ZTF DR23: Synoptic Survey Observations

• ZTF DR23 is a major public release from the 48-inch Schmidt telescope, featuring a 47 deg² camera for high-cadence, time-domain optical observations.
• The data processing pipeline employs modular, parallelized software with advanced image subtraction and machine learning techniques for rapid and precise transient classification.
• The release underpins diverse scientific applications, enabling detailed studies of explosive transients, variable stars, and solar system objects, while supporting multi-messenger follow-up.

The Zwicky Transient Facility (ZTF) DR23 Observations encompass a major public data release from a next-generation optical synoptic survey operating on the 48-inch Samuel Oschin Schmidt telescope at Palomar Observatory. ZTF builds upon the legacy of the Palomar Transient Factory, introducing a 47 deg² camera and an architecture engineered for high-cadence, wide-area time-domain astronomy. DR23 includes calibrated images, source catalogs, alert streams, and light curves, enabling systematic studies of explosive transients, variable stars, Active Galactic Nuclei (AGN), Solar System objects, and electromagnetic counterparts to multi-messenger astrophysical sources. The observational dataset is characterized by its sky coverage, temporal depth, photometric and astrometric precision, and rapid data dissemination mechanisms, making it foundational for both targeted and statistical analyses of the dynamic universe.

1. Survey Design, Instrumentation, and Operations

ZTF is distinguished by its system-level optimization for volumetric transient discovery rate, achieved through hardware, scheduling, and operational innovations (Bellm, 2014, Bellm et al., 2017, Dekany et al., 2020). The survey camera covers the full 47 deg² focal surface of the Schmidt telescope, employing 16 e2v 6k×6k CCDs (600 megapixels total, 1″/pixel)—an increase in areal coverage by a factor of ~6.5 over the prior PTF configuration. Median image quality is ~2″ FWHM, with the plate scale matched for proper Nyquist sampling. A 30-second exposure yields a 5σ limiting magnitude of r~20.4–20.6, and an optimized duty cycle (10 s readout, 15 s net overhead) supports a volumetric survey rate 𝑉̇ ≳ 3×10⁴ Mpc³/s for canonical SNe at M = –19 (Bellm, 2014, Laher et al., 2017, Masci et al., 2019).

Survey operations are a hybrid of rolling public and boutique programs, with the main public surveys imaging the northern sky every three nights in g/r and the Galactic plane nightly (Graham et al., 2019). A novel Integer Linear Programming scheduler maximizes coverage and cadence, factoring in time-varying airmass ($X = \sec z$), seeing, and sky brightness, optimizing both spatial volume and data quality per exposure by weighting fields according to $V = 10^{0.6(m_{\rm lim} - 21)}$ (Bellm et al., 2019). The robotic system includes a KUKA LBR iiwa arm for rapid, multi-band filter exchanges and closed-loop hexapod focus control.

2. Data Processing Architecture and Alert Generation

ZTF’s Science Data System (ZSDS) employs scalable hardware and modular, parallelized software pipelines for real-time and archival processing (Laher et al., 2017, Masci et al., 2019). The ingestion chain splits CCD multi-extension FITS into quadrants, applies bias and flat corrections, and performs astrometric calibration against Gaia DR1, typically achieving 45–85 milliarcsec residuals per axis at S/N ≥ 10. PSF and aperture photometry catalogs are generated for every exposure, with photometric zero-points tied to Pan-STARRS1 via

$m_{\rm diff} = ZP_f + c_f \times \mathrm{PS1clr}$

where $ZP_f$ and $c_f$ are fit for each image.

Reference image stacks (each from 15–40 high-quality epochs) are maintained per field/CCD/filter as subtraction templates. The ZOGY optimal image subtraction algorithm is used to generate difference images and match-filtered S/N maps, maximizing point-source transient detectability (Masci et al., 2019). Candidates are parameterized and classified using machine learning (real–bogus score migration), with alert packets serialized in Avro and distributed via a Kafka backbone; 8–13 min typical end-to-end latency supports prompt electromagnetic follow-up (Graham et al., 2019).

Data products—raw, calibrated, difference, and reference images; catalogs; light curves; and alert packets—are persistently archived with GUI and API programmatic access through IRSA at IPAC (Masci et al., 2019).

3. Science Applications: Transients, Variables, Solar System Objects

Explosive Transients and Supernovae

ZTF DR23 supports discovery and systematic paper of supernovae (SNe), fast-declining objects (“dirty fireballs,” GRB orphans), and electromagnetic counterparts to gravitational wave triggers (Bellm, 2014, Bellm et al., 2017, Graham et al., 2019). Early light curve coverage (with identification often within 24–48 hr post-explosion) enables detailed constraint of progenitor radii, shock-breakout physics, and circumstellar material via “flash spectroscopy” (Bellm et al., 2017). The Bright Transient Survey (BTS) within DR23 defines a magnitude-limited (m<19) sample with >93% spectroscopic completeness at m<18.5, and the SN demographics reveal well-characterized duration–luminosity correlations and host–environment statistics (Perley et al., 2020). Superluminous SNe, TDEs, and lensed SNe are also systematically catalogued (Graham et al., 2019, Brightman et al., 2020).

Variable Stars and Compact Binaries

High cadence and deep sampling yield light curves for >1 billion sources, facilitating population studies of variables, eclipsing binaries, Cepheids (including ~71% recovery of OGLE-registered double-mode Cepheids), and cataclysmic variables—even in crowded Galactic fields (Shah et al., 2022, Szkody et al., 2020). Dedicated filters and photometric flags are employed to identify variable white dwarfs (e.g., ZZ Ceti, GW Vir) via combined Gaia/ZTF analysis and Lomb–Scargle periodogram clustering (Jestin et al., 18 Sep 2025).

Solar System Science

Rapid cadence naturally recovers moving objects. The ZMODE engine links transient “streak” detections to report NEOs and main-belt asteroids, while a dedicated pipeline attributes linear features in difference images. Astrometric accuracy (tens of milliarcsec) facilitates orbit fits and NEO hazard characterization. Twilight images are increasingly affected by LEO satellite streaks (e.g. Starlink), but masking minimizes data loss (<0.2% pixel loss, even under full-constellation scenarios); notably, satellite brightness mitigation via visors produces a >4.6× reduction in g/r/i (Mroz et al., 2022).

4. Systematic Multi-Messenger Follow-up and Survey Synergies

ZTF DR23 provides foundational optical coverage for high-energy and gravitational-wave transient follow-up. Rapid Target-of-Opportunity imaging is routinely triggered for IceCube neutrino alerts (median latency ~12 hr) (Stein et al., 2022) and gravitational-wave triggers (e.g., S250206dm; 68% localization coverage, >10% efficiency for kilonovae with $M < –17.5$) (Ahumada et al., 1 Jul 2025). Constraints on the optical luminosity function of high-energy neutrino sources are set at $M < –21$ (<87%) and $M < –22$ (<58%), with direct limits on bright AGN flare contributions to neutrino flux (<26%) (Stein et al., 2022). Dedicated archival filtering and blind analyses have placed stringent constraints on plausible optical counterparts to IceCube multiplet events, ruling out most TDE and SLSN scenarios absent rapid follow-up (Toshikage et al., 7 Apr 2025).

DR23 is also interoperable with contemporaneous time-domain surveys (e.g. DECam DDFs), sharing Kafka-based alert and broker infrastructure, facilitating cross-matching, and enabling global, real-time multi-facility time-domain astronomy (Graham et al., 2022).

5. Data Quality, Calibration, and Survey Performance

The image quality, calibration, and coverage metrics associated with DR23 set benchmarks for wide-field time-domain imaging. Astrometric calibration achieves 45–65 milliarcsec RMS (down to 30 mas at $S/N>40$) by leveraging Gaia DR1 and two-pass SCAMP solutions (Masci et al., 2019). Photometric repeatability is 8–25 mmag for bright, unsaturated stars, with cross-survey zero-point agreement (to Pan-STARRS1) at 1–2%. Relative photometric refinement via matchfiles can improve photometric scatter by 10–20%. Cadence and depth are consistently maintained by automated scheduling optimizations, with sequence-completion fractions and field revisit rates exceeding most legacy surveys (Bellm et al., 2019). Filtering, differencing, and ML-vetted alert steps yield high purity in detected transient streams.

Survey efficiency remains high (>70% open shutter duty cycle; >3,700 deg²/h surveyed), and the modular, containerized infrastructure supports robust data throughput; nightly volume routinely reaches several terabytes with alert stream rates sustaining >1 million events/night (Laher et al., 2017, Masci et al., 2019).

6. Collaborative Structure and Future Directions

ZTF DR23 is the product of an international collaboration spanning Caltech, University of Washington, Oskar Klein Centre, Weizmann Institute, and partner institutions, with operations co-funded through NSF, DOE, international, and private sources (Graham et al., 2019). Both public and partnership surveys are executed in parallel, with data releases on a regular cadence. Internal infrastructure (brokerage, SEDM for follow-up spectroscopy) is harmonized with community brokers and larger facilities (e.g. LSST, ATLAS). The survey serves both as a science engine and a technology pathfinder, refining classification algorithms, machine learning pipelines, and event-broker protocols essential for scaling to LSST-class data rates.

The design and performance of ZTF DR23 directly inform the next decade of time-domain astronomy, demonstrating both the science impact and system requirements for future surveys with even higher volumetric rates, deeper limiting magnitudes, and lower latencies.

7. Summary Table: ZTF DR23 Core Technical Metrics

Attribute	Typical Value	Context
Camera FoV	47 deg²	16 x 6k × 6k CCD mosaic (Bellm, 2014, Laher et al., 2017)
Pixel Scale	1 arcsecond/pixel	Matches median seeing ~2″ FWHM
Exposure Time	30 s	Optimized by survey speed (Bellm, 2014)
Limiting Magnitude	g~20.8, r~20.6 (5σ, 30s)	(Masci et al., 2019, Roestel et al., 2019)
Readout + Overhead	10 s (read) + 15 s (slew, settle)	(Laher et al., 2017, Dekany et al., 2020)
Survey Rate	~3,760 deg²/h (areal), 3–12 × 10⁴ Mpc³/s (volumetric)	(Bellm, 2014, Bellm et al., 2017)
Astrometric Precision	45–65 mas RMS at S/N ≥ 10	Gaia-based solution, two-pass SCAMP (Masci et al., 2019)
Photometric Repeatability	8–25 mmag (bright sources)	Relative refinement via matchfiles (Masci et al., 2019)
Alerts per Night	~1 million	Kafka Avro stream, 8–13 min latency (Masci et al., 2019, Graham et al., 2019)
Public Release Content	Calibrated images, PSF/aperture catalogs, diff images, light curves, alerts	GUI, API, IRSA (Masci et al., 2019)

ZTF DR23 thus constitutes a flagship data release in time-domain astronomy, integrating volumetric survey strategy, scalable data pipelines, and cross-domain scientific utility, positioning it as a foundation for ongoing multi-messenger and variable sky investigations.