Deep learning for ionospheric electron density with ionograms: A time-series approach to Ionospheric monitoring

Ionograms provide a direct measurement of the ionosphere’s electron density profile and its irregularities. By examining critical frequencies researchers can identify key parameters—such as the F-region critical frequency (foF2), the height of maximum electron density (hmF2), and the presence of Spread-F irregularities—that are vital for understanding signal propagation, space weather effects, and radio-communication reliability. Over the past decades, tools have been developed for the extraction of ionospheric parameters of ionograms: ARTIST-5, CADI, Autoscala, and others. There are approximations in previous works using deep learning for automatic scaling with parameters extraction of great importance as the identification of the E and F2 layer. These tools generally work for relatively quiet days (QD), but not for days with ionospheric irregularities (i.e. Spread-F) where a lot of variability is observed and manual correction is necessary.
In this work, we propose a Deep Learning model that takes a sequence of ionograms as input and returns the corresponding frequency profile. Data were collected from five digisonde stations located at different latitudes, selecting dates randomly for ensure variability in the dataset. An autoencoder was first trained in both unsupervised and supervised modes, optimizing the reconstruction of the ionograms. Afterwards, a fine-tuning stage was performed to obtain the electron density frequency profiles.
The resulting frequency profiles were evaluated and compared with manually scaled profiles and with profiles obtained using ARTIST 5.0, showing significant differences. Ionograms were collected from the Vertical Incidence Pulsed Ionospheric Radar (VIPIR) at Jicamarca, part of the Low-Latitude Ionospheric Sensor Network (LISN), as well as from other digisonde stations available through the DIDBase Web Portal. Our results show that the proposed method can be easily adapted to any ionosonde system.