrons in the bounce loss cone that will precipitate within one bounce period. The stably trapped electrons follow drift orbits that always mirror above the Earth s atmosphere. The electrons in the drift-loss cone mirror above the Earth s atmosphere at the current location of SAMPEX but will eventually drift to a decreased magnetic field region where the mirror point moves into the atmosphere. The electrons in the bounce loss cone mirror above the atmosphere at the current location of SAMPEX but will encounter the atmosphere in the opposite hemisphere as they bounce along the field line. To define which population SAMPEX samples at each geographic location, it is assumed that the nearly omnidirectional detector measures predominantly electrons with 90 pitch angles mirroring locally at the satellite. Determining the magnetic field magnitude and tracing field lines using the IGRF field model then defines the appropriate population assuming the atmosphere extends to an altitude of 100 km. [23] Some uncertainty in the exact flux of each population identified is introduced by the assumption that the instrument measures predominantly electrons with 90 pitch angles even though it is a nearly omnidirectional detector. For example, the flux of electrons identified as being trapped may contain some contribution from precipitating electrons with local pitch angles less than or greater than 90. Likewise, the flux of electrons identified as being in the drift loss cone may contain contributions from more fieldaligned electrons. The estimated flux of these two populations may increase or decrease with time as the pitch angle distribution and the flux of field-aligned electrons changes. However, the flux of precipitating electrons identified as being in the bounce loss cone, with which we are most concerned, is less ambiguous. All electrons in this population, including those with local pitch angles less than 90, are expected to precipitate regardless of how the pitch angle distribution changes with time. [24] Figure 10 shows superposed epoch plots of the electron flux for each population of electrons sampled. For each of the 52 events the data are averaged in 0.5 day and 0.2 L bins. The superposed plots show the median flux value for all events in each bin. Figures 10a, 10b, and 10d, which show all electrons, those in the drift loss cone, and the trapped population, all show depletions occurring after the start of the events. In contrast, Figure 10c shows an increased flux of electrons in the bounce loss cone near the start of the events. The increase suggests that scattering of electrons into the bounce loss cone could be responsible for the permanent loss of electron. However, the nature of the SAMPEX measurements prevents identification of the exact time and local time location of the precipitation. The data are limited because SAMPEX does not continuously measure each population of electrons. Instead, the different populations are sampled only when the satellite is located in specific geographic regions. The sampling of the electrons in the bounce loss cone is especially sparse because these electrons are only measured when the satellite is at the base of a field line whose opposite foot point is in the South Atlantic Anomaly. To ensure adequate statistics for the superposed plots, the data are separated into large half day bins which prevent precise timing. The limited measurements also prevent further separation of the data into local time bins that could indicate the cause of the precipitation. 5. Conclusions [25] We have investigated the cause of electron flux depletions at geostationary orbit by analyzing 52 events measured simultaneously by multiple spacecraft. The following sequence describes observed features commonly associated with the flux decrease events. The electron flux is reduced first in the dusk sector concurrent with the stretching of the dusk magnetic field to a more tail-like configuration. The extreme stretching at dusk is caused by the formation of a partial ring current driven by changing 9of12

10 Figure 10. Superposed epoch plot of median electron flux in.5 day and 0.2 L bins. (a) The median flux of all electrons measured at the altitude of SAMPEX. (b) The median flux of electrons in the drift loss cone. (c) The median flux of electrons in the bounce loss cone. (d) The median flux of trapped electrons. The data are plotted from 5 days before the events begin to 5 days after. The black vertical line marks the start of the events. See color version of this figure at back of this issue. solar wind conditions. One to two days of extremely quiet solar wind and little geomagnetic activity lead to a dense plasma sheet. With the onset of a pressure pulse or elevated AE the now dense plasma sheet moves earthward to form a partial ring current and stretched field at dusk. Analysis of electron phase space density suggests that adiabatic electron motion in response to the stretching of the magnetic field causes the initial dusk location and asymmetry of the flux decrease but cannot explain the permanent loss of electrons. Comparisons between expected magnetopause locations and the region of reduced electron flux suggest that magnetopause encounters are not responsible for the permanent loss of electrons. This conclusion is supported by analysis of high-energy proton flux, which should be similarly affected by magnetopause encounters but shows no flux reduction. Low-altitude observations show increased electron flux in the bounce loss cone near the start of the events and a decrease in the trapped electron flux. This result suggests that precipitation to the atmosphere may be the cause of the flux decrease. [26] Precipitation into the atmosphere may explain the removal of relativistic electrons from the radiation belts but questions remain. Is the flux observed in the bounce loss cone sufficient to explain the entire decrease? More information regarding the local time extent of the precipitation is required to accurately estimate losses and answer this question. Also, what mechanism caused the electrons to be scattered and precipitated into the atmosphere? Several mechanisms have been suggested which could explain the observed precipitation, including scattering due to increased field line curvature and interaction with waves such as electron cyclotron harmonic waves (ECH) [Horne and Thorne, 2000], electromagnetic ion cyclotron waves (EMIC) [Summers and Thorne, 2003; Albert, 2003; Lorentzen et al., 2000; Horne and Thorne, 1998; Thorne and Kennel, 1971], and whistler waves [Lorentzen et al., 2001; Meredith et al., 2002; Horne et al., 2003; Horne and Thorne, 2003; Horne and Thorne, 1998]. Each scattering mechanism makes testable predictions regarding observables such as the location of precipitation, particle energies affected, and timescales. We remark only briefly on the mechanisms and predictions, which can be used in future studies to identify the cause of the increased precipitation. [27] In highly stretched field regions, electron pitch angles may be modified because the field changes significantly on spatial scales less than the electron gyroradii such that the electron s first invariant is not conserved [Young et al., 2002, and references therein]. This type of scattering, at first glance, seems an appealing explanation for the flux depletions because some of the observations appear consistent with the expectations of this mechanism. For example, the flux depletions occur when the magnetic field becomes highly stretched, the precipitation is observed in similar local times as the stretched field region, and only highenergy electrons that have large gyroradii are affected. However, this scattering mechanism should predominantly 10 of 12

11 Table A1. Start Times of All 61 Sharp Flux Decrease Events in the GOES and LANL Data Event Year/Month/Day Time Day of Year Event /03/03 01:52:30 63 Event /03/19 13:00:30 79 Event /04/13 23:52: Event /04/17 04:47: Event /04/27 22:02: Event /08/28 21:37: Event /09/04 17:02: Event /10/22 06:00: Event /03/05 17:00:00 64 Event /03/28 14:00:00 87 Event /06/15 01:07: Event /09/17 20:17: Event /02/17 14:27:30 48 Event /03/04 13:47:30 63 Event /03/14 21:17:30 73 Event /03/20 00:42:30 79 Event /04/04 02:12:30 94 Event /04/20 04:07: SOLAR PROTONS Event /05/02 23:42: SOLAR PROTONS Event /07/21 03:57: Event /10/06 21:00: Event /11/13 02:00: Event /04/16 15:37: Event /05/12 19:00: Event /07/28 09:07: Event /08/09 00:02: Event /08/22 19:57: Event /08/22 22:42: DUPLICATE Event /09/21 15:00: Event /10/21 02:12: Event /12/13 03:47: Event /01/10 05:00:00 10 Event /02/05 20:52:30 36 Event /02/11 11:00:00 42 Event /04/15 20:00: Event /05/28 22:47: Event /07/14 12:57: SOLAR PROTONS Event /08/12 05:52: Event /08/20 23:22: Event /09/17 20:22: Event /09/24 17:00: Event /10/13 01:00: Event /10/28 10:00: Event /11/08 23:32: SOLAR PROTONS Event /03/22 18:00:00 81 Event /04/11 16:00: Event /04/17 20:37: Event /06/17 08:12: Event /07/22 01:12: Event /07/30 15:00: Event /08/03 03:02: Event /08/17 06:07: Event /08/25 12:00: Event /11/04 18:42: SOLAR PROTONS Event /11/22 23:37: SOLAR PROTONS Event /12/26 06:27: SOLAR PROTONS Event /03/29 20:52:30 88 Event /04/22 22:22: SOLAR PROTONS Event /05/18 21:02: Event /07/08 20:47: Event /08/18 23:07: affect particles at large L where the field is most stretched and is not expected to cause losses at L* = 4 such as was observed by the Polar satellite. It is conceivable that the adiabatic response to the development of the ring current distorts the electron drift orbits and moves them outward to distances where such scattering becomes significant. However, the mechanism should also affect protons with similar energy whose gyroradii are much larger than those of the electrons. Yet the flux of protons of similar energy is not depleted. Thus a cursory evaluation suggests that scattering due to the curvature of the field is an unlikely cause of the electron precipitation, but a more thorough evaluation of scattering lifetimes is required to definitively discount this mechanism. [28] Mechanisms which rely on waves to scatter electrons make predictions regarding the expected location and the energy of precipitating electrons. Electrons should precipi- 11 of 12

12 tate only in regions where waves are present. ECH, EMIC, and whistler waves are generally observed in different local time sectors. ECH waves are predominantly seen in the midnight to dawn region of the magnetosphere associated with substorm injections [Koons and Roeder, 1990; Parrot and Gaye, 1994; Meredith et al., 2000]. EMIC waves occur in regions where the ratio between the plasma frequency and electron gyrofrequency is high [Summers and Thorne, 2003; Meredith et al., 2003], and whistler waves are observed in the dawn sector [Meredith et al., 2001]. Thus the local time location of precipitating electrons should distinguish which, if any, of these waves are responsible for the loss of electrons. Measurements of the wave frequency and plasma parameters are needed to evaluate the resonance condition and determine what energy electrons should be affected by the waves. Future work will investigate how field stretching and wave particle interactions contribute to the loss of electrons and determine how these mechanisms are triggered by solar wind and geomagnetic conditions. Appendix A [29] Table A1 lists the start times of all 61 sharp flux decrease events in the GOES and LANL data and labels those events not used in the study as either solar proton or duplicate events. [30] Acknowledgments. We are grateful to the many data providers for the wide variety of datasets used in this analysis. We thank G. D. Reeves for the use of the LANL high-energy particle data and M. F. Thompsen for the use of the low-energy LANL particle data. We thank J. B. Blake of the Aerospace Corporation for the use of the HEO and Polar particle data. We thank H. J. Singer and T. G. Onsager for providing GOES magnetic field and particle data. We are thankful to the caretakers of the NSSDC OMNIWEB database for providing solar wind and geomagnetic indices. This work was supported by the NASA GGS (CEPPAD) grant NAG Work performed by T. Onsager was partially supported by NASA PR [31] Arthur Richmond thanks Geoff D. Reeves and Richard Thorne for their assistance in evaluating this paper. References Albert, J. M. (2003), Evaluation of quasi-linear diffusion coefficients for EMIC waves in a multispecies plasma, J. Geophys. Res., 108(A6), 1249, doi: /2002ja Baker, D. N., G. M. Mason, O. Figueroa, G. Colon, J. G. Watzin, and R. M. Aleman (1993), An overview of the Solar, Anomalous and Magnetospheric Particle Explorer (SAMPEX) mission, IEEE Trans. Geosci. Remote Sens., 31(3), 531. Bame, S. J., et al. (1993), Magnetospheric plasma analyzer for spacecraft with constrained resources, Rev. Sci. Instrum., 64, Blake, J., et al. (1995), CEPPAD: Comprehensive energetic particle and pitch angle distribution experiment on polar, Space Sci Rev., 71, 531. Carpenter, D. L., and R. R. Anderson (1992), An ISEE/Whistler model of equatorial electron density in the magnetosphere, J. Geophys. Res., 97, Contos, A. (1997), A complete description of the high sensitivity telescope (HIST) onboard the polar satellite, M.S. thesis, Boston Univ., Boston, Mass. Cook, W. R., et al. (1993), A proton electron telescope for studies of magnetospheric, solar, and galactic particles, IEEE Trans. Geosci. Remote Sens., 31(3), 1. Dessler, A. J., and R. Karplus (1961), Some effect of the diamagnetic ring currents on Van Allen radiation, J. Geophys. Res., 66, Green, J. C. (2002), Testing relativistic electron acceleration mechanisms, thesis, Univ. of Calif., Los Angeles, Los Angeles, Calif. Green, J. C., and M. G. Kivelson (2004), Relativistic electrons in the outer radiation belt: Differentiating between acceleration mechanisms, J. Geophys. Res., 109, A03213, doi: /2003ja Horne, R. B., and R. M. Thorne (1998), Potential waves for relativistic electron scattering and stochastic acceleration during magnetic storms, Geophys. Res. Lett., 25, Horne, R. B., and R. M. Thorne (2000), Electron pitch angle diffusion by electrostatic electron cyclotron harmonic waves: The origin of pancake distributions, J. Geophys. Res., 105, Horne, R. B., and R. M. Thorne (2003), Relativistic electron acceleration and precipitation during resonant interactions with whistler-mode chorus, Geophys. Res. Lett., 30(10), 1527, doi: /2003gl Horne, R. B., S. A. Glauert, and R. M. Thorne (2003), Resonant diffusion of radiation belt electrons by whistler-mode chorus, Geophys. Res. Lett., 30(9), 1493, doi: /2003gl Koons, H. C., and J. L. Roeder (1990), A survey of equatorial magnetospheric wave activity between 5 and 8 Re, Planet. Space Sci., 38, Lorentzen, K. R., M. P. McCarthy, G. K. Parks, J. E. Foat, R. M. Millan, D. M. Smith, R. P. Lin, and J. P. Treilhou (2000), Precipitation of relativistic electrons by interaction with electromagnetic ion cyclotron waves, J. Geophys. Res., 105, Lorentzen, K. R., J. B. Blake, U. S. Inan, and J. Bortnik (2001), Observations of relativistic electron microbursts in association with VLF chorus, J. Geophys. Res., 106, Meredith, N. P., R. B. Horne, A. D. Johnstone, and R. R. Anderson (2000), The temporal evolution of electron distributions and associated wave activity following substorm injections in the inner magnetosphere, J. Geophys. Res., 105, 12,907. Meredith, N. P., R. B. Horne, and R. R. Anderson (2001), Substorm dependence of chorus amplitudes: Implications for the acceleration of electrons to relativistic energies, J. Geophys. Res., 106, 13,165 13,178. Meredith, N. P., R. B. Horne, R. H. A. Iles, R. M. Thorne, D. Heynderickx, and R. R. Anderson (2002), Outer zone relativistic electron acceleration associated with substorm-enhanced whistler mode chorus, J. Geophys. Res., 107(A7), 1144, doi: /2001ja Meredith, N. P., R. M. Thorne, R. B. Horne, D. Summers, B. J. Fraser, and R. R. Anderson (2003), Statistical analysis of relativistic electron energies for cyclotron resonance with EMIC waves observed on CRRES, J. Geophys. Res., 108(A6), 1250, doi: /2002ja Onsager, T. G., G. Rostoker, H.-J. Kim, G. D. Reeves, T. Obara, H. J. Singer, and C. Smithtro (2002), Radiation belt electron flux dropouts: Local time, radial, and particle-energy dependence, J. Geophys. Res., 107(A11), 1382, doi: /2001ja Onsager, T. G., A. A. Chan, Y. Fei, S. R. Elkington, J. C. Green, and H. J. Singer (2004), The radial gradient of relativistic electrons at geosynchronous orbit, J. Geophys. Res., 109, A05221, doi: /2003ja Parrot, M., and C. A. Gaye (1994), A statistical survey of ELF waves in a geostationary orbit, Geophys. Res. Lett., 21, Reeves, G. D. (1998), Relativistic electrons and magnetic storms: , Geophys. Res. Lett., 25, Reeves, G. D., K. L. McAdams, R. H. W. Friedel, and T. P. O Brien (2003), Acceleration and loss of relativistic electrons during geomagnetic storms, Geophys. Res. Lett., 30(10), 1529, doi: /2002gl Roederer, J. G. (1970), Dynamics of Geomagnetically Trapped Radiation, 166 pp., Springer-Verlag, New York. Selesnick, R., and J. B. Blake (2000), On the source location of radiation belt electrons, J. Geophys. Res., 105, Shue, J.-H., J. K. Chao, H. C. Fu, C. T. Russell, P. Song, K. K. Khurana, and H. J. Singer (1997), A new functional form to study the solar wind control of the magnetopause size and shape, J. Geophys. Res., 102, Summers, D., and R. M. Thorne (2003), Relativistic electron pitch-angle scattering by electromagnetic ion cyclotron waves during geomagnetic storms, J. Geophys. Res., 108(A4), 1143, doi: /2002ja Thomsen, M. F., J. E. Borovsky, R. M. Skoug, and C. W. Smith (2003), Delivery of cold, dense plasma sheet material into the near-earth region, J. Geophys. Res., 108(A4), 1151, doi: /2002ja Thorne, R. M., and C. F. Kennel (1971), Relativistic electron precipitation during magnetic storm main phase, J. Geophys. Res., 76, Young, S. L., R. E. Denton, B. J. Anderson, and M. K. Hudson (2002), Empirical model for m scattering caused by field line curvature in a realistic magnetosphere, J. Geophys. Res., 107(A6), 1069, doi: / 2000JA D. N. Baker and J. C. Green, Laboratory for Atmospheric and Space Physics, University of Colorado, 1234 Innovation Drive, Boulder, CO 80303, USA. (janet.green@lasp.colorado.edu) T. P. O Brien, Aerospace Corporation, M2-260, P. O. Box 92957, Los Angeles, CA , USA. T. G. Onsager, Space Environment Center, National Oceanic and Atmospheric Administration, 325 Broadway Blvd., Boulder, CO 80305, USA. 12 of 12

13 Figure 3. Superposed epoch plot of electron flux, magnetic field elevation angle, and electron phase space density as a function of local time and time. (a) The median >2 MeV (#/cm 2 s) electron flux at geosynchronous measured by GOES and LANL satellites for all 52 decrease events in 1-hour time and local time bins. (b) The median elevation angle (degrees) of the magnetic field at geosynchronous measured by the GOES satellites in the same format. (c) The median phase space density ((MeV-s) 3 )form = 6000 MeV/G and K =0G 1/2 km. Data are plotted from 8 hours before the start time of the events to 8 hours after. Figure 4. Superposed epoch plot of electron flux and magnetic field as a function of local time and time. (a) The median >2 MeV (#/cm 2 s) electron flux measured by GOES and LANL satellites at geosynchronous for all 52 decrease events in 1-hour time and local time bins. (b) The median elevation angle (degrees) of the magnetic field at geosynchronous measured by the GOES satellites in the same format. Data are plotted from 4 days before the start time of the events to 4 days after. 4of12and5of12

14 Figure 6. Superposed epoch representation of plasma data measured at geosynchronous altitude by the LANL satellites. (a) The median density (cm 3 ) of ions with energy between 0.13 and 45 kev in 1-hour time and local time bins. (b) The median perpendicular temperature (kev) of the ions and (c) the median parallel temperature. Figure 7. Superposed epoch plot of HEO1 and HEO3 electron and proton flux as a function of L and time. (a) The median value of the HEO1 >1.5 MeV electron flux (#/cm 2 s). (b) The median value of the HEO3 >1.5 MeV electron flux (#/cm 2 s). (c) The median value of the HEO1 >0.3 MeV proton flux (#/cm 2 s). (d) The median value of the HEO3 >0.3 MeV proton flux (#/cm 2 s). The white trace in all panels shows the minimum magnetopause standoff distance calculated from the Shue et al. [1997] model. 7of12

15 Figure 8. Electron phase space density obtained from the Polar HIST instrument during a flux decrease event (event 9 as listed in Appendix A). (a) Electron phase space density versus L measured over 10 Polar orbits. Phase space density from each orbit is color coded in time. Blue traces represent data before the flux decrease observed at geosynchronous. (b) The ratio of the Roederer L parameter (L*) to the McIllwain L parameter. (c) The >2 MeV electron flux measured by the GOES satellite for reference. Vertical bands show times when Polar was in the radiation belts. Figure 9. (a) The median value of the GOES.8 4 MeV proton flux in 1-hour time and local time bins. (b) The magnetic field elevation angle for reference. The data are plotted from 4 days before to 4 days after the start of events. 8of12and9of12

16 Figure 10. Superposed epoch plot of median electron flux in.5 day and 0.2 L bins. (a) The median flux of all electrons measured at the altitude of SAMPEX. (b) The median flux of electrons in the drift loss cone. (c) The median flux of electrons in the bounce loss cone. (d) The median flux of trapped electrons. The data are plotted from 5 days before the events begin to 5 days after. The black vertical line marks the start of the events. 10 of 12