Solar wind magnetosphere coupling: Predicted and modeled with intelligent hybrid systems

Abstract

The solar wind-magnetosphere coupling is described as a mapping of the solar-solar wind state (input) to the magnetosphere-ionosphere state (output). The input-state is defined by solar wind parameters and the output-state is defined by geomagnetic disturbance indices. Most often are the geomagnetic disturbance indices D st and AE used. However, a new family of geomagnetic disturbance indices, CO, SO and EO, have been constructed which even might be able to describe the cusp and dayside magnetosphere. As mapper, a multi-layer backpropagation network, an Elman neural network and a radial basis function neural network, were used. Predictions of solar wind parameters and geomagnetic indices D st and AE are presented. The mapper represents a coupling function. It was found that the coupling function, learned by the neural networks, was more accurate than predefined theoretical coupling functions. In the case of modeling, it is therefore very important to explain the mapper and the trained neural network. Several methods are available: By extracting rules and decision trees from neural networks. By studying the prediction accuracy for different combinations of solar wind input parameters and for different coupling functions. By comparison with well-known mathematical methods. Integrating both data-driven and theory-driven methods into intelligent hybrid systems further improves the scientific method of studying the solar wind-magnetosphere coupling. Lund Space Weather Model is such a hybrid system. IHSs are also capable of mapping simultaneous multi-satellite observations. Such a mapping is therefore also suggested to improve the understanding of the solar wind-magnetosphere coupling.