Completion of Global Ionospheric TEC Maps Using a Deep Learning Approach. Issue 5 (24th May 2022)

• Record Type: Journal Article
• Authors: Yang, Ding; Fang, Hanxian; Liu, Zhendi
• Is Part Of: Journal of geophysical research. Volume 127:Issue 5(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-05-24
• Subjects: Magnetospheric physics -- Periodicals; Space environment -- Periodicals; Cosmic physics -- Periodicals; Planets -- Atmospheres -- Periodicals; Heliosphere (Astrophysics) -- Periodicals; Geophysics -- Periodicals; 523.01
• DOI: 10.1029/2022JA030326
• Languages: English
• ISSNs: 2169-9380

Abstract

Abstract: Total electron content (TEC) is an important parameter that describes the features of the ionosphere. The International GNSS Service (IGS) has been providing IGS Global TEC maps using analysis algorithms. However, collecting the completed data is difficult because of the lack of ground receivers, and the processing to obtain the completed IGS TEC maps is time‐consuming. The fast development of deep learning brings an effective way to solve these problems. Among the various deep learning methods, the generative adversarial network (GAN) exhibits great potential in recovering missing data. In this paper, we fill the missing data of the global IGS TEC maps using pix2pixhd, which is a novel deep learning method based on GAN. Differing from the traditional GAN, pix2pixhd has two generators and three discriminators. The network enhances the ability of our model to complete images with large‐scale missing areas. The result demonstrates that our model generates the ionospheric peak structures at low latitudes well, while behaving badly (the average correlation coefficient: 0.6857) around the edge of the ionospheric peak region. Comparing different scales of the missing data areas, our model has the best performance with 0%–15% missing data. With the large scale of missing data areas (30%–45% and >45%), the performance is still satisfactory. In addition, the completion effect of our model is slightly affected by geomagnetic and solar activity. Our work demonstrates a new possibility for the application of deep learning to a broader field of geosciences, particularly for problems of missing observational data. Key Points: We make a model to complete International GNSS Service total electron content maps using pix2pixhd based on generative adversarial network The completion effect of our model with the large scale of missing data areas is still satisfactory Our model generates the ionospheric peak structures well while it behaves slightly badly around the edge of ionospheric peak region