A global neural network model of 3D ionospheric electron density based on COSMIC radio occultation measurements

Abstract

Developing more accurate ionospheric empirical models, especially for representing the 3-dementional distribution of electron density (Ne), is important for both ionospheric studies and radio wave propagation applications. In this study, an accurate Ne profile model, named Deep Learning Ionosphere Model (DLIM) was built by utilizing over 3 million Ne profiles derived from radio occultation measured by the Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC). We used the α-Chapman function with five parameters to describe each profile, including the peak density of the F2 layer (NmF2), the corresponding altitude of the peak (hmF2), scale height at F2 peak height (H₀), the scale height gradients α₁ (above and equals hmF2) and α₂ (below hmF2). We then built a sub-model based on deep convolutional neural network for each parameter, incorporating latitude, longitude, day-of-year (DoY), local time, solar activity indices, and geomagnetic indices as inputs. By combining these five sub-models, we constructed this DLIM for predicting the 3-D distributions of electron density. The model shows excellent agreement with observations from COSMIC-1 as well as the independent dataset from COSMIC-2. It also performances well in capturing the ionospheric climatology, such as solar activity dependence, local time variation, and global pattern, e.g., the equatorial ionization anomaly (EIA) and longitudinal four-wave patterns.