Sunquake Detection: A Machine Learning Approach

In this talk, I will discuss the process behind a two-year research project that aimed to automate the detection of sunquakes. As the manual detection of sunquakes poses significant challenges, the goal of this research was to streamline and potentially improve the detection process. The project was conducted under a collaboration between members of the Technical University of Cluj-Napoca, Romania, the Astronomical Institute of the Romanian Academy, and the High Altitude Observatory in Boulder, Colorado, US.

Sunquakes are seismic emissions visible on the solar surface, associated with some solar flares. Although discovered in 1998, they have only recently become a more commonly detected phenomenon. To our knowledge, the astrophysical data produced for sunquakes was previously unexplored territory for machine learning, despite the availability of several manual detection guidelines.

The talk will begin with a brief presentation of a dataset constructed from acoustic egression-power maps of solar active regions obtained for Solar Cycles 23 and 24 using the holography method. The presentation will cover various applications of machine-learning representation methods to this dataset for sunquake detection, including contrastive learning, object detection, and recurrent techniques. The importance of domain-specific data augmentation transformations will be emphasized. The talk will also highlight the main challenges of the automated sunquake detection task, namely the very high noise patterns in and outside the active region shadow and the extreme class imbalance. Finally, the talk will discuss new insights resulting from applying the presented approach to previously unprocessed events, such as the ability to detect very weak solar acoustic manifestations.