Prediction of daily average solar wind velocity from solar magnetic field observations using hybrid intelligent systems

Abstract

A hybrid intelligent system, combining theory driven and data driven models, is used to predict the daily solar wind velocity at 1 AU from solar magnetic field observations. The Potential Field Model (theory driven) is used to calculate the coronal magnetic field up to the source surface placed at 2.5 R ⊙. The Earth's position is projected onto the source surface using a time-lag τ and the corresponding magnetic field strength B s is calculated. The field line is then mapped down to the photosphere and the magnetic field strength B ⊙ is determined. The daily full disc RMS photospheric magnetic field strength B RMS is also determined. An Artificial Neural Network (ANN) (data driven) is trained using B ⊙( t) and B s ( t) as input and the solar wind velocity V( t+ τ) as output. The ANN is trained on data from solar cycle 21 and tested on data from solar cycle 22. The function realised in the ANN corresponds very well to the mapping 1 f s ↦ V , where f s = ( R ⊙ R s ) 2 B ⊙ B s is the flux tube expansion factor. There seems to be no explicit dependence on B s . Also including B RMS ( t) to the input improves the results slightly because now there is some information about the daily magnetic field. Extending the input over a time series t − 2, t − 1, t (days) improves the results further. The predicted velocity is now 3 days ahead ( V( t+3)). In this case the ANN also models the dynamic behaviour of the solar wind. This also means that the inverse mapping problem of structures from 1 AU to the Sun is partially avoided. The final model gives an error of 90 km/s and a correlation of 0.58 on the test data set.