A ConvLSTM‐Based Prediction Model of Aurora Evolution During the Substorm Expansion Phase

Abstract

Abstract Aurora is an important manifestation of solar‐terrestrial physical processes. The aurora activities have rapid changes in spatial and intensity distribution during a substorm, especially the expansion phase. In this paper, a newly developed aurora evolution model is built based on the Convolutional Long Short‐Term Memory network, using the aurora images captured by the ultraviolet imager on the Polar satellite during the substorm expansion phases. Given the images after the onset, the model can predict the evolution of aurora with reasonable accuracy. The structure similarity, Peak Signal‐to‐Noise Ratio (Peak Signal to Noise Ratio), and Root Mean Square Error are used to evaluate the similarity between the predicted and observed images. The results demonstrate that the model performs well at different scales of evolution. Additionally, the model can estimate the evolution of both the aurora intensities and the movement of the poleward boundary. Furthermore, three cases are displayed to illustrate the behavior of the model and its limitation.

Keywords

• aurora
• prediction
• substorm
• ConvLSTM
• deep learning