aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAaron LI <aly@aaronly.me>2018-05-23 20:04:17 +0800
committerAaron LI <aly@aaronly.me>2018-05-23 20:04:17 +0800
commite12c4928dc71f3de2f5568bccfd9e0b9dfb5ac09 (patch)
tree345470d3a17b0649c956d2976fc6a6f867c1a0ef
parentf0657cb84a6d3dea52cc95978c004fc8646805b1 (diff)
downloadatoolbox-e12c4928dc71f3de2f5568bccfd9e0b9dfb5ac09.tar.bz2
Add astro/montage_skypatch.py to create large sky model for oskar
-rwxr-xr-xastro/oskar/montage_skypatch.py251
1 files changed, 251 insertions, 0 deletions
diff --git a/astro/oskar/montage_skypatch.py b/astro/oskar/montage_skypatch.py
new file mode 100755
index 0000000..5da855b
--- /dev/null
+++ b/astro/oskar/montage_skypatch.py
@@ -0,0 +1,251 @@
+#!/usr/bin/env python3
+#
+# Copyright (c) 2018 Aaron LI <aly@aaronly.me>
+# MIT License
+#
+
+"""
+Montage a list of sky patches (e.g., simulated sky maps of radio halos
+of 10x10 deg^2) to create a large map (e.g., hemisphere) for OSKAR
+simulation to help investigate the far side confusion noise (FSCN).
+
+Configuration file (YAML format)
+--------------------------------
+nside: 8 # npix=768, pixsize~7.33[deg]
+
+region:
+ center: [0.0, -27.0] # [deg]
+ rmin: 10.0 # [deg]
+ rmax: 80.0 # [deg]
+
+threshold:
+ min: 0.0001 # [K]
+ max: null
+
+frequency:
+ type: calc
+ start: 154.0 # [MHz]
+ stop: 162.0
+ step: 0.16
+
+input:
+ filename: '{dir}/cluster_{freq:06.2f}.fits'
+ dirlist: 'dir.list' # filename of the directory list file
+ pixelsize: 20 # [arcsec]
+
+output:
+ filename: 'skymodel/cluster_{freq:06.2f}.osm'
+ clobber: true
+"""
+
+import os
+import argparse
+import logging
+from functools import lru_cache
+
+import yaml
+import numpy as np
+import healpy as hp
+import astropy.units as au
+from astropy.io import fits
+from astropy.wcs import WCS
+
+
+logging.basicConfig(level=logging.INFO,
+ format='[%(levelname)s:%(lineno)d] %(message)s')
+logger = logging.getLogger()
+
+
+def get_frequencies(config):
+ config_freq = config['frequency']
+ if config_freq['type'] == 'custom':
+ return np.array(config_freq['frequencies'])
+ else:
+ start = config_freq['start']
+ stop = config_freq['stop']
+ step = config_freq['step']
+ return np.arange(start, stop+step/2, step)
+
+
+def central_angle(points, p0):
+ """
+ Calculate the central angles between the points with respect to the
+ reference point (p0) on the sphere.
+
+ Input parameters:
+ points: (longitude, latitude) [deg]
+ point coordinates, two columns
+ p0: (longitude, latitude) [deg]
+ coordinate of reference point
+
+ Algorithm:
+ (radial, azimuthal, polar): (r, theta, phi)
+ central_angle: alpha
+ longitude: lambda = theta
+ latitude: delta = 90 - phi
+ colatitude: phi
+
+ Unit vector:
+ \hat{r}_1 = (cos(theta1) sin(phi1), sin(theta1) sin(phi1), cos(phi1))
+ = (cos(lambda1) cos(delta1), sin(lambda1) cos(delta1), sin(delta1))
+ \hat{r}_2 = (cos(theta2) sin(phi2), sin(theta2) sin(phi2), cos(phi2))
+ = (cos(lambda2) cos(delta2), sin(lambda2) cos(delta2), sin(delta2))
+
+ Therefore the angle (alpha) between \hat{r}_1 and \hat{r}_2:
+ cos(alpha) = \hat{r}_1 \cdot \hat{r}_2
+ = cos(delta1) cos(delta2) cos(lambda1-lambda2)
+ + sin(delta1) sin(delta2)
+
+ References:
+ [1] Spherical Coordinates - Wolfram MathWorld
+ http://mathworld.wolfram.com/SphericalCoordinates.html
+ Equation (19)
+ [2] Great Circle - Wolfram MathWorld
+ http://mathworld.wolfram.com/GreatCircle.html
+ Equation (1), (2), (4)
+ """
+ points = np.deg2rad(points)
+ if points.ndim == 1:
+ lbd1, delta1 = points[0], points[1]
+ elif points.ndim == 2:
+ lbd1, delta1 = points[:, 0], points[:, 1]
+ else:
+ raise ValueError('invalid input points')
+ lbd0, delta0 = np.deg2rad(p0)
+ dotvalue = (np.cos(delta1) * np.cos(delta0) * np.cos(lbd1-lbd0) +
+ np.sin(delta1) * np.sin(delta0))
+ alpha = np.arccos(dotvalue)
+ return np.rad2deg(alpha) # [deg]
+
+
+def get_healpix_coords(nside):
+ npix = hp.nside2npix(nside)
+ lon, lat = hp.pix2ang(nside, np.arange(npix), lonlat=True)
+ return (lon, lat) # [deg]
+
+
+class Patches:
+ def __init__(self, config):
+ config_input = config['input']
+ self._filename = config_input['filename']
+ self._dirlist = [d.strip()
+ for d in open(config_input['dirlist']).readlines()]
+
+ def get_image(self, freq, i):
+ i = i % len(self._dirlist)
+ fn = self._filename.format(freq=freq, dir=self._dirlist[i])
+ logger.info('Load image: %s' % fn)
+ with fits.open(fn) as f:
+ return (f[0].data, f[0].header)
+
+
+class SkyModel:
+ def __init__(self, image, freq, pixelsize, p0, min_=1e-4, max_=None):
+ self._image = image
+ self._freq = freq # [MHz]
+ self._pixelsize = pixelsize # [arcsec]
+ self._p0 = p0 # (lon, lat) [deg]
+ self._min = min_ # [K]
+ self._max = max_ # [K]
+
+ @property
+ def K2JyPixel(self):
+ pixarea = (self._pixelsize * au.arcsec) ** 2
+ equiv = au.brightness_temperature(pixarea, self._freq*au.MHz)
+ return au.K.to(au.Jy, equivalencies=equiv)
+
+ @property
+ def mask(self):
+ _max = self._max or np.inf
+ return ((np.abs(self._image) >= self._min) &
+ (np.abs(self._image) <= _max))
+
+ @property
+ def wcs(self):
+ shape = self._image.shape
+ delta = self._pixelsize / 3600.0 # [deg]
+ wcs = WCS(naxis=2)
+ projection = 'SIN'
+ wcs.wcs.ctype = ['RA---'+projection, 'DEC--'+projection]
+ wcs.wcs.crval = np.array(self._p0)
+ wcs.wcs.crpix = np.array([shape[1], shape[0]]) / 2.0 + 1
+ wcs.wcs.cdelt = np.array([-delta, delta])
+ return wcs
+
+ @property
+ def sky(self):
+ shape = self._image.shape
+ idx = self.mask.flatten()
+ wcs = self.wcs
+ x, y = np.meshgrid(np.arange(shape[1]), np.arange(shape[0]))
+ pix = np.column_stack([x.flatten()[idx], y.flatten()[idx]])
+ world = wcs.wcs_pix2world(pix, 0)
+ ra, dec = world[:, 0], world[:, 1]
+ flux = self._image.flatten()[idx] * self.K2JyPixel
+ return np.column_stack([ra, dec, flux])
+
+
+if __name__ == '__main__':
+ parser = argparse.ArgumentParser(
+ description='Montage sky patches to make a large OSKAR sky model')
+ parser.add_argument('config', help='configuration file in YAML format')
+ args = parser.parse_args()
+
+ config = yaml.load(open(args.config))
+ patches = Patches(config)
+ frequencies = get_frequencies(config) # [MHz]
+ nfreq = len(frequencies)
+ pcenter = config['region']['center'] # [deg]
+ rmin = config['region']['rmin'] # [deg]
+ rmax = config['region']['rmax'] # [deg]
+ min_ = config['threshold']['min'] # [K]
+ max_ = config['threshold']['max'] # [K]
+ pixelsize = config['input']['pixelsize'] # [arcsec]
+ clobber = config['output']['clobber']
+
+ nside = config['nside']
+ logger.info('Nside: %d' % nside)
+ resol = hp.nside2resol(nside, arcmin=True) # [arcmin]
+ logger.info('HEALPix resolution: %.2f [arcmin]' % resol)
+ imgsize = int(round(resol * 60 / pixelsize))
+ logger.info('Image patch size: %d' % imgsize)
+ plon, plat = get_healpix_coords(nside) # [deg]
+ npatch = len(plon)
+
+ for i, freq in enumerate(frequencies):
+ logger.info('[%d/%d] %.2f[MHz] ...' % (i+1, nfreq, freq))
+ outfile = config['output']['filename'].format(freq=freq)
+ os.makedirs(os.path.dirname(outfile), exist_ok=True)
+ if os.path.exists(outfile) and (not clobber):
+ raise FileExistsError(outfile)
+
+ results = []
+ jj = 0
+ for j, p0 in enumerate(zip(plon, plat)):
+ logger.info('[%d/%d|%d] patch @ (%.2f, %.2f)' %
+ (j+1, jj+1, npatch, p0[0], p0[1]))
+ if central_angle(p0, pcenter) > rmax:
+ logger.info('skip')
+ continue
+
+ image, header = patches.get_image(freq, jj)
+ skymodel = SkyModel(image[:imgsize, :imgsize],
+ freq, pixelsize=pixelsize,
+ p0=p0, min_=min_, max_=max_)
+ sky = skymodel.sky
+ points = sky[:, 0:2] # (lon, lat)
+ angles = central_angle(points, pcenter) # [deg]
+ idx = ((angles >= rmin) & (angles <= rmax))
+ sky = sky[idx, :]
+ logger.info('Source counts: %d' % len(sky))
+ results.append(sky)
+ jj += 1
+
+ sky = np.row_stack(results)
+ logger.info('Total source counts: %d' % len(sky))
+ header = ('Frequency = %.2f [MHz]\n' % freq +
+ 'Source counts = %d\n\n' % len(sky) +
+ 'R.A.[deg] Dec.[deg] flux[Jy]')
+ logger.info('Writing sky model to file: %s ...' % outfile)
+ np.savetxt(outfile, sky, fmt='%.10e, %.10e, %.10e', header=header)
+ logger.info("Wrote OSKAR sky model to file: %s" % outfile)