Construction of global IGS-3D electron density (Ne) model by deep learning

Abstract

In this study, we construct a global IGS-3D N_e model that generates global 3-D electron density (N_e) from International Global Navigation Satellite Systems (GNSS) Service (IGS) total electron content (TEC) data through deep learning. As a first step towards this, we make a model to generate a vertical electron density profile from a TEC value using Multi-Layer Perceptron (MLP). In this process, we use the vertical electron density profiles and the corresponding TEC values of the IRI-2016 model from 2001 to 2008 for training, 2009 and 2014 for validation, and 2010 to 2013 for a test. The next step is to generate global IGS electron density profiles using the global IGS TECs as input data for the model, which is called the global IGS-3D N_e model. We evaluate the IGS-3D N_e model by comparing the electron density profiles from the incoherent scatter radars (ISRs) at three stations with the IGS-3D N_e model from 2010 to 2013. The evaluation shows that the electron density profiles from the IGS-3D N_e model are closer to the ISR data than those of the IRI model, especially at high latitudes. The IGS-3D N_e model shows that the averaged root mean square error (RMSE) values between IGS and ISR electron density profiles are 0.37 log(m⁻³), 0.22 log(m⁻³), and 0.34 log(m⁻³) for all test datasets at Jicamarca, Millstone Hill, and EISCAT stations, respectively. These results suggest that our method has sufficient potential to enhance the ability to predict global electron density profiles.