Home

Interconnection and valorisation of long-term solar datasets via deep learning

Context

Sunspots are dark spots appearing in groups on the surface of the Sun as a manifestation of solar magnetism. The magnetic field embedded in sunspots is the driving force behind the solar variability that influences the Earth space environment on a day-to-day basis. Studying sunspots evolution on a long-term basis is a keystone to several areas of Solar Physics, from helioseismology to irradiance modelling and the prediction of space weather.

The Royal Observatory of Belgium (ROB) is a key player in sunspot observations: In 1939, the ROB Uccle Solar Equatorial Table (USET) station started up a solar observing program in collaboration with the Zürich Observatory consisting of daily drawings of the sunspot configuration. As of July 2019, this collection counted 23000 sunspot drawings and is still expanded every day. In this project, we will use these sunspot drawings, along with two other datasets produced by the USET facility: White light images, taken since 2002, and CaIIK images, taken since 2012. The co-temporal and co-spatial acquisition of drawings, white light, and CaIIK images makes it favourable to interconnect these datasets using novel image processing techniques.

Objectives

The goal of DeepSun is to produce high-level data products from the various USET datasets, with a view on:

• Going back in the past by exploiting the connection between drawings and WL pairs of images on one hand, and white light and CaIIK sequences of images on the other hand. The goal is to reconstruct important information when no direct data is available (magnetic information, solar irradiance).
• Advertising the resulting USET data products to the scientific community and the general public.

Methods

In order to use the wealth of today's information to improve our knowledge on past solar activity, we will leverage on advancement happening in signal processing methods, and in particular on image-to-image translation method based on convolutional neural networks (CNN). This will require a careful preparation of the data, with in particular the precise delimitation of sunspot group boundaries and the tracking of these groups over time. CNN will also be used to devise an automated classification of sunspot groups based on photospheric information.

Importance of the Deepsun project

From the drawings, we will generate white light images, which constitute the benchmark for photospheric observations. We will also attempt to reconstruct chromospheric information from photospheric and sunspot group dynamics information. Sunspot group classification is a key element in space weather forecasting, and hence our automated classification will be used by the space weather forecaster team at ROB.

This project has a research and a valorization component. In the research part deep learning methods will be customized to our particular application, and high level science products will be produced. In the dissemination component, a valorization of the dataset by a citizen-science initiative will help assessing the performance of the sunspot group classification task. The high level products will also be made accessible via virtual observatory and standard access protocol, with as primary target groups: colleagues from the scientific community and space weather operators.

Developing new national collaborations

This project will benefit from the collaboration between the "Solar physics and Space Weather" Operational Directorate at ROB and researchers from the Institute of Information and Communication Technologies, Electronics and Applied Mathematics (ICTEAM), Université Catholique de Louvain (UCL).

Funding

This project receives funding from the Belgian Federal Science Policy Office (BELSPO) through the BRAIN2 framework, from the Royal Observatory of Belgium, and from the Université catholique de Louvain.