Adaptive Systems for Detection and Forecasting of Coronal Mass Ejections From Solar Mean Magnetic Field

Coronal mass ejections (CME) are important sources of dynamical phenomena that collectively influence geo-space weather. For example, large, nonrecurrent geomagnetic storms are caused by interplanetary disturbances driven by fast CMEs. Therefore efficient techniques for detection and forecasting of CME events can significantly increase the performance of any realistic space weather forecasting system. Recently it has been shown that wavelet analysis of the high-resolution solar mean magnetic field (SMMF) data can provide valuable information for CME detection. However not all CME events produce easy detectable signatures in the wavelet transformed SMMF time series. Moreover CME forecasting would require extracting information from the data prior to CME onset where these signatures are even less obvious. Therefore to achieve acceptable accuracy in CME detection/forecasting, multiple features from the wavelet spectrum or raw SMMF data should be processed with a powerful classifier based on statistical or machine learning techniques. We applied neural network and support vector machine for this purpose. Performance of the obtained systems will be discussed.