Data science techniques: the challenges of space physics
I intend to understand and explore the nature of space physics data and the novel adaptations this will require for the artificial intelligence and machine learning techniques used in data analysis and space weather forecasting. By incorporating geometry, topology and uncertainty will we be able to solve the combination of challenges inherent to space physics data (eg sparse and biased sampling; wide scales of interest; extreme events; varied sources of uncertainty; strong interdependence) Where these properties challenge fundamental assumptions in typical techniques, developments to overcome these challenges can feed back into applied to other fields of data science. I am investigating the use of manifold regression techniques, distance metrics and graphs to better represent the characteristics of each dataset. My focus in mainly on regression techniques but I am also interested in exploring classification methods such as topological data analysis in the space domain.
Earth’s radiation belts
My diverse research interests stem from a PhD predicting magnetospheric plasma waves and their impact on Earth’s radiation belts. In this theme I am still pursuing questions in space weather forecasting and in the theory behind radial diffusion, a process that results in the energisation and transport of radiation belt electrons. Radial diffusion theory used in models today has many approximations that we no longer believe are valid. Finding the limit of validity of our current theories and implementing more accurate depictions of the wave particle interactions would allow us to model these processes better. As current predictions often vary by orders of magnitude, accurate models are necessary to predict the geospace environment we are ever more dependent on.