A Machine-learning Multispectral Time Series Data Set Prepared from the Solar and Heliospheric Observatory Mission

Abstract:

We present a flexible framework that allows user-defined input selection criteria and a range of pre-processing steps, for generating ready-to-use machine-learning SOHO (Solar and Heliospheric Observatory) multispectral time series data cubes. SOHO is a collaborative international project between ESA and NASA and this mission’s data is selected for both its high spatial and temporal resolution, covering Solar Cycles 23 and 24. The framework utilizes SunPy’s Federated Internet Data Obtainer (Fido) interface with the Virtual Solar Observatory (VSO) tool. We illustrate the use of this data with an application of a deep convolutional neural network (CNN) to a subset of the full data set in an effort to provide a 3-5 day-ahead forecast of the north-south component of the of the interplanetary magnetic field (IMF) observed at L1. Baselines for future model comparison are also provided.