Photometric Redshifts



  • Chris Morrison (University of Washington)
  • Will Hartley (UCL)

Animation showing a template elliptical galaxy spectrum (solid line) at various redshifts, overlaid on the transmission functions of LSST filters (u, g, r, i, z and Y).  As the spectrum is redshifted, the amount of flux through each filter changes, providing information on a galaxy's redshift from photometric measurements alone.



All LSST probes of dark energy and related physics rely on determining the behavior of some quantity as a function of redshift z. Distances, the growth rate of dark matter fluctuations, and the expansion rate of the Universe are all functions of redshift that can be readily calculated given a cosmological model; dark energy experiments then constrain cosmological parameters by measuring observables dependent upon these functions. However, it is completely infeasible with either current or near-future instruments to obtain redshifts via spectroscopy for so large a number of galaxies, so widely distributed, and extending to such faint magnitudes, as those studied by LSST.

Hence, LSST will primarily rely on photometric redshifts – i.e., estimates of the redshift (or the probability distribution of possible redshifts, p(z)) for an object based only on imaging information, rather than spectroscopy (Spillar 1985; Koo 1999). Effectively, multiband (e.g., ugrizy) imaging provides a very low-resolution spectrum of an object, which can be used to constrain its redshift. Because flux from a relatively wide wavelength range is being combined for each filter, imaging provides a higher signal-to-noise ratio than spectroscopy; however, broader filters provide cruder information on the spectrum, and hence on z or p(z).

For most LSST probes of dark energy (BAO is the primary exception), what will matter most is not the precision with which individual photometric redshifts are measured, but rather the degree to which we understand the actual redshift distributions of LSST samples; if photo-z ’s are systematically biased or their errors are poorly understood, dark energy inference will be biased as well (e.g., because we are effectively measuring the distance to a different redshift than is assumed in calculations). For LSST, it is estimated that the mean redshift and z dispersion for samples of objects in a single photo-z bin must be known to ∼2 × 10−3(1 + z) (Zhan & Knox 2006; Zhan 2006b,b; Tyson 2006) for dark energy inference not to be systematically degraded. As a result, this working group’s efforts will be primarily focused on making sure that the photometric redshifts we obtain are accurate and well-understood.


Featured Projects