A Neural network model of Electron density in Earth’s Topside ionosphere (NET)

The ionosphere is an ionized part of the upper atmosphere, where the number of free electrons is large enough to affect the propagation of radio signals, including those of the GNSS systems. The knowledge of electron density values in the ionosphere is crucial for both industrial and scientific applications. Here, we develop a novel empirical model of electron density in the topside ionosphere using the radio occultation profiles collected by the CHAMP, GRACE, and COSMIC missions. We assume a linear decay of scale height with altitude and model four parameters, namely the F2-peak density and height (NmF2 and hmF2) and the slope and intercept of the linear scale height decay (dHs/dh and HO). The resulting model (NET) is based on feedforward neural networks. The model inputs include the the geographic and geomagnetic position, the solar flux and geomagnetic indices. The resulting density reconstructions are validated on more than a hundred million in-situ measurements from CHAMP, CNOFS and Swarm satellites, as well as on the GRACE/KBR data, and the developed NET model is compared to several topside options of the International Reference Ionosphere (IRI) model. The NET model yields highly accurate reconstructions of the topside ionosphere and gives unbiased predictions for different locations, seasons, and solar activity conditions.