• For the first time, we simulate the bursty bulk flow events with kinetic physics embedded in a global magnetosphere model. • The electron velocity distributions demonstrate different anisotropy features at different locations surrounding the BBF. • Energy dependent electron pitch angle distribution evolutions are identified in the simulation.

We consider the flare prediction problem that distinguishes flare-imminent active regions that produce an M- or X-class flare in the future 24 hours, from quiet active regions that do not produce any flare within $\pm 24$ hours. Using line-of-sight magnetograms and parameters of active regions in two data products covering Solar Cycle 23 and 24, we train and evaluate two deep learning algorithms—CNN and LSTM—and their stacking ensembles. The decisions of CNN are explained using visual attribution methods. We have the following three main findings. (1) LSTM trained on data from two solar cycles achieves significantly higher True Skill Scores (TSS) than that trained on data from a single solar cycle with a confidence level of at least 0.95. (2) On data from Solar Cycle 23, a stacking ensemble that combines predictions from LSTM and CNN using the TSS criterion achieves significantly higher TSS than the “select-best” strategy with a confidence level of at least 0.95. (3) A visual attribution method called Integrated Gradients is able to attribute the CNN’s predictions of flares to the emerging magnetic flux in the active region. It also reveals a limitation of CNN as a flare prediction method using line-of-sight magnetograms: it treats the polarity artifact of line-of-sight magnetograms as positive evidence of flares.

The magnetohydrodynamics with embedded particle-in-cell (MHD-EPIC) model has been successfully applied to global magnetospheric simulations in recent years. However, the PIC region was restricted to be one or more static boxes, which is not always sufficient to cover the whole physical structure of interest efficiently. The FLexible Exascale Kinetic Simulator (FLEKS), which is a new PIC code and allows a dynamic PIC region of any shape, is designed to break this restriction. FLEKS is usually used as the PIC component of the MHD with adaptively embedded particle-in-cell (MHD-AEPIC) model. FLEKS supports dynamically activating or deactivating cells to fit the regions of interest during a simulation. An adaptive time-stepping scheme is also introduced to improve the accuracy and efficiency of a long simulation. The particle number per cell may increase or decrease significantly and lead to load imbalance and large statistical noise in the cells with fewer particles. A particle splitting scheme and a particle merging algorithm are designed to limit the change of the particle number and hence improve the accuracy of the simulation as well as load balancing. Both particle splitting and particle merging conserve the total mass, momentum, and energy. FLEKS also contains a test-particle module to enable tracking particle trajectories due to the time-dependent electromagnetic field that is obtained from a global simulation.

We perform a geomagnetic event simulation using a newly developed magnetohydrodynamic with adaptively embedded particle-in-cell (MHD-AEPIC) model. We have developed effective criteria to identify reconnection sites in the magnetotail and cover them with the PIC model. The MHD-AEPIC simulation results are compared with Hall MHD and ideal MHD simulations to study the impacts of kinetic reconnection at multiple physical scales. At the global scale, the three models produce very similar SYM-H and SuperMag Electrojet (SME) indexes, which indicates that the global magnetic field configurations from the three models are very close to each other. We also compare the ionospheric solver results and all three models generate similar polar cap potentials and field aligned currents. At the mesoscale we compare the simulations with in situ Geotail observations in the tail. All three models produce reasonable agreement with the Geotail observations. At the kinetic scales, the MHD-AEPIC simulation can produce a crescent shape distribution of the electron velocity space at the electron diffusion region which agrees very well with MMS observations near a tail reconnection site. These electron scale kinetic features are not available in either the Hall MHD or ideal MHD models. Overall, the MHD-AEPIC model compares well with observations at all scales, it works robustly, and the computational cost is acceptable due to the adaptive adjustment of the PIC domain. It remains to be determined whether kinetic physics can play a more significant role in other types of events, including but not limited to substorms.

We use machine learning methods to predict whether an active region (AR) which produces flares will lead to a solar energetic particle (SEP) event using Space-Weather Michelson Doppler Imager (MDI) Active Region Patches (SMARPs). This new data product is derived from maps of the solar surface magnetic field taken by the Michelson Doppler Imager (MDI) aboard the Solar and Heliospheric Observatory (SOHO). We survey the SMARP active regions associated with flares that appear on the solar disk between June 5, 1996 and August 14, 2010, label those that produced SEPs as positive and the rest as negative. The AR SMARP features that correspond to each flare are used to train two different types of machine learning methods, the support vector machines (SVMs) and the regression models. The results show that the SMARP data can predict whether a flare will lead to an SEP with accuracy (ACC) {less than or equal to}0.72{plus minus}0.12 while allowing for a competitive leading time of 55.3{plus minus}28.6 minutes for forecasting the SEP events.

Magnetospheric sawtooth oscillations are observed during strong and steady solar wind driving conditions. The simulation results of our global MHD model with embedded kinetic physics show that when the total magnetic flux carried by constant solar wind exceeds a threshold, sawtooth-like magnetospheric oscillations are generated. Different from previous works, this result is obtained without involving time-varying ionospheric outflow in the model. The oscillation period and amplitude agree well with observations. The simulated oscillations cover a wide range of local times, although the distribution of magnitude as a function of longitude is different from observations. Our comparative simulations using ideal or Hall MHD models do not produce global time-varying features, which suggests that kinetic reconnection physics in the magnetotail is a major contributing factor to sawtooth oscillations.

The MHD with embedded PIC (MHD-EPIC) model makes it feasible to incorporate kinetic physics into a global simulation. Still, this requires a large enough box-shaped PIC domain to accommodate the movement and changes of the magnetic reconnection regions over time. This wastes computational resources on simulating regions with the expensive PIC model where MHD would be sufficient to describe the physics. We have developed a new MHD with Adaptively Embedded PIC (MHD-AEPIC) algorithm that couples the BATS-R-US MHD model with the new FLexible Exascale Kinetic Simulator (FLEKS) PIC code. In the new coupled model the PIC domains can move with the magnetic reconnection regions and adapt to them with an arbitrary shape. In this work, we will first introduce the algorithms for selecting the reconnection regions in the MHD model that need to be resolved with the kinetic PIC model. Then we will compare simulations obtained with MHD-EPIC using fixed PIC regions versus MHD-AEPIC employing adaptive PIC regions to verify that the new model generates reliable results. Finally, we will apply the MHD-AEPIC model to a global magnetic storm simulation and demonstrate the improved efficiency.