Improving solar storm modeling with machine learning
Improving solar storm modeling with machine learning
Disciplines
Computer Sciences (30%); Physics, Astronomy (70%)
Keywords
-
Space Weather,
Coronal Mass Ejections,
Machine Learning,
Computer Vision,
Ensemble Modeling
Solar storms, also known as coronal mass ejections (CMEs), are large eruptions of plasma and magnetic field from the Sun`s corona. Statistically, one CME per week hits Earth during solar maximum and can then cause disturbances in the Earth`s magnetic field, known as geomagnetic storms. These geomagnetic storms can affect power grids, satellite operations, and communication systems. In extreme cases, severe geomagnetic storms can damage transformers in power grids, causing widespread power outages. For predicting an arrival of a CME at Earth, it is important to have sufficient observations to be able to model evolution of the storm on its way towards Earth. In real- time such coronagraph observations are only available for observations up to 30 solar radii, which is just a little over thirteen percent of the Sun-Earth distance. However, there are the so-called heliospheric imagers (HI) that observe the whole space between Sun and Earth making it possible to follow a CME from its origin up to its impact. These observations are ideal to model CME kinematics and predict arrival times and speeds at Earth. Unfortunately, such observations are only available in real-time in a low spatial and time resolution. Additionally, they suffer from many data gaps. HI data in sufficient quality is only available some days later making it impossible to use them for real- time predictions. In this project, we aim to combine heliospheric imager observations with machine learning methods to improve HI-based CME arrival prediction. We work on two different tasks. The first task aims on improving HI real-time data. Based on HI data of good and bad quality, machine learning algorithms will discover how they are related. These algorithms should then be able to produce artificial data with an improved quality based on real-time data. With these improved data we will test if our HI-based prediction model is able to forecast CME arrivals with higher accuracy than with low quality real-time data. The second task is the development of an automatic detection and tracking tool based on HI data. These tools are only available for coronagraph observations that often miss Earth- directed CMEs. These two approaches should lead to an improvement of todays prediction accuracy and help reducing the number of false alarms. With regard to ESAs Vigil mission, this project is an important contribution to space weather prediction based on heliospheric imager data.
- GeoSphere Austria (GSA) - 100%
- Christian Möstl, GeoSphere Austria (GSA) , national collaboration partner
- Andreas Windisch, Technische Universität Graz , national collaboration partner
Research Output
- 61 Citations
- 17 Publications
- 1 Policies
- 3 Methods & Materials