Modeling Equatorial to Mid-Latitudinal Global Night Time Ionospheric Plasma Irregularities Using Machine Learning

This study focuses on modeling the characteristics of nighttime topside Ionospheric Plasma Irregularities (PI) on a global scale. We utilize Random Forest (RF) and a one-dimensional Convolutional Neural Network (1D-CNN) model, incorporating data from the Swarm A, B, and C satellites, space weather data from the OMNIWeb data center, as well as zonal and meridional wind model data. Our objective is to simulate monthly global PI characteristics using a multilayer 1D-CNN model trained on 12 space weather and ionospheric parameters. In addition, we investigate the most influential input parameters for predicting global nighttime PI characteristics. Our findings indicate that non-equinox months exhibit the highest equatorial PI magnitude over the American-African longitudinal sector, contrary to the expected higher Rayleigh-Taylor instability growth rate during equinox months. Winter months display the most intense and widespread vertically and horizontally distributed equatorial PI patterns. We also observe double peaks across geomagnetic latitudes and longitudinally varying wavelike irregularity structures, particularly in May, August, and predominantly in September. Furthermore, north-south hemispherical asymmetry in PI observed across different seasons. Through the RF parameter importance analysis method, we determine that temporal, geographical, and magnetic disturbance-related factors play a crucial role in predicting global PI variabilities. These findings emphasize the significance of these variables in controlling the strongest PI characteristics observed in the Atlantic sector, which has garnered considerable attention in PI research. The employed 1D-CNN model demonstrates exceptional predictive capabilities, exhibiting a strong correlation of 0.98 for global PI characteristics across all months and satellites.