Space weather modelling with intelligent hybrid systems: Predicting the solar wind velocity

• Published in: Advances in space research 1998-01, Vol.22 (1), p.59-62
• Main Authors: Wintoft, P., Lundstedt, H.
• Format: Article
• Language: English
• ISSN: 0273-1177; 1879-1948
• DOI: 10.1016/S0273-1177(97)01100-9

We are developing a space weather model to predict disturbances of the Earth's magnetosphere/ionosphere on four different time-scales: minutes, hours, 1–3 days, and 27 days. The minutes to hours predictions are made from solar-wind measurements, while predictions days in advance are made from solar observations. In this work we concentrate on predictions of solar wind velocities 3 days ahead. Using daily solar magnetograms we show how current predictions of the solar wind velocity at 1 AU can be improved with an intelligent hybrid system. The intelligent hybrid system consists of the potential field model and an artifical neural network. Solar magnetic field strengths are used as input and the output is the velocity 3 days ahead. The networks are trained and optimised on data from solar cycle 21. The hybrid system is tested on data from cycle 22 and the RMS error of the predicted daily velocity compared to the observed velocity is 90 km/s, the linear correlation is 0.58, and the average relative variance is 0.68.