Flare Transformer: Solar Flare Prediction using Magnetograms and Sunspot Physical Features

Kanta Kaneda, Yuiga Wada, Tsumugi Iida, Naoto Nishizuka, Yûki Kubo, Komei Sugiura; Proceedings of the Asian Conference on Computer Vision (ACCV), 2022, pp. 1488-1503

The prediction of solar flares is essential for reducing the potential damage to social infrastructures that are vital to society. However, predicting solar flares accurately is a very challenging task. Existing methods predict flares using either physical features or images, but the main bottleneck is that they sometimes incorrectly predict a class that is smaller than the actual solar flare. In this paper, we propose the Flare Transformer, a solar flare prediction model that handles both images and physical features through the Magnetogram Module and the Sunspot Feature Module. The transformer attention mechanism is introduced to model the temporal relationships between input features. We also introduce a new differentiable loss function to balance the two major metrics of the Gandin--Murphy--Gerrity score and Brier skill score. We validate our model on a publicly available dataset. The results show that the Flare Transformer outperformed the baseline methods in terms of the Gandin--Murphy--Gerrity score and true skill statistic, and achieved better performance than those given by human experts.