A Machine Learning Approach to Classify Ionospheric Scintillation Signatures in High Latitudes According to Irregularities

This study focuses on ionospheric scintillation signatures in signal phase and amplitude from two distinct regions of the high-latitude ionosphere, the auroral oval and the polar cap. As radio signals like GNSS propagate through the high-latitude ionosphere, they encounter diverse ionospheric structures in various scale sizes. Three major mechanisms have been identified to cause these irregularities: shear instabilities, steep density gradients, and electron precipitation (among others). The question of how much these irregularity mechanisms contribute to the scintillation signatures recorded in each region has yet to be definitively answered. For this work, we formulate two basic hypotheses: Different irregularity mechanisms are predominant in different high-latitude characteristic regions, and secondly, we expect different irregularity mechanisms to produce different scintillation signatures. Following those assumptions, scintillation event data from stations in the auroral oval and the polar cap was prepared for analysis spanning five geomagnetic storms. Using a Machine Learning approach, we clustered the event time series according to their correlation in phase and amplitude and classified them with a supervised decision tree. A training database was created based on successful distinctions of events from the auroral oval versus the polar cap. The second step of the study aims to extend the database with parameters that can link the time series signatures back to the ionospheric irregularity structures that caused them, like estimated irregularity drift velocity and spectral slopes. After identifying the most likely irregularity origin of the scintillation event based on observations from other instruments, we perform a principal component analysis to find the most influential factors that characterize ionospheric scintillation signatures.