Zwicky Transient Facility Survey

Updated 23 January 2026

ZTF is an optical time-domain survey using a 576-megapixel camera on the 48-inch Schmidt telescope to capture a 47 deg² field per exposure.
It processes up to 10^6 real-time alerts per night with advanced image-differencing and machine learning algorithms for transient detection and classification.
The survey's innovative design and data handling set a pathfinder model for future large-scale projects like LSST, driving advances in astronomy and cosmology.

The Zwicky Transient Facility (ZTF) is an optical time-domain survey operating at the Palomar Observatory with the goal of systematically discovering and characterizing astrophysical transients and variables across the northern sky. ZTF utilizes a dedicated 576-megapixel camera on the 48-inch Samuel Oschin Schmidt telescope, covering an instantaneous field of view of approximately 47 deg² per exposure. With a median limiting magnitude of $g\approx20.8$ , $r\approx20.6$ (AB, 5 $\sigma$ , 30 s), ZTF produces up to $10^{6}$ real-time alerts per night and is a pathfinder for future large-scale synoptic surveys such as LSST by pioneering high-throughput data processing, image-differencing pipelines, and public alert distribution infrastructure (Bellm et al., 2019, Dekany et al., 2020).

1. Survey Architecture and Instrumentation

The ZTF optical assembly is based on the modified Samuel Oschin 48-inch Schmidt telescope, now equipped with a purpose-built, cryogenic 16-CCD mosaic camera (6144×6160 px per device, 1.01″/px plate scale), optimized for aberration-free imaging over a flat 47 deg² focal plane. Each exposure consists of a 30 s integration, followed by an 8.25 s readout using 64 parallel amplifiers, with the system achieving a net duty cycle of $\approx$ 75% under survey operation (Dekany et al., 2020). The delivered image quality (FWHM) is typically 2.2″ (g), 2.0″ (r/i). A robotic filter exchanger enables near-seamless cycling among $g$ , $r$ , $i$ filters, supporting both rolling and targeted survey modes (Dekany et al., 2020, Roestel et al., 2019).

Key instrumental parameters are summarized below:

Parameter	Value	Reference
FoV per exposure	47 deg²	(Dekany et al., 2020)
Integration time	30 s	(Dekany et al., 2020)
Readout time	8.25 s	(Dekany et al., 2020)
Limiting mag (5σ)	g ≈ 21.1, r ≈ 20.9, i ≈ 20.2	(Dekany et al., 2020)
PSF FWHM	2.2″ (g), 2.0″ (r/i)	(Dekany et al., 2020)
Pixel scale	1.01″/px	(Dekany et al., 2020)

2. Survey Strategy, Cadence, and Alert Stream

ZTF employs a hybrid survey design: a public Northern Sky Survey (NSS) imaging ~13,000 deg² north of declination −30° every three nights in $g$ and $r$ , complemented by a nightly Galactic Plane Survey (GPS) over |b|<7°, and specialized partnership "boutique" surveys at higher cadence or customized footprints (Graham et al., 2019).

Each night, fields are scheduled via an integer-linear programming scheduler to maximize areal coverage while obeying magnitude, airmass, and follow-up constraints (Bellm et al., 2019). The result is a mapping speed of $\sim3760$ deg² hr⁻¹ and a typical revisit time of 1–3 days per field, with denser cadence in the Galactic plane. Image differencing is performed on a per-CCD-quadrant basis using the ZOGY algorithm for optimal PSF-matched subtraction, followed by both detection and astrometry/photometry pipelines (Masci et al., 2019).

Every alert, packaged as a Kafka/Avro object, contains photometric measurements, $63\times63$ px cutouts (science, reference, difference), contextual cross-matches, and machine-learned real/bogus metrics, and is distributed within 10–15 minutes of shutter close. Alert rates routinely reach $10^6$ per night, supporting public and private brokers for downstream classification and follow-up (Mahabal et al., 2019, Patterson et al., 2019).

3. Data Processing, Machine Learning, and Real-Time Discovery

The ZTF Science Data System at IPAC supports massively parallelized, near-real-time calibration, extraction, and archival. Instrumental and global calibrations achieve photometric precision of 8–25 mmag, and astrometric RMS per axis of 45–85 mas (Gaia frame) for S/N≥10 sources (Masci et al., 2019).

Machine learning (ML) is central at all stages. The primary Real/Bogus classifier (ExtraTrees random forest) filters artifacts at ~90% accuracy (Mahabal et al., 2019). Star/galaxy separation uses Pan-STARRS features, with true-positive rates of 0.7 at 0.005 false-positive rate. Transient typing leverages both feature-based and deep learning methods, including convolutional neural networks ingesting full image stamps and LSTM-based RNNs for irregular time series classification (Mahabal et al., 2019, Rehemtulla et al., 2024).

The Bright Transient Survey (BTS) is fully integrated with ML-based automation ("BTSbot" MM-CNN), achieving 100% completeness and 93% purity in the identification and spectroscopic triggering of extragalactic transients $m_\mathrm{peak} \leq 18.5$ mag across the full survey footprint (Rehemtulla et al., 2024). BTSbot processes both imaging and extracted features, enabling sub-hour latency from discovery to SEDM follow-up (Rehemtulla et al., 2024).

4. Scientific Programs and Key Results

ZTF addresses a wide range of astrophysical science. Principal programs include:

Supernovae and Fast Transients: ZTF yields >3000 SNe Ia candidates in the first 2.5 years, with >700 spectroscopically confirmed per year and a median early-detection epoch of −13.5 d relative to $B$ -band maximum (Dhawan et al., 2021). The public data enable cosmological measurements (Hubble flow, absolute SN Ia calibration), host-galaxy studies, and systematic early-phase coverage for core-collapse and exotic transients (Perley et al., 2020). The BTS approach enables volumetric and luminosity function measurements with spectroscopic completeness >93% for $m<18.5$ mag (Fremling et al., 2019).

Multi-Messenger Astrophysics: The rapid, wide-field tiling is exploited for electromagnetic counterparts to neutrinos and gravitational-wave triggers, enabling optical limits or discoveries for events with localization areas of hundreds–thousands of deg² (Graham et al., 2019).

Active Galactic Nuclei and Tidal Disruption Events: ZTF is optimal for identifying changing-look AGN, rare TDEs, and high-amplitude nuclear variability, with statistically significant samples for ensemble studies (Graham et al., 2019).

Stellar and Compact Object Science: The high cadence and photometric precision enable discovery of variable stars, exoplanet transits (notably for white dwarfs), and compact binary phenomena. ZTF has systematically monitored ~220,000 WDs with a median of ~870 epochs each, enabling the first searches for exoplanet and planetesimal transits on WDs and setting upper bounds on planetary occurrence rates (Bell, 2019).

Solar System Science: Dedicated pipelines (FindStreaks, ZMODE) enable real-time discovery of NEOs, main-belt asteroids, comets, and transient events on small bodies. ~3×10³ NEO trails per night are ML-vetted, and ZTF has submitted >600,000 astrometric observations to the MPC by early 2018 (Masci et al., 2019).

Microlensing and Galactic Structure: ZTF's sub-arcsec accuracy and all-sky cadence are leveraged to detect ~1,100–2,400 microlensing events over 3–5 years within |b|<10°, probing outer-disk structure, the IMF, and compact object populations, as well as constraining primordial black hole dark matter (Medford et al., 2020).

5. Data Products, Archive, and Survey Legacy

ZTF provides raw, calibrated, and difference images; PSF and aperture catalogs; reference coadds; HDF5 matchfiles; and real-time Avro alert streams. The Infrared Science Archive (IRSA) hosts interactive and programmatic access; all public survey data products are released on an annual cadence, while alerts are immediate (Masci et al., 2019, Bellm et al., 2019).

The survey's machine-readable catalogs include standardized SN Ia fits, spectroscopic logs, host cross-matches, and light-curve metadata. ZTF maintains strict calibration and documentation, with PS1 and Gaia cross-matching for photometric and astrometric anchoring (Dhawan et al., 2021).

ZTF is a principal pathfinder for Rubin Observatory LSST, demonstrating scaling and integration of alert systems, machine learning for transient vetting, real/bogus discrimination, and broker infrastructure, with direct transfer for classifier development and pipeline optimization (Mahabal et al., 2019).

6. Impact, Challenges, and Future Directions

ZTF's survey design and data-handling serve as an operational model for next-generation time-domain projects. The unprecedented real-time alert rate (>1 million/night) has driven significant development in astroinformatics, public broker systems (AMPEL, ANTARES, ALeRCE, Lasair), and active learning frameworks (Mahabal et al., 2019, Rehemtulla et al., 2024). The BT Survey's comparison of SN host catalogs has revealed that RCF (redshift completeness fraction) is only 0.6 for $z<0.05$ , motivating deeper all-sky spectroscopic campaigns to enable unbiased multi-messenger follow-up (Fremling et al., 2019).

Automated tools such as BTSbot now deliver sub-hour latency and human-comparable performance on spectroscopic targeting, establishing the first fully automated end-to-end extragalactic transient discovery and classification system (Rehemtulla et al., 2024). Ongoing improvements target earlier spectroscopic confirmations, sub-typing (e.g., kilonovae, SLSNe), and robust adaptation to survey-systematics shifts.

ZTF's ongoing and future legacy includes precision cosmology (~3% statistical H₀ with current SN Ia samples), large-scale variable star catalogs, white dwarf planetary system demographics, detailed Solar System object census, and a foundational ML and software infrastructure for LSST-scale time-domain astronomy. The survey's data, methods, and operational paradigms are broadly transferrable across upcoming transient, variable, and time-domain experiments (Graham et al., 2019, Rehemtulla et al., 2024).