Collins Conference Room
  US Mountain Time
Lauren Orr (University of Warwick)

This event is private.

Abstract.  We present the first dynamic directed network for space weather events. Although networks are commonly used in societal data there are still many challenges to overcome when applying the knowledge to physical systems. One such system is the Earth’s magnetosphere, a dissipating, non-equilibrium system which responds non-linearly to the solar wind driver and therefore is difficult to predict.  Space weather is the term used to describe these time varying conditions which can damage satellite systems, cause widespread blackouts, aviation disruption and impact navigation.  A central aspect of the earth’s magnetospheric response to space weather events are large scale and rapid changes in ionospheric current patterns. Space weather is highly dynamic and there are still many controversies about how the current system evolves in time. The SuperMAG initiative collates ground-based vector magnetic field time series from over 100 magnetometers, which are non-uniformly spread throughout the auroral regions. We can apply network theory to this rich dataset of observations in order to characterize the time dynamics of the full spatiotemporal pattern of the ionospheric current system. A pair of magnetometer stations are connected in our space weather network if their canonical cross correlation exceeds a station and event specific threshold, as correlated disturbances across the magnetometers capture transient currents.  Further, by calculating the cross correlation for lags of 1-15 minutes, and taking the maximum we produce a directed network that can determine the timings and direction of information propagation. This raises the open question of how the directed network relates to causality. Once the dynamic network has been established we can represent the full spatiotemporal pattern by one or two parameters and hence compare hundreds of events. We present new methodology which addresses the complications with handling real world data and in particular performs a dynamic normalization of the physical time series in order to form the network- ideas which are not specific to this system. For network science this provides new ideas in quantifying the network properties of physical data and the challenges faced in doing so.

Dods et al, J. Geophys. Res 120, doi:10.1002/2015JA02 (2015).
Dods et al, J. Geophys. Res. 122, doi:10.1002/2016JA02 (2017).

Research Collaboration
SFI Host: 
Sandra Chapman