An Algorithm for the Determination of Coronal Mass Ejection Kinematic Parameters Based on Machine Learning

Abstract

Abstract Coronal mass ejections (CMEs) constitute the major source of severe space weather events, with the potential to cause enormous damage to humans and spacecraft in space. It is becoming increasingly important to detect and track CMEs, since there are more and more space activities and facilities. We have developed a new algorithm to automatically derive a CME’s kinematic parameters based on machine learning. Our method consists of three steps: recognition, tracking, and the determination of parameters. First, we train a convolutional neural network to classify images from Solar and Heliospheric Observatory Large Angle Spectrometric Coronagraph observations into two categories, containing CME(s) or not. Next, we apply the principal component analysis algorithm and Otsu’s method to acquire binary-labeled CME regions. Then, we employ the track-match algorithm to track a CME’s motion in time-series images and finally determine the CME’s kinematic parameters, e.g., velocity, angular width, and central position angle. The results of four typical CME events with different morphological characteristics are presented and compared with a manual CME catalog and several automatic CME catalogs. Our algorithm shows some advantages in the recognition of CME structure and the accuracy of the kinematic parameters. This algorithm can be helpful for real-time CME warnings and predictions. In the future, this algorithm is capable of being applied to CME initialization in magnetohydrodynamic simulations to study the propagation characteristics of real CME events and to provide more efficient predictions of CMEs’ geoeffectiveness.