By measurements of Magnetospheric Multiscale (MMS) spacecraft in the magnetotail, we studied electron distribution function and Nonideal electric field across an electron diffusion region. The energy of the non-gyrotropic distribution increased with growth of the vertical distance from the mid-plane. For the electrons within certain energy range, they exhibited the non-gyrotropic distribution at the distance further away from the mid-plane than that expected from the meandering motion. The correlation between the crescent-shaped distribution with multiple stripes and the large Hall electric field was established. It appears that the measured non-gyrotropic distribution and the crescent-shaped distribution were caused by the meandering motion and the Hall electric field together. Moreover, we found the electron pressure tensor is a fundamental aspect in breaking the frozen-in condition, and makes a great contribution to the energy dissipation.

Electrostatic solitary waves (ESWs) are ubiquitously observed in magnetic reconnection. In this paper, two-dimensional (2-D) particle-in-cell (PIC) simulations are performed to investigate the characteristics of ESWs in asymmetric magnetic reconnection. ESWs with bipolar structures of the parallel electric field can be generated near the separatrices only on the magnetosphere side, and propagate to reconnection downstream direction along the magnetic field. These structures corresponding to electron phase-space holes can cross both the electron outflow and inflow channels although their main part is located in the electron outflow. When there is no guide field, the ESWs are generated by electron two-stream instability. When there is a guide field, the amplitude of the ESWs are different on the right and left side of the X line. On the left side, the ESWs are weaker and generated by the electron two-stream instability. On the right side, the ESWs are stronger, and there are two kinds of ESWs with distinct phase speed. The faster one is generated by electron two-stream instability, while the slower one is generated by Buneman instability.

Recently, several papers have discussed the Kelvin-Helmholtz (K-H) instability and its related waves and vortex structures by using global 2D/3D MHD simulations [Guo et al., 2010; Li et al., 2012; Li et al., 2013; Merkin et al., 2013].  Merkin et al. (2013) found a double-vortex sheet in which a vortex train propagates along the inner and outer edges of the magnetopause, which is so-called a forward Karman vortex. The vortex trains in the inner edge continues to north and south into ionospheres to form an inverse Karman vortex in the ionospheres, and form a consistent coherent structure in the magnetosphere.  Global magnetospheric coherent structures related to the dynamical coupling between the dayside magnetopause frontier and ionosphere in a northward IMF configuration are analyzed using the concept of forward and inverse Karman vortex in a global 3D MHD simulation.

It is known that the earth magnetic field interacts with the solar wind and forms a complex magnetic manifold so-called the “magnetosphere.” This interaction involves a large-scale energy transfer at some locations, where the magnetic reconnection takes place. The magnetic reconnection has been studied for a long time and is still one of the hottest research topics in the magnetospheric physics. The two-dimensional and one-dimensional stable or unstable manifolds, respectively, are generated and visualized from the critical points (CPs) or magnetic nulls in north and south in a 3D global MHD simulation with a northward IMF using Geodesic Level Set (GLS) method. The 2D unstable and stable manifold generated from the north and south CPs, respectively, forms cylinder-like topologies and cross each other from north to south near the meridian plane. We, first time, visualize the terrestrial magnetospheric topology from a 3D global MHD simulation and discuss how the magnetospheric topology relates to the magnetospheric reconnection.   

The Global‐scale Observation of Limb and Disk (GOLD) mission, for the first time, provides synoptic two‐dimensional (2D) maps of OI 135.6 nm observations. These maps describe the unambiguous and dynamic evolution of nighttime ionospheric F₂‐peak electron densities (N_mF₂) as the 135.6 nm airglow emission radiance correlates well with N_mF₂ at night_. On November 19, 2018, the 135.6 nm radiance measured by GOLD, N_mF₂measured by a digisonde, and GPS total electron content (TEC) measurements at Cachoeira Paulista (CP) all showed a postsunset enhancement, with an increase near 22:30 universal time. The 135.6 nm radiance map showed that this enhancement was due to the southward movement of the southern equatorial ionization anomaly (EIA) crest. Therefore, the GOLD observation showed the linkage between postsunset enhancement of NmF2 and EIA movement. Furthermore, unlike the southward movement of the southern crest, the corresponding EIA northern crest, however, did not show northward motion. This is the first time that the EIA hemispheric asymmetry, which included both different densities and movement of two crests in a short time period (<2‐hour), was captured. The cause of this asymmetric movement of the two crests is not clear and requires further investigation.

COSMIC-2 has multiple space weather instruments for ionospheric measurements.  The IVM (Ion Velocity Meter) instrument measures in-situ electron densities.  The TRGS (Tri-GNSS Radio Occultation System) instrument measures slant TEC, which can be used to derive electron density profiles.  There have been comparisons of these two types of measurements, which were done indirectly.  Since COSMIC-2 has 6 satellites and each has an IVM instrument, validating all six instruments for consistency will be invaluable for future studies of global scale structures in the ionosphere arising from tides and planetary waves.  IVM observations have traditionally been validated with ground-based observations; however, the in-situ electron density can be derived from TGRS slant TEC observations providing an additional validation data source. The in-situ electron density can be derived from the TGRS slant TEC data when the observational geometry is such that the GNSS-to-LEO ray-path is aligned with the COSMIC-2 satellite velocity in the forward or backward directions.   As the COSMIC-2 satellite moves, the distance between the GNSS and COSMIC-2 LEO satellite changes resulting in variations in the slant TEC.   By taking the ratio between the change in the slant TEC and the distance travelled by the LEO satellite, the in-situ electron density can be derived.  As a specific geometry is required, we expect to be able to derive the in-situ electron density from the TGRS observations for about 5 cases per day for each LEO satellite.  Since both instruments are continuously operating, we have more coincidences.  We compared the IVM density values with the TGRS derived in-situ densities for all 6 COSMIC-2 LEO satellites.   There are some small differences in the scaling factor for IVMs of different satellites, which we will investigate further to see if there is a justification for further refinement in data processing. 

On 23 November 2019 NASA GOLD mission observed equatorial plasma bubbles (EPBs), whose southernmost edges are bordered with the enhancement of 135.6 nm airglow over the background. Madrigal TEC maps also show EPBs with the southernmost boundaries having higher TECs than the ambient. One hour later both Swarm-Alpha and Swarm-Charlie encountered low-latitude plasma blobs near those EPB boundaries. The blob signatures in Madrigal TEC maps are elongated from northwest to southeast, and the structures do not move in the westward/equatorward direction. Continental-scale coverage of the TEC data reveal that the blob event in the Northern Hemisphere is near-conjugate to EPBs in the Southern Hemisphere. We are looking forward to the synergy between GOLD, Madrigal TEC, Swarm, ICON and FORMOSAT-7/COSMIC-2 for studying low-latitude plasma blobs and their interconnections to the EPBs or other ionosphere instabilities.

During the rare Antarctic sudden stratospheric warming (SSW) in September 2019, the quasi-6‐day wave (Q6DW) was dramatically enhanced in the mesosphere and lower thermosphere (MLT) region, and strong quasi-6‐day oscillations (Q6DO) are observed in ionospheric equatorial electrojet and electron density.  However, model simulations demonstrated that the Q6DO in the ionosphere is not driven directly by the Q6DW during the September 2019 SSW.  Instead of the secondary waves generated by nonlinear interaction between the Q6DW and semi-diurnal migrating tides.  In this study, we examine the amplitude variations of Q6DO and shorter period child waves by using data-assimilation analysis of electron density from three-dimensional Global Ionosphere Specification (GIS).  The GIS assimilates the slant total electron content (TEC) observations from FORMOSAT-7/COSMIC2 and worldwide ground-based GNSS receivers.  We focus on the behaviors of W1_13h and W3_11h, which are the largest in the secondary waves and are the possible contributors to the ionospheric Q6DO.  The day-to-day variabilities of child waves show that both the W1_13h and W3_11h are dramatically enhanced after the SSW peak, growing in amplitude at the EIA region with poleward extension from negligible amplitude.  The amplitudes reach maximum and contribute about 6% to the zonal and diurnal mean background TEC, which coincides with the day of maximum amplitude of Q6DO.  The amplitude distributions of two child waves are also overlapped with comparable amplitude.  These results suggest that the two-child waves might interfere with each other at the EIA regions, resulting in the change of the amplitudes and phases of Q6DO at the different local times if their contributions are not uniform in local time.  Additionally, child waves' activity during the same period in the non-SSW year of 2020 is also examined for comparison.

Coronal mass ejections (CMEs) are frequently observed on the Sun. However, compared to their solar counterparts, CMEs are seldom observed on other stars. Using the spectroscopic data from LAMOST, we have attempted to search for possible CME signatures by conducting a survey of asymmetric H-alpha line profiles. From the public data archive of LAMOST, we have identified obviously asymmetric H-alpha line profiles from observations of seven late-type stars. The seven stars have various spectral types, including F type and M type. All events reveal an emission enhancement at the blue or red wing that corresponds to a velocity of 100 km/s to 200 km/s, which is far less than the escape velocities of the stars. In one of the seven events we observed an intensity variation similar to that of solar flares. The H-alpha line has a clear Voigt profile in the impulsive phase of the flare. Obvious enhancement was identified at the red wing of the H-alpha line in the decay phase. The red wing enhancement corresponds to plasma moving away from the Earth at a velocity of 100-200 km/s, which might be signatures of cooling materials draining back to the stellar surface or backward ejected CMEs from the edge of the stellar disk.

Light bridges (LBs) are relatively bright structures that divide sunspot umbrae into two or more parts. Chromospheric LBs are known to be associated with various activities including fan-shaped jet-like ejections and brightenings. Although magnetic reconnection is frequently suggested to be responsible for such activities, not many studies present firm evidence to support the scenario. We carry out magnetic field measurements and imaging spectroscopy of an LB where fan-shaped jet-like ejections occur with co-spatial brightenings at their footpoints. We study LB fine structure and magnetic field changes using TiO images, Near-InfraRed Imaging Spectropolarimeter, and Hα data taken by the 1.6m Goode Solar Telescope. We detect magnetic flux emergence in the LB that is of opposite polarity to that of the sunspot. The new magnetic flux cancels with the pre-existing flux at a rate of 5.6×10¹⁸ Mx hr⁻¹. Both the recurrent jet-like ejections and their base brightenings are initiated at the vicinity of the magnetic flux cancellation, and show apparent horizontal extension along the LB at a projected speed of up to 18.4 km s⁻¹ to form a fan-shaped appearance. Based on these observations, we suggest that the fan-shaped ejections may have resulted from slipping reconnection between the new flux emerging in the LB and the ambient sunspot field.

We report on the successive occurrence of 0.5 asc wide photospheric vortices with strong transverse shear flows at the edge of a sunspot light bridge (LB), and the subsequent ejection of chromospheric surges observed using a Visible Interferometry Spectrograph, a broadband TiO filter, and a Near InfRared Imaging Spectrograph of the Goode Solar Telescope operating at Big Bear Solar Observatory. The Hα surges ejected at the location of the vortices often appeared in a hollow cylindrical structure. We also observed quasi-periodic vortex-associated bright Hα plasma blobs moving upward with a speed of up to 4 km s−1. In view of the strong shear flow at the edge of the LB, it is likely that the vortices form under the Kelvin–Helmholtz instability. The surges may result from either the magnetic tension generated after magnetic reconnection or an acoustic impulse of a fast photospheric transverse flow. Otherwise, the surges could also be associated with Alfvénic waves, in which case their origin could be torsional magnetic fields generated in the process of the vortex formation.

Observations from the GST telescope have revealed unprecedented details and fast dynamics in sunspot light bridges. (1) We examined the high-resolution TiO images and discovered a new type of fine structures in light bridges: striking knot-like dark structures within the central dark lane. These dark knots divide the central dark lane into multiple sections, and they appear to be very common in narrow light bridges. The evolution of these highly dynamic dark structures could provide detailed information about the magnetoconvection in light bridges. (2) Through a detailed analysis of the spectropolarimetric data taken by the NIRIS instrument, we found evidence of small-scale flux emergence in a light bridge. An Halpha jet is clearly triggered by the interaction between the emerging flux and the background field. (3) Through joint GST and IRIS observations, we identified unambiguous evidences of magnetic reconnection in light bridges: frequently occurring inverted Y-shaped jets in the Halpha wing images and UV burst-type profiles of the transition region lines. We also demonstrated that the surge-like activity above light bridges has two components: the ever-present short and slow surges likely to be related to the upward leakage of magnetoacoustic waves from the photosphere, and the occasionally occurring long and fast surges that are obviously caused by the intermittent reconnection jets.

A generation process of plasmaspheric hiss-like emissions has been successfully reproduced in a recent electromagnetic particle simulation. The result possesses a preliminary agreement with the analysis in accordance with nonlinear wave growth theory. We further examine the applicability of the theory to the generation process of hiss-like emissions in this study. The main approach is to vary key parameters in both simulation and theory, and to inspect their correspondence. We firstly change hot electron density to different levels. In nonlinear wave growth theory, as the hot electron density decreases, the overlap between optimum and threshold amplitudes which represent the existence of nonlinear wave growth process will diminish quickly. We find that the simulation results hold the same tendency that wave amplitudes turn to a small magnitude and lose their obvious hiss-like structures. We provide a consistent analysis that a low hot electron density may lead to a small linear growth rate at the initial stage, thus the inhomogeneity factor S may have a large value, denoting that nonlinear wave growth process is not involved. We then vary gradient of background magnetic field. In the theoretical analysis, as we increase the gradient from the homogenous condition gradually, the optimum amplitude remains constant, while the threshold amplitude becomes larger, making nonlinear wave growth difficult to occur. We find a relatively similar tendency in simulation results that wave amplitude is depressed for a larger background field gradient condition. Some aberrant wave packets seem to undergo nonlinear growth process even though the overlap in the theoretical result has vanished. We give discussions on the depth of electron hole, that the occurrence of relatively deep electron hole may explain the discrepancy between theory and simulation.

Global navigation satellite systems (GNSS) or satellite navigation is a significant technological advancement; however, they can be greatly impacted by the effects of space weather, such as ionosphere scintillation. Ionosphere scintillation is one of the causes of errors in the GNSS signals and also has the potential to lead to a loss of access to GNSS. Ionosphere scintillation often impacts the polar region; however, the cause is not always known. One potential source of scintillation is polar cap patches. In Ren et al. [2018], a polar cap patch database was created based on the incoherent scatter radar measurements at Resolute Bay (RISR). Using ionosphere scintillation data provided by the CHAIN Network near Resolute Bay in 2016, we studied how the polar cap patches impact ionosphere scintillation by looking at the scintillation changes at the leading edge, the trailing edge, and the patch center separately. Statistical analysis, as well as event analysis, have been performed. Scintillation data from GNSS satellites with an elevation angle over 40 degrees were collected from each patch in the database and were compared to the daily average. It is found that statistically there is no obvious phase scintillation or amplitude scintillation increase associated with the center of the patch in the polar cap, but obviously enhanced phase scintillations near both edges. The highest phase scintillations associated with patches occur around the noon MLT. Three different patch events with and without enhanced scintillation were chosen for in-depth analysis and cross-comparison for the event analysis. Other datasets, including AMPERE FAC and RISR, are used to understand the plasma characteristics and geomagnetic activity conditions during these events.

The earth’s magnetosphere is thought to be an area of space controlled by a distorted dipole magnetic field with a long tail extending to hundreds of earth radius (RE), which protects the earth’s environment and prevent the plasma escaping from outer atmosphere into the interplanetary space. However, how far the magnetotail extends during different interplanetary magnetic field conditions is still not known, especially during northward IMF condition. Here we report a nearly pure dipole magnetosphere formed with a tail only extending to 28 RE during a long-lasting northward IMF, simulated by a high-resolution data-driven 3-D global magnetohydrodynamics (MHD) code. This magnetosphere is generated due to the quasi-balance between the denudation of the open and closed magnetic field lines by the single high latitude lobe magnetic reconnection at one hemisphere and the supplement of reclosed field lines by the dual lobe reconnections at both hemispheres. This results in a dominated two reversed convection cells with pairs of arc-like poleward-moving field-aligned currents both from dawnside and duskside auroral oval regions and merging together around the noon-midnight meridian, confirmed by the global convection and auroral observations. This study gives a new insight into the understanding of the solar wind-magnetosphere-ionosphere coupling processes under northward IMF conditions.

This paper is a statistical survey of polar cap patches in relation to solar and geomagnetic activity. 13199 patches have been identified from in-situ plasma observations of the Defense Meteorological Satellite Program (DMSP) F16 satellite for 14 years (2005-2018). These patches are divided into two groups: 1) cold patches, which consist of dense but cold plasma; and 2) hot patches, which consist of dense but hot plasma. The statistical results indicate that (1) the occurrence of cold patches is clearly dependent on solar and geomagnetic activity, but hot patches do not show such dependence; (2) solar activity reduces the spatial size of both cold and hot patches; (3) geomagnetic activity increases the spatial size of both cold and hot patches.

In this study, we investigate the variations in the D-region ionosphere during a fireball occurred in Hokkaido at 11:55:55 UT on 18 October, 2018, using VLF / LF transmitter signals. The transmitters used in this study were JJY40kHz (Fukushima, Japan, 37.37˚ N, 140.85˚ E), JJY60kHz (Saga, Japan, 33.47˚ N, 130.18˚ E), and JJI (Miyazaki, Japan, 22.2 kHz, 32.05˚ N, 130.82˚ E). The receiver was located at RKB (Rikubetsu, Hokkaido, Japan, 43.45˚ N, 143.77˚ E). Periodic variations of 100-200 s were identified by a wavelet transformation of the signal intensities for the JJY40kHz-RKB, JJY60kHz-RKB, and JJI-RKB paths at about five minutes (12:01 UT) after the fireball. We consider that these variations of intensity were caused by the D-region variations due to acoustic waves in the atmosphere excited by the fireball. If the acoustic waves were excited at the luminous point (118 km altitude) or vanishing end point (25 km altitude) of the fireball, the propagation times of the acoustic waves from the exited point to the LF reflection point at 90 km height over RKB were calculated to be 138 s or 311 s, respectively. The arrival time (311 s) of the acoustic waves excited from the vanishing point roughly matched with the onset of the VLF/LF variations. From the onset of the VLF/LF variations, we estimated the location where the variations in the VLF/LF intensity along the paths. The estimated location was close to the RKB. The VLF/LF variations would be caused by acoustic waves excited at the vanishing end point. The acoustic waves obliquely propagated from the vanishing end point up to the D-region height at the south point of the RKB receiver.

In the D-region ionosphere, oscillations of LF (low frequency, 30-300 kHz) transmitter signals with a period of 100s were reported about five minutes after mainshock of the 2011 Tohoku earthquake [Ohya et al., JGR, 2018].  This is only one report for coseismic disturbances in the D-region ionosphere. In this study, we investigate the D-region ionospheric variations associated with the 2015 Nepal earthquake using VLF (very low frequency, 3-30 kHz)/LF transmitter signals that reflect in the D-region ionosphere. The mainshock of the Nepal earthquake (Mw 7.9) occurred at 06:11:26 UT on April 25, 2015. The propagation path was BPC (68.5 kHz, 34.63°N,115.83°E) - TKN (Takine, Fukushima, 37.34°N, 140.67°E). Intensity and phase were observed with a sampling time of 0.1s. We compared between the VLF/LF transmitter signals and vertical velocity data of seismometers provided by IRIS (Incorporated Research Institutions for Seismology), USA. Based on wavelet analysis, a periodic component of about 100-300 s was seen in both the VLF/LF signals and seismic velocity at arrival time of acoustic waves excited by Rayleigh waves. The coherences between the VLF/LF variation and the seismic velocity were 0.91 and 0.85 for amplitude and phase, respectively, which were significant at the 95% confidence level. If the acoustic waves were excited at the midpoint of BPC-TKN propagation path by the Rayleigh waves that propagated horizontally from the epicenter, and propagate upward up to the D-region height 70 km, the propagation time from the ground to 70 km height would be 225s. The propagation time of the Rayleigh wave calculated from seismograph data was 1057s, and the whole propagation time of the Rayleigh and acoustic waves was 1252s . The arrival time of the acoustic waves was in good agreement with the VLF/LF oscillations. 

We present a statistical study of the thickness of the interplanetary shocks, by using WIND 3D plasma instrument and Magnetospheric Multiscale (MMS) mission measurements. For these cases, we use the Rankine-Hugoniot (RH) shock fitting technology to determine the normal direction and shock speed of the interplanetary shocks, and calculate the thickness of the interplanetary shock by formula V^sc_sh-N·Δt_ramp. In these cases, the thickness of the interplanetary shocks are quite different, all in the range of tens to thousands of kilometers. Compared to previous studies, this study provides a more accurate value for the thickness of the interplanetary shock. Furthermore, by processing the data of the interplanetary shock observed by the MMS mission and the wind spacecraft, we can compare the changes in the parameters of the same interplanetary shock during the propagation process, which is helpful for us to understand the evolution process of the interplanetary shock.

In this study, we develop a model, which is called DeepIRI, to generate improved IRI global TEC and electron density (N_e) maps by deep learning. We develop a DeepIRI TEC model based on conditional Generative Adversarial Networks (cGANs). For this we consider the IRI TEC maps and IGS TEC maps for training the model. We evaluate the model by comparing IGS TEC maps and DeepIRI TEC ones. The DeepIRI TEC maps are much more consistent with the corresponding IGS TEC maps than the IRI TEC ones. Especially, ionospheric peak structures are successfully generated in DeepIRI TEC maps, while they are not in IRI-2016 ones. Then we construct a global DeepIRI-3D N_e model from DeepIRI TEC data. For this we develop a model to generate 1-D N_e profile from a TEC value using Multi-Layer Perceptron (MLP). For training and test, we use five input data (day of year, time, latitude, longitude, IRI TEC). Output data of the model are the 3D global N_e profile with temporal and spatial resolutions of 2 hours in time, 2.5° in latitude, 5° in longitude, and ~25 km in altitude. For test data sets, the model shows that N_e profiles with IRI TEC input are consistent with those of IRI model. We use the DeepIRI TEC as input data of the model to generate the DeepIRI N_e profile data, which is called DeepIRI-3D N_e model. We evaluate the model by comparing the observed N_e profiles and DeepIRI ones. The evaluation shows that the N_e profiles from the DeepIRI-3D Ne model are closer to the observed data than those of the IRI model. These results suggest a sufficient possibility that our DeepIRI model can improve the global TEC and N_e profile prediction ability of the IRI model.           

The semantic segmentation has been a task on ionosonde measurements. Here we choose Spatial Attention U-Net to perform the task on Hualien VIPIR dataset. The dataset contains 4071 ionograms, and is separated into 2606, 651, 814 ionograms as training, validation and test set. Each ionogram is manually idenified as ground truth with 7 classes: Eo, Ex, Esa, Fao, Fax, Fbo, and Fbx. The resulting IoU of Esa and Fbo on the test set are 73.53% and 71.29%. We furthermore collected 2109 ionograms from the original dataset to form an indicative dataset called "golden set". The resulting IoU of Esa and Fbo on the golden set are 77.66% and 72.16%.

We present the challenges upon applying image segmentation in ionograms for space weather monitoring. In ionogram recovery, determination of the critical frequency is needed to extract the amount of ionization in the ionosphere. The distribution of the critical frequencies for each layer versus the frequency shows that the coverage is non-uniform wherein several deep and wide gaps can be observed. These gaps are a direct consequence of the strong quasi-static signal contaminations from ground-based transmitters that are present in the ionograms. Because of the dynamic behavior of the E and F layers, the minimum and maximum frequencies also overlap over the distribution. To resolve this problem, the ionogram recovery model must operate on a wide frequency range and provide a smart interpolation of the layers. The signal-to-noise ratio (SNR) is computed to evaluate the quality of the outputs of our model. We have devised a new definition of SNR specifically for ionograms in order to account for the signal, contamination, and also the background. The mean SNR calculated in this study is ~ 1.4. The signal shape is also important in evaluating the output of the model. We have defined a shape parameter known as circumference over area (C/A) to distinguish whether the signals are elongated or point-like. The elongated signals are expected to have a higher Intersection over Union (IoU) compared to the point-like signals because the area of signals are higher compared to the contaminations due to labelling. In both cases, the accuracy of the model is still limited by the quality of the labeled ionograms.  The work was performed as part of NCU AI group: Chang Y.-C., Dmitriev A., Hsieh M.-C., Hsu H.-W., Huang G.-H., Li Y.-H., Lin C.-H., Lin Y.-C., Mendoza M., Tsai L.-C., Tsogtbaatar E.

We apply algorithms of segmentation based on deep learning for monitoring the upper atmosphere and space weather by ionozonde product of ionograms. This machine learning technique has impressive independent adaptability consequences for fast and robust erasing of the noise from ionospheric signals that makes possible complete recovery of ionograms. The technique provides recognition systems and straight influence the ionograms confirmation, feature map visualization and classification results. This is the most significant and very challenging issue in this field. We used two segmentation algorithms based on popular technology of convolutional neural networks, named as Autoencoder, and on dense-fully convolutional network (DFCN) algorithms. We apply them to two different data sets: the so-called Peru data set of 816 ionograms and Taiwan data set of 4071 ionograms for training. Also, we accurately selected 2024 out of 4071 ionograms for testing of models. The Autoencoder model achieves the accuracy of 73 % for the Peru data and 75% for the Taiwan data. The DFCN model achieves the accuracy of 79% and 80% for the Peru and Taiwan data, respectively. These high-level accuracies outperform the existing approaches on challenging the ionogram data bases. The work was performed as part of NCU AI group: Chang Y.-C., Dmitriev A., Hsieh M.-C., Hsu H.-W., Huang G.-H., Li Y.-H., Lin C.-H., Lin Y.-C., Mendoza M., Tsai L.-C., Tsogtbaatar E.

To perform effective long-term predictions for the regional ionosphere, we developed a Long Short Term Memory deep-learning algorithm for geomagnetic quiet days (LSTM-quiet) in the previous study (Moon et al., 2020). However, our previous model could not predict geomagnetic storm days effectively at all. In this study, we developed an LSTM model suitable for geomagnetic storms using the new training data set and re-designing input parameters and hyper-parameters. We collected 131 days of geomagnetic storm cases from 1 January 2009 to 31 December 2019 and obtained the IMF Bz, Dst, Kp, and AE indices related to the geomagnetic storm corresponding to each storm date. These indices and F2 parameters of Jeju ionosonde (33.43˚N, 126.30˚E) were used as input parameters for the LSTM model. To test and verify the predictive performance and the usability of the LSTM model for geomagnetic storms developed in this manner, we created and diagnosed the 0.5, 1, 2, 3, 6, 12, and 24-hour predictive LSTM models. As the results new the LSTM storm model for 24-hour developed in this study achieved a predictive performance during the geomagnetic storms about 32% (10%), 34% (17%), and 37% (5%) better in RMSE of foF2 (hmF2) than the LSTM quiet model (Moon et al., 2020), SAMI2, and IRI-2016 models. We propose sufficient potential for prediction models of less than 3 hours for short and long-term predictions. It shows an RMSE of less than ~1MHz for foF2 and 25km for hmF2, which are sufficiently competitive compared to other traditional models. Our study is meaningful work because it is the first attempt to collect and train only geomagnetic cases using the deep learning models. Thus, this study suggests that our model can be used for short-term prediction and monitoring of the regional mid-latitude ionosphere.

Downstream of the Earth’s bow shock, resulting from the interaction of the supersonic solar wind with the geomagnetic field is the Magnetosheath region. This highly disturbed environment contains several transient phenomena (e.g. magnetosheath jets) that further contribute to the plasma turbulence. Most of the properties of the shock and its associated magnetosheath originate from the angle (Θbn) between the interplanetary magnetic field (IMF) and the bow shock normal vector (n). Angles less than 45° are associated to the so called “quasi-parallel” shock and these with more than 45° to the “quasi-perpendicular”. Closely related is the concept of the ion foreshock. Ion foreshock is the region upstream of the bow shock where wave-particle interactions, defuse ions and ULF wave activity governs the dynamics of the environment. Typically foreshock becomes significant at ~45 degrees and gets intensified closer to the bow shock and when  becomes smaller.While shock crossings have been known for a while and theoretical descriptions for idealized cases exist, there is limited work on how specific plasma properties change between the upstream and downstream between the different plasma environments (Q-par/Q-perp shock, Foreshock/no-Foreshock). This is primarily because of the Qpar shock complexity but also due to the unavailability of reliable measurements upstream and downstream of the shock.We present the progress done in the modeling, classification and characterization of the different upstream and downstream plasma along with the classification of the localized dynamic pressure enhancements (jets). To accurately characterize the different environments we use a plethora of different spacecraft and datasets. The input of the models consist of solar wind values (OMNIweb), upstream and downstream measurements (Cluster) and magnetosheath data (MMS). These measurements are imported in neural networks that provide the modeling and the final characterization through a variety of classification and regression tasks.