def cloudyness(diff_image,
fwhm=5.,
sigma_cut=3.,
sigma_second=3.,
airmass_map=None,
airmass_limit=None,
skyRMS_max=None,
grow_iter=3,
grow_fwhm=5.,
grow_lower_limit=0.1):
"""
Parameters
----------
diff_image: np.array
An array that is the difference of two all-sky images
sigma_cut: float
blah
Returns
-------
out_area: float
area (sq degrees) that are flagged as cloudy
cloud_mask: np.array
Array same size as diff_image with values of -1, 0, 1 to show which
pixels have been flagged as negative or positive.
"""
# XXX. Maybe the solution is to set a minimum angular scale, but then filter on the
# scale that has the most power (if that's larger)?
# Or, there should be some relation between the brightness level of the frame and the
# RMS of the frame--such that if the median of the frame - dark_sky is ~0, the sigma ~= 0.06,
# but if frame - dark >> dark, then sigma should scale down. So that's how I could set skyRMS_max.
# Also look at imposing some continuity
unmasked = np.where(diff_image != hp.UNSEEN)[0]
skyRMS = robustRMS(diff_image[unmasked])
if skyRMS_max is not None:
skyRMS = np.min([skyRMS, skyRMS_max])
# Try to find the angular scale of any clouds?
smooth_map = hp.sphtfunc.smoothing(diff_image,
fwhm=np.radians(fwhm),
verbose=False,
iter=1)
smooth_map[unmasked] = smooth_map[unmasked] - np.median(
smooth_map[unmasked])
# outliers = np.where(np.abs(smooth_map[unmasked]) > sigma_cut*skyRMS)[0]
# Can think about going back to the original map and growing the region that got flagged
cloud_mask = np.zeros(diff_image.size, dtype=int)
highOutliers = np.where((smooth_map > sigma_cut * skyRMS)
& (diff_image > sigma_second * skyRMS)
& (diff_image != hp.UNSEEN))
lowOutliers = np.where((smooth_map < -1 * sigma_cut * skyRMS)
& (diff_image < -1 * sigma_second * skyRMS)
& (diff_image != hp.UNSEEN))
cloud_mask[highOutliers] = 1
cloud_mask[lowOutliers] = -1
# rather than loop, let's just use smoothing in clever ways!
if np.max(np.abs(cloud_mask)) != 0:
for ack in np.arange(grow_iter):
expanded_mask = hp.sphtfunc.smoothing(cloud_mask,
fwhm=np.radians(grow_fwhm),
verbose=False,
iter=1)
cloud_mask[np.where((expanded_mask > grow_lower_limit)
& (diff_image > 0))] = 1
cloud_mask[np.where((expanded_mask < -1. * grow_lower_limit)
& (diff_image < 0))] = -1
# What is the most efficient way to loop over this stuff?
# if np.max(np.abs(cloud_mask)) != 0:
# changed_pix =1
# while changed_pix != 0:
# clear_pix = np.where(cloud_mask == 0)[0]
# for i in clear_pix:
nside = hp.npix2nside(np.size(diff_image))
pix_area = hp.nside2pixarea(nside)
outliers = np.where(cloud_mask != 0)[0]
out_area = outliers.size * pix_area * (180. / np.pi)**2
return out_area, cloud_mask
raw_flux = aperture_photometry(data, apertures)
bkgflux_table = aperture_photometry(data, annulus_apertures)
phot_table = hstack([raw_flux, bkgflux_table], table_names=['raw', 'bkg'])
bkg_mean = phot_table['aperture_sum_bkg'] / annulus_apertures.area()
bkg_sum = bkg_mean * apertures.area()
final_sum = phot_table['aperture_sum_raw'] - bkg_sum
phot_table['residual_aperture_sum'] = final_sum
print(phot_table['residual_aperture_sum'])
ack = np.array(positions).T
#plt.plot(ack[0], ack[1], 'o')
fluxes.append(phot_table['residual_aperture_sum'].data.tolist())
backgrounds.append(
[robustRMS(data[10:30, 10:30]),
robustRMS(data[1000:1100, 1500:1600])])
fluxes = np.array(fluxes)
backgrounds = np.array(backgrounds)
print 'stars'
for i in np.arange(2):
print 'mean, std, mean^0.5 = %f, %f, %f' % (
fluxes[:, i].mean(), fluxes[:, i].std(), fluxes[:, i].mean()**0.5)
print 'background stats (unilluminated section and blankish area)'
for i in np.arange(2):
print 'mean (of robust RMS), std = %f, %f' % (backgrounds[:, i].mean(),
backgrounds[:, i].std())
# Take away message, I think the standard deviation of the flux values is a factor of 2-3 more than what
alt, az = stupidFast_RaDec2AltAz(ra, dec, site.latitude_rad,
site.longitude_rad, mjd)
out = np.where((np.isnan(frame)) | (np.isnan(previous))
| (frame == hp.UNSEEN) | (previous == hp.UNSEEN)
| (alt < np.radians(10.)))
previous = frame.copy()
diff[out] = hp.UNSEEN
frame[out] = hp.UNSEEN
diff_frac[out] = hp.UNSEEN
diff_frac[np.where(previous == 0)] = hp.UNSEEN
# maybe rotate based on LMST and latitude?
gdiff = np.where((diff != hp.UNSEEN) & (~np.isnan(diff)))[0]
if np.size(gdiff) > 0:
rms = robustRMS(diff_frac[gdiff])
median_value = np.median(diff_frac[gdiff])
nout = np.size(
np.where((
np.abs(diff[gdiff] - np.median(diff[gdiff])) > outlier_mag)
& (alt[gdiff] > alt_limit))[0])
nabove = float(np.size(np.where(alt[gdiff] > alt_limit)[0])) * 100
cf, cloud_mask = cloudyness(diff_frac, skyRMS_max=0.05)
cf = cf / (gdiff.size * hp.nside2pixarea(nside) *
(180. / np.pi)**2)
if nabove != 0:
nout = nout / nabove
else:
nout = -666
else:
