High-Resolution Total Electron Content (TEC) Maps Using Deep Learning Based Image Inpainting

The total electron content (TEC), as the number of vertical electrons in the ionosphere, serves an important role in the study of ionospheric environment. Though there are over 6,000 GNSS receivers worldwide that provide the TEC data with fine temporal cadence, the limited coverage of GNSS causes approximately 52% missing data on the global TEC maps. We treat the recovery of the missing TEC values as an image inpainting task, for which various deep learning algorithms have been developed. In this work, we adapt the spectral normalized patch generative adversarial network (SNP-GAN) for TEC map inpainting. The international GNSS Services (IGS)-TEC maps are used to train SNP-GAN for inpainting of high-resolution MIT-TEC maps. The results show better qualities and the potential of identifying meso-scale structures in completed TEC maps. We plan to build an online repository of high-resolution TEC maps in global view and north-polar view, as an important space physics product to cross validate different simulation results, or check TEC structures at a particular time.