Automatic detection of field line resonance frequencies in the Earth’s plasmasphere

Abstract

Ground-based magnetometer stations represent a multi-viewpoint and easy-to-access system for sounding Earth’s magnetic field disturbances in the inner magnetosphere. Using Ultra-Low Frequency (ULF) measurements recorded from pairs of meridionally aligned stations, it is possible to determine the Field Line Resonance (FLR) frequencies, which are directly related to the equatorial magnetospheric plasma mass density. Recently, it has been shown by Foldes et al. (J Geophys Res 126(5):e2020JA029008. https://doi.org/10.1029/2020JA029008 , 2021) that the Machine Learning (ML) algorithms are valuable tools for detecting FLRs by exploiting the useful information provided by cross-phase Fourier spectra, which are at the heart of the ULF technique for inferring the magnetospheric mass density. The main shortcoming of this approach is that it is not possible to discriminate between active and quiet times in terms of resonances. It is commonly known that detecting FLRs using cross-phase spectra may often be unfeasible due to data gaps, noisy signals, and/or quiescent ULF wave periods. To handle these situations, we implement an ML classification algorithm to identify periods when the resonance frequencies are observable and thus easily estimated. Our algorithm can distinguish samples into three main classes: periods with observed frequency (“Freq" class) from others (“NoFreq"), and, in addition, it can determine whether the considered field line crosses the plasmasphere boundary layer (PBL or plasmapause) at a given time. The results of our method are validated for a particular pair of stations (at $$L=2.9$$ ) along the Equatorial quasi-Meridional Magnetometer Array (EMMA), using a large dataset comprising different geomagnetic conditions. The proposed approach might be combined with a regression algorithm (such as those proposed in Foldes et al. (J Geophys Res 126(5):e2020JA029008. https://doi.org/10.1029/2020JA029008 , 2021)) in a two-stage ML pipeline, with the ultimate goal of implementing a completely automated system for the real-time monitoring of the plasmasphere dynamics from ground-based magnetometer stations.