Solar eruptions: predicting geomagnetic storms

Astronomical observation missions collect data in such large quantities that they can no longer be analysed manually, but only automatically. The present project used methods of machine learning in solar research to better understand and predict solar eruptions.

  • Portrait / project description (completed research project)

    Dropdown Icon

    The IRIS Mission has been collecting data from various layers of the sun’s atmosphere since 2013. The resulting data set will provide us with a deeper knowledge of the physics of the sun. But first a way must be found to automatically characterise and search the enormous quantities of data. We are developing methods that enable computers to learn to recognise patterns in the IRIS archive and to characterise the temporal progression of the solar eruptions observed. Building on this, we would like to significantly improve our understanding of and ability to predict solar eruptions.

  • Background

    Dropdown Icon

    The regular eruptions that take place on the sun can cause disturbances on Earth – for instance in radio and GPS positioning systems – as well as power outages. To date, scientists have been unable to explain the physical cause of solar eruptions, nor can they reliably predict them. The fact that solar eruptions occur in diverse complex spatial and temporal patterns makes it considerably more difficult to systematically analyse these phenomena.

  • Aim

    Dropdown Icon

    The objective of this project was to gain a better understanding of the physics of the sun and to develop methods for predicting solar eruptions. We used the huge data archive compiled by IRIS (Interface Region Imaging Spectrograph), NASA’s latest solar satellite, and wanted to create machine learning algorithms to evaluate the data for spatial and temporal patterns.

  • Relevance/application

    Dropdown Icon

    Since solar eruptions can cause widespread interference on Earth, a great deal of importance is attached to being able to predict them. This would be helpful for flight planning and for the operation of satellites and power grids and, accordingly, reduce any damage from eruptions. Moreover, the algorithms and image processing methods we developed could be used for the analysis of other data sets in science or industry.

  • Results

    Dropdown Icon

    According to the main objective of the project to elucidate the physics underlying solar flares and to develop capabilities to predict them, we have developed a number of methods to address this objective. The developed methods were applied to the real data produced by the IRIS NASA mission. The main obtained results are reflected in seven published papers and presented at eleven international conferences and workshops.

    The main focus of our study was on the analysis of spatio-temporal patterns observed in the IRIS data. It should be pointed out that the targeted problem is very challenging due to the numerous factors such as dimensionality and volume of data, data multimodality, lack of labels as such, unbalanced representation of different events in the data such as the regions of quiet sun, pre-flare and flare. The quiet sun event represents a dominating majority in the data. To address these challenges, we needed to adapt the existing Machine Learning (ML) approaches that led us to new insights in solar physics.

    Along this study we have addressed the following main research questions:

    1. Identification of typical Mg II Flare spectra using machine learning,
    2. Exploration of mutual information between IRIS spectral lines,
    3. Real-time flare prediction based on distinctions between flaring and non-flaring active regions spectra,
    4. Solar flares detection on IRIS data using DCT-Tensor-Net,
    5. Investigation of solar activity classification based on compressed Mg II spectra and
    6. Information bottleneck classification in extremely distributed system.

    The obtained results also gave us an impetus to extend our original plans to novel prediction models, where several papers should appear in the nearest future.

  • Original title

    Dropdown Icon

    Machine Learning based Analytics for Big Data in Astronomy