Deep learning-based prediction of CME-driven shock standoff distances in metric type II radio emissions

Abstract

Type II radio emissions are events mostly found to be associated with coronal mass ejections (CMEs) and accelerated by the CME-driven shock in the heliosphere. This study reports on the estimation of the CME-shock standoff distance at the commencement of metric type II radio emissions by combining the CME-deprojected speed and spectral features of radio bursts using a robust TensorFlow Deep-Learning Sequential (TFDLS) technique. The dataset of 96 CMEs at the commencement of type II radio bursts was used between Solar cycle 24 and the ascending phase of Solar Cycle 25. The measured root mean squared error (RMSE) was 0.145 (Rs), with an average height difference of 0.096 Rs between the observed and predicted CME-shock heights. Five (5) CMEs/radio bursts energetic events associated with solar flares were selected from the test data, and the CME shock stand-off heights were forecasted using the TFDLS and flare-onset (FL) methods. The data were used to compare the leading-edge (LE) and dynamic spectra (DS) methods. The RMSE measured between the FL and LE was 0.35 Rs, and the RMSE estimated between the TFDLS and LE approaches was 0.04 Rs. The RMSE between FL and DS was 0.34. Rs, and the RMSE between the TFDLS and the DS was 0.04 Rs. We also used the findings gained from the five selected events and compared them to the 3D shock-fitting (3D-SF) approach. The RMSE found between the TFDLS and the 3D-SF was 0.18 Rs, while the RMSE estimated between the FL and the 3D-SF was 0.23 Rs. This shows that the TFDLS has satisfactory performance and can be used as an alternative technique.