Combination of Optimization Machine Learning and Statistical Tests to Determine MIT Key Parameters for Ionospheric Forecast Models

The main Earth spheres interact with each other in various ways, attributed to the dynamic and complex nature of the Earth’s environment. For example, the interaction between the solar wind plasma and Earth’s magnetic field can cause phenomena in the atmosphere, such as auroras, geomagnetic storms, ionospheric disturbances, and other natural hazards. Previous studies used indices of solar wind plasma and geomagnetic fields in their ionospheric Total Electron Content (TEC) forecast models. However, the selected indices among the studies were inconsistent. Therefore, our study combines optimization Machine Learning (ML) techniques with statistical and hypothesis tests to determine features for ionospheric TEC forecast models. Together with three time-related parameters (Hour, Day of Year, and Year), 38 indices of solar activities and geomagnetic fields were tested to identify the best-suited parameters. Our investigation results reveal the Earth's Magnetosphere, Ionosphere, and Thermosphere (MIT) key parameters, including Plasma temperature (or Plasma speed), Proton density (or Flow pressure), Kp index (or ap, Dst index), F10.7 index, and Lyman alpha. Depending on specific processing strategies and computer capacity that can handle big data, more or less features can be included in mathematical models. Still, the essential parameters just mentioned should be prioritized. Using these MIT key parameters enables an optimization of both processing speed and accuracy in training the ML-based ionospheric forecast models.