Euclid: Galaxy SED reconstruction in the PHZ processing function: impact on the PSF and the role of medium-band filters

TL;DR

This paper addresses how galaxy SED reconstruction within Euclid's PHZ processing function impacts the chromatic PSF used in weak lensing analyses. It compares data-driven Gaussian processes and physics-based template-fitting approaches, introducing a template-based flux correction and a novel SED metric $\,\mathcal{M}_{\lambda}\,$ that directly maps reconstruction bias to PSF quadrupole errors. A hybrid strategy leveraging both GP and TF meets the Euclid PSF accuracy in most redshift bins, and inaccuracies are analyzed against a target bias of $b = 3.5 \times 10^{-4}$. The study also explores how a new set of evenly spaced Subaru/HSC medium-band filters could improve SED reconstruction, offering a practical path toward quasi-spectroscopic VIS coverage for future improvements.

Abstract

Weak lensing surveys require accurate correction for the point spread function (PSF) when measuring galaxy shapes. For a diffraction-limited PSF, as arises in space-based missions, this correction depends on each galaxy SED. In the Euclid mission, galaxy SED reconstruction, a tasks of the photometric-redshift processing function (PHZ PF), relies on broad- and medium-band ancillary photometry. The limited wavelength sampling of the Euclid VIS passband and signal-to-noise ratio may affect the reconstruction accuracy and translate into biases in the weak lensing measurements. In this study, we present the methodology, which is employed in the Euclid PHZ PF, for reconstructing galaxy SEDs at 55 wavelengths, sampling the VIS passband every 10 nm, and we assess whether it fulfils the accuracy requirements imposed on the Euclid PSF model. We employ both physics- and data-driven methods, focusing on a new approach of template-based flux correction and Gaussian processes, and we introduce an SED metric whose bias propagates into PSF quadrupole moment errors. Our findings demonstrate that Gaussian processes and template fitting meet the requirements only in specific, but complementary, redshift intervals. We therefore propose a hybrid approach, which leverages both methods. This solution proves to be effective in meeting the Euclid accuracy requirements for most of the redshift range of the survey. Finally, we investigate the impact on the SED reconstruction of a new set of 16 evenly-spaced medium-band filters for the Subaru telescope, providing quasi-spectroscopic coverage of the VIS passband. This study shows promising results, ensuring accurate SED reconstruction and meeting the mission PSF requirements. This work thus provides not only the methodological foundation of galaxy SED reconstruction in the Euclid PHZ PF, but also a roadmap for future improvements using a new medium-band survey.

Figures (10)

• Figure 1: Median fractional bias in the SED metric arising from the interpolation of galaxy SEDs from coarse sampling down to 1 nm sampling using linear (pink line) and cubic spline (dark green) interpolation. The dotted line represents the requirement set on the PSF model.
• Figure 2: Set of baseline SED templates used to build the galaxy population model, displayed in the optical to infrared window. Each galaxy is simulated through a linear combination of the elements of this basis.
• Figure 3: Coverage of the filter obtained using the Subaru SC medium band filters, and the broad pass-bands $u$ and $z$ from DECam and $Y$ from VISTA VIRCam. The vertical dashed lines are placed at the weighted central wavelength of each filter.
• Figure 4: Example of SED reconstruction for a galaxy randomly drawn from the simulated sample. The true SED is plotted in green. Left panel: example of SED reconstruction using template-based flux correction (TFC). The black points are the true fluxes known from simulations, and the orange-contoured white diamonds represent their best-fit estimates. The grey line shows the SED reconstructed with TF only, and the pink line is the colour correction. True and reconstructed SEDs are shifted in flux to enhance readability. Right panel: example of SED reconstruction using GP regression starting from the subset of medium-band simulated fluxes. The blue line (with the respective $95\%$ confidence intervals) shows the inferred galaxy SED using the Matérn kernel.
• Figure 5: Bias in SED metric as a function of photometric redshift, calculated using different methods for SED reconstruction: colour-space corrected TF (TFC; magenta) and GP (blue). The plot displays the bias averaged over single redshift bins. The dashed grey line represents the results obtained with TF before colour correction. The dotted horizontal line highlights the contribution of the PSF wavelength-dependency to the bias accounted for the PSF model.