Deep reconstruction of synthesised ensemble coronal hole images #

Yang Zhou, Chris J Budd OBE, Tom S F Haines, Siegfried Gonzi, David R Jackson

11:10 Tuesday in 4Q07.

Part of the Geophysics and climate session.

Abstract #

The UK Met Office uses the ADAPT-WSA-Enlil model to simulate the ambient solar wind. The accuracy of solar wind has direct impacts on forecasting CMEs and other extreme space weather events.This paper is the first part of the work in using deep learning model to reconstruct coronal holes images. As part of the operational requirements in Met Office, the model needs to be explainable as well should be able to analyse those Magneto-hydrodynamics model (MHD) outputs and reconstruct coronal holes based on satellite observations of solar corona from the Atmospheric Image Assembly (AIA) onboard the Solar Dynamics Observatory(SDO). Due to the complexity in numerical ensemble outputs from ADAPT-WSA simulation, this paper used synthesised data to analyse how the deep learning structure is analysing the data and why the purpose could be met by this method.

Deep learning and machine learning had been applied in many research and industrial applications. Recently, those techniques has been applied in solar physics and space weather research projects. However, most of the models have being tested as black-box models. Even though many mechanisms of deep learning and machine learning models on space weather was not theoretically explainable nor justifiable, most of the empirical results outperformed many existing numerical simulations.

Our objective in this paper is to present a deep learning model which is capable of analysing coronal holes. The results are demonstrated on various types of synthesised data. Generated data allows us to analyse how our neural network model processes data and how valid is the image construction process.