Prediction of magnetic substorms using a state space model

Nonlinear dynamical models of the magnetosphere derived from observational time series data using phase space reconstruction techniques have yielded new advances in the understanding of its dynamics. Considering the solar wind–magnetosphere interaction to be a natural input–output system its dynamical features can be reconstructed on the storm time scale by using the method of time delay embedding. Here, fourteen magnetic storm intervals belonging to low/moderate and high solar activity periods are considered and a suitable state space model has designed by performing training and validation tests, for which dawn to dusk electric field (VBz) is chosen as the input, and the AL time series as the output. The percentage of the output variations that is reproduced by the model is termed as fit_model and a higher number of fit_model means a better model. The number of components m used in the state space model is varied from 1–9 and the best prediction is obtained when m=4. The fit_model values of time series used for validation are 67.96, 67.2, 72.44, and 70.89, with m=4. In the present study most of the storms considered are having Dstmax in between −100 and −300 nT, and they can be predicted well with this procedure. To reveal the prediction capability of the proposed state space model the 30 steps ahead outputs for the storm events are generated, which reasonably reproduce the observed values.

► Solar wind–magnetosphere interaction is considered as a natural input–output system.
► A state space model was designed by choosing VBz as input and AL time series as the output.
► State space model exhibits best performance when the number of components m=4.
► Storms with Dst max in between −100 nT and −300 nT, can be predicted well with this procedure.