AbstractPlasmaspheric hiss waves make great significance on the loss of electrons in the Earth's radiation belts. The prediction and reconstruction for global evolution plasmaspheric hiss are critical to investigate the dynamic process of radiation belt. In this study, the realistic variation model of plasmaspheric waves is established based on deep learning technology. Using Light Gradient Boosting Machine and Permutation Importance methods, we first select optimal indices and corresponding historical time series associated with variation of plasmaspheric hiss waves. Then the geomagnetic indices in corresponding time series are considered as input of deep neural network model, and the samples of plasmaspheric hiss waves observed from Van Allen Probe A are utilized for training and validation. The result indicates that the model can roughly reproduce the variation of integrated hiss wave amplitude during geomagnetic storm. Based on the model, we analyze the global evolution of hiss waves during a geomagnetic storm event. In addition, we find that the prediction performance has been greatly improved with the addition of in situ density as an auxiliary input parameter, it provides a new idea of establishing higher accuracy model with auxiliary input of other characteristic parameters.
Read full abstract