The DECADE cosmic shear project III: validation of analysis pipeline using spatially inhomogeneous data

TL;DR

This work presents a robust end-to-end cosmic shear analysis pipeline for the DECADE dataset, combining Metacalibration shapes, SOMPZ redshift estimates, and an analytical CosmoCov covariance within a HMCode2020-based nonlinear power spectrum framework and a TATT intrinsic alignment model. Through comprehensive validation on simulated data and an extensive suite of forty-six subset tests that explore spatial inhomogeneities and galaxy-property splits, the authors show that cosmological constraints, particularly in the $S_8$ direction, remain consistent within $1\sigma$ to $2\sigma$ across all cases. They demonstrate the effectiveness of scale cuts in mitigating baryonic and IA systematics and confirm the resilience of their analysis to modeling choices, priors, and sampling methods. The study confirms that existing weak lensing analysis frameworks can perform reliably on spatially inhomogeneous datasets, supporting broader use of heterogeneous image data for cosmology and informing the design of future survey pipelines.

Abstract

We present the pipeline for the cosmic shear analysis of the Dark Energy Camera All Data Everywhere (DECADE) weak lensing dataset: a catalog consisting of 107 million galaxies observed by the Dark Energy Camera (DECam) in the northern Galactic cap. The catalog derives from a large number of disparate observing programs and is therefore more inhomogeneous across the sky compared to existing lensing surveys. First, we use simulated data-vectors to show the sensitivity of our constraints to different analysis choices in our inference pipeline, including sensitivity to residual systematics. Next we use simulations to validate our covariance modeling for inhomogeneous datasets. Finally, we show that our choices in the end-to-end cosmic shear pipeline are robust against inhomogeneities in the survey, by extracting relative shifts in the cosmology constraints across different subsets of the footprint/catalog and showing they are all consistent within $1σ$ to $2σ$. This is done for forty-six subsets of the data and is carried out in a fully consistent manner: for each subset of the data, we re-derive the photometric redshift estimates, shear calibrations, survey transfer functions, the data vector, measurement covariance, and finally, the cosmological constraints. Our results show that existing analysis methods for weak lensing cosmology can be fairly resilient towards inhomogeneous datasets. This also motivates exploring a wider range of image data for pursuing such cosmological constraints.

Figures (11)

• Figure 1: A set of maps from the DECADE (top) and DES Y3 (bottom) datasets. From left to right, we show the (apparent) magnitude limit map in the $i$-band (estimated at $S/N = 10$ within a $2\hbox{$^{\prime\prime}$}$ aperture; see paper1), the overdensity in source galaxy number counts $\delta_g$, and the same after randomizing the galaxy positions within the survey mask. All maps are shown at $\texttt{NSIDE}=1024$. While the DECADE magnitude limit map has significantly more structure relative to that of DES Y3, the DECADE source-galaxy sample has much less inhomogeneity as the DECADE and DES Y3 source-galaxy samples are defined with certain cuts (specifically $m_i < 23.5$, where $m_i$ is the $i$-band magnitude of the source galaxy) that make the variation in $\delta_g$ similar across the two surveys. The remaining large-scale variations, found towards the eastern/western edges of the footprint, are due to the closer proximity of the DECADE footprint to the Galactic plane. The right-most panels show, as a simple reference, the number-density fluctuations for a sample with randomized positions. There are still some small fluctuations due to the mask structure on small scales; the mask is defined at $\texttt{NSIDE}=4096$paper1.
• Figure 2: Forecasted constraints on $\Omega_{\rm m}$, $\sigma_8$ and $S_8$ using a simulated data vector corresponding to the DECADE dataset (red). The cross-hair indicates the input cosmology, showing that our pipeline adequately recovers the input values. The DES Y3 constraints from Secco2021 and Amon2021 are overlaid for comparison. To enhance comparisons of constraining power, we shift the DES posteriors to be centered on the simulated data vector's input cosmology. Both the fiducial and the optimal constraints from DES include information from shear ratios Sanchez2022, resulting in tighter constraining power and slightly different degeneracy directions.
• Figure 3: Constraints on $S_8$, $\Omega_{\rm m}$ and $\sigma_8$ from simulated data vectors, under different analysis choices. This series of tests (detailed in Section \ref{['sec:validation']}) ensures that the modeling choice we adopted for the data is robust, including the mitigation of small-scale baryonic effects, the assumptions of the nonlinear power spectrum and intrinsic alignment, the priors on the nuisance parameters, and the choice of sampling approaches in the inference process. In this plot, as well as later Figures \ref{['fig:SurveySplits']} and \ref{['fig:SurveySplitsProperties']}, the constraints shown are the marginalized mean, 1$\sigma$ and 2$\sigma$ uncertainties. The vertical shaded region correspond to the 1$\sigma$ (dark) and 2$\sigma$ (light) intervals of the fiducial case. Here, we also show the true input value to the simulated data vectors with the vertical dotted line. These are $\Omega_{\rm m} = 0.27, \sigma_8 = 0.846, S_8 = 0.80$. Note that line 1 ("Fiducial") corresponds to the marginal constraints shown in Figure \ref{['fig:fiducial']}.
• Figure 4: The constraints on $S_8$ from different subsets of the DECADE footprint, while consistently rederiving all calibrations, redshift distributions, covariance matrices, etc. The first, "Fiducial" constraint is using data the entire dataset. The rest, from top to bottom, split the survey based on: airmass, seeing (PSF width), differential chromatic refraction (DCR) effects in the R.A., Dec., and $e_{1,2}$ coordinates, the east/west regions, the $\rm Dec. \gtrless -30$ region where photometric calibrations used Pan-STARRS/SkyMapper, the sky background brightness and variations, exposure time, magnitude limit, number of exposures, interstellar extinction coefficients, and Gaia stellar number density. Analyses with $X^+$ ($X^-$) use the area with higher (lower) values than the median. The data vectors in each subset have essentially independent noise realizations and so the cosmology constraints can exhibit relative shifts of up to $3\sigma$ due to just statistical fluctuations. We show the parameter constraints relative to the fiducial estimate, with thick and thin horizontal lines denoting the $68\%$ and $95\%$ intervals for each constraint, respectively. For visibility reasons, we do not show the $95\%$ interval for the "SkyMapper" subset alone. Most (all) deviations are within $1\sigma$ ($2\sigma$) of the fiducial results. The constraints from a subset and its complement are always within $2\sigma$ as well. No split exhibits shifts in cosmology that are larger than the expected variation from shape noise and cosmic variance. The footprints corresponding to the splits are shown in Figure \ref{['fig:SplitsMaps']}.
• Figure 5: The distribution of magnitude limits in the $i$-band (from a HealPix map of $\texttt{NSIDE} = 4096$) for the two subsets of the DECADE footprint split by magnitude limit, and for the DES Y3 footprint. The ranges in the figure denote the $1\%$ and $99\%$ values of the distribution. The Maglim$^+$ region (blue) has twice the width --- defined as the difference between the 99% and 1% intervals shown in the Figure --- compared to those of the other two samples. There is a slight, minimal overlap in the two DECADE distributions as we plot values from the $\texttt{NSIDE} = 4096$ maps, whereas the selection mask is defined using the downgraded $\texttt{NSIDE} = 1024$ maps (see Section \ref{['sec:SpatialInhomog:splits']} for details). The overlap has no impact on our qualitative and quantitative discussions.