Forecasting auroral electrojet activity from solar wind input with neural networks

Neural networks are developed for reconstructing the chaotic attractor in t he nonlinear dynamics of the solar wind driven, coupled magnetosphere-ionos phere (MI) system. Two new methods which improve predictive ability are con sidered: a gating method which accounts for different levels of activity an d a preconditioning algorithm which allows the network to ignore very;short time fluctuations during training. The two networks are constructed using the Bargatze et al. [1985] substorm database that contains solar wind speed and interplanetary magnetic field (IMF) along with ionospheric electrojet index, AL. Both networks are found to produce improvements in predictabilit y, and the significance of the performance increase of the gated network is demonstrated using the bootstrap model testing method.