| Presentation Title: | DeepFiltering: Utilising the Power of Generative Networks to Produce Quality Images of Coronal Rain |
| Author(s): | Luke McMullan, Dr Patrick Antolin, Dr Brandon Panos, Prof. Lucia Kleint |
Abstract:
Coronal rain can be a key indicator of coronal heating taking place. If the coronal heating problem is going to be resolved it behoves us to fully investigate
this link across the full disk of the Sun. There is no lack of observational data,
but currently this data is inadequate for a complete analysis of the phenomenon
to be carried out. The AIA 304 channel provides the best dataset for coronal
rain observations. However, besides the cool component at 10^5 K from He II
emission, the passband also includes hotter coronal emission from other ions.
The contribution of this hotter emission can become comparable to that of the
cool emission in off-limb observations, leading to ambiguity when determining
the temperature of structures. Conversely, IRIS/SJI 1400 dominated by Si IV
emission at 10^4.8 K provides higher resolution images with far less ambiguity
between hot and cool emission, and therefore higher contrast between both the
rain and the surrounding corona. Unfortunately, the small field-of-view of the
satellite makes it ill-suited for large scale statistical analysis of the phenomenon.
We present a novel approach to this problem by training a CycleGAN based
algorithm to undertake a style translation between AIA 304 images and those
belonging to IRIS 1400. This produces a model which can optimally, and with-
out the need of additional data, convert AIA 304 images into those unhampered
by the large temperature ambiguity. The structures in these images are then
compared to the original IRIS 1400 images, as well as those produced from alternative methods, to show the reliability of this method going forwards. This
method opens the path for large-scale statistical analysis of coronal rain, and a
greater understanding of the phenomenon and its relation to coronal heating.