Continuous 3D model of ionospheric electron density based on machine learning

Earth's ionosphere represents a complex and dynamic region characterized by increased concentration of charged particles. Changes in ionospheric density can affect the propagation of electromagnetic signals thus disrupting navigation and positioning. The existing ionospheric models rarely meet the accuracy requirements due to either being only climatological, or using spatial and temporal averaging. Here we present a newly developed global continuous three dimensional (3D) density model at heights 100-1500 km. Previous studies have indicated that empirical ionospheric models often suffer from the limited data coverage, and therefore it is essential to use all of the collected observations. We use radio occultation data from various spacecraft (e.g., COSMIC), together with in-situ data by CHAMP, GRACE, ROCSAT mission and ionosonde measurements. We add data from the topside sounders aboard Alouette and ISIS satellites to account for the topside variability. The entire database thus spans from 1960s up to 2019. We discuss the application of deep learning to efficiently handle this data set, comprising billions of data points. The resulting model gives accurate predictions of electron density in the Earth's ionosphere and yields >90 percent correlation on the validation data. The model has a wide range of applications for scientific purposes, space weather monitoring and industrial applications such as positioning and navigation. Furthermore, several output parameters can be directly incorporated into the International Reference Ionosphere (IRI) model.