Automated Magnetospheric Boundary Detection Using Threshold and Deep Learning methods

Date:

Abstract: We utilize data from the CAPS and MAG instruments onboard Cassini spacecraft to present a algorithms for automatic bow shock and magnetopause boundary detection.

An automatic classification method of the bow shock and magnetopause boundaries at Saturn based on in-situ magnetic field and plasma data measured by Cassini is presented. The algorithm shows promising results with over 90% f1 score (harmonic mean of recall and precision) for detecting bow shock crossings but significantly lower for MP crossings with an f1 score of around 65%. Traditional threshold-based algorithms for boundary detection and modern deep learning method using Convolutional Neural Networks (or CNN) were investigated for this task. The threshold method uses two sliding windows with fixed separation over the data to create an array of parameters from which to apply thresholding to determine the presence of a boundary. The CNN method uses images of the electron energy spectrogram as input to classify the presence of boundaries. 2012 data was used as test data to compare the performance of these two methods, showing that the CNN method outperforms the threshold method. From a data analysis perspective, automation improves reproducibility, discovery potential in unlabeled data, reusability in existing and future planetary missions. “Human-in-the-loop” combined with partial automation would speed up data selection many folds by only correcting a manageable number of misclassifications. Automation would facilitate larger crossing datasets and could also have implications for the future development of onboard data-processing protocols in the pre-downlink stage. From a more fundamental perspective, the boundary survey provides an invaluable dataset to understand the boundary structures like the distribution of radial distances, local times, latitudes, under different phases of the solar cycle throughout the mission lifetime. This offers a wealth of information about the interface between Saturn and the solar wind. Reliable detection of boundary crossings could enable future spacecraft missions like the PEP instrument on the upcoming JUICE spacecraft mission to dynamically adapt the best observing mode based on rapid detection of the boundary crossings as soon as it appears yielding higher quality data and improve potential for scientific discovery.

Cheng, I K., Achilleos, N. “Automated Bow Shock and Magnetopause Boundary Detection With Cassini Using Threshold and Deep Learning Methods”, COSPAR. July, 2022. Athens, Greece.

My immense gratitude to the ML4PSP committee for inviting me to speak at their monthly seminars. The recording can be found on the Machine Learning for Planetary and Space Physics YouTube channel here.