import os import galsim import random import numpy as np import h5py as h5 import healpy as hp import astropy.constants as cons import traceback from astropy.coordinates import spherical_to_cartesian from astropy.table import Table from scipy import interpolate from datetime import datetime from ObservationSim.MockObject import CatalogBase, Star, Galaxy, Quasar from ObservationSim.MockObject._util import tag_sed, getObservedSED, getABMAG, integrate_sed_bandpass, comoving_dist from ObservationSim.Astrometry.Astrometry_util import on_orbit_obs_position # (TEST) from astropy.cosmology import FlatLambdaCDM from astropy import constants from astropy import units as U try: import importlib.resources as pkg_resources except ImportError: # Try backported to PY<37 'importlib_resources' import importlib_resources as pkg_resources NSIDE = 128 def get_bundleIndex(healpixID_ring, bundleOrder=4, healpixOrder=7): assert NSIDE == 2**healpixOrder shift = healpixOrder - bundleOrder shift = 2*shift nside_bundle = 2**bundleOrder nside_healpix= 2**healpixOrder healpixID_nest= hp.ring2nest(nside_healpix, healpixID_ring) bundleID_nest = (healpixID_nest >> shift) bundleID_ring = hp.nest2ring(nside_bundle, bundleID_nest) return bundleID_ring class Catalog(CatalogBase): def __init__(self, config, chip, pointing, chip_output, filt, **kwargs): super().__init__() self.cat_dir = os.path.join(config["data_dir"], config["catalog_options"]["input_path"]["cat_dir"]) self.seed_Av = config["catalog_options"]["seed_Av"] self.cosmo = FlatLambdaCDM(H0=67.66, Om0=0.3111) self.chip_output = chip_output self.filt = filt self.logger = chip_output.logger with pkg_resources.path('Catalog.data', 'SLOAN_SDSS.g.fits') as filter_path: self.normF_star = Table.read(str(filter_path)) self.config = config self.chip = chip self.pointing = pointing self.max_size = 0. if "star_cat" in config["catalog_options"]["input_path"] and config["catalog_options"]["input_path"]["star_cat"] and not config["catalog_options"]["galaxy_only"]: star_file = config["catalog_options"]["input_path"]["star_cat"] star_SED_file = config["catalog_options"]["SED_templates_path"]["star_SED"] self.star_path = os.path.join(self.cat_dir, star_file) self.star_SED_path = os.path.join(config["data_dir"], star_SED_file) self._load_SED_lib_star() if "galaxy_cat" in config["catalog_options"]["input_path"] and config["catalog_options"]["input_path"]["galaxy_cat"] and not config["catalog_options"]["star_only"]: galaxy_dir = config["catalog_options"]["input_path"]["galaxy_cat"] self.galaxy_path = os.path.join(self.cat_dir, galaxy_dir) self.galaxy_SED_path = os.path.join(config["data_dir"], config["catalog_options"]["SED_templates_path"]["galaxy_SED"]) self._load_SED_lib_gals() if "AGN_cat" in config["catalog_options"]["input_path"] and config["catalog_options"]["input_path"]["AGN_cat"] and not config["catalog_options"]["star_only"]: AGN_dir = config["catalog_options"]["input_path"]["AGN_cat"] self.AGN_path = os.path.join(config["data_dir"], config["catalog_options"]["input_path"]["AGN_cat"]) self.AGN_SED_path = os.path.join(config["data_dir"], config["catalog_options"]["SED_templates_path"]["AGN_SED"]) self.AGN_SED_wave_path = os.path.join(config['data_dir'], config["catalog_options"]["SED_templates_path"]["AGN_SED_WAVE"]) self._load_SED_lib_AGN() if "rotateEll" in config["catalog_options"]: self.rotation = float(int(config["catalog_options"]["rotateEll"]/45.)) else: self.rotation = 0. # Update output .cat header with catalog specific output columns self._add_output_columns_header() self._get_healpix_list() self._load() def _add_output_columns_header(self): self.add_hdr = " model_tag teff logg feh" self.add_hdr += " bulgemass diskmass detA e1 e2 kappa g1 g2 size galType veldisp " self.add_fmt = " %10s %8.4f %8.4f %8.4f" self.add_fmt += " %8.4f %8.4f %8.4f %8.4f %8.4f %8.4f %8.4f %8.4f %8.4f %4d %8.4f " self.chip_output.update_ouptut_header(additional_column_names=self.add_hdr) def _get_healpix_list(self): self.sky_coverage = self.chip.getSkyCoverageEnlarged(self.chip.img.wcs, margin=0.2) ra_min, ra_max, dec_min, dec_max = self.sky_coverage.xmin, self.sky_coverage.xmax, self.sky_coverage.ymin, self.sky_coverage.ymax ra = np.deg2rad(np.array([ra_min, ra_max, ra_max, ra_min])) dec = np.deg2rad(np.array([dec_max, dec_max, dec_min, dec_min])) # vertices = spherical_to_cartesian(1., dec, ra) self.pix_list = hp.query_polygon( NSIDE, hp.ang2vec(np.radians(90.) - dec, ra), inclusive=True ) if self.logger is not None: msg = str(("HEALPix List: ", self.pix_list)) self.logger.info(msg) else: print("HEALPix List: ", self.pix_list) def load_norm_filt(self, obj): if obj.type == "star": return self.normF_star elif obj.type == "galaxy" or obj.type == "quasar": # return self.normF_galaxy return None else: return None def _load_SED_lib_star(self): self.tempSED_star = h5.File(self.star_SED_path,'r') def _load_SED_lib_gals(self): pcs = h5.File(os.path.join(self.galaxy_SED_path, "pcs.h5"), "r") lamb = h5.File(os.path.join(self.galaxy_SED_path, "lamb.h5"), "r") self.lamb_gal = lamb['lamb'][()] self.pcs = pcs['pcs'][()] def _load_SED_lib_AGN(self): from astropy.io import fits self.SED_AGN = fits.open(self.AGN_SED_path)[0].data self.lamb_AGN = np.load(self.AGN_SED_wave_path) def _load_gals(self, gals, pix_id=None, cat_id=0): ngals = len(gals['ra']) # Apply astrometric modeling # in C3 case only aberration ra_arr = gals['ra'][:] dec_arr = gals['dec'][:] if self.config["obs_setting"]["enable_astrometric_model"]: ra_list = ra_arr.tolist() dec_list = dec_arr.tolist() pmra_list = np.zeros(ngals).tolist() pmdec_list = np.zeros(ngals).tolist() rv_list = np.zeros(ngals).tolist() parallax_list = [1e-9] * ngals dt = datetime.utcfromtimestamp(self.pointing.timestamp) date_str = dt.date().isoformat() time_str = dt.time().isoformat() ra_arr, dec_arr = on_orbit_obs_position( input_ra_list=ra_list, input_dec_list=dec_list, input_pmra_list=pmra_list, input_pmdec_list=pmdec_list, input_rv_list=rv_list, input_parallax_list=parallax_list, input_nstars=ngals, input_x=self.pointing.sat_x, input_y=self.pointing.sat_y, input_z=self.pointing.sat_z, input_vx=self.pointing.sat_vx, input_vy=self.pointing.sat_vy, input_vz=self.pointing.sat_vz, input_epoch="J2000", input_date_str=date_str, input_time_str=time_str ) for igals in range(ngals): # # (TEST) # if igals > 100: # break param = self.initialize_param() param['ra'] = ra_arr[igals] param['dec'] = dec_arr[igals] param['ra_orig'] = gals['ra'][igals] param['dec_orig'] = gals['dec'][igals] param['mag_use_normal'] = gals['mag_csst_%s'%(self.filt.filter_type)][igals] if self.filt.is_too_dim(mag=param['mag_use_normal'], margin=self.config["obs_setting"]["mag_lim_margin"]): continue param['z'] = gals['redshift'][igals] param['model_tag'] = 'None' param['g1'] = gals['shear'][igals][0] param['g2'] = gals['shear'][igals][1] param['kappa'] = gals['kappa'][igals] param['e1'] = gals['ellipticity_true'][igals][0] param['e2'] = gals['ellipticity_true'][igals][1] # For shape calculation param['ell_total'] = np.sqrt(param['e1']**2 + param['e2']**2) if param['ell_total'] > 0.9: continue param['e1_disk'] = param['e1'] param['e2_disk'] = param['e2'] param['e1_bulge'] = param['e1'] param['e2_bulge'] = param['e2'] param['delta_ra'] = 0 param['delta_dec'] = 0 # Masses param['bulgemass'] = gals['bulgemass'][igals] param['diskmass'] = gals['diskmass'][igals] param['size'] = gals['size'][igals] if param['size'] > self.max_size: self.max_size = param['size'] # Sersic index param['disk_sersic_idx'] = 1. param['bulge_sersic_idx'] = 4. # Sizes param['bfrac'] = param['bulgemass']/(param['bulgemass'] + param['diskmass']) if param['bfrac'] >= 0.6: param['hlr_bulge'] = param['size'] param['hlr_disk'] = param['size'] * (1. - param['bfrac']) else: param['hlr_disk'] = param['size'] param['hlr_bulge'] = param['size'] * param['bfrac'] # SED coefficients param['coeff'] = gals['coeff'][igals] param['detA'] = gals['detA'][igals] # Others param['galType'] = gals['type'][igals] param['veldisp'] = gals['veldisp'][igals] # TEST no redening and no extinction param['av'] = 0.0 param['redden'] = 0 param['star'] = 0 # Galaxy # NOTE: this cut cannot be put before the SED type has been assigned if not self.chip.isContainObj(ra_obj=param['ra'], dec_obj=param['dec'], margin=200): continue # TEMP self.ids += 1 # param['id'] = self.ids param['id'] = '%06d'%(int(pix_id)) + '%06d'%(cat_id) + '%08d'%(igals) if param['star'] == 0: obj = Galaxy(param, self.rotation, logger=self.logger) # Need to deal with additional output columns obj.additional_output_str = self.add_fmt%("n", 0., 0., 0., param['bulgemass'], param['diskmass'], param['detA'], param['e1'], param['e2'], param['kappa'], param['g1'], param['g2'], param['size'], param['galType'], param['veldisp']) self.objs.append(obj) def _load_stars(self, stars, pix_id=None): nstars = len(stars['sourceID']) # Apply astrometric modeling ra_arr = stars["RA"][:] dec_arr = stars["Dec"][:] pmra_arr = stars['pmra'][:] pmdec_arr = stars['pmdec'][:] rv_arr = stars['RV'][:] parallax_arr = stars['parallax'][:] if self.config["obs_setting"]["enable_astrometric_model"]: ra_list = ra_arr.tolist() dec_list = dec_arr.tolist() pmra_list = pmra_arr.tolist() pmdec_list = pmdec_arr.tolist() rv_list = rv_arr.tolist() parallax_list = parallax_arr.tolist() dt = datetime.utcfromtimestamp(self.pointing.timestamp) date_str = dt.date().isoformat() time_str = dt.time().isoformat() ra_arr, dec_arr = on_orbit_obs_position( input_ra_list=ra_list, input_dec_list=dec_list, input_pmra_list=pmra_list, input_pmdec_list=pmdec_list, input_rv_list=rv_list, input_parallax_list=parallax_list, input_nstars=nstars, input_x=self.pointing.sat_x, input_y=self.pointing.sat_y, input_z=self.pointing.sat_z, input_vx=self.pointing.sat_vx, input_vy=self.pointing.sat_vy, input_vz=self.pointing.sat_vz, input_epoch="J2000", input_date_str=date_str, input_time_str=time_str ) for istars in range(nstars): # # (TEST) # if istars > 100: # break param = self.initialize_param() param['ra'] = ra_arr[istars] param['dec'] = dec_arr[istars] param['ra_orig'] = stars["RA"][istars] param['dec_orig'] = stars["Dec"][istars] param['pmra'] = pmra_arr[istars] param['pmdec'] = pmdec_arr[istars] param['rv'] = rv_arr[istars] param['parallax'] = parallax_arr[istars] if not self.chip.isContainObj(ra_obj=param['ra'], dec_obj=param['dec'], margin=200): continue param['mag_use_normal'] = stars['app_sdss_g'][istars] # if param['mag_use_normal'] >= 26.5: # continue self.ids += 1 param['id'] = stars['sourceID'][istars] param['sed_type'] = stars['sourceID'][istars] param['model_tag'] = stars['model_tag'][istars] param['teff'] = stars['teff'][istars] param['logg'] = stars['grav'][istars] param['feh'] = stars['feh'][istars] param['z'] = 0.0 param['star'] = 1 # Star obj = Star(param, logger=self.logger) # Append additional output columns to the .cat file obj.additional_output_str = self.add_fmt%(param["model_tag"], param['teff'], param['logg'], param['feh'], 0., 0., 0., 0., 0., 0., 0., 0., 0., -1, 0.) self.objs.append(obj) def _load_AGNs(self): data = Table.read(self.AGN_path) ra_arr = data['ra'] dec_arr = data['dec'] nAGNs = len(data) if self.config["obs_setting"]["enable_astrometric_model"]: ra_list = ra_arr.tolist() dec_list = dec_arr.tolist() pmra_list = np.zeros(nAGNs).tolist() pmdec_list = np.zeros(nAGNs).tolist() rv_list = np.zeros(nAGNs).tolist() parallax_list = [1e-9] * nAGNs dt = datetime.utcfromtimestamp(self.pointing.timestamp) date_str = dt.date().isoformat() time_str = dt.time().isoformat() ra_arr, dec_arr = on_orbit_obs_position( input_ra_list=ra_list, input_dec_list=dec_list, input_pmra_list=pmra_list, input_pmdec_list=pmdec_list, input_rv_list=rv_list, input_parallax_list=parallax_list, input_nstars=nAGNs, input_x=self.pointing.sat_x, input_y=self.pointing.sat_y, input_z=self.pointing.sat_z, input_vx=self.pointing.sat_vx, input_vy=self.pointing.sat_vy, input_vz=self.pointing.sat_vz, input_epoch="J2000", input_date_str=date_str, input_time_str=time_str ) for iAGNs in range(nAGNs): param = self.initialize_param() param['ra'] = ra_arr[iAGNs] param['dec'] = dec_arr[iAGNs] param['ra_orig'] = data['ra'][iAGNs] param['dec_orig'] = data['dec'][iAGNs] param['z'] = data['z'][iAGNs] param['appMag'] = data['appMag'][iAGNs] param['absMag'] = data['absMag'][iAGNs] # NOTE: this cut cannot be put before the SED type has been assigned if not self.chip.isContainObj(ra_obj=param['ra'], dec_obj=param['dec'], margin=200): continue # TEST no redening and no extinction param['av'] = 0.0 param['redden'] = 0 param['star'] = 2 # Quasar param['id'] = data['igmlos'][iAGNs] if param['star'] == 2: obj = Quasar(param, logger=self.logger) # Append additional output columns to the .cat file obj.additional_output_str = self.add_fmt%("n", 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., -1, 0.) self.objs.append(obj) def _load(self, **kwargs): self.objs = [] self.ids = 0 if "star_cat" in self.config["catalog_options"]["input_path"] and self.config["catalog_options"]["input_path"]["star_cat"] and not self.config["catalog_options"]["galaxy_only"]: star_cat = h5.File(self.star_path, 'r')['catalog'] for pix in self.pix_list: try: stars = star_cat[str(pix)] self._load_stars(stars, pix_id=pix) del stars except Exception as e: self.logger.error(str(e)) print(e) if "galaxy_cat" in self.config["catalog_options"]["input_path"] and self.config["catalog_options"]["input_path"]["galaxy_cat"] and not self.config["catalog_options"]["star_only"]: for pix in self.pix_list: try: bundleID = get_bundleIndex(pix) file_path = os.path.join(self.galaxy_path, "galaxies_C6_bundle{:06}.h5".format(bundleID)) gals_cat = h5.File(file_path, 'r')['galaxies'] gals = gals_cat[str(pix)] self._load_gals(gals, pix_id=pix, cat_id=bundleID) del gals except Exception as e: traceback.print_exc() self.logger.error(str(e)) print(e) if "AGN_cat" in self.config["catalog_options"]["input_path"] and self.config["catalog_options"]["input_path"]["AGN_cat"] and not self.config["catalog_options"]["star_only"]: try: self._load_AGNs() except Exception as e: traceback.print_exc() self.logger.error(str(e)) print(e) if self.logger is not None: self.logger.info("maximum galaxy size: %.4f"%(self.max_size)) self.logger.info("number of objects in catalog: %d"%(len(self.objs))) else: print("number of objects in catalog: ", len(self.objs)) def load_sed(self, obj, **kwargs): if obj.type == 'star': _, wave, flux = tag_sed( h5file=self.tempSED_star, model_tag=obj.param['model_tag'], teff=obj.param['teff'], logg=obj.param['logg'], feh=obj.param['feh'] ) elif obj.type == 'galaxy' or obj.type == 'quasar': factor = 10**(-.4 * self.cosmo.distmod(obj.z).value) if obj.type == 'galaxy': flux = np.matmul(self.pcs, obj.coeff) * factor # if np.any(flux < 0): # raise ValueError("Glaxy %s: negative SED fluxes"%obj.id) flux[flux < 0] = 0. sedcat = np.vstack((self.lamb_gal, flux)).T sed_data = getObservedSED( sedCat=sedcat, redshift=obj.z, av=obj.param["av"], redden=obj.param["redden"] ) wave, flux = sed_data[0], sed_data[1] elif obj.type == 'quasar': flux = self.SED_AGN[int(obj.id)] * 1e-17 # if np.any(flux < 0): # raise ValueError("Glaxy %s: negative SED fluxes"%obj.id) flux[flux < 0] = 0. # sedcat = np.vstack((self.lamb_AGN, flux)).T wave = self.lamb_AGN # print("sed (erg/s/cm2/A) = ", sed_data) # np.savetxt(os.path.join(self.config["work_dir"], "%s_sed.txt"%(obj.id)), sedcat) else: raise ValueError("Object type not known") speci = interpolate.interp1d(wave, flux) lamb = np.arange(2000, 11001+0.5, 0.5) y = speci(lamb) # erg/s/cm2/A --> photon/s/m2/A all_sed = y * lamb / (cons.h.value * cons.c.value) * 1e-13 sed = Table(np.array([lamb, all_sed]).T, names=('WAVELENGTH', 'FLUX')) if obj.type == 'quasar': # integrate to get the magnitudes sed_photon = np.array([sed['WAVELENGTH'], sed['FLUX']]).T sed_photon = galsim.LookupTable(x=np.array(sed_photon[:, 0]), f=np.array(sed_photon[:, 1]), interpolant='nearest') sed_photon = galsim.SED(sed_photon, wave_type='A', flux_type='1', fast=False) interFlux = integrate_sed_bandpass(sed=sed_photon, bandpass=self.filt.bandpass_full) obj.param['mag_use_normal'] = getABMAG(interFlux, self.filt.bandpass_full) # if obj.param['mag_use_normal'] >= 30: # print("obj ID = %d"%obj.id) # print("mag_use_normal = %.3f"%obj.param['mag_use_normal']) # print("integrated flux = %.7f"%(interFlux)) # print("app mag = %.3f"%obj.param['appMag']) # np.savetxt('./AGN_SED_test/sed_objID_%d.txt'%obj.id, np.transpose([self.lamb_AGN, self.SED_AGN[int(obj.id)]])) # print("obj ID = %d"%obj.id) # print("mag_use_normal = %.3f"%obj.param['mag_use_normal']) # print("integrated flux = %.7f"%(interFlux)) # print("app mag = %.3f"%obj.param['appMag']) # print("abs mag = %.3f"%obj.param['absMag']) # mag = getABMAG(interFlux, self.filt.bandpass_full) # print("mag diff = %.3f"%(mag - obj.param['mag_use_normal'])) del wave del flux return sed