Modeling of TEC over the Iberian Peninsula using PCA decomposition, linear regression and neural networks

Anna Morozova - University of Coimbra, CITEUC, Portugal; Tatiana Barlyaeva - University of Coimbra, CITEUC, Portugal; Teresa Barata - University of Coimbra, CITEUC, Portugal

Session: Thermosphere-Ionosphere Interactions and Effects

Abstract

The total electron content (TEC) over the Iberian Peninsula was modeled using the principal component analysis (PCA) decomposition and a reconstruction of the PCA modes using space weather parameters as regressors.  

As the first step, the principal component analysis applied to the TEC data allowed us to extract two main modes. Each mode is characterized by daily TEC variation of a certain type (PC) and its amplitude for each of the studied days (given by the empirical orthogonal functions, EOF).  

Afterwards, multiple regression models (MRM) were constructed to fit EOFs using combinations of the space weather parameters with different lags. As space weather parameters we considered proxies for the solar UV and XR fluxes, number of the solar flares, parameters of the solar wind and the interplanetary magnetic field, and geomagnetic indices. Combining MRMs of EOFs with the corresponding PCs we reconstructed TEC variations as a function of space weather parameters observed in previous days. The possibility to use such reconstructions for the TEC forecasting was also studied. 

Another approach with reconstruction of EOFs using neural networks (NN) and the same set of space weather parameters was tested. The results of the PCA-MRM and PCA-NN reconstructions of TEC are compared.