CE23 - Intelligence Artificielle

PREdicting Solar Activity using machine learning on heteroGEneous data – PRESAGE


PREdicting Solar Activity using machine learning on heteroGEneous data


Our project concerns itself with the activity of the Sun, those events (e.g. flares, coronal mass ejections (CME)) are dynamical phenomena that may have strong impacts on the solar-terrestrial environment. Events of solar activity seem to be strongly associated with the evolution of solar structures (e.g. active regions, filaments), which are objects of the solar atmosphere that differ from the “quiet Sun” and which appear, evolve, and disappear over a period of days to months. The exact mechanisms of solar activity, and the links between solar structures and activity events, are still ill-understood.<br /><br />Our project has three objectives related to solar physics, namely: <br /><br />1) To improve our understanding of the mechanisms of solar activity <br /><br />2) To enable the prediction of solar activity events such as flares, CMEs, radiation emission levels, and the fluxes of ionized particles likely to reach the Earth environment <br /><br />3) To investigate the existence of typical temporal behaviours for 2D and 3D solar structures, i.e. filaments, active regions (AR), sunspots, and coronal holes <br /><br />These objectives will be supported by two central objectives in machine learning (ML): <br /><br />4) To develop new ML methods that can exploit the heterogeneous data generated by the many solar physics and space weather observation missions <br /><br />5) To develop new ML algorithms that are guided by prior physics knowledge to increase their robustness and interpretability, and to reduce their need for large training datasets

The work is structured into three scientific Work Packages (WP) that: 1) detect 3D solar structures, 2) model their behaviours, and 3) combine these new indicators with historical ones to predict solar activity. A multimodality approach and the integration of domain knowledge will support the design of new ML methods throughout these WPs.

Work package #1: 3D detection of solar structures
Goals: Localising solar structures and recovering their 3D shapes from visual observations of the Sun.

Work package #2: Behavioural study of solar structures
Goals: Determining the existence of typical behaviours and modelling them, through a study of the lifetime evolution of solar structures.

Work package #3: Prediction of solar activity events
Goals: To deepen our understanding of solar activity, and to predict its events, by combining traditional activity indicators and new indicators of WPs 1&2. The combination of indicators will need new ML methods to handle heterogeneous data. The activity predictor will integrate prior knowledge from WPs 1&2 for greater robustness and interpretability.

Our five objectives will be achieved through the production of the following results and outputs:

#1: An identification of the correlations between the various indicators of solar activity considered in the study, and of their relevance to predicting events of solar activity.

#2: An ML method for predicting events of solar activity, including flares, CMEs, radiation emission levels, and the fluxes of ionized particles likely to reach the Earth environment.

#3: New high-level indicators of solar activity:
- Information on the existence of behavioural patterns for solar structures (filaments, ARs, sunspots, and coronal holes), both in isolation and in interaction, and their modelling if they do exist.
- Information on the 3D geometry of solar structures (filaments, ARs, and coronal holes).

#4: Multimodal ML methods for:
- studying the correlations between multiple types of solar activity indicators
- combining multiple observations of a same solar structure at different layers of the solar atmosphere
- exploiting jointly different indicators from multiple instruments to predict solar activity

#5: New ML methods that integrate prior physics knowledge for increased robustness, interpretability, and easier training from fewer examples.

The research will create new knowledge in CV, ML, and DL, by tackling the new challenges offered by heliophysics data. It will also advance the young field of DL, whose large potential impact on technology and on our lives is currently limited by its black-box aspect that reduces usability for sensitive applications. Addressing this limitation is a strong focus of the project.

The project also aims to create new knowledge on the mechanisms of solar activity through enabling the exploitation of big observational data, hence contributing to improving our understanding of the Sun and stars in general.

The end users of the project are researchers on solar physics and space weather. New methods and tools will allow them to analyse the increasingly large and heterogeneous collection of data from new observation missions, and they will transform their current practices. The project will also contribute new knowledge and tools to advancing the astrophysics and space science disciplines in general, by addressing multimodal image analysis challenges that are common to many observation scenarios. It will support their academics to deliver high quality research outputs based on big observational data.

These impacts will be realised through a wide dissemination of outputs to the AI, heliophysics, and astrophysics communities. Image analysis tools will be released publicly as an open-source toolkit on GitHub. New image analysis methods will be published in solar physics / astrophysics and AI journals. The toolkit and methods will be presented at workshops, seminars (e.g. at the partner institute), and conferences.

The goal of our multidisciplinary project is to develop new machine learning methods that support a deeper understanding of the mechanisms of solar activity in order to predict its events. The solar physics community is currently facing a deluge of data, which is too widely varied and complex to allow an overall analysis leading to a global understanding of solar activity. We propose to solve this problem by developing new machine learning algorithms that exploit these heterogeneous data, to: 1) study the properties in 3D of objects of the solar atmosphere (filaments, sunspots…), 2) model their evolutions and behaviors, 3) study the correlations between many indicators of solar activity (inc. solar objects and their behaviors), solar activity events (flares, CMEs...), and their resulting terrestrial impacts (geomagnetic indices...), and 4) use these new insights to predict the events of solar activity and their effects on Earth.

Project coordinator

Madame Adeline Paiement (Laboratoire d'Informatique et Systèmes)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.


LIS Laboratoire d'Informatique et Systèmes

Help of the ANR 226,119 euros
Beginning and duration of the scientific project: August 2021 - 36 Months

Useful links

Explorez notre base de projets financés



ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter