Normalizing Flows Evaluated for Tomographic Redshift Calibration Using DESI DR1 and Legacy Surveys DR10
What Was Observed (Empirical Data)
Researchers evaluated conditional normalizing flows against random forest and quantile random forest models to reconstruct tomographic redshift distributions, N(z), for weak lensing analysis. The models were trained on 1.1 million galaxy and quasar targets and tested on 2.0 million sources in a non-overlapping sky region, combining spectroscopic redshifts from the Dark Energy Spectroscopic Instrument (DESI) Data Release 1 with optical and infrared photometry from the Legacy Surveys Data Release 10.
The Reality Check (Anti-Hype Analysis)
This study is a methodological calibration benchmark for statistical redshift estimation rather than a new observational discovery of cosmic structure or dark energy properties. Because the models are trained and tested strictly on DESI target selections, their accuracy remains bound by spectroscopic sample completeness and photometric noise limits, and does not eliminate the need for spectroscopic calibration in fainter, unrepresented galaxy populations.
PRIMARY SOURCE:
arXiv Astrophysics (astro-ph)