Using a neural network-based TEC model to reproduce the small-scale Nighttime Winter Anomaly feature

Abstract

In this work a Neural Network (NN) based TEC model is proposed which is trained on vertical Total Electron Content (TEC) data from Global Ionospheric Maps (GIMs). The GIMs comprise a period of almost 20 years (2001-2020). The model was tested with data from the High Solar Activity (HSA) year 2015 and the Low Solar Activity (LSA) year 2020. The test data was excluded from the training dataset. The model performance was compared with the Neustrelitz TEC model (NTCM) and an improvement of approximately 1 TEC Unit was found during both solar activity periods for the NN based TEC model. Not only the performance of the NN based TEC model was analyzed but also its capability in predicting TEC containing large- and small-scale features of the ionosphere, e.g. the seasonal-, solar activity-, diurnal variations, equatorial anomalies and the Nighttime Winter Anomaly (NWA) was evaluated. Other NN-based networks are also capable of showing large-scale features in their predictions but the capability of reproducing the small-scale NWA feature is new. The NWA is only visible at certain locations in the Northern Hemisphere at the American sector and in the Southern Hemisphere at the Asian longitude sector under LSA conditions. This feature is caused by a higher mean ionization level during the winter nights compared to the summer nights.