Technical Article

Assessment of models for the prediction of the Travelling Ionospheric Disturbance activity index in mid-latitude Europe

¹ Institute for Solar-Terrestrial Physics, German Aerospace Center, 17235 Neustrelitz, Germany
² Department of Electrical Engineering, University of Brasília, 70910-900 Brasília, Brazil

^* Corresponding author: arthur.ferreira@dlr.de

Received: 11 April 2024
Accepted: 10 April 2025

Abstract

Given the impact the ionosphere electron density has on radio wave propagation, understanding, characterizing and predicting its behaviour and associated perturbations is of high importance. One type of perturbation commonly observed during geomagnetic storm events is the Large Scale Travelling Ionospheric Disturbance (LSTID). LSTIDs correspond to the ionospheric signature of large-scale atmospheric gravity waves that propagate in the thermosphere. Such waves, which are typically generated due to the input of energy from the solar wind into the Magnetosphere-Ionosphere-Thermosphere (MIT) system, are an essential component contributing to the development of ionospheric storms. Recently, the A_TID index, which has been introduced for statistical analyses of TIDs, has been shown to correlate well with solar wind energy input in Europe in mid-latitude regions. The feasibility of predicting LSTIDs in this region has been demonstrated using a linear regression model. Here, an assessment of more advanced modelling approaches is presented to demonstrate their applicability and improvement of the predictions. This work applies methodologies based on artificial neural networks and multi-model ensembles. The persistence model is taken as a baseline for the performance assessment of the different methodologies. A given challenge for the generation of LSTID prediction models is the limited number of observations available. Still, the results show that all proposed methodologies outperform the baseline model when predicting the level of LSTID activity during geomagnetic storms over mid-latitude Europe for predictions beyond 1 h. The linear regression model shows in most cases the best performance among the investigated methodologies, evidencing that more complex techniques could not educe their capabilities in the application of LSTID prediction. For prediction times of 30 min, however, the ensemble of the linear regression and the persistence models presented the best performance overall. The presented assessment of LSTID prediction models/approaches contributes to the development of strategies for predicting LSTID activities over the European region and, it enhances the understanding of the strengths and limitations of different modelling methodologies for this use case.