A kernel-based technique for forecasting geomagnetic activity and prediction of Dst Original Research Article

A novel technique for estimating the non-linear dynamic relationship between VBs and Dst is presented. The method is demonstrated through forecasting on a six month data sample and found to give adequate performance, especially in storm regimes. The non-stationary nature of such processes is also highlighted in these results. RMS errors of 15.44nT (winter-spring) and 16.88nT (spring-summer) are achieved for the for the Volterra Series model in comparison to 11.680nT (winter-spring) and 11.357nT (spring-summer) for the nonlinear auto-regressive model with exogenous input.