Predicting GICs from L1 solar wind data using geophysical methods in combination with machine learning

  • Rachel Louise Bailey (Speaker)
  • Roman Leonhardt (Contributor)
  • Christian Moestl (Contributor)
  • Martin Reiss (Contributor)
  • Andreas Weiss (Contributor)
  • Albert, D. (Contributor)
  • Schachinger, P. (Contributor)
  • Georg Achleitner (Contributor)

Activity: Talk or presentationPoster presentationScience to science


We aim to predict the local maximum geomagnetically induced currents (GIC) from solar wind data using geophysical methods in combination with machine learning methods. A multiyear collection of GIC measurements at multiple locations in Austria is available for model validation. The prediction of GICs from solar wind data is carried out using a deep learning method, specifically a Long-Short-Term-Memory (LSTM) neural network, which will be trained to output the regional geoelectric field at the Earth’s surface. A prediction of this kind allows for lead times of half an hour on average, although we plan to extend this lead time with solar wind forecasts reaching further into the future. From the geoelectric field, the GICs in the Austrian power grid will be calculated using a power transmission network model. Furthermore, measurements of GICs are available for six substations in the Austrian power grid, on which a deep learning model will be trained for each station. The output of the models trained directly on measurements will be compared to the GICs calculated from the geoelectric field predictions to determine differences in performance, and a validation study on both approaches will be carried out.
Period8 Dec 2020
Event titleAGU Fall Meeting 2020
Event typeConference
LocationVirtuellShow on map
Degree of RecognitionInternational

ASJC Scopus subject areas

  • Earth and Planetary Sciences (miscellaneous)

Fields of Expertise

  • Sustainable Systems

Treatment code (Nähere Zuordnung)

  • Basic - Fundamental (Grundlagenforschung)