Machine Learning Detection of Radio Occultation Electron Density Profiles Perturbed by the Equatorial Plasma Bubbles

Abstract

The FORMOSAT-7/COSMIC-2 (F7C2) constellation consists of six small satellites that provide high temporal and spatial resolutions of ionosphere observations at mid- and low-latitudes using radio occultation (RO) technology. While having the advantage of such dense radio soundings, ensuring the quality of the derived electron density profiles (EDPs) is crucial for applications such as data assimilation forecasting models or monitoring of the ionosphere status. However, after the Hunga Tonga-Hunga Ha’apai volcano erupted on January 15, 2022, more than 70% of the F7C2 EDPs from the Pacific to the Indian Ocean exhibited significant fluctuations, indicating possible data quality degradation. In addition to the extreme event giving such a high proportion of EDPs with quality uncertainties, the fluctuated EDPs are also observed in daily RO soundings. More than 40% of EDPs fluctuate during premidnight hours from October to December within $90~^{\circ }$ W– $0~^{\circ }$ E, while 70% fluctuate during postmidnight hours from May to July. This study presents, for the first time, a comprehensive investigation into the fluctuating EDPs during usual and event days. The statistics indicate that these fluctuating or irregular EDPs primarily occur during nighttime (>70%). The high correlation (0.82) between the longitudinal and seasonal variations of irregular EDPs and the ion velocity meter (IVM) climatological occurrence of equatorial plasma bubbles (EPBs), as observed in previous studies, indicates that irregular EDPs during postmidnight hours primarily result from EPBs. The machine learning models utilizing the Bagged Trees classification are developed to classify the normal and irregular EDPs across varying times, locations, and solar activity levels showing the clear connection between the EPBs and the irregular EDPs.