1 / 24

Photometric Redshifts

Photometric Redshifts. PHAT Meeting Pasadena 3-5 Dec 2008. Christian Wolf. data. model. Farb- bibliothek. estimator. Schätzer/ Klassifikator. Frequentist precision statistics: = “Using what IS there: N(z)!”. result. Bayesian frontier exploration: = “What do we (not) know: p(z)=?”.

Download Presentation

Photometric Redshifts

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Photometric Redshifts PHAT MeetingPasadena3-5 Dec 2008 Christian Wolf

  2. data model Farb-bibliothek estimator Schätzer/Klassifikator Frequentist precision statistics:= “Using what IS there: N(z)!” result Bayesian frontier exploration:= “What do we (not) know: p(z)=?” Photo-z Ingredients & Application 2-fitting artificial neural netlearning algorithms empirical data or external template spectral energydistribution PDF: p(z)

  3. Model + Estimator Combinations • 2 • PDF  Ambiguity warning • NN • No PDF, no warning • Template model • Can be extrapolated in z,mag • Calibration issues • Priors’ issues • Empirical model • Good priors • No calibration issues • Can not be extrapolated Code 2 NN Model Template  Empirical 

  4. PStar type ML PGalaxy redshift MEV PQuasar redshift Class Decision and z-Estimation <z> ± z Can include morphology etc.in statistics or neural net

  5. Examples from COMBO-17 Classification ~98% complete at R<23 Stars (~3,000) White Dwarfs (~30) Ultra-cool WD (1) Galaxies (~30,000) QSOs (~300)

  6. E colour z=0.52 z=0.56 Sb colour Redshift Error Regimes • Three regimes in photo-z quality • Saturation • Model-data calibration offsets; intrinsic dimensionality of class • Transition • Locally linear colour(z) grid • Breakdown • Globally nonlinear colour(z) grid mag

  7. Galaxies: Saturation & Transition R=20 R=22 R=23.7 Galaxies at z~0.45

  8. R=21.5 R=22.9 R=23.8 QSOs: Saturation at R<24 rms 0.008 7%-20%outlier QSOs at z~2.8

  9. Catastrophic failures & misclassifications Large z errors Local z bias Unrealistic z errors Model ambiguities in colour space (spotted?) PDF too unconstrained PDF wrong (calib, prior) Mismatch between data and model Photo-Z Trouble

  10. Catastrophic failures & misclassifications Large z errors Local z bias Unrealistic z errors Model ambiguities in colour space PDF too unconstrained PDF wrong (calib, prior) Mismatch between data and model Add more data Add model priors Repair modelsor data Realistic data errors Common Fixes

  11. Add More Data: Wider  Range • Wider  coverage • Covers spectral features across wider z range • Add NIR data • For z > 1 galaxies • Only weak & high-variance features in rest-frame UV • Red z>1 galaxies with noisy optical data • Add UV data (e.g. GALEX) • For QSOs (Ball et al. 2007) • Lyman break at z < 2-3 Abdalla et al. 2007

  12. z/(1+z) ~ 0.006 QE (%) Wolf, Gray & Meisenheimer 2005 λ/ nm Add More Data: Narrow Filters • Improve localization and contrast of features • QSO line detection avoids catastrophic failures at z < 3 • Galaxies+QSOs: improve z • Galaxy, star, QSO, WD,… ? Wolf 2001

  13. z/(1+z) ~ 0.015 QE (%) Wolf et al. 2004 λ/ nm Add More Data: Narrow Filters • Improve localization and contrast of features • QSO line detection avoids catastrophic failures at z < 3 • Galaxies+QSOs: improve z • Galaxy, star, QSO, WD,… ? Wolf 2001

  14. S/N high p prior q p S/N medium q p S/N low q Add Priors… • Impact: ptot = pprior pcolour • Reduce rate of bimodal PDFs • Reduce larger z (up to √2) • Explicit for template models • Luminosity function, range • Mag / z extrapolation ~”ok” • Implicit for empirical models • Restricted in mag & z • Mag extrapolation wrong • Z extrapolation impossible • Representative sample?

  15. Repair (or Make) Templates Budavari et al. 2000,2001

  16. Kinney et al. templates 25% z outliers Age  Abell 901 z~0.16 Dust  Wolf, Gray & Meisenheimer 2005 Repair Models: Uncommon Objects • COMBO-17 field Abell 901/2 • Super-cluster at zspec~ 0.16 • 800 members with R < 21 • Photo-z 2002: using SEDs by Kinney et al. (1996) • 25% of S/N~100 members outliers with zphot~ 0.06 • z ~ 0.1  SHOCK! • Red spirals! • Photo-z 2003: include dust-reddened old SEDs • ~1% outliers

  17. Photometry: PSF-matched Calibration (obs. frame) Artefacts: instrumental, data reduction Error distribution, non-Gaussian systematics in Gaussian error floor Source variability Stars: RR Lyr, long-term Galaxies: supernovae Photometric Blends Transient blends by moving objects Close neighbours Line-of-sight projections, strong lenses Binaries, WD+M etc. SED composition AGN component Composite stellar populations Photometry & SEDs

  18. Basic method: Assume Gaussian PSF Convolve to worst PSF Photometry in aperture A Problems: Local PSF variations Non-Gaussian PSF (Capak et al. 2007) Special case: Gaussian aperture & PSF Stronger weight to brighter object centre Aeff = PSF A(“space-based aperture”) If A and PSF Gaussian, then Aeff Gaussian as well Minimize computations:Fix Aeff &adjust A to PSF 2A = 2eff - 2PSF (Röser & Meisenheimer 1991) Non-gaussian aperture & PSF Shapelet-based method(Kuijken 2008) PSF-Matched Photometry

  19. What We Understand by Now • Origin of local z bias • Observed-frame + rest-frame calibration • Non-flat priors: p(zph|zsp) vs. p(zsp|zph), z>0 • Origin of catastrophic outliers • Unrecognised z ambiguity in colour space • Wrong data / errors: blends, instrumental issues • Minimum z variance levels • Intrinsic SED variations • Spectral resolution

  20. E colour z=0.52 z=0.56 Sb colour Local Z Bias: Calibration • N = number of filters, i.e. independent data points • Calibration offsets in N=3 D • 1-D normalisation • 1-D z-bias • 1-D restframe SED bias • 1 out of N offset dimensions causes a photo-z bias z • More filters  smaller z (proj. component ~1/√N) • Narrow filters  small z (larger col/z on feature) • Spectroscopy with N~102..3: z without flux calibration • Few-filter photo-z’s limited by calibration accuracy • Many-filter photo-z’s limited by number and resolution of filters

  21. Catastrophic Outliers • Result from undetected ambiguities: Also: wrong data/errors • Example: see shrinking training sample • 20% sample in 1:20 ambiguities causes overall 1% unflagged outliers

  22. Intrinsic Variety: Z Error Support • Example: • QSO near g-r~1 or z~3.7 • Main signal: Ly forest in g, but SEDs >0-D family • Training sample in box • Redshift distribution:mean 3.66, rms 0.115 • RMS/(1+z) = 0.024 • Testing sample in box • RMS/(1+z) error 0.023

  23. What To Work On: Data • Define most effective & efficient data sets: • From simulations (…which don’t rule out outliers) • Describe data correctly: • Consistent apertures across bands • True photometric scatter by object • Minimiseunrecognised error sources in data: • Error floor from photometric blends & transients

  24. What To Work On: Models • Templates etc.: • Best templates, rare objects with “different SEDs” • Best priors, best extrapolation in (mag, z) • Training samples: • Discretization effects, confidence limits on random-ness • Propagation into n(z) errors and outlier risks • Size matters: “What do I need?” • Combine all approaches? • Empirical + extrapolated template model for all kinds of use

More Related