Filament Chirality Detection using Machine Learning

Space weather monitoring and prediction efforts are growing in importance with the increasing interest in commercialization of the space sector and as it opens space for public. From our previous analysis (Aparna & Martens, 2020) of about 2 solar cycles of CME and ICME data we have shown that the CME Bz can be predicted by monitoring the regions of eruptions or the chromospheric filaments on the Sun. This analysis requires manual identification of chirality in the cases of filament eruptions and the skew of the overlying arcade fields in active regions where filaments might not be visible or may not be fully formed. Once the chirality is obtained, we get the axial field direction of the polarity inversion line of that region using photospheric magnetograms. Hence, so far, we have been manually determining the chiralities of these filaments. Due to the potential of this method in determining a non-geo-effective CME from a geo-effective one while a filament is still on the Sun, further efforts in automating the process seems worthwhile. As a first step, we automate the filament chirality identification using computer vision. We use chromospheric filament data between 2003 and 2013 from Helio Research, Inc., run by Mrs. Sara Martin - one of the pioneers in filament chirality and filament eruptions research, taken in the center line of Halpha and its red and blue wings. The images have a resolution of 0.9 per pixel with a narrow field of view. We use LabelBox1, a proprietary online tool to label the various features in the images and Google Cloud2 for storing and easy access to our images. We will present the details of the machine learning algorithm that we use for classifying the images and the results from our model at the AGU 2021 Fall meeting. References Aparna, V., & Martens, P. C. H., 2020ApJ, 897, 68 1https://labelbox.com 2https://cloud.google.com/edu/