Machine Learning for Space Physics

Lead Research Organisation: University of Southampton
Department Name: School of Physics and Astronomy


Machine learning is a very hot topic in computer science these days. As a world we are generating ever greater volumes of
data, and we need to find effective ways to gather and analyse that data, often by searching for regular patterns in data
sets. The human eye is very good at picking out patterns either from images or from simple time series graphs. However,
the human eye comes with its own biases: if you are trying to pick out blips in a single line trace on a screen your selection
may not always be the same, but may depend on what has come before. Reproducibility is a huge issue here and one
which impacts any kind of data science: if we are to do an experiment, or pick out interesting features from data, we want to
make sure we get the same result every time given the same initial input. Furthermore, as our input data streams get
bigger and bigger, it is extremely time consuming (and a bit boring!) to look through all the data by eye to pick out the kind
of features that we want. This is where the extremely powerful tool known as machine learning can help. In this work we
propose to use machine learning to pick out particular signatures from large catagloues of Space Physics data - but the
computer analysis methods that we will develop will be applicable across multiple disciplines.

The Space Physics problem we are interested in is called magnetic reconnection: it is a very energetic process which can
take place when two oppositely directed magnetic field lines meet, come together, and break. Right before reconnection
happens the field lines are holding lots of energy, but as soon as they break this energy can be released into multiple forms
including kinetic energy and thermal energy (heating). The field lines change shape after they break and these newly shaped
field lines can "ping" away from the site of reconnection, much like an elastic band that has been snapped. The
field lines also carry with them charged particles, and these particles can heat up or change their flow direction as a result
of the transfer of energy.

In the solar system everything happens on a giant scale, and magnetic reconnection can involve the magnetic field lines
and plasma of the Sun and of several magnetised planets, including, but not limited to Mercury, Earth, Jupiter and Saturn.
Spacecraft flying through the solar system have instruments which can measure magnetic fields and plasmas, and thus
can sample any changes associated with reconnection.

The changes in the shape and orientation of magnetic fields and in the temperature and flow characteristics of charged
particles can be observed by spacecraft. When scientists examine spacecraft data to search for evidence of this
reconnection process, they know what they are looking for in the field and plasma data. There is a huge amount of
spacecraft data: years and years' worth, with measurements taken several times a second. Reconnection can happen
every few minutes at some planets. It would be impossible for a human being to look through all the data and pick out
every time reconnection happened in our enormous catalogue.

The purpose of this research is to teach the computer what reconnection signatures look like to a human eye, and to train
the computer to pick these signatures out itself. This technique is called machine learning, and it has many advantages,
because computers can be taught to work more quickly than humans, to give the same answer every time, and to not show

The ultimate goal at the end of this project is to have trained the computer to select reconnection signatures, and to be able
to roll out this technique on multiple data sets from the solar system. This will be particularly useful for scientists who want
to conduct large studies of the behaviour of magnetic fields and plasma across the solar system, under different conditions
and over multiple years.

Planned Impact

Several different groups will benefit from this research. As mentioned in the Academic Beneficiaries section, the machine
learning algorithms that will be developed will be of use to those interested in studying reconnection and those who wish to
conduct statistical studies of field and plasma fluctuations in planetary magnetospheres and the solar wind. The work may
feed directly into improved knowledge of Space Weather.

Furthermore, the economic and social beneficiaries of this research will primarily include industrial and governmental
groups linked to space technology and space weather. Such groups include:
UK Space - trade association of the UK space industry
The Cabinet office and National Security and Intelligence, who maintain the National Risk Register, of which Space
Weather is a part:

Modern society is increasingly reliant on space-based technologies for their everyday lives and reconnection events which
transfer huge amounts of energy have consequences on modern technological infrastructure in the Space and Energy
sectors. These include damage to satellites, especially from surface charging, and disruptions to satellite communications
and navigation due to ionospheric absorption and scintillation, to electricity supply due to electrical currents induced in the
ground from ionospheric currents, and to oil and mineral prospecting due to geomagnetic field fluctuations. Such so-called
'space weather' hazards are now considered to be sufficiently important to have been included in the latest UK Government
National Risk Register. Providing information directly relevant to predictive space weather modelling efforts is the first step
towards providing advance warning for low-frequency, but high-consequence events such as those identified by the top UK
and US Science Advisors Holdren and Beddington who warn "The potential total cost of an extreme Space Weather event
is estimated as $2 Trillion in year 1 in the U.S. alone, with a 4-10 year recovery period".

The UK Meteorological Office is responsible for providing space weather predictive capability and any large statistical
studies, enabled by the use of machine learning algorithms such as those we propose, will feed into this predictive
capability. More generally, the effects of space weather can be felt in all activities that use space-related assets. For
example, during the October-November 2003 geomagnetic storms the effects of space weather were included in a US
National Weather Service report for the first time.

The research and professional skills that the PDRA will develop during this project will be in
computational programming, processing large datasets, and scientific reasoning, with written skills in the form of reports
and publications. All are applicable to many employment sectors.


10 25 50
Description JHUAPL research collaboration 
Organisation Johns Hopkins University
Department Applied Physics Laboratory (APL)
Country United States 
Sector Charity/Non Profit 
PI Contribution Dr. Tadhg Garton and Dr. Caitriona Jackman have worked alongside staff of the JHUAPL to apply their machine learning methods to Cassini spacecraft observations. Tadhg and Caitriona provided the JHUAPL members with experience and knowledge of underlying plasma and magnetic physical phenomena occuring in planetary magnetosphere. This was achieved through numerous meetings and presentations of work together. Furthermore, Tadhg and Caitriona provided proof reading and editing for a shared two shared publications currently in process.
Collaborator Contribution The collaborators at JHUAPL provided Tadhg and Caitriona with experience, knowledge and recommendations on machine learning to large datasets to prevent typical traps and pitfalls associated. Furthermore, they provided insight into typical metrics of validation of model robustness needed to create a succesful publication. Similarly, these collaborators provided prrof reading and editing of the aforementioned two shared scientific publications.
Impact Machine learning model and associated reconnection catalogue for Cassini's observations at Saturn. Machine learning model and associated boundary crossing catalogue for Cassini's observations at Saturn. Publication (Accepted) : "Machine Learning Applications to Kronian Magnetospheric Reconnection Classification" Publication (in prep): "Machine Learning Classification of Jovian Magnetic Boundary Crossings"
Start Year 2020
Description MSSL collaboration 
Organisation University College London
Department Mullard Space Science Laboratory
Country United Kingdom 
Sector Academic/University 
PI Contribution Dr. Tadhg Garton and Dr. Caitriona Jackman provided MSSL with expertise in machine learning, as well as supplied a collaborative network for machine learning planetary research.
Collaborator Contribution MSSL members supplied Tadhg with a background knowledge of planetary magnetospheric reconnection and an understanding of the state-of-the-art for it's identification and classification. Furthermore, MSSL members provided proofing for scientific publications, and editing.
Impact Machine learning model to classify reconnections in Cassini observations of Saturn and subsequent catalogue of identifications Machine learning model to classify boundary crossings in Cassini observations of Saturn and subsequent catalogue of identifications Publication (Accepted): "Machine Learning Applications to Kronian Magnetospheric Reconnection Classification" Publication (in prep): "Machine Learning Identification of Magnetospheric Boundary Crossings" Publication (in prep): "Kronian Magnetospheric Reconnection Statistics Across Cassini's Lifetime"
Start Year 2019
Title GartontT/SaturnML: Base ML code 
Description ML code used to create a database of reconnection events from Cassini data 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
Impact This software enables rapid and full classification of magnetospheric reconnection events from Cassini observations alongside a full catalogue of events across Cassini's full lifetime. This model now enables cross training for other spacecraft or other planets with in-situ observations of their magnetospheres.