Extreme solar storms can have very destructive effects on communications and the electrical infrastructure, like large power outages. To protect us from harmful effects of magnetic storms, we need to upgrade our space weather forecasting systems. During his PhD research Mandar Chandorkar from CWI and Inria studied how AI and data from space missions can be used to get better forecasts for space activity. Chandorkar defends his thesis ‘Machine Learning in Space Weather: Forecasting, Identification & Uncertainty Quantification’ at Eindhoven University of Technology today, on 14 November. His results can help decision makers to adequately act on extreme space weather.
Using Machine Learning techniques, Chandorkar and his colleagues developed models for space activity that can give forecasts and their uncertainties. For space agencies and satellite operators, it is important to know the chance of an extreme event happening, so that sensitive equipment can be protected.
Chandorkar says: “Earlier examples of such extreme events are the Quebec power grid failure, which left 7 million people without electricity for 9 hours, and an even larger event in 1859, which killed many telegraph operators. These kinds of events are even dangerous if electricity is turned off, since the large magnetic fields can cause electrical currents in conducting materials by themselves”.
He continues: “Nowadays, such events are able to stop GPS and other communications via satellites. The electricity grid can shut off and, for instance, flights might have to change, as the radioactivity in the sky near the poles is very high during those storms. Therefore it is important to improve the forecasts. When I started, the prediction that could be made was for about 1 hour. By combining models based on Gaussian processes with neural networks with models from physics and real data, we can now make predictions - from pictures of the sun - for 3 to 5 days when the harmful radiation and particles will hit the earth, and what the uncertainties are.”
Chandorkar’s research was conducted as a part of the CWI-INRIA International Lab. He was hosted by the Multiscale Dynamics group at CWI and the TAU group in INRIA Paris-Saclay. His supervisors are Prof. P.D. Grünwald (CWI and UL) from CWI's Machine Learning research group and Prof. Ute Ebert (CWI and TU/e) from CWI's Multiscale Dynamics group. Co-supervisor is dr. Enrico Camporeale (CWI and University of Colorado).
More information
- Multiscale Dynamics research group
- Machine Learning research group
- Mandar Chandorkar's personal homepage
Mandar Chandorkar, PhD student at CWI and Inria.
