Distinguishing Auroral-E and Sporadic-E in GNSS Radio Occultation Measurements

GNSS radio occultation (RO) is widely used to detect metallic-ion sporadic-E layers on a global scale, but in auroral environments, particle precipitation-driven ionization can produce auroral-E structures with similar RO signatures. This overlap complicates reliable identification of sporadic-E and may bias high-latitude occurrence rate estimates. In this study, we quantify auroral contamination in RO-based sporadic-E detection and evaluate the performance of both threshold-based and multivariate classifier approaches. A dataset of 106 conjunction events was constructed using RO observations from the Spire and COSMIC-1 constellations, together with co-located ionosonde and magnetometer measurements at two high-latitude sites and an ionosonde at a mid-latitude site. Events were labeled as sporadic-E, auroral-E, or non-event from manual ionogram inspection. Two commonly used RO detection methods were applied: an excess phase perturbation Δϕ approach and an amplitude-based S4 approach. Results show that true sporadic-E events are primarily identified when both amplitude- and phase-based thresholds are satisfied. In contrast, more than 80% of auroral-E events satisfy at least one sporadic-E detection criterion, with false triggering of auroral-E and non-events dominated by the phase-based method. A Random Forest classifier was developed as a proof-of-concept model showing success at classifying sporadic-E and non-events, although ambiguity between auroral-E and sporadic-E persists likely due to overlapping phase-perturbation distributions. A Multiple Phase Screen Model is implemented to provide physical insight into the RO observations, where the vertically thicker auroral-E layers are shown to produce significant phase perturbations but negligible amplitude scintillation. These results demonstrate that RO-based detection methods identify E-region structure rather than formation mechanism, implying that high-latitude sporadic-E occurrence rates derived from phased-based RO observations are likely overestimated without additional context.