A Lightweight Solar Activity Prediction Model compatible with CubeSat

The prediction of Solar Activity is critical in the context of space weather and associated technological implications. Many methods have been proposed towards this and for identifying precursors of solar activity. Primarily, only whole 2D magnetograms have been used for this purpose. While physics-based models are important, we show that using light-curves reduces the dimensionality of the problem and provide interesting and useful results. Using data from Chandrayaan-2’s XSM from the ISSDC archive and GOES from NCEI, prediction of flares, background and event timing accurate to 99% has been demonstrated.
The data-driven nature of the model allows it to be applied on other wavelengths such as EUV. We aim to extend the testing and applications to a wider range of data from other missions like PROBA-3 and Aditya-L1 and from others stellar sources. Such input from Kepler, TESS and in preparation for PLATO could contribute towards habitability studies as well. Magnetogram (3D) inputs will be integrated to create a comprehensive model, and comparison of these with ground-based white light and H-alpha images conducted.
We also intend concept development of a CubeSat bus (3 or 6 U) for imaging and monitoring. Upon comparison of the main model (LSTM) with a lighter-weight and more resource-efficient network (GRU), we find the GRU provides comparable predictions to the LSTM and thus would be more amenable with CubeSat applications and parts of onboard processing where resource usage becomes critical.