UCLA Library

The detection and mapping of Coronal Holes (CH) in solar Extreme Ultraviolet (EUV) images is useful for a variety of scientific and space weather prediction applications. Long-term averages of EUVI/STEREO A/B (195 Å) and SDO/AIA (193 Å) images are used to compute data-derived corrections for center-to-limb variations in images and intensity differences among instruments. The calculation of these corrections has been greatly simplified through modern database storage and querying techniques implemented in the image processing pipeline. After image processing, CH regions are detected using two primary machine learning methods: a convolutional neural network (CNN) and K-means clustering. Results are mapped on an image-by-image basis and merged to create EUV and CH maps. The merge-mapping process favors lower intensity data in areas of overlap, and an arbitrary number of images may be used, helping to mitigate the systematic issues of obscuration and evolution. With this flexible new framework and data-driven approach to CH detection, we explore methods for creating time-dependent coronal hole maps and evaluate their use for model comparisons.