Bayesian Photometric Redshifts (BPZ) Narciso Benítez 1,2 (2000) Narciso Benítez 1,2 et al. (2004) Dan Coe 1,2,3 et al. (2006) Johns Hopkins University 1 Instituto de Astrofísica de Andalucía 2 JPL/Caltech 3 Science Team Science Team

Photo-z Methods  Spectral Energy Distribution (SED) Template Fitting  Empirical Training Set (Neural Networks)  Spectral Energy Distribution (SED) Template Fitting  Empirical Training Set (Neural Networks)

Coleman, Wu, Weedman ‘80 Kinney ‘96 Bruzual & Charlot ‘03 Spectral Energy Distribution (SED) templates BPZ v1.99b Benítez ‘00, ‘04 Coe ‘06 recalibrated with real photometry Normally interpolate 2 between adjacent templates

Flux Wavelength SED template fit

Redshift Probability prior: I = 26 without prior with prior Bayesian use of priors Benítez00 Output:

Benítez00 Redshift Inaccuracy (photo-z vs. spec-z) Poorness of Fit Poorest fits yield most accurate redshifts!

 2 = 4.27  2 = 0.11 Wavelength Flux  2 mod = 0.03  2 mod = 0.19

PHAT GOODS BPZ results (training set) Important to plot error bars and goodness-of-fit

PHAT GOODS BPZ results (training set) Single-peaked P(z) [ODDS  0.95] no error bars plotted

Most GOODS objects have good photometry ACS ground IRAC

…but some are bad ACS ground IRAC

ACS ground IRAC …some are ugly

Robust photo-z’s require Robust photometry One of the best methods (even if Peter doesn’t like it ;)

PSF-corrected aperture-matched photometry What is the best method?

PHOTEST  Photometry Testing  PSF Degradation vs. Model Fitting  Magnitude Uncertainties  Zeropoint Calibration  Object Detection & Deblending  …  Sounds like a job for a new group  Let’s meet in Greece 2009  Photometry Testing  PSF Degradation vs. Model Fitting  Magnitude Uncertainties  Zeropoint Calibration  Object Detection & Deblending  …  Sounds like a job for a new group  Let’s meet in Greece 2009

UDF NICMOS fluxes too low

NICMOS flux recalibration Objects w/ spec-z

Comprehensive Segmentation Map Forced into SExtractor

Wish List (Goals for PHAT?)  Improve SED library  more galaxy types  broader wavelength coverage  SED uncertainties  derived from population synthesis models??  Improve Priors  using UDF, surveys  Improve SED library  more galaxy types  broader wavelength coverage  SED uncertainties  derived from population synthesis models??  Improve Priors  using UDF, surveys

Optimal Filter Choice for a given amount of observing time Benítez et al. (2008) A&A submitted  filters is sub-optimal !  addition of near-IR helps somewhat  > 8 filters performs much better  filters is sub-optimal !  addition of near-IR helps somewhat  > 8 filters performs much better

Filters tested  = const   contiguousoverlapping

Photo-z completeness Best is > 8 overlapping filters Depth to which 80% of objects have ODDS ≥ 0.99

Photo-z accuracy for ODDS ≥ 0.99 objects Best is many non-overlapping (contiguous) filters

lab including CCD, atmosphere, mirror reflectivity ALHAMBRA Survey (Moles08) 20 medium-band (310Å wide) filters Å, supplemented by JHK s

ALHAMBRA Survey 1.5’ x 1.5’ 14-filter color image to cover 4+ sq deg

 8, ,000 sq deg  z <  years  6 sq deg camera  new 2-3m telescope to be built in Aragon, Spain  8, ,000 sq deg  z <  years  6 sq deg camera  new 2-3m telescope to be built in Aragon, Spain

PAU Survey : Å-wide filters (~ Å) + SDSS u & z

PAU Survey :  z/(1+z) L*, I < 23 LRGs

PAU Survey: BAO cosmological constraints

PAU Survey: relative w constraints