Redshift Assessment Infrastructure Layers (RAIL): Rubin-era photometric redshift stress-testing and at-scale production

Abstract

Virtually all extragalactic use cases of the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) require the use of galaxy redshift information, yet the vast majority of its sample of tens of billions of galaxies will lack high-fidelity spectroscopic measurements thereof, instead relying on photometric redshifts (photo-$z$) subject to systematic imprecision and inaccuracy best encapsulated by photo-$z$ probability density functions (PDFs). We present the version 1 release of Redshift Assessment Infrastructure Layers (RAIL), an open source Python library for at-scale probabilistic photo-$z$ estimation, initiated by the LSST Dark Energy Science Collaboration (DESC) with contributions from the LSST Interdisciplinary Network for Collaboration and Computing (LINCC) Frameworks team. RAIL's three subpackages provide modular tools for end-to-end stress-testing, including a forward modeling suite to generate realistically complex photometry, a unified API for estimating per-galaxy and ensemble redshift PDFs by an extensible set of algorithms, and built-in metrics of both photo-$z$ PDFs and point estimates. RAIL serves as a flexible toolkit enabling the derivation and optimization of photo-$z$ data products at scale for a variety of science goals and is not specific to LSST data. We thus describe to the extragalactic science community, including and beyond Rubin the design and functionality of the RAIL software library so that any researcher may have access to its wide array of photo-$z$ characterization and assessment tools.

Figures (6)

• Figure 1: Overview of structure of the RAIL codebase. The core module provides building blocks for RAIL with dependencies on several existing DESC software packages, i.e., ceci, qp, and tables_io, as well as basic utilities and tools shared across all RAIL modules. The main functionality of RAIL has a tripartite structure enabling experiments to optimize photo-$z$ data products, namely, creation, estimation, and evaluation (bold blocks). Along with these modules, we also introduce the core functionality and examples in the main body of the paper (orange blocks). Utilities, tools, and the major dependencies of RAIL (green blocks) are introduced in the Appendices.
• Figure 2: The workflow of the RAIL.creation forward modeling subpackage. Input and output data are represented by rectangles, and RAIL stages are represented by ovals. A typical creation pipeline starts with training a creation ('generative') model from either a reference catalog (often simulations), such as in the case of PZFlow, or from the template SEDs drawn from a prior, such as in the case of DSPS/FSPS. The empirical or explicit joint distribution of true redshifts with other catalog properties ($\mathbf{p}$) is then computed. A mock truth catalog ('underlying galaxy catalog') is then drawn from the creation model. This catalog is then degraded by the degradation stages, such that it mimics a noisy observed catalog ('biased galaxy sample').
• Figure 3: The workflow of a typical estimation RAIL pipeline. The training data and prior information are fed into the Informer, which generates the photo-$z$ model. Then the model is combined with the test dataset to produce the photo-$z$ PDFs. Optional point estimate of the PDF can be requested during the estimation. Similarly, to Fig. \ref{['fig:creation']}, input and output data are represented by rectangles, and RAIL stages are represented by ovals.
• Figure 4: The color-redshift scatter of CosmoDC2 galaxies before (blue) and after (red) applying a series of degraders, which are described in Section \ref{['sec:examples:gs']}. We can see that the population shown in red has a different distribution in color-redshift space compared to the population shown in blue.
• Figure 5: The probability density function of the redshift of a single galaxy from CosmoDC2, as estimated by three methods. The vertical line shows the true redshift.