Specification and forecast ionospheric parameters, such as ionospheric electron density (Ne), have been an important topic in space weather and ionosphere research. Neural networks (NNs) emerge as a powerful modeling tool for Ne prediction. However, heavy manual attention costs time to determine the optimal NN structures. In this work, we propose to use neural architecture search (NAS), an automatic machine learning method, to address this problem of NN models. NAS aims to find the optimal network structure through the alternated optimization of the hyperparameters and the corresponding network parameters. A total of 16-year data from Millstone Hill incoherent scatter radar (ISR) are used for NN models. One single-layer NN (SLNN) model and one deep NN (DNN) model are trained with NAS, namely SLNN-NAS and DNN-NAS, for Ne prediction and compared with their counterparts without NAS from previous studies, denoted as SLNN and DNN. Our results show that SLNN-NAS and DNN-NAS outperformed SLNN and DNN, respectively. NN models can reveal more finer details than the empirical ionospheric model developed using traditional data fitting approaches. DNN-NAS yields the best prediction accuracy measured by quantitative metrics and rankings of daily pattern prediction. The limited improvement of NAS is likely due to the network complexity and the limitation of fully connected NN without a memory mechanism.

Auroral bright spots have been observed at Earth, Jupiter, and Saturn in regions that map to the boundary layer. It has been suggested that the bright spots are associated with Kelvin-Helmholtz instability. We utilize a quasistatic magnetosphere-ionosphere coupling model driven by a vortex in the boundary layer to determine how the field-aligned current structure depends on ionospheric and boundary layer parameters. We compare vortex induced currents with shear-flow induced currents. We find that the strength of the maximum currents are comparable, but the structure is significantly different. For a vortex, the current and electron precipitation maximize when the vortex size mapped to the ionosphere is approximately 1.5 L, where L=(Σp/κ)1/2 is the auroral scale length, Σp is the Pedersen conductivity, and κ is the Knight parameter. For a vortex, the current width provides a direct measure of the size, Δ, of the boundary layer structure, while shear-flow aurora generally are determined by the larger of Δ or L. For comparison with observations, an event is considered where auroral bright spots in the ionosphere are detected by DMSP SUSSI UVI when Kelvin-Helmholtz structures are observed on the dusk flank by THEMIS.

Many processes throughout the heliosphere such as flares, CMEs, storms and substorms have abrupt onsets. The waiting time between these onsets provides key insights as to the underlying dynamical processes. We explore the tail of these waiting time distributions in the context of random processes driven by the solar magnetic activity cycle, which we approximate by a sinusoidal driver. Analytically, we find that the distribution of large waiting times of such a process approaches a power law slope of -2.5 at large enough waiting time, and we find that this power law is primarily controlled by the conditions when the driving is minimum. We find that the asymptotic behavior of the waiting time distributions of solar flares, coronal mass ejections, geomagnetic storms, and substorms exhibit power laws are in reasonable agreement with a sinusoidally driven nonstationary Poisson process.

An empirical model of radiation belt relativistic electrons (m = 560–875 MeV G–1 and I = 0.088–0.14 RE G0.5) with average energy ~ 1.3 MeV is developed. The model inputs solar wind parameters (velocity, density, interplanetary magnetic field (IMF) |B|, Bz, and By), magnetospheric state parameters (SYM-H, AL), and L*. The model outputs radiation belt electron phase space density (PSD). The model is operational from L* = 3 to 6.5. The model is constructed with neural network assisted by information theory. Information theory is used to select the most effective and relevant solar wind and magnetospheric input parameters plus their lag times based on their information transfer to the PSD. Based on the test set, the model prediction efficiency (PE) increases with increasing L*, ranging from –0.043 at L* = 3 to 0.76 at L* = 6.5. The model PE is near 0 at L* = 3–4 because at this L* range, the solar wind and magnetospheric parameters transfer little information to the PSD. This baseline model complements well a class of empirical models that input data from Low Earth Orbit (LEO). Using solar wind observations at L1 and magnetospheric index (AL and SYM-H) models solely driven by solar wind, the radiation belt model can be used to forecast PSD 30–60 min ahead.

Many solar wind parameters correlate with one another, which complicates the causal-effect studies of solar wind driving of the magnetosphere. Conditional mutual information (CMI) is used to untangle and isolate the effect of individual solar wind and magnetospheric drivers of the radiation belt electrons. The solar wind density (nsw) negatively correlates with electron phase space density (PSD) (average energy ~ 1.6 MeV) with time lag (t) = 15 hr. This effect of nsw on PSD has been attributed to magnetopause shadowing losses, but when the effect of solar wind velocity (Vsw) is removed, t shifts to 7–11 hr, which is a more accurate time scale for this process. The peak correlation between Vsw and PSD shifts from t = 38 to 46 hr, when the effect of nsw is removed. This suggests that the time scale for electron acceleration to 1–2 MeV is about 46 hr following Vsw enhancements. The effect of nsw is significant only at L* = 4.5–6 (L* > 6 is highly variable) whereas the effect of Vsw is significant only at L* = 3.5–6.5. The peak response of PSD to Vsw is the shortest and most significant at L* = 4.5–5.5. As time progresses, the peak response broadens and shifts to higher t at higher and lower L*, consistent with local acceleration at L* = 4.5–5.5 followed by outward and inward diffusion. The outward radial diffusion time scale at L* = 5–6 is ~40 hr per RE.

\justify The location of the polar cap boundary is typically determined using low-orbit satellite measurements in which the boundary is identified by its unique signature of a sharp decrease in energy and particle flux poleward of the auroral oval. In principle, this decrease in precipitating particles should appear as a concomitant sharp change in auroral luminosity. Based on a few events, \cite{Blanchard_1995} suggested that a dramatic gradient in redline aurora may also be an indicator of the polar cap boundary. In recent years, advances in capabilities and the deployment of ground-based all-sky imagers have ushered in a new era of auroral measurements. Auroral imaging has moved well beyond the capabilities of the instrumentation in the previous study in terms of both spatial and temporal resolution. We now have access to decades of optical data from arrays spanning a huge spatial range, enabling a fresh examination of the relationship between redline aurora, particle precipitation, and the polar cap open closed boundary. In this study, we use data from the DMSP satellites in conjunction with the University of Calgary’s REGO (630.0nm) data to assess the viability of automated detection of the 2-dimensional polar cap boundary. Our results exhibit good agreement between the optical and particle polar cap boundary and suggest that a luminosity in redline emission could serve as a reasonable proxy for the location of the the electron poleward boundary during, while providing both high temporal and spatial resolution maps of the open-closed boundary.

The present paper demonstrates the first observations by the Magnetospheric Multiscale (MMS) mission of the counter-streaming energetic electrons and trapped energetic protons, localized in the magnetic field depressions between the mirror mode peaks, in the Earth’s dusk sector high-latitude magnetosphere. This region is characterized by high plasma beta, strong ion temperature anisotropy and intermediate plasma density between magnetospheric and magnetosheath plasma. We show that these plasma conditions are unstable for the drift mirror instability. The counter-streaming electron feature resembles those of the previously reported energetic electron microinjections, but without the energy-time dispersion signature. This suggests that MMS is passing through one of the potential microinjection source regions. The energetic ion data in the present study is mainly used to estimate the scale size of the mirror mode structures.