Investigating Pre-flare Signatures with K-means Clustering

We present the results of a large statistical study of pre-flare activity using spectroscopic data from the IRIS spacecraft, analysed using K-means clustering. Solar flares are large energy releases whose effects are observed throughout the solar atmosphere. They are also heavily correlated with eruptions and CMEs, which extend their influence into the greater heliosphere. Many models exist that attempt to explain the triggering of flares and eruptions, but not all have clear observational signatures related to them. The identification of reliable pre-flare signatures is therefore highly important to not only furthering our understanding the processes that lead to flaring and eruptions, but also to efforts to predict the occurrence of such events.

We compiled a data set of over 100 flare events with pre-flare coverage, comprising a range of GOES classes and eruptivities. The individual spectra from these data were then standardised for direct comparison, and the unsupervised machine learning technique K-means clustering was run upon them. This allowed us to investigate the types of spectra that were observed prior to flaring, and how these spectra are distributed both spatially and temporally. Additionally we investigated how these spectral clusters are related to the eruptivity of the flares that they precede.