PhD candidate Advancing Space Weather Predictability

Space weather, similar to weather on Earth, has a significant effect on infrastructure in space and on Earth. For example, solar storms may obstruct radio communication on Earth and space. Severe space weather events may even lead to failures of the electric grid, causing significant economic damage and posing a threat to societal stability. As weather forecasts on Earth, space weather forecasts may help to mitigate and prepare for such space weather events. Recently, the use of machine learning for improved space weather forecasting has gained in importance.

The Luxembourg-based company Mission Space and SnT, University of Luxembourg are joining forces to develop a space weather center of excellence. The first building block is to improve the accuracy of space weather forecasting using a combination of existing space weather models and data collected in space. For the latter, Mission Space will soon launch a dedicated satellite constellation for acquiring space weather data in space.

You will perform research on applying machine learning to space weather forecasting. This includes:

• Geomagnetic indices analysis for time series behavior extrapolation into the near future (flux, indicators, indices). The objective is to discover patterns in the time series that are correlated with solar particle events in the magnetosphere. Then, we would use these patterns as a learning base to estimate the probability of occurrence of solar particle events connected to these patterns. Thus, it is desirable that the to-be-developed machine learning models are interpretable, or at least that their prediction can be justified through explainable AI techniques. The end goal is to enhance event criteria, which are currently based on threshold values (NOAA). Open-source data are available to support the research. To facilitate the start-up of the research project, Mission Space can also provide terminology table and “fresh publication” that the student can consult to increase his knowledge of the field
• Use of secondary indices: new combinations, based more on LEO/ground measurements than on solar wind parameters. The use and the automated forecasting of secondary indices could improve the time series and lead to better predictions