Autoprediction of Dst index using neural network techniques and relationship to the auroral geomagnetic indices

Abstract

The possibility of prediction of Dst variations using previous Dst values has been studied using a feedforward multi-layer perceptron. It was found that the Dst index can be autopredicted a few hours ahead. Both main and recovery phases of geomag- netic storms are accurately predicted up to 3 hours in advance. But, for more advanced predictions, a time shift between observed and predicted Dst minima is observed. The use of auroral electrojet indices as input has shown that there exists a slight relationship between these indices and Dst variation at least one hour ahead. Weak and moderate geomagnetic storms are predicted well, but the predicted Dst values for more intense storms are less negative than the observed minima, this may be related to the known saturation of auroral electrojet indices due to intense storm development. A prediction based on the PC index shows better corre- lation with Dst. Although the amplitude of Dst variation is not reproduced correctly, there is no time shift between measured and predicted location of Dst minima.