Item

Abstract

Due to a blocking of the Earth's magnetosphere, the bow shock is created by the supersonic solar wind in front of it. The solar wind suddenly drops to subsonic after passing the bow shock. This interaction results in a very complex magnetic region that fills with high-temperature and high-density plasma, called the magnetosheath. Physical quantities of the magnetosheath environment is important for studying of the solar wind-magnetosphere interaction that affects the dynamics of the magnetosphere. In past studies, researchers used a global simulation to predict the physical quantities of the magnetosheath. However, these efforts are often limited by computing resources, space-time resolution, model accuracy and so on. Using magnetopause crossing events from the THEMIS probes and solar wind data from the OMNI database, two different neural network models, Feedforward Neural Network (FNN) and Generalized Regression Neural Network (GRNN), were trained to forecast the physical quantities of the magnetosheath just outside the magnetopause from particular solar wind conditions. On our training results, GRNN performs better than FNN in terms of prediction accuracy. With an additional inclusion of the locations of the THEMIS probes as inputs, GRNN's performance became much better. Correlation coefficients were 0.81 and 0.84 for the particle density and temperature predictions, respectively, and the model has moderate accuracy for the prediction on the magnetic fields. This work allows us to predict the magnetosheath environment with low cost and high accuracy, which can advance our understanding of interactions between the solar wind and magnetosphere.