High-latitude ionospheric electrodynamics specification using the machine-learning based ionosphere models

The prediction of high-latitude ionospheric electrodynamics plays a crucial role in the space weather impact analysis and mitigation of space weather hazards. Due to the complex and dynamic coupling between the solar wind and the Magnetosphere-Ionosphere-Thermosphere (M-I-T) system, the high latitude electrodynamics are extremely challenging to predict. The empirical models (e.g., Weimer 2005 and Fuller-Rowell & Evans 1987) consider static ionosphere, thus failing to reproduce the dynamic response of ionosphere to the time-varying solar wind conditions. In recent times, ML models supported by the long-term heliosphere missions' data enabled researchers to understand and model the dynamics of the M-I-T system. The ML models are easy to implement and computationally fast to run after training. This study takes advantage of such advanced machine learning model and introduces a framework of high-latitude ionospheric electrodynamics model that utilizes ML-based models to solve the current continuity equation. The current version of this framework uses a ML model of Field Aligned Currents (FAC) of Kunduri et al. (2020) and the recent empirical conductance model of Robinson et al. (2021). Kunduri et al. (2020) developed a deep learning based FAC model by using the Convolution Neural Networks with 7 Years of the AMPERE FAC data. Their model captured the dynamic FAC response to the 1-hour time history of solar wind conditions and geomagnetic indices. Robinson et al. 2021 derived a statistical relationship between the FAC and the conductance using the Poker Flat Incoherent Scatter Radar data and the AMPERE FAC observations. We use the two recent models as an initial tool of our framework and simulate the ionospheric electrodynamics of a few geomagnetic events. The ionospheric electric potentials predicted by our framework are comparable to the ones from the Weimer2005 model and the SuperDARN observations. This suggests promising future of our ML-based model framework as we include more machine-learned models (e.g., ionospheric conductance model of McGranagan (2021), ionospheric convection model of Bristow et al. 2022).