Tracking Solar Phenomena from the SDO

Abstract

This paper focuses on the problem of tracking solar phenomena by creating spatiotemporal trajectories from solar event detection reports. Though tracking of multiple objects in video sequences has seen much research and improvement in recent years, there has been relatively little focus on the domain of tracking solar phenomena (events). In this work, we improve on our previous endeavors by eliminating offline model training requirements and utilizing crowd-sourced human labels to evaluate our performance. We apply our method to the metadata of two solar event types spanning 4 yr of detection reports from the automated detection modules for the Solar Dynamics Observatory mission. We compare our results with those produced by the detection module for active regions and coronal holes by using a crowd-sourced trajectory database as the ground truth. We show that our results are as good as or better than the event-specific detection module for these two event types. This is especially promising because our tracking algorithm is a generalized module for all solar events, and not specific to a single event type, allowing it to be applied to other solar event types reported to the Heliophysics Event Knowledgebase that do not contain tracking information.

A Curated Image Parameter Dataset from Solar Dynamics Observatory Mission

“…One already established use case for this dataset is tracking solar events that have been reported to the HEK (Kempton et al 2018;Kempton & Angryk 2015) where the parameters are used to perform visual comparisons of detections forming different possible paths a tracked event could take. Another is the use of the pa-rameters to perform whole image comparisons for similarity search in the context of content based image retrieval (Kempton et al 2016b).…”

“…Schuh et al also employed these ten image parameters for the development of a trainable module for use in the CBIR system (Schuh et al 2015), along with a thorough analysis on three years of SDO data (from Jan 1, 2012 through Dec 31, 2014). Yet another sequence of studies benefits from the same set of image parameters for tracking of the solar phenomena in time (Kempton & Angryk 2015;Kempton et al 2016aKempton et al , 2018. In that work, their tracking model utilize sparse coding to classify solar event detections as either the same detected event at a later time or an entirely different solar event of the same type.…”

Unsupervised Anomaly Detection With Variational Autoencoders Applied to Full‐Disk Solar Images

“…Several studies make use of traditional machine learning methods to detect and track solar events and forecast space weather (e.g., Kempton et al., 2018; Reiss et al., 2015; Schuh et al., 2017). These studies often rely on a set of image parameters capturing the distribution of pixel intensities and describing shape parameters but do not make use of deep learning to directly learn a feature representation from the raw data.…”

Revisiting the Solar Research Cyberinfrastructure Needs: A White Paper of Findings and Recommendations

“…This 1TiB-per-year dataset provides a much lighter version of the SDO's 0.55 petabytes per year (Martens et al, 2012), tailored for tasks related to CBIR, including feature recognition. Tracking solar events is one of the direct contributions of this dataset (Kempton et al, 2018). A thorough discussion on the collection, curation, and integration of this dataset, as well as the data analytics and feature recognition experiments are authored by , and the project is made open-sourced as part of the DMLab Library (2021).…”