CLASH/HST photoz estimation: the challenges & their quality Stephanie Jouvel, Ofer Lahav, Ole Host
Presentation overview -CLASH observations and photoz codes and photoz quality definitions -Photoz for cluster galaxies -Photoz for high-z galaxies : Arcs for the strong lensing analysis -Conclusion
Data and photoz quality CLASH/HST observations = > 16 filters covering 2000 to AA => detect up to z~4 gal (Balmer) We currently have 3 template fitting codes : -BPZ -Le Phare -Munich’s code (Seitz et al.) We use these codes to estimate the photoz quality in terms of : -Photoz accuracy : NMAD -Number of catastrophic redshift -Median of the zphot-zspec (zp-zs) distribution
How do we assess the photoz quality Robust estimator of the photoz scatter : NMAD Normalised Median Absolute Deviation => 1.48*median[|zp-zs|/1+zs] Fraction of catastrophic zp-zs>0.1 Median (zp-zs)/(1+zs)
Identify the photoz problems The data : – Photometry Calibration which will result in big 0pts corrections => Produce biases in photoz results Photometric errors => Impact the photoz scatter and PDZ – Number of filters => Impact photoz accuracy Photoz codes : – Template representativity and diversity – Priors in redshift/template => More likely to produce catastrophic redshifts and bias
Template representativity We use Le Phare with the template optimised for the COSMOS survey The COSMOS template fill the color-color space defined by the CLASH observation which is a first validation of the template representativity
The CLASH/HST specz data oct 2011 Specz sample of 271 galaxies covering the first 6 CLASH clusters completed Specz catalogue mainly composed by cluster members 0.1<z<0.65. Need to separate the specz catalogue in 2 samples : foreground structure and cluster members (z<0.65) Arcs z>0.65
Photoz for cluster galaxies
Le Phare’s photoz results for specz cat ACS only (7 filters) UVIS+ACS (11 filters) ACS+NIR (12 filters) UVIS+ACS+NIR (16 filters) NMAD 4.3 to 5.4%(1+z) Median to
Le Phare photoz uncertainty Photoz errors underestimated => We then modify the photometric errors bands to achieve this.
Le Phare photoz uncertainty With 0.03 photometric errors added in quadrature at all bands Validation of the photoz uncertainties on the whole mag-redshift range
BPZ and Le Phare 0pts #FILTERS zp_offset err_zp F225W F275W F336W F390W F435W F475W F606W F625W F775W F814W F850LP F105W F110W F125W F140W F160W Problems with UVIS ? Or with templates photoz codes are using ?
Steph’s photoz results z <0.1 filters (chisq2 0.9) NMAD abs(zp-zs) 0.1 ACS 74 objects3.9%(1+z) 64 objects 3.0%(1+z) 23% ACS+IR 105 objects 4.4%(1+z) 78 objects 2.8%(1+z) 25% ACS+UVIS 97 objects 4.0%(1+z) 81 objects 3.1%(1+z) 16% ACS+UVIS+IR 109 objects 3.9%(1+z) 83 objects 2.9%(1+z) 23% # median NMAD n_obj ACS ACS+IR ACS+UVIS ACS+UVIS+IR Txitxo’s results Dan’s photoz results "ACS" % "ACS_NIR" % "UVIS_ACS" % "UVIS_ACS_NIR” %
summary We do not need both UVIS and NIR data to find good photoz for galaxies z<0.65 which is expected since the color gradient produced by the Balmer break is in the optical for these redshift range The UVIS data seem to have a calibration problem since both BPZ and Le Phare find big 0pt in this wavelength range Le Phare derives good uncertainty after an addition factor of 0.03 in the photometric errors Both BPZ and Le Phare have consistent results. We reach an NMAD 2.8 to 4%(1+z), median and low catastrophic redshift rate for confident redshift
Photoz for high-z galaxies – Arcs for strong-lensing analysis
Photoz for arcs : Strong-Lensing Photoz for arcs => High-z galaxies 1<z<6 Balmer break at 8000 AA Lyman break at 2400 AA Need of UVIS or NIR data to detect a color gradient which will help the photoz estimation What photoz quality is necessary for the strong- lensing analysis ?
Redshift and Cosmology Lens Efficiency: For a fixed lens redshift, the efficiency increase with source redshift Weak cosmology dependence Bartelmann & Schneider
Photoz for Strong-Lensing O. Host, D. Coe
Le Phare photoz for specz catalogue Catastrophic redshift for high-redshift galaxies
Le Phare photoz for specz catalogue
Catastrophic redshift for high-redshift galaxies Le Phare photoz for specz catalogue
Catastrophic redshift for high-redshift galaxies c c We can mistake cluster galaxies for background galaxies Le Phare photoz for specz catalogue
Catastrophic redshift for high-redshift galaxies Le Phare photoz for high-z galaxies Photoz uncertainty well estimated using ACS only
Catastrophic redshift for high-redshift galaxies Le Phare photoz for high-z galaxies
Catastrophic redshift for high-redshift galaxies Le Phare photoz for high-z galaxies Photoz uncertainty not as well estimated than using ACS only => Add information that do not help at the color gradient
Catastrophic redshift for high-redshift galaxies => can’t tell from photoz uncertainty Le Phare photoz for high-z galaxies Filt NMAD median %outlier ”A" ”UA" ”AN" ”All” Having all 16 filters improves the statistics and %outliers 41 galaxies at z>0.65
Summary For the cluster galaxies, you only need optical data to derive unbiased redshifts. Since most of our specz catalogue are composed by cluster galaxies UVIS/NIR do not make a big difference. For high-z galaxies z>0.65, the full HST filters does make a difference. It allows to reduce the number of catastrophic redshifts, the scatter, and gives less skewed distribution. Need to understand the number of catastrophic redshift for high-z galaxies.
Le Phare : photoz errors validation
Specz cat 22 A383 8 MACS A MACS RXJ MACS2129
Le Phare photoz errors uncertainty With 0.03 (ACS) 0.04 (NIR) and 0.05 (UVIS)photometric errors added in quadrature at all bands Validation of the photoz uncertainties on the whole mag-redshift range