Auroral substorms, dynamic phenomena that occur in the upper atmosphere at night, are caused by global reconfiguration of the magnetosphere, which releases stored solar wind energy1,2. These storms are characterized by auroral brightening from dusk to midnight, followed by violent motions of distinct auroral arcs that suddenly break up, and the subsequent emergence of diffuse, pulsating auroral patches at dawn1,3. Pulsating aurorae, which are quasiperiodic, blinking patches of light tens to hundreds of kilometres across, appear at altitudes of about 100 kilometres in the high-latitude regions of both hemispheres, and multiple patches often cover the entire sky. This auroral pulsation, with periods of several to tens of seconds, is generated by the intermittent precipitation of energetic electrons (several to tens of kiloelectronvolts) arriving from the magnetosphere and colliding with the atoms and molecules of the upper atmosphere4,5,6,7. A possible cause of this precipitation is the interaction between magnetospheric electrons and electromagnetic waves called whistler-mode chorus waves8,9,10,11. However, no direct observational evidence of this interaction has been obtained so far12. Here we report that energetic electrons are scattered by chorus waves, resulting in their precipitation. Our observations were made in March 2017 with a magnetospheric spacecraft equipped with a high-angular-resolution electron sensor and electromagnetic field instruments. The measured13,14 quasiperiodic precipitating electron flux was sufficiently intense to generate a pulsating aurora, which was indeed simultaneously observed by a ground auroral imager.
Theories and computer simulations have provided a promising explanation for the origin of electron precipitation: the interaction between electrons and whistler-mode chorus waves near the magnetospheric equator7,8,15,16. After chorus waves are generated at the equator, they propagate towards higher latitudes (grey wavy arrows in Fig. 1) and interact with bouncing, counter-streaming electrons in both hemispheres. As these electrons are scattered by the Lorentz force of the chorus waves, some of them become nearly aligned with the field, fall into the empty electron loss cone (blue open arrow in Fig. 1a), fill it (blue filled arrows in Fig. 1a), and precipitate into the upper atmosphere (top left in Fig. 1a).
Data from the THEMIS (Time History of Events and Macroscale Interactions during Substorms) spacecraft and ground-based imagers have shown a correlation between chorus-wave modulation and auroral patch pulsation9,10. Geosynchronous satellite measurements have revealed a correlation between electron flux modulation and auroral pulsation17. However, direct evidence of interaction between chorus waves and precipitating electrons at the source region—the connection needed to verify the whistler-mode chorus–electron interaction theory of pulsating aurorae—has not been obtained. This is principally because of the large acceptance angle of earlier space plasma instruments, which prohibited them from distinguishing small loss cones in the equatorial magnetosphere. Thus, it has been difficult to directly establish the association between precipitating electrons hidden inside loss cones and the chorus waves. The high angular resolution of our instrumentation has enabled us to overcome this limitation.
On 2017 March 27, auroral pulsations were observed by the ground-based auroral All-Sky Imagers (ASIs) of the THEMIS mission18 (Fig. 2) when a magnetic storm with auroral substorm characteristics developed. Auroral snapshots illustrate the appearance and disappearance of pulsating auroral patches. The ionospheric footprint of the Exploration of energization and Radiation in Geospace (ERG, also called Arase) spacecraft13,14, which was launched in December 2016 and started observations in March 2017, was traced with an empirical magnetic field model19 and plotted over the auroral images (red crosses in Fig. 2). The model footprint is well within the fields of view of the ASIs in which the auroral pulsations can be clearly detected (see also Supplementary Video 1). Depending on the instantaneous magnetospheric configuration, however, the actual footprint may be displaced from the model by a few degrees in magnetic latitude and longitude20 (fewer than several hundred kilometres in both directions). Therefore, the model footprint should be regarded as approximately located in a wide area where pulsating aurorae were observed (we note that the field of view of an ASI is about 1,000 km across, larger than the mapping uncertainties). Electrons with sufficient energy flux to cause a visible aurora were observed near the equator, closely correlated with chorus waves when the footprint of ERG was located in a region replete with pulsating auroral patches.
During morning hours in magnetic local time, ERG captured a correlation between wave activity and particle flux modulations near the magnetospheric equator. During the 30-min duration of the observation near the equator, waves are seen intermittently in the power spectra of the magnetic (Fig. 3a) and electric fields (Extended Data Fig. 1a) in both the lower-band (<0.5fce) and upper-band (>0.5fce) frequency ranges (where fce is the electron gyrofrequency). Despite dynamic chorus wave activity, northward-streaming electron fluxes outside the loss cone (with pitch angle, the angle between the particle velocity and the local magnetic field, from 20° to 40°) are relatively stable and do not show any apparent correlation with the waves (Fig. 3c). This is consistent with the absence of correlation between electron flux and chorus waves in previous observations.
Dramatic modulations of the electron flux in concert with chorus waves do occur in the loss cone, however. The northward loss-cone electron flux (Fig. 3b), as measured by the Medium-Energy Particle experiments – electron analyser (MEP-e) onboard ERG with a 3.5° field of view (full-width at half-maximum), shows repeated enhancements and depletions over a broad energy range from 10 keV to 30 keV, in striking correspondence with bursts of lower-band chorus waves (Fig. 3a).
The correlation between chorus waves and loss-cone electron flux is more clearly presented in Fig. 4, which focuses on the flux of 24.5-keV electrons (selected because of the good loss-cone coverage) and a magnetic wave power spectral density of 0.38–0.64 kHz. This combination of energy and frequency satisfies the cyclotron resonance condition21 (thus enabling considerable electron scattering) if we assume an electron density of 3 cm−3, which is typical of this region22, and it is also consistent with the observed frequency of the faint signature of an upper hybrid resonance wave (Extended Data Fig. 1a). The excellent correlation between loss-cone electrons (black solid lines in Fig. 4) and chorus waves (blue dash-dotted lines in Fig. 4) indicates that wave–particle interaction is indeed taking place. Upon close examination of Fig. 4, we see that the loss-cone flux is strongly modulated on a timescale of 10 s (or less, although the time resolution of the loss-cone measurement is 8 s, and thus shorter periods cannot be resolved), within the typical range of auroral pulsations. As expected from low-altitude observations5,11, the flux is modulated by a factor of 2 to 3. Furthermore, when the loss cone is filled, the measured electron energy flux in the atmosphere is sufficient to excite visible aurorae. Specifically, it is several times 109 keV cm−2 s−1 or several erg cm−2 s−1—comparable to, but above, the threshold for visible aurorae23 (Extended Data Fig. 2 further illustrates the correlation between auroral patch pulsation and the loss-cone electron flux, as well as the chorus waves).
Figure 4 also illustrates that the variability in the quasiparallel flux outside the loss cone (shown by black dashed lines) is nearly stable compared to the dynamic modulation of the loss-cone flux and chorus wave power. This indicates that the observed electron flux just outside the loss cone does not provide clear evidence of precipitation, which explains why previous measurements did not reveal the wave–particle correlation that causes pulsating aurorae.
The interpretation of this observation is illustrated in Fig. 1 (bottom left box). The spacecraft was in the Southern Hemisphere but quite close to the magnetospheric equator. The loss cone of the northward-streaming electrons, which is empty during quiet periods, fills with scattered electrons as southward-propagating chorus waves emerge. The evolution of the pitch-angle distribution (PAD) of the northward-streaming electron flux is schematically shown in the right panels of Fig. 1b–d. On the other hand, the electron flux in the southward-streaming loss cone (red arrows) is also seen in the observation (Extended Data Fig. 1d). Although this flux is also modulated (that is, varies with time), the correlation with the chorus waves is apparently weaker than that of the northward loss-cone flux (Fig. 3b and Extended Data Fig. 1c). For example, chorus waves were absent from 10:56 to 11:00 universal time (ut), while the southward-streaming loss cone continued to be filled. This can be expected because southward-streaming electrons must interact with northward-propagating chorus waves (presumably present in the Northern Hemisphere), which are not necessarily symmetrical with southward-propagating chorus waves.
We determined the location and physical mechanism of electron scattering into the loss cone and subsequent precipitation of energetic electrons, which results in a pulsating aurora. Whether these results can be generalized across radial distances from Earth, different local times and geomagnetic activity conditions is yet to be determined. Comparing loss-cone fluxes with other potential scattering mechanisms, such as electrostatic electron cyclotron harmonic waves7 and large-amplitude parallel electric fields12, for which the correlated energy range is typically lower than that discussed here, would help to determine whether other wave types make a similar contribution to scattering. The physical process presented here should apply to the aurorae of Jupiter and Saturn, where chorus waves have been detected24,25, and may also operate at other magnetized astrophysical objects.
The key observation in our dataset is the electron measurement made by the MEP-e sensor onboard the ERG spacecraft. The sensor comprises an electrostatic analyser (ESA) specifically designed for energetic particles26 and avalanche photodiodes27 (APDs). The ESA is an energy filter covering 7–87 keV with an energy resolution of 8%. The angular resolution is determined by combining the focusing property of the ESA with the size of the detector (or the APD). The ESA has a disk-like, 2π-rad field of view divided into 16 narrow elements, each of which corresponds to an APD. Using the spacecraft’s spin motion, a full solid angle is swept (with certain dead angles due to gaps between detectors). Depending on the geometry (more specifically, the angle between the spacecraft’s spin axis and the magnetic field), MEP-e may see the magnetic field direction (and thus observe the loss cone) once or twice in one spacecraft spin, as in the presented case. The sensor may also miss the magnetic field directions (the magnetic field direction drops at the gaps) or the loss cone may be observed for a longer time (if the spin axis is along the magnetic field direction). Neither of these scenarios occurred in the time period reported here.
The most important property of the MEP-e sensor for this study—the angular resolution in two orthogonal directions—is shown in Extended Data Fig. 3. These calibration data were obtained by pre-flight laboratory experiments. Profiles are essentially the same for all 16 channels. Here the data are modelled with a Gaussian shape and a full-width at half-maximum of 3.5° for use in the evaluation of the loss-cone determination discussed below.
Loss cone determination
In Figs 3b and 4, as well as in Extended Data Fig. 1c and d, we have plotted the loss-cone electron flux. In these analyses, we used the following criterion for loss-cone measurements: when the angle between the centre of the field of view of the MEP-e and the magnetic field is smaller than 2°, the detected electrons are inside the loss cone. This approach needs caution regarding contamination from electrons outside the loss cone, because a fraction of the field of view can extend outside the loss cone. We carefully evaluated this effect (Extended Data Fig. 4) and found it to be insignificant (although not negligible), confirming the robustness of our analyses.
Estimation of precipitating electron energy flux
The downward electron energy flux F at the ionosphere, which contributes to the auroral illumination, is estimated as F ≈ (Bi/Beq) E Jeq ΔΩ ΔE, where the parameters are the magnetic field strength at the ionosphere (Bi ≈ 55,000 nT) and at the equator (Beq), the electrons’ characteristic energy (E), the differential flux at the equator (Jeq), the solid angle of the loss cone (ΔΩ) and the energy range of the precipitating electrons (ΔE). On the basis of our in situ observation, we take Beq ≈ 100 nT, E ≈ 20 keV, Jeq ≈ 4×106 keV cm−2 sr−1 s−1 keV−1 (here we take the peak value), ΔΩ ≈ 6×10−3 sr and ΔE ≈ 20 keV to obtain a downward electron energy flux of about 5×109 keV cm−2 s−1, or 8 erg cm−2 s−1.
The ERG data presented here are publicly available from the ERG science centre website (https://ergsc.isee.nagoya-u.ac.jp/dataset/2017.l-1001.shtml.en). THEMIS-ASI data that support the findings of this study are publicly available at http://themis.ssl.berkeley.edu/data_all.shtml.
The software used to read and analyse the data, part of the publicly available SPEDAS software package, is available at http://themis.ssl.berkeley.edu/software.shtml and can be used without any restrictions. All data have a common data file format (https://cdf.gsfc.nasa.gov/).
The observations presented here were obtained with the help of Mitsubishi Heavy Industries, Ltd, Meisei Electric Co., Ltd, Hamamatsu Photonics Co. Ltd, YS DESIGN Co., Ltd, NIPPI Co. Ltd, Sumitomo Heavy Industries, Ltd and TIERRA TECNICA Co. Ltd. We acknowledge the work of the members of the ERG project team over several years. Y.M. is supported by JSPS Kakenhi (15H05747, 15H05815 and 16H06286). Y. Kasahara is supported by JSPS Kakenhi (16H04056 and 16H01172). H.U.F. is supported by grant AGS-1004736 from the National Science Foundation (NSF) of the USA. I.S. is supported by JSPS Kakenhi (17H06140). We thank NASA for contract NAS5-02099, S. Mende and E. Donovan for use of the ASI data, the Canadian Space Agency for logistical support in fielding and data retrieval from the ground-based observatory stations, and the NSF for support of the Ground-based Imager and Magnetometer Network for Auroral Studies programme through grant AGS-1004736. The ERG (Arase) satellite science data is available from the ERG Science Centre operated by the Institute of Space and Astronautical Science of the Japan Aerospace eXploration Agency and the Institute for Space–Earth Environmental Research of Nagoya University (https://ergsc.isee.nagoya-u.ac.jp/index.shtml.en). We are grateful to J. Hohl for assistance in editing the manuscript. We also thank N. Umemura for assistance in source data archiving. S. Kasahara thanks T. Mukai and M. Fujimoto for discussions.
Extended data figures
Successive clear sky images from two ground stations (Fort Simpson at the upper left, and The Pas at the lower right) are shown as a video. The red cross indicates the nominal spacecraft footprint. Dashed lines illustrate magnetic coordinates every 10o in latitude and 15o in longitude. The presented time period covers that of Fig. 3.