Speaker
Description
Coronal Mass Ejections (CMEs) are among the most significant events impacting space weather. Their role in triggering severe geomagnetic disturbances has prompted ongoing research to predict their velocities and arrival times on Earth. During their eruption, CMEs undergo rapid acceleration at lower altitudes in the solar atmosphere, reaching speeds ranging from 100 km/s to over 3000 km/s. Accurately tracking CME evolution over time is essential yet challenging. Effective tracking enables the identification of multiple CMEs and the precise estimation of their properties. In this study, we present CMETNets (CME Tracking Networks), a novel framework designed for efficient CME tracking using Temporal Convolutional Networks (TCNs). Leveraging the power of Deep Neural Networks (DNNs), this approach analyses and detects spatio-temporal patterns in both short- and long-term time series, making it a robust solution for CME tracking. By utilising segmented CME maps from LASCO C2 and LASCO C3 instruments, our method demonstrates high performance in tracking CME regions over time, achieving a loss of 0.002. This allows for accurate estimation of CME properties, including angular width, size, and location. To the best of our knowledge, this is the first time that TCN has been proposed for Multiple CMEs tracking and property estimation.
