aasensio/farside

Improved detection of farside active regions using deep learning analysis

Introduction

The analysis of waves in the visible side of the Sun allows the detection of active regions in the farside through local helioseismology techniques. The knowledge of the magnetism in the whole Sun, including the non-visible hemisphere, is fundamental for several space weather forecasting applications. During the last years, the Solar TErrestrial Relationship Observatory (STEREO) has been monitoring the farside of the Sun, providing Extreme UltraViolet (EUV) images of that hemisphere. However, STEREO spacecrafts are currently returning to the Earth-side of their orbit, and there are no guaranties that they will be operative ten years from now, when they will be back at the farside, as contact with STEREO-B is already lost. Thus, farside helioseismology is the only available tool to obtain a continuous monitoring of the non-visible solar hemisphere. Seismic identification of farside active regions is challenged by the reduced signal-to-noise, and only large and strong active regions can be reliable detected. Here we develop a new methodology to improve the identification of active region signatures in farside seismic maps using a deep learning approach. Our results show that this method can significantly increase the number of detected farside active regions.

Getting started

Once the dependencies are installed, the code is able to get farside probability maps from farside phase-shift maps remapped onto a Carrington coordinate grid. Farside phase-shift maps computed from HMI Doppler maps and HMI magnetograms are available through the JSOC (http://jsoc.stanford.edu/).

The code is simply run with:

python farside_to_magnetogram.py -i INPUT -o OUTPUT [-b MAXBATCH] [-v VERBOSE]

where the parameters are:

• INPUT : IDL save file or HDF5 file that should contain a single dataset with name phases of size [n_cases,11,nx,ny]. n_cases is often 1 but if many maps are available, they can be done in parallel. The number of time steps is 11, in steps of 12 hours, centered on the desired time.

• OUTPUT : HDF5 or FITS file with the resulting probability maps.

• MAXBATCH : maximum batch size, useful when running in GPUs with reduced amount of memory.

• VERBOSE : True/False for adding verbosity in the output

Dependencies

- pytorch (1.0 or later)
- h5py
- scipy
- astropy