DECam Local Volume Exploration Survey (DELVE)

• DELVE is a wide-area optical survey using DECam to systematically map faint stellar systems and substructures in the Local Volume.
• It applies advanced data processing, including machine learning for photometric redshifts and weak lensing analysis, ensuring precise calibration and source detection.
• The survey integrates new and archival data to uncover ultra-faint satellites, stellar streams, and strong lens candidates, enhancing our understanding of galaxy evolution.

The DECam Local Volume Exploration Survey (DELVE) is a wide-area, multi-epoch optical imaging campaign executed with the Dark Energy Camera (DECam) on the Blanco 4-meter telescope at CTIO, designed to systematically characterize faint resolved stellar systems and substructures in the Local Volume (LV). DELVE integrates new observations with archival DECam data, comprising a major data resource for studying the census and properties of ultra-faint galaxies, stellar streams, globular clusters, and lensing phenomena across 21,000 deg² of high-Galactic-latitude sky to sensitive depths in the griz filters.

1. Survey Design and Data Management

DELVE leverages DECam, a wide-field imager with 62 science CCDs covering 3 deg² per exposure, to achieve deep photometric coverage ($g=24.3$, $r=23.9$, $i=23.5$, $z=22.8$\,mag at $5\sigma$) over 17,000 deg² in all four bands (Drlica-Wagner et al., 2022). Data acquisition combines targeted observations and the assimilation of more than 270 archival programs, with ~160,000 individual exposures processed for DR2.

Raw data undergo parallelized detrending and calibration steps in the DES Data Management (DESDM) pipeline (Mohr et al., 2012). This includes:

• CCD-level bias and overscan subtraction, crosstalk correction, flat-fielding, flux normalization, and pupil ghost mitigation.
• Astrometric alignment utilizing the AstrOmatic tool SCAMP, implementing third-order distortion models for individual CCDs and overall focal plane, cross-matched to reference catalogs (2MASS).
• Photometric calibration using a dual approach: standard star zeropoints and self-calibration through overlapping stellar loci, anchored to NIR absolute scales and adjusting for per-exposure systematics.

Point Spread Function modeling employs PSFEx, capturing spatial PSF variations via second-order polynomial expansions. Coaddition of exposures is preceded by PSF homogenization, convolving each image to a common seeing kernel to facilitate uniform cataloging and optimized star-galaxy separation (Mohr et al., 2012).

2. Satellite Census and Discovery of Faint Systems

DELVE has enabled the detection and characterization of numerous ultra-faint satellites and star clusters in the Milky Way periphery. Discovery methodology employs matched-filtering in color–magnitude space, dividing the sky into HEALPix pixels, applying isochrone filters (e.g., PARSEC models with ages $\sim12$–$13$\,Gyr, [Fe/H] $<-1.4$), and identifying statistically significant overdensities (Cerny et al., 2022, Mau et al., 2019, Cerny et al., 2021).

Recently discovered systems span a wide range in luminosity and size:

• Extended candidates such as Eridanus IV ($M_V=-4.7\pm0.2$, $r_{1/2}=75^{+16}_{-13}$\,pc, $\epsilon=0.54\pm0.10$, $D_\odot\sim77$\,kpc) (Cerny et al., 2021).
• Compact clusters (e.g., DELVE 3, 4, 5) with $r_{1/2}<10$\,pc and $M_V\sim+0.4$ to $-1.3$; more diffuse galaxies (Boötes V, Virgo II, Leo Minor I) with $r_{1/2}>15$\,pc and $M_V>-3.2$ (Cerny et al., 2022).

Property estimation combines maximum-likelihood fitting to Plummer or Sersic profiles and Markov Chain Monte Carlo sampling, comparing observed color-magnitude diagrams to population synthesis isochrones.

Proper motions from Gaia DR3 confirm the dynamical associations and possible group infall signatures for several satellites (e.g., Centaurus I, DELVE 2) (Mau et al., 2019, Cerny et al., 2020). Tidal disruption features and extra-tidal envelopes are investigated through spatial overdensity mapping and dynamical modeling.

3. Multi-Scale, Multi-Wavelength Science

DELVE's catalog (2.5 billion objects in DR2) supports a broad range of analyses:

• Mapping stellar substructures and streams which record the accretion and tidal disruption history of the Local Group.
• Cross-correlation with radio HI surveys (e.g., LVHIS (Koribalski et al., 2019)) reveals the spatial and dynamical relationship between gas distribution and stellar populations, crucial for understanding star formation in dwarf galaxies and environmental quenching.
• Synergies with auxiliary datasets (WISE, SDSS, Gaia, NSC) enable robust photometric, astrometric, and proper motion diagnostics.

Data products, including cutout services and cross-matched catalogs, are accessible via the NOIRLab Astro Data Lab platform (Drlica-Wagner et al., 2022).

4. Advanced Computational Methods: Photometric Redshifts and Lensing Candidates

DELVE has pioneered probabilistic photometric redshift (photo-$z$) estimation using deep learning, combining Recurrent Neural Networks (Legendre Memory Unit architectures) with Mixture Density Networks (MDN) (Teixeira et al., 27 Aug 2024). The model produces redshift PDFs as mixtures of Gaussians:

$$PDF(z) = \sum_{i=1}^{C} \alpha_i \cdot \mathcal{N}(z; \mu_i, \sigma_i)$$

Performance is evaluated using bias ($-0.0013$), scatter ($0.0293$), and outlier fraction ($5.1\%$), plus calibration diagnostics such as the Probability Integral Transform and odds distributions. Data storage challenges are addressed with autoencoders, reducing PDF parameter arrays by a factor of six and accelerating generation time by a factor of eight.

DELVE data underpins the identification of strong lens candidates using convolutional neural networks. A pipeline trained on simulated and real images flags lensing morphologies in galaxy cutouts, yielding 581 candidates (55 Grade A, 149 Grade B, 377 Grade C; 562 new) (Zaborowski et al., 2022). Einstein radii are estimated from image separations; quadruply lensed quasar candidates are highlighted.

5. Weak Lensing, Cosmic Shear, and Cosmology

The DELVE Early Data Release 3 (EDR3) presents the DECADE weak lensing shape catalog of 107 million galaxies over 5,412 deg², processed with DESDM and calibrated using Metacalibration (Anbajagane et al., 24 Feb 2025). Shear estimation is corrected for selection and noise biases via image simulations:

$$\gamma_i^{\rm est} = (1+m)\,\gamma_i^{\rm true} + c$$

with $m = (-2.454 \pm 0.124)\times10^{-2}$, and effective source density $4.59\, \textrm{arcmin}^{-2}$ for cosmology.

Systematic control is achieved through PSF tests (brighter-fatter, color effects), null tests (B-modes, star cross-correlations) and aggressive masking of affected regions. The inhomogeneity of the catalog is mitigated by statistical calibration and simulation-based response modeling.

6. Future Directions and Impact

DELVE data, via its combination of depth, area, and precision calibration, establishes a critical foundation for Galactic and extragalactic science in the LSST era, especially for ultra-faint satellite searches, weak lensing cosmology, stellar stream mapping, and strong lens demographic studies. The methodological innovations in photometric redshift PDFs and machine learning classification provide scalable solutions for massive datasets anticipated from Rubin Observatory and the Roman Space Telescope.

The survey’s results continue to inform models of galaxy formation and evolution, constrain small-scale dark matter structure, and refine our understanding of the quenched fraction and abundance of satellites among Milky Way analogs. In tandem with auxiliary LV surveys (ELVES, LVHIS, DESI Legacy Imaging), DELVE advances a comprehensive and systematic census of Local Volume structure, offering a statistically complete, multi-wavelength framework for investigating the physics at play in nearby galaxies and star clusters.