Prediction of Geomagnetic Field Disturbance across Alaska using Long Short-Term Memory Neural Networks trained from OMNI and superMAG 2000-2015 datasets

Abstract:

The dynamic solar wind interaction with the Earth’s magnetosphere creates a disturbed time varying magnetic field that can cause Geomagnetically Induced Currents (GICs) at the surface. These GICs can impact sensitive electronic devices, and in an extreme case cause harmful damage to the power systems and computer systems that our society rely on. In an effort to lower the risk of damage from GICs, the science community has focused its efforts on the prediction of when and where these events will occur. This study aims to model geomagnetic disturbances in the Earth’s magnetic field from multiple ground magnetometer stations setup across the state of Alaska provided by the superMAG database. These models are driven by solar wind inputs, the direct source of these disturbances, as provided by the NASA OMNI database. Each of the models are constructed using data samples spanning from 2000-2015 and predict the individual spatial components of the magnetic field at these stations, which are essential in the prediction of GICs. The models are LSTM Neural Networks fine tuned for each station for maximum effectiveness across all three magnetic field spatial components. We aim to provide our process at creating these models, as well as the choices made in data presented to the models and the parameters used in the models for better understanding given to the community, as well as long term reproducibility, which is essential in the growth of these techniques.