Solar proxies pertaining to empirical ionospheric total electron content models

Abstract

Solar proxies and indices exhibiting extreme ultraviolet (EUV) irradiance that affects the ionospheric total electron content (TEC) were examined through training an artificial neural network (ANN). A TEC database was constructed from a dense GPS receiver network over Japan from April 1997 to March 2008, covering an entire 11 year solar activity period. In empirical models of upper atmospheric parameters, such as the International Reference Ionosphere model and the Mass Spectrometer and Incoherent Scatter thermospheric model, the 10.7 cm solar radio flux (F10.7) or the sunspot number (R) is used as a proxy for determining the solar activity. In the present study, ANN training for predicting TEC as a target parameter was done by including new solar proxies/indices in the input space that were based on direct measurements of solar EUV/UV flux, SOHO_SEM26–34 (the integrated 26–34 nm EUV emission), and Mg II cwr (the core‐to‐wing ratio of Mg II 280 nm line), as well as the traditional indices F10.7 and R. Root mean square errors (RMSEs) of TEC were compared after the training was completed using a variety of combinations of solar proxies. When a single proxy was used, SOHO_SEM26–34 yielded the smallest RMSE, or it was the best proxy for modeling ionospheric TEC. Further, general improvements were obtained by combining different types of proxies and short‐ and long‐term means of them. The best combination was the 3 day smoothed daily, 7 day and 27 day backward mean values of Mg II cwr, SOHO_SEM26–34, and the 10.7 cm radio flux.