Authors: Stanley Dodds

Affiliations: Institute for Astronomy, Univ. of Hawaii Manoa

Date: 2018-10-13

URL: https://www2.hao.ucar.edu/csac/csac-data/sp-data-description Citation:

"Inverting Solar Spectropolarimetric Observations with Deep Learning", Curt Dodds, Ian Cunningham, Lucas Tarr, Sarah Jaeggli, Tom Schad, Peter Sadowski and Xudong Sun, [pdf]

Each FITS file is a spectral inversion using the MERLIN code implementation of Milne-Eddington inversion from the level 1 Stokes params

The Y (512) dimension is Y on the solar disc in 0.317 arcsec pixels.

The X (112) dimension is X on the solar disc in #### arcsec pixels.

The Z (3) dimension are the magnetic field strength, inclination and azimuth. 0 Field Strength 1 Field Inclination 2 Field Azimuth

Each tar file contains 875 FITS files. Each FITS file is a point in time slit image on the solar disc in x arcsec pixels.

The Y (512) dimension is Y on the solar disc in 0.317 arcsec pixels.

The W (112) dimension is wavelength in 0.021549 Angstrom pixels centered (pixel 56.5) on 6302.08A.

The Z (4) dimension is the Stokes vector. 0 I 1 Q 2 U 3 V

The 875 FITS files comprises the X (875) dimension.

The X (875) dimension is X on the solar disc where x=56 (6302A) for each of 875 FITS files.

The Y (512) dimension is Y on the solar disc for all y in Y.

The Z (1) dimension is the Stokes I parameter (z=0).

The X (875) dimension is X on the solar disc where x=56 (6302A) for each of 875 FITS files.

The Y (512) dimension is Y on the solar disc in 0.317 arcsec pixels.

The Z (4) dimension is the Stokes vector.

prep.py - Load and visualize (plot) each FITS file Edit parameters at the top of prep.py to set: basePath = Path to scan for SP3D Level 1 and 2 images in FITS format spNum is the Stokes paramater to use 0123 => IQUV wlOffset = Offset in pxiels from the central line

fitsinfo.py - Print out the FITS header values Edit parameters at the top of fitsinfo.py to set: basePath = Path to scan for FITS files

hinode2tfr.py - Convert raw FITS files to TensorFlow TFRecord format and split into train, valid and test sets Outputs 3 files: ../data/train.tfr ../data/val.tfr ../data/test.tfr Edit parameters at the top of hinode2tfr.py to set: basePath = Path to scan for SP3D Level 1 and 2 images in FITS format wlOffset = Offset in pxiels from the central line

train.py - Train a model using training data in TFRecord format Inputs 2 files: ../data/train.tfr ../data/val.tfr Outputs the "best" model as ../data/test1.h5 Edit parameters at the top of train.py to set: sizeBatch is the batch size nEpochs is the number of epochs to train

invert.py - Invert arbitrary size Hinode images using sliding windows of 64x64 patches Input directory containing Hinode SOT SP FITS files (level 1) ./invert Outputs magnetograms for each level 1 file found in the input directory Edit image position and size parameters in invert.py Optionally plot kernel density estimates Optionally convert cartesian to spherical magnetic field coordinates Model,Version,Checkpoint control which trained SPIN model to use. nPatch=2 use sliding window (64/2 = 32 pixels); nPatch=1 no sliding windows but faster plotWL[iquv] determines which wavelength bin to plot Around line 797 xlim, ylim control the crop rectangle to plot showColorbar, showTitle, showTicks controls plotting outputs colormap controls the plot color scheme

run inference on 2 Hinode samples from SICON paper (were they held out of training? validation?)