Photometric Redshifts from Empirically Derived SED Templates

Photometric Redshifts from
Empirically Derived SED Templates
Roberto J. Assef
The Ohio State University
C.S. Kochanek, M. Brodwin, M.J.I. Brown, N. Caldwell,
R.J. Cool, P. Eisenhardt, D. Eisenstein, A.H. Gonzalez,
B.T. Jannuzi, C. Jones, E. McKenzie, S.S. Murray, D.
Stern, J. Moustakas, C. Onken and J. Kollmeier

Introduction
● Large and deep photometric surveys are becoming
increasingly more common
– SDSS
– LSST
– DES
– Pan-STARRS, etc.
● Spectroscopic follow-up is expensive
● Most analysis must come from photometric data
– Redshifts
– Spectral Classifications
– K-Corrections

● Major development in photometric redshift techniques in the
last decade.
– Empirical methods.
– SED fitting.
● Empirical methods are only good within training set
boundaries. Cannot do spectral classification, K-Corrections,
etc.
● SED fitting methods rely on spectral templates. Can do
spectral analysis.
– Spectral templates are empirical or come from models.
– No templates reproduce mid-IR or UV reliably.
● Best method would be a combination of both.

● Budavari et al. (2000) and Csabai et al. (2000) proposed a
method that combines both approaches.
● Idea is to fit for the templates from a broad-band photometry
training set.
● Training set must be deep enough in redshift to optimize
wavelength coverage.
● We have determined low resolution templates from NUV to
mid-IR that accurately reproduce galaxy properties.
– Training set from NOAO Deep Wide-Field Survey
● Most results in this talk already published in Assef et al. 2008,
ApJ, 676, 286.

Photometric Data
● NOAO Deep Wide-Field Survey (NDWFS)
– Cetus and Boötes fields
● Available Imaging
– Bw, R, I and K (NDWFS)
– J and Ks (FLAMEX)
– 4 IRAC bands (Shallow Survey, SDWFS)
– z-Band (zBoötes)
– FUV and NUV (GALEX)
– MIPS 24µm (MAGES)
– X-Ray (XBoötes)
– Radio (FIRST, NVSS, WENSS)

Spectroscopic Data
● AGN and Galaxy Evolution Survey (AGES,
Kochanek et al. in prep.)
– Low resolution spectra
– ~17000 Galaxies
● I

Fitting the Templates
● Based on the method proposed by Budavari (2000)
and Csabai (2000).
– Find PCA components from photometry
● Assumption: Every galaxy SED is a non-negative
linear combination of:
– Old stellar population (Elliptical)
– Continuously star-forming (Spiral)
– Starburst
– Post-Starburts (E+A, alternative model)
● Non-negative combinations avoid unphysical
SEDs

Fitting the Templates
● Low Resolution
– Each template is divided in 160 wavelength bins
– Each bin is convolved with the filter band pass and fit
to the data
● Initial guesses
– Coleman, Wu & Weedman (CWW,1980)
– Bruzual & Charlot (2003) UV & mid-IR
– M83 & VC1003 PAH features (Devriendt et al. 1999)
● Smoothing Templates
– Resulting templates must be smooth
● Photometric zero point / aperture corrections

Resulting Templates
● Significant
deviations from
initial SEDs
● E has lower UV
● SF templates have
much higher PAHs
● E+A gives a much
better overall fit,
but...

Photo-zs for NDWFS Galaxies
● 4T slightly worse
then 3T
● 90% of objects
within blue lines
● Simple
luminosity prior
from LCRS
luminosity
function

● Error distribution has very non-gaussian tails.
● When clipping to 95%, dispersion similar to 68.3% interval.
● Average median deviation small.

Planned and Current Updates
● GALEX and MIPS data added
– Effective wavelength from 0.03 to 30µm.
● Added upper
limits for most bands
● Deeper IRAC
imaging (SDWFS)
● Variable IGM
absorption

AGN Template
● Added an AGN template following the same
procedure
– Assumption: All AGNs have the same SED + variable
reddening
● Initial guess
– Power-law (UV-optical) + black body (IR)
● Goals
– Photo-zs for AGNs
– AGN identification
– Luminosity Functions
– Eddington Ratios

Templates

Photo-zs for AGNs in Extended
Sources
● Bright hosts
provide photo-z
accuracy
● σ = 0.2
σ 95% = 0.1
● Outliers
dominate the
formal
dispersion

Photo-zs for AGNs in Point
Sources
● σ = 0.75
σ 95% = 0.60
● SED is too flat
for good photo-z
● Work in
progress!

Applications
● AGN selection
– Recover most AGNs found by other methods (IRAC
colors, MIPS, X-Ray, Radio, line-ratios)
– Classified ~100 new objects as AGNs in AGES data set
● AGN LF
– Limit to I

Summary
● We have derived a basis of empirical SED
templates for galaxies and AGNs.
● Photometric redshifts and K-corrections code is
publicly available.
● Accurate galaxy photo-zs can be calculated with
this technique.
● Templates from Assef et al. (2008) are being
updated to include GALEX, MIPS and new IRAC
data.