With society’s increasing reliance on ground- and space-based technology such as electricity networks, communications and GPS, we are ever more exposed to the risk of disruption from space weather events in the form of solar flares, coronal mass ejections and geomagnetic storms. Although there are a number of real-time monitoring systems in space measuring the solar wind and its magnetic field, forecasting the space weather impacts on both ground-based technology and the satellites essential for observing the Earth, remains problematic.
There are many potential sources of information which could be used to predict a space weather event including human-made forecasts (e.g. reports issued by the Space Weather Prediction Centre, and the Met Office Space Weather Operations Centre), magnetic field measurements (from the worldwide network of geomagnetic observatories), solar disc imagery and data from geostationary satellites (e.g. the Solar Dynamic Observatory, and the Solar and Heliospheric Observatory). We seek to combine such information to improve the real-time automation of forecasts of space weather between 1 to 72 hours ahead of time, aiding efforts to reduce the space weather hazard to ground- and space-based technological infrastructure. This will be achieved using advanced machine learning techniques including nonlinear time series forecasting, Bayesian modelling, text mining, and image processing. Models of the Earth’s magnetic field, built using satellite data (e.g. ESA’s Swarm mission) will also be used to better understand the wider context for space weather impacts, for example, the increased risk of electronic damage to satellites in the region of the South Atlantic Anomaly.
Year 1: The first step of this project will involve forecasting geomagnetic activity based on existing geophysical knowledge and machine learning techniques. This will include analysing satellite data to identify shocks in the solar wind (using real-time data from ACE and DSCOVR), that often result in heightened geomagnetic activity. Alongside this, an appropriate system for evaluating the forecasts will be developed, particularly taking into account extreme events which are rare but have the most impact.
Year 2: By including other data sources such as forecast reports, solar disc images and coronagraphs, the aim is to improve the longer-term (1-3 days ahead) forecasts of geomagnetic activity which will also allow identification and investigation of the underlying mechanisms that are most significant for geomagnetic activity. This will require applying advance machine learning methods, such as text mining and image processing to extract features from the data sources, and integrating this information using multi-source learning.
Year 3: The student will build software for real-time automated forecast and alerting systems based on the methods explored and investigate the uncertainty at each stage of the forecasting process, and develop mechanisms for communicating those uncertainties effectively. These products and methodologies will be compared with those from international teams working within the space weather community.
The project will suit an applicant with a background in informatics, computer science or geophysics, physics or mathematics and a keen interest in machine learning.
The student will be based in the University of Edinburgh School of GeoSciences, which has a large portfolio of postgraduate research skills training, and will also spend time with the world leading deep Earth research group in Leeds, and a minimum of 3 months based in British Geological Survey (BGS) offices, where they will have access to training through the BGS Learning and Development programme. Specific training will be in techniques related to space weather forecasting and machine learning and the student will be encouraged to take relevant taught courses. There will also be opportunities to present at conferences in both the UK (e.g. the annual Magnetosphere, Ionosphere and Solar-Terrestrial meeting) and internationally (e.g. European Space Weather Week, and the European Geosciences Union General Assembly).
Camporeale (2019), The Challenge of Machine Learning in SpaceWeather: Nowcasting and Forecasting, Space Weather, 17, doi:10.1029/2018SW002061