Solar Radio-Burst Forecast Based on a Convolutional Neural Network

Abstract

A solar radio burst is the enhancement of radio emission during the release of solar magnetic energy. It is an important indicator of the level of solar activity. In this paper, we propose a solar radio-burst forecast model that takes the Solar and Heliospheric Observatory (SOHO)/Michelson Doppler Imager (MDI) full-disk solar magnetograms as inputs. The model takes advantage of a Convolutional Neural Network (CNN) to automatically extract the effective feature information from the input images. Through multiple trainings, the relationship between the magnetic-field characteristics of the full-disk solar magnetograms and the solar radio bursts is established, so it is possible to predict the presence or absence of a solar radio burst on that day. The experimental results demonstrate that the forecast Accuracy of the proposed model is 0.875 ± 0.007. The True Skill Statistic (TSS) is 0.723 ± 0.026, and the Heidke Skill Score (HSS) is 0.713 ± 0.019. These results indicate the strong reliability and wide applicability of the forecast model proposed in this paper. The proposed model is also used to predict solar type-II and type-III bursts, respectively. It is found that the prediction performance for type-III bursts is better than that of type-II bursts. The result is well explained from the differences of their formation mechanisms.