A storm-time global electron density reconstruction model in three-dimensions based on artificial neural networks

Abstract

We present results of a dedicated global storm-time model for the reconstruction of ionospheric electron density in three-dimensions. Using the storm criterion of |Dst| ≥ 50 nT and Kp ≥ 4, the model is constructed using a combination of radio occultation and ionosonde data during the periods of 2006–2021 and 2000–2020, respectively. From the ionosonde data, only the bottomside electron density profiles up to the maximum height of the F2 layer (hmF2) are considered. In addition to the selection of storm-time data only for the model development, we have investigated the inclusion of time history for the geomagnetic storm indicator Kp at 9 and 33 h in an attempt to take into account the delay of physical processes related to atmospheric gravity waves or traveling ionospheric disturbances and thermospheric composition changes which drive varying ionospheric storm effects during storm conditions. Based on incoherent scatter radar data and in comparison with the IRI 2020 model, the developed storm-time model provides foF2 modelling improvement of above 50% during the storm main phase over Millstone Hill (42.6°N, 71.5°W) and Tromsø (69.6°N, 19.2°E) for the storm periods of 3–6 November 2021 and 23–25 March 2023, respectively. Modelled results for Jicamarca (11.8°S, 77.2°W) show that the storm-time model estimates foF2 by an improvement of over 20% during the main phase of the 07–10 September 2017 storm period. As the ionospheric conditions return to quiet time levels, the IRI 2020 model perform better than the constructed storm -time model.