A Deep Learning Approach for an Online Ionospheric Delay Forecasting Map System

André L. A. Silva, Moisés Santos Freitas, Paulo R. P. Silva, Clodoaldo Faria Jr, João F. G. Monico, Bruno C. Vani, Jonas Sousasantos, Alison O. Moraes

Abstract: This paper discusses an architecture proposal for a real-time forecasting system of Total Electron Content (TEC) maps for ionosphere over low-latitudes. While previous works have focused on TEC map forecasting, this study aims to discuss the feasibility of implementing a practical system, with user-friendly service. The proposed system utilizes deep learning techniques, specifically a Multilayer Perceptron (MLP) neural network, to forecast TEC maps 24-hours in advance for the South American region. To generate accurate forecasts, the system incorporates inputs such as TEC maps from the previous 5 days, capturing the spatiotemporal distribution of TEC over low-latitudes. It also considers geophysical indexes as solar flux and geomagnetic activity during the same period, which provide insights into the underlying space weather conditions affecting TEC configuration. The input TEC maps used in this proof of concept were sourced from the MAGGIA Laboratory in Argentina. The performance of the system is evaluated across different hours of the day and various locations over low-latitudes. A comparison is made between the TEC maps generated by the proposed system and the real maps produced by the MAGGIA Laboratory. Additionally, the forecast maps are applied to Global Positioning System (GPS) single point positioning estimation. The results indicate that the proposed system significantly reduces positioning errors compared to well-established conventional inputs like the Global Ionospheric Maps (GIM). Notably, the average 3D errors for two evaluated locations of Petrolina and Sao Cristovao were 4.66 m and 3.89 m, respectively, demonstrating improvements of 9.4% and 10.5% over the GIM. Therefore, the proposed system demonstrates precise TEC maps for Global Navigation Satellite System (GNSS) users in South America, surpassing conventional data sources like GIM, with featured 24-hour in advance forecasting, thus improving positioning application reliability in low-latitude region.
Published in: Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023)
September 11 - 15, 2023
Hyatt Regency Denver
Denver, Colorado
Pages: 149 - 158
Cite this article: Silva, André L. A., Freitas, Moisés Santos, Silva, Paulo R. P., Jr, Clodoaldo Faria, Monico, João F. G., Vani, Bruno C., Sousasantos, Jonas, Moraes, Alison O., "A Deep Learning Approach for an Online Ionospheric Delay Forecasting Map System," Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023, pp. 149-158. https://doi.org/10.33012/2023.19293
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In