The application of non-linear filtering methods to the forecast of geomagnetic indices

A new method to forecast the evolution of terrestrial magnetospheric dynamics and geomagnetic indices under the influence of the solar wind is presented. It is based on the derivation of an adaptive Volterra Series model of the dynamics from the analysis of simultaneous measurements of the solar wind (VBs) and the D-st index. The method uses a novel technique to compute, implicitly, the Volterra kernels, avoiding the numerical and computational difficulties associated with their direct estimation. An adaptive approach is used to combat non-stationarity inherent in the data. The resulting model may be used for the medium-term, on-line forecast of D-st index under the assumption that measurements of solar wind parameters are available further in advance than is now practicable, the prediction horizon being determined by the lead-time of the solar wind measurements. The data are drawn from the six month period January-June 1979 and an RMS forecasting error of similar to13nT is achieved for an eight hour ahead forecast. The significant events are considered in detail.