The Young Supernova Experiment: Survey Goals, Overview, and Operations

In the hours after a stellar explosion, an SN is still roughly the size of its progenitor star. The shock breakout, which produces a large X-ray and UV photon flux, will "flash ionize" the immediate circumstellar medium (CSM; e.g., Gal-Yam et al. 2014). During this time, the SN shock will also interact with the nearby CSM and any potential companion star, ablating material from its surface. As the ejecta expands, it will smooth out the initial inhomogeneities caused by asymmetric explosions and the initial conditions of the progenitor star.

Flash ionization provides a unique window to examine the CSM before the SN ejecta sweeps it up. If the ionizing spectrum is known and the CSM is optically thin, one can convert measured fluxes from lines of different atomic species into densities and abundances at different radii. Combined with multiwavelength observations (e.g., Chomiuk et al. 2012; Margutti et al. 2012), one can create a holistic picture of the progenitor star's circumstellar environment (Jacobson-Galán et al. 2020a).

Any radioactive elements in the outermost layers of the SN ejecta will also result in additional flux before energy produced in the deeper layers of the ejecta has time to diffuse out. As a result, this can produce "excess" flux relative to the later rising light curve (Piro & Nakar 2013). Excess flux has been seen for several SNe Ia (Marion et al. 2016; Hosseinzadeh et al. 2017; Dimitriadis et al. 2019a; Shappee et al. 2019; Miller et al. 2020) and other peculiar thermonuclear transients (Cao et al. 2015; Jiang et al. 2017), but because of either a limited number of data points or limited color information, the interpretation is often unclear (e.g., Dimitriadis et al. 2019b; Tucker et al. 2019). Earlier detections, higher cadence, additional color information, and early spectroscopy could break such degeneracies.

Luminous mass-loss episodes prior to the SN explosion may also be detectable given a survey with sufficient depth and cadence (Ofek et al. 2014; for a review, see Smith 2014). A particularly spectacular example was a luminous outburst 2 yr before the explosion of the SN Ibn SN 2006jc (Foley et al. 2007; Pastorello et al. 2007). Slightly more common are outbursts likely associated with luminous blue variables (LBVs) before a terminal explosion, such as in SN 2009ip (Smith et al. 2010b; Foley et al. 2011; Pastorello et al. 2013; Margutti et al. 2014) and SN 2015bh (Elias-Rosa et al. 2016; Thöne et al. 2017). Early bumps in the light curves of superluminous SNe (SLSNe) have also been found (e.g., Leloudas et al. 2012; Nicholl et al. 2015, 2017; Smith et al. 2016) and may be due to outbursts prior to the explosion or interaction between the SN and shells of the CSM. A survey with deep time-domain imaging would provide enormous legacy value for analysis of future transients—perhaps discovered even decades later—by enabling deep searches for such outbursts.

There is considerable uncertainty in our understanding of the nature of rare classes and subclasses of transients, including peculiar thermonuclear SNe (e.g., SN Iax, SN Ia-CSM, Hamuy et al. 2003; Foley et al. 2013; Ca-rich SNe, Perets et al. 2010; SN 2000cx-like, Li et al. 2001; SN 2002es-like, Ganeshalingam et al. 2010; SN 2006bt-like, Foley et al. 2010; various He-shell explosions, De et al. 2019; Jacobson-Galán et al. 2020b; Miller et al. 2020; Siebert et al. 2020), SLSNe (e.g., Gal-Yam 2012), TDEs (Gezari et al. 2013), low-luminosity transients (e.g., Foley et al. 2009; Valenti et al. 2009; Rodney et al. 2018; Srivastav et al. 2020), fast-evolving blue optical transients (FBOTs; Drout et al. 2014), LBV outbursts (e.g., Van Dyk et al. 2000; Kilpatrick et al. 2018a), and kilonovae (Coulter et al. 2017; Drout et al. 2017; Tanvir et al. 2017).

Transients from many of these classes are rarely discovered because they are intrinsically low luminosity, fast evolving, or red (Siebert et al. 2017). Traditional surveys such as LOSS (Leaman et al. 2011) were designed to detect common SNe and so had cadences of 5–15 days with limiting magnitudes set to detect SNe II in targeted galaxies. These surveys were usually performed with blue-sensitive filters. Such a survey would therefore miss many transients of these uncommon classes even if the intrinsic volumetric rate was high. Surveys such as Gattini-IR (De et al. 2020a), with J-band observations every 2 days and a median limiting mag J = 15.7, are just now beginning to explore a redder discovery space. Transients such as intermediate-luminosity red transients (ILRTs; Bond et al. 2009) and luminous red novae (LRN; e.g., Mould et al. 1990; Mason et al. 2010; Nicholls et al. 2013), associated with stellar outbursts and mergers, are some of the faint transient classes that may have been fRequently missed by previous surveys.

Additionally, these small field-of-view (FOV) surveys targeted luminous galaxies to increase their overall discovery yield and rejected potential transients at the centers of galaxies to avoid subtraction false positives and active galactic nucleus (AGN) activity. However, some rare classes occur preferentially in low-luminosity galaxies (Lunnan et al. 2015) and nuclear regions (Bloom et al. 2011), and previous surveys would miss these transients even if they were luminous, blue, and long-lived.

A new, untargeted, red-sensitive survey would not only increase rare transient discoveries from individual epochs but also have the potential to increase discoveries of faint but long-duration transients by stacking multiple search epochs obtained over a period of several nights. This would have a significant impact, for example, on the discovery of gravitationally lensed SNe (glSNe) and SLSNe, where the SNe are luminous and long-lived but their volumetric rates are low. Both classes are predominantly found at high redshift and extend the cosmological eras that can be probed spectroscopically. Red bands are advantageous for glSN discoveries in particular (Wojtak et al. 2019), and systematic detections of glSNe could pave the way for constraining the Hubble constant (H₀) from gravitational time delay measurements (Refsdal 1964). For SLSNe (Knop et al. 1999; Smith et al. 2010a; Quimby et al. 2011), many questions remain about their progenitors, explosion mechanisms, luminosity sources, intrinsic rates, interaction with circumstellar material, and how feedback from these sources affects their host galaxies (Gal-Yam 2019). Surveys capable of discovering SLSNe before peak with high-cadence multiband light curves will enable rapid identification, providing sources for multiwavelength follow-up and a large statistical sample.

Perhaps counterintuitively, it is the nearest SN Ia samples that are responsible for most of the systematic uncertainty on measurements of the dark energy equation-of-state parameter, w (Brout et al. 2019b; Jones et al. 2019). Most existing low-z SN Ia observations were compiled on more than 13 different, though partially correlated, photometric systems at a time when cosmological analyses were not yet limited by millimagnitude-level uncertainties. It is critical to replace these legacy data with large, well-calibrated samples of low-z SNe Ia for next-generation cosmological analyses. Large statistical samples of low-z SNe are also ideal for measuring the growth of structure from their peculiar velocities to test general relativity (Howlett et al. 2017; Huterer et al. 2017; Kim et al. 2019; Boruah et al. 2020) and reducing the statistical uncertainty of H₀ (Riess et al. 2016).

The 2020s will have two groundbreaking facilities for measuring cosmological parameters with SNe Ia: the Rubin Observatory and the Roman Space Telescope. However, neither will provide an optimal low-z sample for cosmological parameter measurements. Roman will find few low-z SNe (Hounsell et al. 2018), while the nominal-cadence strategy of Rubin will create large (∼20-day) single-filter gaps in their low-z light curves that would make SN Ia standardization less accurate. Therefore, cosmological measurements with either observatory will likely rely on external low-z data sets. Large, high-cadence, unbiased low-z samples could also be used to refine and extend new models for SN Ia standardization (e.g., Guy et al. 2007; Burns et al. 2011; Saunders et al. 2018; Léget et al. 2020; Mandel et al. 2020), to better constrain relationships between SN Ia distance measurements and host galaxy properties (Rigault et al. 2013, 2020; Jones et al. 2018a; Roman et al. 2018), to improve our understanding of common sources of core-collapse SN (CC SN) contamination in SN Ia cosmology measurements (Jones et al. 2018b; Popovic et al. 2020; Vincenzi et al. 2020), and to increase the number of independent photometric systems on which SNe are observed in order to reduce photometric calibration uncertainties—typically the dominant systematic uncertainty in dark energy measurements.

Varying accretion rates in AGNs and QSOs are reflected in their optical variability. By discovering and monitoring a large number of quasars, it is possible to revisit tests of the damped random walk model of quasar variability (Kelly et al. 2009; MacLeod et al. 2010) and study correlations between quasar variability and physical properties (e.g., Sánchez-Sáez et al. 2018; Kimura et al. 2020). Optical variability is also a useful tool for identifying the relatively elusive AGNs in low-mass galaxies (e.g., Butler & Bloom 2011; Baldassare et al. 2018, 2020; Sánchez-Sáez et al. 2019).

A small fraction of these AGNs and QSOs will evolve dramatically over a period of a few years (MacLeod et al. 2016; Ruan et al. 2016), changing their brightness, the hardness of their spectrum, their ionization state, and the presence of broad spectral lines (LaMassa et al. 2015; Runnoe et al. 2016). Prompt spectroscopic observations immediately after photometric changes are detected may then reveal the underlying physics of these "changing-look quasars" (e.g., MacLeod et al. 2019). By discovering changing-look AGNs in the early stages of their transition, one can monitor the change in black hole accretion physics and the possible buildup of an accretion disk (Gezari et al. 2017).

A survey that can reliably find transients in the cores of galaxies and identify them early also allows TDE discoveries. Detailed multiwavelength follow-up observations of TDEs can trace their full evolution, including the formation of accretion disks (Hung et al. 2020), the formation of jets (Zauderer et al. 2011), and the properties of the black holes themselves (e.g., Auchettl et al. 2017; Hung et al. 2017; Holoien et al. 2019; Mockler et al. 2019; Alexander et al. 2020; van Velzen et al. 2021). Studying the diversity of TDEs constrains the black hole mass function and extends it to lower masses than is possible with AGN observations alone (MacLeod et al. 2012; Kochanek 2016; French et al. 2020). Observations of AGNs or TDEs from wandering black holes could constrain their dynamics, a key phase before the formation of black hole binaries, and track the ways in which SMBHs settle to the centers of their host galaxies (Bellovary et al. 2019; Reines et al. 2020).

One of the key transient science challenges of the forthcoming Rubin Observatory era is how to identify unusual or scientifically valuable transients from samples of hundreds of thousands. This could include classifications within the first days or hours of discovery, allowing subsequent spectroscopic or photometric follow-up observations, or full light-curve classifications, necessary for a census of SN rates and luminosity functions, as well as for cosmological parameter measurements with SNe Ia. Recent efforts to classify a diverse sample of transients have been successful (e.g., Boone 2019; Muthukrishna et al. 2019; Villar et al. 2019; Möller & de Boissière 2020), including those that exclusively use contextual host galaxy information (Foley & Mandel 2013; Baldeschi et al. 2020; Gagliano et al. 2020), but the sensitivity of these classifiers to biases and nonrepresentative training samples is often unclear. Training data sets built from existing transient surveys will typically have significantly different photometric filters or cadences than Rubin. Multicolor light curves for thousands of transients within a magnitude-limited but otherwise unbiased discovery space would serve as a vital test bed for training transient brokers and designing efficient follow-up strategies. Light curves with iz coverage in particular are missing from most current time-domain surveys. SNe from the Pan-STARRS Medium Deep Survey will assist with these goals (Hosseinzadeh et al. 2020; Villar et al. 2020), but even that sample lacks examples of several rare classes and rarely has detections within a few days of explosion.

Building a survey that coordinates observations between multiple telescopes would also be useful logistical and conceptual preparation for the Rubin Observatory era, which will pre-announce its pointing plan. This will allow supplemental observations from other telescopes that improve the effective cadence of Rubin.

In spite of thousands of new SN discoveries per year, the rates and luminosity functions of many CC SN classes—especially fainter and redder classes—are highly uncertain (Li et al. 2011b; Perley et al. 2020). This propagates to uncertainty in photometric classification of SNe (Jones et al. 2017), stellar evolution, the physics of SN explosions, and the chemical enrichment of the universe (Maoz & Graur 2017). The most often used local rate measurements were measured from a galaxy-targeted survey (Li et al. 2011b), and therefore the rates and luminosity functions are heavily biased by the correlations between SNe and their galaxy environments (Smith et al. 2007; Quimby et al. 2011; Sanders et al. 2012; Taggart & Perley 2019). Blue-sensitive surveys will also not discover highly reddened transients, resulting in a systematic uncertainty for their rate. Since some SNe, particularly those with short-lived progenitor stars, will preferentially occur in dusty environments, the level of bias will be different for each class.

A number of teams, including ASAS-SN (Holoien et al. 2019) and the ZTF Bright Transient Survey (Fremling et al. 2020), are carrying out large surveys to spectroscopically classify all transients within a given magnitude for an improved census of transients that does not rely on galaxy associations, redshift catalogs, and assumed distances. However, a survey with greater sensitivity at redder wavelengths would increase the sample of rare, red SNe that can be discovered, and an extended volume-limited survey, similar to, e.g., De et al. (2020b) from ZTF, would ensure that low-luminosity transients are well represented in the measured rates and luminosity functions. Combining magnitude- and volume-limited surveys leverage the statistics of luminous events in the magnitude-limited sample while preserving the low-luminosity transients in the volume-limited sample. A combined sample would constrain the stellar initial mass function and the progenitor systems of transient classes and subclasses (Strolger et al. 2015).

YSE observations use the two 1.8 m Pan-STARRS telescopes (Pan-STARRS1 and Pan-STARRS2), each with 1.4 gigapixel cameras (GPC1 and GPC2; Kaiser et al. 2002). The Pan-STARRS telescopes have an approximately 7 deg² FOV, with that of Pan-STARRS2 (PS2) slightly larger owing to additional CCD chips at the corners of the FOV and the GPC2 detector's reduced masking compared to GPC1. Pan-STARRS1 (PS1) commenced formal survey operation in 2010 May and has currently imaged over 3π sr of the sky (Chambers et al. 2016). PS2 recently finished commissioning. There are subpercent color transformations between the PS1 and PS2 photometric systems.

Pan-STARRS observations are taken through one of six broadband filters, grizyw_P1 (hereafter grizyw). The filter transmissions and total system throughput for these filters have been measured via a calibrated photodiode and tunable laser (Tonry et al. 2012). Combined with the well-measured filter functions, the 3π sr of Pan-STARRS sky coverage allowed Schlafly et al. (2012) to solve for the relative calibration of the PS1 photometric system to a precision of better than 5 mmag. Scolnic et al. (2015) then improved the absolute calibration of PS1 by comparing observations of secondary standard stars across multiple photometric systems. The excellent photometric calibration of PS1 is critical for SN cosmology and helpful for many of the other YSE science goals discussed in Section 2.

Pan-STARRS data are processed using the Image Processing Pipeline (IPP) at the University of Hawaii's Institute for Astronomy (Magnier et al. 2020). The IPP is used to download, process, and archive all Pan-STARRS images and includes a difference imaging pipeline to search for transients. Once data are taken and reduced and a template image is convolved and subtracted from the data, the Transient Science Server at Queens University Belfast (Section 4; Smith et al. 2020) uses a combination of massive catalog cross-matching and a machine-learning algorithm to discover transient phenomena in the survey images.

The Pan-STARRS design and infrastructure have enabled a number of innovative transient science programs, notably the Medium Deep Survey (MDS), the Pan-STARRS survey for Transients (PSST; Huber et al. 2015), and recently the Pan-STARRS Search for Kilonovae (McBrien et al. 2021). The Medium Deep Survey observed 70 deg² of sky at an average of six observations per 10 days from 2010 to 2014, discovering over 5000 SNe. Approximately 500 of these SNe were classified spectroscopically (Rest et al. 2014; Scolnic et al. 2018), and spectroscopic host galaxy redshifts for over 3000 were obtained (Jones et al. 2017). Science highlights from these data include cosmological parameter measurements (Rest et al. 2014; Jones et al. 2018b), TDE discoveries (Gezari et al. 2013; Chornock et al. 2014), fast-evolving luminous SNe (Drout et al. 2014), SLSNe (Chomiuk et al. 2011; Lunnan et al. 2013; McCrum et al. 2014), SNe Iax (Narayan et al. 2011), and transient classification tools for the Rubin Observatory (Villar et al. 2019).

PSST commenced after the end of the MDS and used the data taken on PS1 through NASA's near-Earth object (NEO) observation program to discover new transients. PSST combines the IPP difference imaging pipeline at the IfA in Hawaii and the Transient Science Server at Queen's University Belfast to search for transients. The PSST team has discovered the majority of the ∼9000 publicly reported PS1 transients. The Pan-STARRS Search for Kilonovae is using these same NEO observations to conduct a volume-limited search for intrinsically faint transients within ∼200 Mpc (McBrien et al. 2021). Though YSE will not carry out LIGO counterpart searches, Pan-STARRS is also a powerful facility for searching for gravitational wave counterparts, due to both the FOV and the existence of reference sky templates for immediate difference images above δ ≈ − 30. Pan-STARRS has has been employed for this work during the LIGO Scientific Collaboration and Virgo Collaboration's observing runs O1–O3 (e.g., Smartt et al. 2016; Ackley et al. 2020).

The Pan-STARRS telescopes allow for a large-area, high-cadence, well-calibrated survey but place a number of constraints on the exposure time and area coverage of YSE. Our team considered only exposure times greater than 15 s; for practical purposes, this is the minimum allowed exposure time to limit systematic uncertainties in the photometry due to the GPC1 shutter. However, the Pan-STARRS cameras also have a relatively high read noise of 8e⁻ (Chambers et al. 2016) and an overhead of approximately 11 s per exposure, which makes it advantageous to increase the YSE exposure time beyond the 15 s minimum.

Figure 2 demonstrates the trade-off between area and volume for different PS1 exposure times. The read noise and sky noise for PS1 during dark time are given by Chambers et al. (2016) and agree with measured depths from Foundation Supernova Survey data (Foley et al. 2018) and PSST. We wish to survey a large volume while still covering a large-enough area to discover many brighter, nearby events and while surveying a significant fraction of the ZTF area. ²⁶ Overlap with ZTF enables high-cadence light curves from combined YSE and ZTF data.

Zoom In Zoom Out Reset image size

After evaluating a number of different strategies (Appendix A), our team chose griz exposure times of 27 s. This choice of exposure time satisfies a number of useful criteria. First, YSE observations are ∼0.4–0.8 mag deeper than ZTF, enabling day-before observations—both detections and nondetections—that put useful limits on the ages of newly discovered SNe. Second, while YSE data are read noise dominated in g observations taken during dark time, this exposure time ensures that the rest of the data will be background dominated. Finally, a 27 s exposure time nearly maximizes the possible YSE survey volume (Figure 2) but still covers a large area such that more than 100 SNe yr^–1 will have r < 18.5 mag, making them easy to follow spectroscopically using 2 and 3 m class telescopes from the ground.

We considered several different filter sequences for YSE, a number of which are discussed in Appendix A. We decided on a strategy that emphasizes the iz filters while also having at least one g- or r-band observation per epoch, as gr observations allow us to measure rise times for young SNe from the combination of YSE and ZTF data without any assumptions on the color of the event. YSE observes in only two filters per night to maximize the survey area while still preserving color information for transient observations. During dark time, we alternate observations of gr and gi, and in bright time, due to the ∼1 mag lower g-band depth, we alternate ri and rz observations. During brief periods of gray time (moon illumination between 33% and 66%), we alternate gi and gz.

Finally, YSE observes each field with a 3-day cadence to increase synergy with the ZTF observing strategy and to allow well-sampled light curves for relatively fast-evolving transients.

In Table 2 we compare the YSE survey characteristics with those of other ongoing time-domain surveys. Each of these surveys has unique advantages, with ASAS-SN discovering nearby SNe with extremely fast cadence, ATLAS having a slightly slower cadence compared to ASAS-SN but increased depth and excellent calibration, and ZTF covering less area than ASAS-SN/ATLAS but having greater depth and both high-cadence and i-band subsurveys. In terms of area observed at a given time, YSE will observe significantly less area than ASAS-SN, ATLAS, and ZTF. However, YSE's depth is 0.4–0.8 mag deeper than any other survey, and therefore YSE will cover more volume than ASAS-SN, 50%–70% as much volume as ATLAS, depending on the transient luminosity, and ∼20% as much volume as ZTF, with the i-band data in particular probing a unique volume equal to 76%–90% of the ZTF i-band volume. In terms of the photometric calibration in particular, the excellent PS1 calibration outperforms other systems (and PS2 calibration will be tied to PS1 for comparable accuracy), which is particularly important for SN Ia cosmology. The oversampled PS1 point-spread function (PSF) may allow for improved image subtraction, which benefits black hole variability and TDE studies by facilitating discoveries near the centers of galaxies. The combined ZTF/YSE cadence is particularly advantageous for fast-cadence studies, helping with discoveries of young SNe, and the z-band coverage is unique to YSE and will aide in the discovery of very red transients.

We select YSE survey fields based on the following criteria:

• 1.  
  To increase transient discovery volume, we choose fields with high Galactic latitude and low Milky Way (MW) extinction.
• 2.  
  We prioritize fields with a large amount of available archival data, primarily redshifts and deep multiwavelength imaging.
• 3.  
  We prioritize equatorial fields, as they can be observed with follow-up facilities from both hemispheres. However, in most cases we require fields to be at least 20° from the ecliptic plane to minimize observing gaps caused by the Pan-STARRS moon avoidance angle of 30°.
• 4.  
  We avoid fields with declinations less than −30°, as YSE requires deep template images from previous Pan-STARRS observations, which are scarce at large negative declinations.
• 5.  
  We choose fields that, given their history of observations, ZTF will typically observe on a 3-day cadence or greater and in two filters.
• 6.  
  Our observations are interwoven with Pan-STARRS NEO observations, requiring us to space our fields in R.A. so that they can be observed throughout the night.
• 7.  
  When possible, we place pointings in minimum schedulable blocks of six pointings each, as Pan-STARRS slews of more than 15° require approximately 1 minute of overhead to refocus the telescope.
• 8.  
  We prioritize new fields with rising or scientifically interesting transients that have been discovered by other time-domain surveys and for which YSE data would be beneficial. We will attempt to follow such transients using as many fields as possible in the first year of the survey and using 50% of fields in subsequent years. The remaining 50% will be kept the same across multiple years for long-term monitoring.
• 9.  
  We prioritize fields that have a larger number of nearby galaxies within 150 Mpc, with galaxies at <10 Mpc given the highest weight. This will result in a slight bias to the nearby rates in exchange for an increase in SN discoveries, but we expect to be able to correct for this bias in future analyses given the large fraction of galaxies that are untargeted.

We use a semi-arbitrary field selection metric to take these priorities into account, which is discussed in Appendix B. We consider the total amount of HST exposure time when selecting fields, which is important for identifying or constraining progenitor systems in pre-explosion imaging (e.g., Li et al. 2011a; McCully et al. 2014; Foley et al. 2015; Kilpatrick et al. 2018b; Kilpatrick & Foley 2018), as well as the number of active SNe—particularly SNe before maximum light—in a given field, but do not formally include these quantities in the metric. The initial set of YSE pointings is shown in Figure 1.

3.3.1. Daily Survey of Virgo

When the Virgo Cluster is observable at air mass < 1.5, YSE will also dedicate two Pan-STARRS pointings to a daily survey of Virgo. Our team may choose to undertake similar mini-surveys of other nearby clusters, such as Coma, in the future. The Virgo region of the sky produces an extraordinary number of SNe; in the past 15 yr, 30 SNe were spectroscopically classified within the radii of the two YSE Virgo pointings our team has adopted ²⁷ (not accounting for detector masking). Most recently, this includes three SNe Ia (SN 2018bgb, SN 2018axs, and SN 2017eea) and one SN IIn (SN 2017jfs).

YSE will be able to detect pre-explosion outbursts for Virgo transients to an approximate absolute magnitude of −10. We have chosen our pointings to maximize the archival HST exposure time, thereby simultaneously including the majority of Virgo galaxies. A 4 deg² deep stack from the first season of observation for our daily M87 Virgo field is shown in Figure 3, along with the locations of six transients observed in the field (five are located in the background of Virgo, and one is in the cluster). These are YSE-observed SN 2020pf (a ZTF discovery but with simultaneous YSE observations) and YSE-discovered SN 2019yub, SN 2020ndg, SN 2020lhy, SN 2020mvx, and finally AT 2020iuy, which is likely a stellar outburst with peak M_i ≈ − 11.9 mag.

Zoom In Zoom Out Reset image size

Although we wish to keep 50% of our fields fixed for long-term monitoring, the other 50% of YSE fields can be adjusted each year to follow new transients that meet our science objectives. We hope to obtain targeted observations of ∼15 SNe per month. This includes rare transients, young transients, or low-z SNe Ia to build an anchor sample for next-generation cosmology surveys. Of these 50% that can be adjusted, our survey requirements specify that 40% of new fields should be within 15° of existing fields to limit refocus overheads. To move an existing YSE survey field to target a new SN, the survey field should meet these criteria (below, a setting field must have air mass greater than 1.5):

• 1.  
  To relocate an existing field to a new location, the existing field either (a) should have no clearly rising transients and be within 10 days of setting or (b) should have no transients within either 2 weeks of maximum light or within a month of discovery. Certain rising transients that would have very poor light-curve coverage (e.g., if the field is near setting) would also be reasonable candidate fields to move. Conversely, a current field with a post-max SN that is of high scientific value to the collaboration should not be discontinued unless it is near setting.
• 2.  
  If YSE science will substantially benefit from continued follow-up observations of the transients in the existing field, the field will not be moved.
• 3.  
  The location of the new field should be within 15° of existing survey fields unless the target is of exceptionally high priority.
• 4.  
  In most cases, a new field should be 3 months or more from setting.

To maximize discoveries of young SNe and increase the effective cadence for many YSE and ZTF transients, we attempt to schedule YSE observations such that they precede ZTF observations by one calendar day. Because our survey is ∼0.4–0.8 mag deeper than ZTF during dark time, we hope to discover fast-rising SNe or newly exploded SNe by observing a >1 mag rise between the Pan-STARRS observation and the ZTF observation approximately 21 hr later. As ZTF reports transient discoveries with a median delay of just ∼10 minutes from the time the image is taken, our team is able to quickly combine the data streams to trigger rapid spectroscopic and multiwavelength follow-up observations of fast-rising SNe.

In practice, predicting ZTF observations can be nontrivial, as ZTF uses a complex scheduling metric to avoid large coverage gaps rather than a fixed 3-day cadence, and weather makes it difficult to anticipate whether an observation will be scheduled or carried out on a given night. However, ZTF generally prioritizes fields by those with the longest time since they were last observed (Bellm et al. 2019b). Over the first 3 months of YSE, during full nights of observing with clear weather at both Haleakala and Palomar, 61% of our observations were taken a day before ZTF observations of the same field (compared to 33% for uncoordinated observations; see Section 5 for examples of our data). However, we find that during longer periods of good weather we are more able to more successfully predict the ZTF observations. Near the beginning of YSE, most fields were also concentrated within a narrow R.A. window, and these may be less likely to have ZTF observations evenly spread over 3 days. We note that ZTF will soon begin making their nightly observing plans public using the International Virtual Observatory Alliance Observation Locator Table Access Protocol, ²⁸ which will significantly improve our ability to plan overlapping observations. Rubin will also announce their observing schedule via the same protocol, preparing us for potentially allowing YSE to supplement the Rubin cadence in 2023.

YSE reports all discoveries of likely transients, omitting likely variable stars or AGNs, to the Transient Name Server ²⁹ (TNS), the official International Astronomical Union mechanism for recording new transient discoveries. Our discovery reports include the first epoch of photometry for each transient (we will provide the first two photometric epochs in the near future), and whenever a team member obtains a spectroscopic classification of an unclassified YSE transient, the spectrum and classification are sent to TNS as well.

Obtaining a second epoch before reporting a discovery is sometimes necessary for asteroid rejection; to avoid substantial ∼1-minute refocusing overheads for Pan-STARRS when slewing more than 15°, our observations in different filters are spaced only ∼4 minutes apart. This makes asteroid rejection more difficult, occasionally necessitating a second observation or confirmation from a second survey in order to confirm that a YSE alert is a bona fide transient. When YSE observations for a new field begin, AGN contamination is also significant; as our team has gained vetting experience, we have become more conservative in promoting candidates during the first epochs of a given field to avoid reporting AGNs. In most cases, however, the colors and presence of nearby host galaxies make transient discoveries unambiguous. In cases where it is unclear whether a transient is a true SN versus an asteroid, variable star, or image artifact, we wait to send these events to TNS until additional epochs of data can clarify whether or not they are bona fide transient phenomena.

To evaluate the effectiveness and expected SN discoveries from the survey strategy presented above, we simulated the YSE survey using the SNANA software (Kessler et al. 2010). We include models of different SN types and subtypes from the PLAsTiCC SN classification challenge (Kessler et al. 2019), including SN Ia, SN Iax, SN II, SN Ib/c, ILOT, and CART. We omit TDEs, SLSNe-I, and kilonovae owing to large uncertainties on their rates, as well as purely theoretical models and nontransient models (e.g., variable stars and AGNs). We use anticipated survey depths from Chambers et al. (2016), moon-phase-dependent sky noise to match 3π observations, SN rates from Strolger et al. (2015) and Dilday et al. (2008), normalization of those rates at z = 0 from PLAsTiCC (and references therein; most rates originate from Li et al. 2011b), and a nominal ZTF survey to match their reported depths with ZTF observations scheduled 21 hr after YSE. Details of the simulation methodology and the effect of choosing alternate survey strategies are given in Appendix A.

Summary statistics for the nominal simulated YSE survey design are shown in Figure 4. Due to its depth—YSE is able to find SNe Ia at redshifts up to nearly ∼0.3—simulations predict 5149 SNe per year having at least three detections with a signal-to-noise ratio (S/N) > 5, including 3920 SNe Ia, 277 SNe Ib/c, 705 SNe II, and 67 SNe IIn. Our magnitude- and volume- limited samples, comprising SNe with m < 18.5 mag and D < 250 Mpc, respectively, are expected to contain a total of 217 SNe per year (Figure 5; 49 SNe are members of only the magnitude-limited survey, 109 are members of only the volume-limited survey, and 59 are members of both).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Appendix A, we discuss the results of several other survey designs explored by our team, including variations in exposure time, three or four filters per day, and a survey design focused on either blue or red filters. We caution that some simulated yields are affected by small number statistics. As exposure time increases, the number of discovered SNe tends to increase, but begins to flatten after 25 s, as would be expected from Figure 2. Surveys consisting of three or four filters per day would be excellent for some science cases, but would drastically reduce the discovery space for transients, resulting in a 40%–50% reduction in the number of transient discoveries. A blue-focused survey was primarily ruled out because it offers a less unique discovery space compared to existing time-domain surveys, while a survey focusing only on riz observations would be expected to reduce young SN yields, as most young SNe are expected to be blue. Our nominal survey design—27 s exposures with a gr, gi observing sequence during dark time—was settled on as the best choice for YSE, especially given the characteristics of the Pan-STARRS system and our allocated 7% fraction of time on each telescope.

Finally, Figure 6 and Table 5 show the anticipated yields of SNe less than 3 days old assuming that PLAsTiCC models are correct realizations of young SN brightness and color. For example, PLAsTiCC simulations of SNe Ia are based on the SALT2 model, which has a slower rise at early times than observations of SN 2017cbv, SN 2018oh, and a number of SNe discovered by Jiang et al. (2020). Since SNe IIb often show luminous shock breakout cooling (e.g., Richmond et al. 1994; Arcavi et al. 2011; Kilpatrick et al. 2017; Fremling et al. 2019), they are often easy to detect early as well, yet there are no SN IIb templates in the PLAsTiCC set. It is therefore difficult to truly predict the number of young SNe that will be discovered by YSE, but we note that our team has already observed a handful of young SN candidates, including SNe Ia SN 2020pf, SN 2020fci, SN 2020ioz, SN 2020juq, and SN 2020nbo and CC SNe SN 2020pni, SN 2020kpz, and SN 2020tlf, that are almost certainly within 3 days of explosion or less (see Section 5).

Zoom In Zoom Out Reset image size

Figure 7 also shows the simulated colors of young SNe; we find that a simple color cut could identify a subset of young SNe with ∼40% accuracy, which is a reasonably high confidence given the follow-up resources available. We find that g − r colors are useful for separating young versus old CC SNe and YSE r − i colors appear to be excellent indicators of young SNe Ia, complementing g − r observations from YSE and ZTF. Our team is working on more sophisticated classification methods using existing observations of young SNe.

Zoom In Zoom Out Reset image size

While these simulations are state of the art for transient survey planning, they still lack several details for precise estimates. Although early YSE survey yields are returning the expected number of volume- and magnitude-limited SNe, just 26% of transients to date are above the median predicted survey redshift of 0.19. This is likely because our algorithm for measuring YSE photometry has been obtaining inflated uncertainties, effectively reducing the S/N of YSE photometry and our detection efficiency. We identified this issue by implementing a second photometric reduction algorithm and are currently working to implement a fix to our pipeline photometry. Other contributing factors may include biases in the photo-z determinations, as reliable redshifts are more difficult to estimate for faint or undetected host galaxies. Additional stacked observations may also reduce the noise in the template images, which simulations assume to be negligible. Finally, efficiency losses in the machine-learning "real/bogus" algorithm (Section 4) may also reduce the number of discovered SNe as a function of S/N. Nevertheless, we have observed or discovered 916 transients over the first 10 months of the survey while observing at half of our nominal area and experiencing ∼2 months of downtime.

Given the YSE survey design and simulations described above, we briefly summarize a number of ways in which the YSE data will address the science drivers identified in Section 2:

• 1.  
  Young SNe. YSE will discover at least two SNe within 3 days of explosion per month, with a larger sample likely due to known deficiencies in the PLAsTiCC models used to generate YSE simulations. Many of these SNe will have i or z photometry at very early times and can be discovered by YSE ∼ 1 day before they are found by other surveys. Several examples from early YSE data are discussed in Section 5.
• 2.  
  Red and Rare Transients. The iz coverage of YSE will aid the identification of very red transients. YSE's additional ∼1.2 mag i-band depth compared to ZTF will probe a unique volume equal to ∼76%–90% of the ZTF i-band survey volume (Table 2). YSE is also the only time-domain survey currently observing in the z band. Within our volume-limited sample out to 250 Mpc, we will discover transients down to M_i  ≈ −15.5 mag and M_z  ≈ −16.5 mag. We also expect YSE to observe two glSNe per year in regular sky scans and up to five glSNe per year in stacked images based on projections from Wojtak et al. (2019). YSE fields to date include ∼1400 galaxies near enough for YSE to discover luminous red novae, though their volumetric rate is not well known (Pastorello et al. 2019).
• 3.  
  Cosmology. The YSE SN Ia sample will be a mix of targeted SN Ia observations to increase the z ≲ 0.05 sample size, using a strategy modeled after the Foundation Supernova Survey (Foley et al. 2018), and untargeted observations, which will have a median redshift of z ≈ 0.12 (Appendix A). The untargeted sample will include up to 350 z < 0.1 SNe per year with a well-understood, magnitude-limited selection and lower uncertainties from peculiar velocities than other low-z samples. The large sample with z > 0.1 will serve as an important sample for measuring w, testing general relativity, and helping to understand CC SN contamination in future cosmology analyses. The photometric calibration of the sample will be more precise than any non-Pan-STARRS low-z sample.
• 4.  
  Black Holes and TDEs. We did not include TDEs in survey simulations, as their redshift-dependent rates are uncertain, as is the efficiency of detecting them in the bright cores of galaxies; based on the survey magnitude limit alone with the rates and luminosities from van Velzen et al. (2021), and including an approximate mean M_g = −17.5 mag, we would expect ∼20 per year—and YSE has already observed two TDEs in untargeted observations just in early survey data (a third, AT 2020nov, was targeted; Section 5)—but due to the difficulty of detecting transients in the cores of galaxies, it is unclear whether the expected number from these order-of-magnitude estimates is overly optimistic. YSE survey data will also contain tens of thousands of AGNs for variability studies, and its oversampled PSF may make it easier to detect transients at the bright cores of galaxies compared to other low-z surveys.
• 5.  
  Preparation for the Rubin Observatory. YSE will measure griz light curves for hundreds of spectroscopically classified transients and thousands of transients in total, including hundreds of detections within the first few days of explosion. This will allow YSE to serve as a unique and invaluable training sample for the Rubin Observatory, adding fidelity to early-time and full light-curve classifications and constraining the rates of rare classes that will exist in Rubin survey data.
• 6.  
  Magnitude- and Volume-limited Census of SNe. The iz coverage of YSE will improve the census of red transients, and the YSE redshift range will allow a longer lever arm on the redshift-dependent rates of many transients compared to other ongoing surveys. We plan to spectroscopically classify every SN that is either brighter than r ≈ 18.5 mag at peak or nearer than 250 Mpc, a sample of ≈217 SNe per year (Section 3.7) to constrain both the rates and luminosity functions of SN types and subtypes. Approximately 61 of these SNe (36% of the volume-limited sample and 28% of the total) will be both fainter than r ≈ 18.5 and more distant than 200 Mpc, the limits of the ZTF magnitude-limited and CLU samples. Additionally, we include photometric redshift estimates in building our volume-limited sample in order to avoid completeness limitations that affect spectroscopic catalogs of the local volume (and the CLU sample to some degree).

In addition to addressing these time-domain science questions, YSE will create 1 yr stacked images over approximately 7000 deg² of sky and 3 yr stacked images over an additional ∼2000 deg² of sky. Using the typical read-noise and moon-dependent sky-noise values from Chambers et al. (2016) and assuming 30% loss due to weather and 24% due to pixel masking, we roughly estimate that in 1 yr stacks YSE+3π will reach depths of griz ≈ 23.6, 23.7, 23.6, 22.8 mag. In the 3 yr stacks YSE+3π will reach depths of griz ≈ 24.0, 24.2, 24.1, 23.3 mag. Comparisons to the depths of several other large-area surveys are shown in Table 3.

The Transient Science Server removes YSE detections from known AGNs, variable stars, and moving objects using a library of catalogs (see Smith et al. 2020 for details) and employs both pixel-based and catalog-based machine-learning algorithms—i.e., the algorithm is based on both the pixels themselves and photometric metadata such as PSF shape—to remove likely image artifacts and assign a "real-bogus" score to each detection. Candidates are included on the by-eye vetting list if the real-bogus score considers them to be real with at least 32% confidence; we choose this threshold because the training procedure estimates that it will yield a high fraction of real candidates (75%), while rejecting just 1% of bona fide transients. We require 3.5σ detections in at least two separate images (possibly over multiple epochs) to add a source to the list of candidates to be vetted. The machine-learning algorithm is trained using slow-moving asteroids as the "good" objects; we plan to retrain using real transients in more realistic host galaxy environments in the future. Additional details of the algorithm are given in Smith et al. (2020).

If members of our team determine that a candidate is likely a bona fide transient, they are immediately sent to the TNS along with the first epoch of photometric data. If we are unsure, the transients remain on a "possible" list until more data have been collected. Both the transients on the "good" and "possible" lists are ingested to YSE-PZ for combining with data from other surveys and making follow-up decisions.

The YSE-PZ application (YSE-PrioritiZe) is designed to ingest every transient reported to the TNS (see footnote 29) to facilitate queries of the data, combine YSE transients with external data, store follow-up data obtained by our team, and enable easy communication among team members for vetting and following new transients. YSE-PZ is a Django-based application built on a MySQL database. The code base is publicly available at https://github.com/davecoulter/YSE_PZ, and we encourage collaboration and new contributors. An example of the web application interface is shown in Figure 9.

Zoom In Zoom Out Reset image size

YSE-PZ will be described in detail in D. A. Coulter et al. (2021, in preparation). Here, we summarize key features that allow our team to prioritize and organize follow-up observations of transients:

• 1.  
  Data ingestion from ZTF and TNS. We ingest ZTF alerts using the ANTARES broker (Narayan et al. 2018) for events in our survey fields that have not yet been reported to TNS. We also ingest every transient reported to TNS using the TNS API. For transients that have been reported to TNS, we ingest ZTF photometry from MARS. ³⁰
• 2.  
  Dashboard and Initial Transient Vetting. Transients are ingested into YSE-PZ with a status of "New" and then parsed into categories of "Watch," "Interesting," "FollowupRequested," "Following," or "Ignore" based on characteristics such as brightness, redshift, classification, or color. Transients that meet the YSE volume- or magnitude-limited survey criteria are prioritized for follow-up by moving them to "Interesting" or "FollowupRequested," and subsequent queries are used to flag transients that brighten to the point where they are included in the magnitude-limited criteria. YSE-PZ compiles a large amount of data and metadata to aid in the vetting process, including survey images, archival images, other external data (e.g., HST or Chandra imaging), MW extinction, photometry, spectra, classifications, and discovery details. Our team sorts through new transient discoveries daily.
• 3.  
  Follow-up Requests. YSE-PZ allows follow-up observations to be requested on any telescope for which our team has access. Follow-up requests are added to a custom page for a given observing night and resource, where observers can prioritize the requests and schedule observations. Finder charts using Pan-STARRS 3π imaging may be generated through the web interface, and we plan to add additional tools to facilitate scheduling in the near future.
• 4.  
  Queries. YSE-PZ allows users to query on any field in the database. A query, once completed and saved, can be added to a user's personal "dashboard" to allow each user to be alerted to transients that are relevant to their science interests. Example queries include the YSE volume-limited sample of transients within 250 Mpc, or transients rising by more than 0.5 mag in a day in a given filter. As new transients are ingested, users can choose to be alerted via email or text message to transients that match a given query to enable extremely rapid follow-up observations.

YSE-PZ is under active development, and new features are continually being added. A well-developed, open-source, robust transient management framework is vital to our science goals and a key part of developing tools for the next decade of transient science.