HETDEX Public Data Release 1: Source Catalog 2 and Data Cubes from ∼90 deg² of Integral-field Optical Spectroscopy

The data cubes and catalogs released in PDR1 consist of the full main HETDEX survey (K. Gebhardt et al. 2021), drawn from the fifth internal data release (HDR5) from the HETDEX Collaboration. It contains 6778 observations obtained starting in 2017 January, when the VIRUS IFU assembly contained just 16 operational IFUs, and concluding on 2014 July 31 with the full 78 IFUs installed within the VIRUS array in 2021 July.

The HETDEX footprint consists of two primary fields that allow for full-year surveying as shown in Figure 2. The HETDEX-Spring Field, labeled as the dex-spring field throughout this paper and in the associated catalog, covers 390 deg² of high-decl. (δ ∼ 51°) sky, while the HETDEX-Fall Field, labeled as dex-fall in the catalog, covers 150 deg² along the celestial equator (see K. Gebhardt et al. 2021 for full details on field selection). The project does not need complete coverage within this sky area to accomplish its scientific goals, as discussed by C.-T. Chiang et al. (2013); a fill factor of 0.22 (1/4.6), which optimizes the number of IFUs given the area of the focal plane of the HET, is sufficient.

Zoom In Zoom Out Reset image size

In addition to the two main survey fields, this HETDEX data release also includes coverage from multiple legacy fields, as presented in Figure 3. The most extensive coverage comes from collaborative observations with the Texas Euclid Survey for Lyα (Ó. A. Chávez Ortiz et al. 2023) of the north ecliptic pole (NEP). Nearly full field coverage of the central 1 deg² of the Cosmic Evolution Survey survey (COSMOS; N. Scoville et al. 2007) is included, as well as sparse coverage of the SA22 (C. C. Steidel et al. 1998) and GOODS-N (M. Dickinson et al. 2002) fields. Field coordinates, coverage area, and the number of IFU observations contained in each field are summarized in Table 1.

Zoom In Zoom Out Reset image size

Table 1. Catalog Release Survey Statistics

Field ID	Center	N	Area	N(Source)	N(LAE)	N(OII)	N(AGN)	N(LzG)	N(Star)
	(deg, deg)	(IFU)	(deg2)
dex-spring	(195.00∘,  +51.00∘)	244,176	49.00	640,603	253,458	287,017	11,093	10,807	77,522
dex-fall	(22.50∘,  +0.00∘)	136,829	27.46	305,915	114,419	140,133	5,531	6,586	38,886
nep	(270.00∘,  +66.00∘)	34,269	6.88	100,872	25,649	45,035	1,232	1,381	27,359
cosmos	(150.12∘,  +2.21∘)	11,271	2.26	24,662	7,794	11,521	431	498	4,390
ssa22	(336.50∘,  +0.00∘)	4,393	0.88	10,905	3,975	4,036	146	93	1,930
goods-n	(189.18∘,  +62.24∘)	775	0.16	2,172	1,025	877	42	27	195
Total	​​⋯	431,713	86.64	1,085,129	406,320	488,619	18,475	19,392	150,282

Note. Listed is the number count of IFU observations observed and the count included after observation quality inclusion criteria. Reported area coverage includes repeat observations and therefore does not represent unique sky coverage. Field centers are given in equatorial coordinates (ICRS; epoch J2000).

Download table as:  ASCIITypeset image

HETDEX uses VIRUS (G. J. Hill et al. 2021), the fiber-fed, multispectrograph instrument of the upgraded Hobby–Eberly Telescope (L. W. Ramsey et al. 1998; G. J. Hill et al. 2021). Each IFU connects to a pair of VIRUS spectrographs with 448 15 diameter fibers positioned on a rectangular array with fiber center separations of 25. In a single HETDEX observation, three exposures, each typically lasting 6–7 minutes (the exposure times range from 3.6–12 minutes, depending on observing conditions), are captured; the telescope is dithered in a triangular pattern to ensure complete fill factor for each of the 51^″ × 51^″ IFU fields (see G. J. Hill et al. 2021; K. Gebhardt et al. 2021). The bottom-right panel in Figure 2 provides an example IFU fiber layout for this three-dither pattern. In a single IFU observation, 1344 fiber spectra are collected, providing full sky coverage of the IFU. At full completion of the VIRUS instrument, its 78 IFUs cover approximately 21.7% (a factor of 4.6) of the HET’s 18′-diameter FOV. The exact layout and numbering scheme for the array of IFUs is shown in Figure 3 in K. Gebhardt et al. (2021).

HETDEX data processing is described in detail in K. Gebhardt et al. (2021). Briefly, reduction begins with CCD-level calibrations, including bias subtraction, pixel flats, and twilight sky flats. These are followed by fiber extraction, wavelength calibration, and sky subtraction to produce a calibrated spectrum for each fiber in the array.

Sky subtraction is performed using a local sky model constructed at the amplifier level from fibers within each dithered exposure. This local sky subtraction is optimized for the detection of low signal-to-noise emission lines, which are the primary targets of HETDEX.

In addition to this approach, the internal HETDEX data model also supports a full-frame sky estimate constructed using all IFUs across the VIRUS focal plane within a given observation. While this full-frame model can better capture large-scale sky structure, it introduces low-level systematics in the calibrated spectra that can degrade the flux calibration on small spatial scales. For this reason, the public data products are calibrated using the local sky model. As a consequence, flux measurements near bright or extended sources, or on spatial scales comparable to the IFU FOV, may be affected by oversubtraction of extended emission.

Astrometric calibrations are achieved by measuring the centroid of each field star using the signal between 4400 Å and 5200 Å and comparing their IFU positions to the stars’ equatorial coordinates in SDSS (D. G. York et al. 2000) and Gaia (Gaia Collaboration et al. 2018) catalogs. This process typically results in global solutions that are accurate to .

Absolute flux calibrations are achieved using g  <  24 SDSS field stars as in situ flux standards. Their ugriz colors (N. Padmanabhan et al. 2008), Gaia parallaxes (Gaia Collaboration et al. 2018), and foreground reddening estimates (E. F. Schlafly & D. P. Finkbeiner 2011) are used to determine their most likely spectral energy distribution in a grid of model spectra (A. J. Cenarro et al. 2007; J. Falcón-Barroso et al. 2011). The final system throughput curve is derived from the most likely flux distribution of ∼20 stars, and is generally accurate to ∼5% (K. Gebhardt et al. 2021). Uncertainties in the calibrated fiber spectra are derived by propagating the photon noise through the reduction steps and tracking the uncertainties that come with each step.

Each dither in an HETDEX observation is flux calibrated independently, as there may be small differences in their relative throughput due to variations in the observing conditions. For inclusion in the HETDEX survey, we require that a nominal throughput (assuming a 360 s exposure time) exceeds 08, and that the relative throughput of each dithered exposure cannot differ by more than a factor of 3. The most common reason for rejection by this criterion is a significant drop in transparency during the third dithered exposure when clouds drifted into the FOV.

The data are organized into a database of flux-calibrated, 1D fiber spectra each with their own corresponding sky coordinate. This forms the fundamental data product of the internal HETDEX data model. The internal data model stores not only the calibrated fiber data but also a number of intermediate data products including the raw and calibrated CCD images from each dithered observation. Internally, these intermediate data are used for quality control and artifact identification. Cutouts of these products are used in machine learning–based classification and vetting of detections (L. R. House et al. 2023, 2024; S. Mukae et al. 2026).

Internally, HETDEX data are stored in HDF5 files (The HDF Group 2024), with the primary data model constructed around a database of calibrated fiber spectra. Direct use of the fiber data is nontrivial: because the HET and is not equipped with an atmospheric dispersion corrector, each fiber samples a slightly different sky position as a function of wavelength. This atmospheric differential refraction (ADR) is corrected in the internal pipeline with dedicated software (hetdex-api³¹ ), which is also required for subsequent interpolation and spectral extraction.

To simplify data access and avoid imposing these software dependencies, the public release is distributed as interpolated IFU data cubes rather than the native fiber database. This format also allows users to download only data cubes of interest, each of modest size (< 20 MB), and analyze the data using standard astronomical software tools.

In contrast to the internal release, the public release excludes the raw and intermediate data products. This decision is motivated primarily by the large data volume and complexity of the internal data model: the full internal dataset, including raw CCD frames, calibrated products, and intermediate processing files, exceeds ∼350 TB, while the processed internal data products alone comprise ∼150 TB. By comparison, the public release of data cubes and catalogs requires <10 TB of storage. At present, we do not plan a full public release of the intermediate data products. However, we are happy to provide access to these data upon request for users interested in reproducing specific aspects of the internal processing or investigating instrumental effects in greater detail. The publicly released data cubes and catalogs are sufficient for the majority of scientific investigations with the HETDEX survey.

The database of calibrated fibers is interpolated using custom software from hetdex-api³² into a set of IFU data cubes in a format that is similar to other IFU survey data from Calar Alto Legacy Integral Field Area (CALIFA; S. F. Sánchez et al. 2012), Mapping Nearby Galaxies at APO (MaNGA; K. Bundy et al. 2015; N. Drory et al. 2015; D. R. Law et al. 2015), and MUSE (T. Urrutia et al. 2019). As VIRUS is composed of an array of many IFUs with 1′ gaps between them, the natural unit of data cube storage is that of a single IFU. A given HETDEX exposure is thus composed of 16–78 IFU data cubes depending on how many units were active on the instrument at the time of observation. Each IFU is stored in an individual FITS (D. C. Wells et al. 1981) file with the format described in Table 2. The spectral sampling of each cube is set to 2 Å, the same as the spectral sampling of the internal fiber spectra. Pixel spatial sampling is set to 05. With a median image quality of FWHM = 18, this is sufficient to sample the point-spread function (PSF) of HETDEX observations while keeping storage requirements manageable.

The IFU data cubes cover the same spectral range and resolution as the internally calibrated fibers, sampling a linear wavelength grid from 3470 to 5540 Å in steps of 2 Å. For each wavelength slice, the apparent sky position of every fiber is adjusted to account for ADR using the astrometric solution from the internal pipeline (K. Gebhardt et al. 2021). The resulting irregularly spaced fiber positions are then resampled onto a 52″ × 52″ square grid with spatial sampling. Interpolation is performed with the griddata routine from SciPy (P. Virtanen et al. 2020), which linearly interpolates the flux and associated uncertainties of all contributing fibers to the output coordinate grid. The WCS of each cube follows standard FITS conventions, with the CRPIXn values specifying 1-indexed reference pixel coordinates at pixel centers rather than pixel edges. The data cubes are oriented according to the position angle of the telescope and VIRUS focal plane to minimize any unnecessary data storage space. Thus, the projected orientation is not constrained to the conventional “north up, east left” display; users should rely on the WCS headers when visualizing the data.

The resulting data cube spaxels are per-bin flux values sampled in 2 Å wide spectral bins in units of 10⁻¹⁷ erg s⁻¹ cm⁻² per 2 Å bin. Flux values are stored in Header Data Unit 1 (HDU1), and their associated flux uncertainties are stored in HDU2. Spaxels that fall outside the region with valid fibers or within fully masked regions are assigned NaN. To prevent spurious interpolation across masked fibers, a Boolean coverage map is propagated by resampling the good-pixel indicator with nearest-neighbor interpolation and blanking any output spaxels where this indicator is false.

In addition, a bitmask array, described in detail in Section 2.3, is interpolated from the internal data model to the data cube wavelength slices following the same methodology as the flux interpolation except that, for the bitmask, the nearest-neighbor method is used for interpolation.

The calibrated fiber spectra in the internal HETDEX data model include a single mask for problems automatically identified by the reduction pipeline. This includes detector artifacts, poor-quality fibers yielding unusable spectra, transient cosmic rays, and various calibration failures. Such cases are flagged through a fill value of 0.0 in the calibrated fiber error arrays. While the fiber spectrum itself may still report a flux value, it is considered invalid when flagged in the error array.

In practice, given the complexity and sheer volume of HETDEX data, a wide range of previously undiscovered issues must also be masked from the data. Over the years, these have been identified through an extensive manual effort. In particular, data reports generated by the Emission Line eXplorer (ELiXer) tool developed by D. Davis et al. (2023a) have proved invaluable in diagnosing artifact features that are clear in the view of the raw CCD science frames but may not be obvious in the interpolated fiber spectrum. This manual inspection led to the development of a number of automated procedures to remove spurious artifacts from our source catalogs and fiber spectral data.

For each flux-calibrated spectrum, these masks are stored as bitmask values for every spectral element in the array. In the public data model, these bitmasks are interpolated into HDU3 of the data cube FITS files as listed in Table 2. The interpolation is performed such that the nearest bitmask value is stored in the cube spaxel.

Table 3 provides a reference for the bits used to flag various issues in the data. The bitmask format offers the compact representation for a large array of flagging options. A user can select which flags they wish to include, or simply use the default setting where any nonzero value in the mask is flagged. In this section we provide additional detail on each bitmask flag, in the order they are encoded into the model.

2.3.1. MAIN (Pipeline Reduction Flags)

2.3.2. FTF (Fiber-to-fiber Outliers)

2.3.3. CHI2FIB (High χ² Fibers)

2.3.4. BADPIX (Bad Pixel Regions)

2.3.5. BADAMP (Amplifier Variability)

2.3.6. LARGEGAL (Large Galaxy and Planetary Nebula Masks)

We mask bright galaxies whose optical diameters D₂₅ (measured at the 25 mag arcsec⁻² isophote) exceed 1′. Their locations and shapes are provided by the Third Reference Catalog of Bright Galaxies (RC 3; G. de Vaucouleurs et al. 1991) and the Uppsala General Catalog of Galaxies (UGC; P. Nilson 1973). The Spring field contains 644 such galaxies; the fall field, 447. All detections that fall within 1.5 ×  the D₂₅ scale of a bright galaxy’s elliptical isophote are removed. This factor was determined by examining the HETDEX spectra at different scalings and ensuring that all detections related to the bright galaxy were encompassed in the aperture mask. These galaxy masks are consistently applied to the HETDEX Source Catalogs and the PDR1 data cubes.

Multiple planetary nebulae (PNe) were identified in the dataset due to an excess of bright emission lines found in the emission-line detection search. An example of a resolved PN is shown in Figure 4. Spanning several arcminutes in diameter, the PN G136.7+61.9 is seen in emission at 5007 Å across several VIRUS IFUs. We mask these in a similar manner as the large galaxy mask. The masked regions are listed in Table 4, with their central coordinate information and angular sizes provided by the HASH PN catalog³³ from Q. A. Parker et al. (2016). Note that all three PNe extend over multiple HETDEX observations due to their large angular extent. Flagged detections are removed from our catalog and masked as circular regions similar to the large galaxy masking.

Zoom In Zoom Out Reset image size

Table 4. Planetary Nebulae in HETDEX PDR1

PNG ID	Coordinates (R.A., Decl.)	shotid	r (arcmin)
PNG 085.3+52.3		20210510014	4.08
PNG 085.3+52.3		20220131013	4.08
PNG 144.8+65.8		20240314023	3.00
PNG 144.8+65.8		20240403008	3.00
PNG 136.6+61.9		20210309019	2.83
PNG 136.6+61.9		20210401013	2.83

Download table as:  ASCIITypeset image

Users interested in studying these masked galaxies or PNe can ignore this flag. Similarly, there are background sources that can be seen in the outskirts of these galaxies that are masked conservatively at 1.5 ×  the isophotal sizes. With visual inspection, this mask can be ignored, but the calibration may be off in the vicinity of bright sources. Users ignoring this flag should proceed with caution.

2.3.7. METEOR (Meteors)

2.3.8. BADSHOT

2.3.9. THROUGHPUT

2.3.10. BADFIB (Known Bad Fibers)

2.3.11. SAT (Satellites)

2.3.12. BADCAL (Calibration Residual Issues)

2.3.13. PIXMASK (Spectrum==0)

2.3.14. BADDET

The internal HETDEX data model is shown to be well matched to SDSS spectra (K. Gebhardt et al. 2021; E. Mentuch Cooper et al. 2023). In this section, we perform two tests to ensure that the extracted spectra from HETDEX data cubes are consistent with the internal model and with SDSS external spectral data.

2.4.1. Comparison with Internal Pipeline Spectra

HETDEX pipeline spectra are composed of spectral extractions using several fibers and adding their flux values based on PSF-weighted extraction (K. Horne 1986; K. Gebhardt et al. 2021). For point-source objects, this will be well approximated by a circular aperture extraction. In Figure 5, we provide examples of LAE spectra, focusing on a 100 Å window surrounding their emission-line wavelength. These panels display examples for sources with decreasing in S/N and compare spectra generated by the HETDEX pipeline in dashed purple to extracted circular (r = 35) aperture spectra generated by the data cubes. We see that for these emission-line sources, the spectra between the public and internal model are well matched.

Zoom In Zoom Out Reset image size

2.4.2. Comparisons with SDSS

We consider spectral comparisons between a sample of 8076 AGNs that are found in both HETDEX and SDSS in Figure 6. Spectral extractions are performed on the HETDEX IFU data cubes with 35 circular apertures. The median SDSS/HETDEX flux ratio is consistent with unity with a scatter in the agreement of under 20%. A modest offset of <10% is evident in the blue region on the spectrum.

Zoom In Zoom Out Reset image size

Two independent, but complementary, object-detection search techniques are performed as part of the main HETDEX reduction pipeline: one to identify emission lines, the other to detect continuum sources. In the fifth internal data release for HETDEX (HDR5), a search was performed across more than >6 × 10⁸ flux-calibrated fibers. The search method details can be found in K. Gebhardt et al. (2021). We briefly summarize the procedures here.

During this process, no imaging preselection is used; the HETDEX data itself provides object detection. Both the emission-line and continuum detection algorithms are designed to identify point sources and account for the variable image quality, or PSF of each independent HETDEX three-dither observation.

The raw detections are stored in HDF5 files. We include these files in the PDR1 data release and describe their content and access in Appendix A.

3.1.1. Emission-line Detection

3.1.2. Continuum Detection

3.1.3. AGN Catalog Selection

The internal HETDEX pipeline produced 32,643,783 raw detections for HDR3/4/5 line and continuum emission all combined. These files are included in this release and are described in Appendix A. This raw database is processed to a quality-controlled catalog (internal version “5.0.2”) of 4,627,812 detections. The outputs from the three object-detection methods (continuum, line, and AGN) are down-selected from the raw detection database through two primary steps: (1) artifact mitigation via the application of the mask model, and (2) filtering based on detection parameters.

1. 1.  
   Detection masking. Internal masking of HETDEX data is done at a detection level. For each issue described in Section 2.3, the relevant detection parameters are searched to determine if the source candidate should be flagged. For example, for meteors and satellites, a combination of observation name (e.g., shotid) and sky coordinates are used to decide whether to exclude the detection. In contrast, for calibration, amplifier, or fiber issues, the relevant instrument info is queried for an affected time range.
2. 2.  
   Detection parameter selections. For the line search, the following line-parameter ranges are required for catalog inclusion:

   

   Here, S/N and χ² are the signal-to-noise ratio and reduced quality of fit of the emission line, respectively, and sigma is the line width measured from the Gaussian line fit. continuum measures the local continuum around the HETDEX emission-line detection. It is derived from a linear fit to two 50 Å spectral windows placed 3× sigma on either side of the detected line. is the quality of fit of the fiber profile of the highest weighted fiber for the detection. High values of are associated with detection artifacts. We only include detections with an aperture correction factor, apcor, >0.45. This factor is the fraction of the 35 radius circular aperture with fiber coverage.For continuum sources, the only exclusion is based on apcor and g_HETDEX, the magnitude derived from the detection spectrum by folding the detected spectrum through the SDSS-g filter’s bandpass and summing the flux.

   

Both the line- and continuum-emission pipelines are optimized for point-source detection. For LAEs and many faint [O ii] emitters, this assumption is generally valid, although neither LAEs nor [O ii] emitters are truly pointlike. As shown in E. Mentuch Cooper et al. (2026), more than 50% of LAEs with S/N > 6 are better described by a two-component surface-brightness model consisting of a compact core and an extended exponential halo. Roughly 25% of HETDEX sources, such as z  <  0.5 galaxies, AGNs, and stars, are composed of multiple detections, arising either from extended emission or from multiple emission lines. In addition, bright sources can be independently detected in both the line- and continuum-emission searches.

To robustly associate detections and mitigate confusion from extended low-z [O ii] emission, we apply a 3D friends-of-friends (FOF) clustering algorithm. Using spatial and spectral linking lengths of 6″ and 8 Å, we combine emission features into coherent source groups as described in E. Mentuch Cooper et al. (2023). This procedure ensures multiline galaxies are properly grouped, although in some blended cases, background sources may be merged into a foreground object. This will cause some background sources in gravitational lenses to be merged with the foreground object; although, these can be discovered through independent searches (I. H. Laseter et al. 2022) within the expanded catalog described in Appendix B.

For each source, we designate a single representative detection, identified by the value in the selected_det column of the expanded catalog Detection Information Table (Appendix B). In the main catalog table, only this detection is printed for each source (including its exact detection position at RA_det, DEC_det); all other detection entries are removed. Non-LAEs are assigned the brightest g_HETDEX member, while LAEs are assigned the highest-S/N Lyα detection. This procedure produces a quality-controlled source catalog in which multiple detections are linked through the unique source identifier source_id. In summary, source_id provides a common identifier for all detections associated with the same astrophysical object, enabling the association of extended emission features, multiple emission lines, and the assignment of a consistent redshift and source classification.

Each HETDEX source group is assigned a classification and redshift using one of three methods depending on brightness and spectral features. For sources with g  <  22, we apply Diagnose (M. H. Debski et al. 2025; M. Debski & G. Zeimann 2024), a PCA-based template-fitting algorithm developed for the Hobby–Eberly Telescope VIRUS Parallel Survey (HETVIPS; G. R. Zeimann et al. 2024). Diagnose assigns classifications such as star, galaxy, or qso for bright sources, along with a redshift derived from spectral template matching. It recovers ∼97% of classifications reliably at g_HETDEX < 22. However, for fainter objects, confusion between Lyα and low-z lines demands that we rely on alternative methods.

Objects with little to no continuum emission are classified using ELiXer (D. Davis et al. 2023a). This program is based on the initial Bayesian work presented in A. S. Leung et al. (2017), incorporating equivalent widths, broadband imaging, and luminosity priors to infer line identity and redshift. Each detection is assigned a likelihood P(Lyα) of being Lyα, with a threshold of 0.4 chosen to balance recovery of true LAEs (96%) and [O ii] contamination in the LAE candidate sample (<3%). For high-redshift sources, ELiXer assigns the redshift of the most central detection, considering clustering of lines and multiline associations.

Redshifts and classifications for AGNs are adopted from the HETDEX AGN catalog (C. Liu et al. 2022, 2025), which provides redshift assignments for single broad-line detections (generally assuming the line is Lyα), for emission-line pairs indicative of AGN activity, and for broad-line sources identified through crossmatching with SDSS DR14.

A logical sequence is implemented to assign each source a best redshift and object classification. Priority is given to AGN matches, followed by Diagnose redshifts for brighter objects (g_HETDEX < 22), and ELiXer results otherwise. A histogram of spectroscopic redshifts is presented in Figure 7.

Zoom In Zoom Out Reset image size

We assess the accuracy of HETDEX redshifts through a comparison to two independent spectroscopic samples. The first is the COSMOS field spectroscopic redshift compilation created by A. A. Khostovan et al. (2026); this sample is skewed toward brighter sources, and, when restricted to objects with high-quality redshifts (flag ≥ 3), the crossmatched sample (with a matching radius of 10) consists of 395 LAEs, 303 AGNs, 347 low-redshift galaxies, and 4767 [O ii] emitting sources. We note that since COSMOS was used by HETDEX as a science verification field, the region was visited multiple times in the course of the survey. The data acquired during each visit were reduced and analyzed independent of the other visits; as a result, a single COSMOS source can potentially have different HETDEX redshifts and object classifications. The numbers listed above therefore represent the total number of redshift comparisons, not the number of unique sources on the sky.

The second sample comes from objects observed by the DESI–HETDEX survey (M. Landriau et al. 2025). This program concentrated on sources with HETDEX classifications that straddled the decision boundary between LAE and [O ii] galaxy, and is composed of 851 LAEs, 67 AGNs, and 139 [O ii] galaxies, all with high-quality (VI_QUALITY> 3) redshifts. Thus, our redshift comparison involves over 6800 measurements in both the local and high-redshift Universe. Note that both DESI and HETDEX measure the LAE redshifts assuming that it is a single Gaussian emission-line profile, even though it is not uncommon to observe a double-peak feature (M. Landriau et al. 2025). Using the mock LAE spectral profile in H. Khoraminezhad et al. (2025), we have checked that the impact of this assumption is given by Δz/(1 + z) ∼ 0.0005 for DESI and Δz/(1 + z) ∼ 0.0001 for HETDEX, where HETDEX has a smaller bias due to its slightly poorer spectral resolution than DESI.

Figure 8 shows the direct comparison between z_HETDEX and z_spec, as well as the distribution of normalized residuals, defined as (z_spec − z_HETDEX)/(1 + z_spec). The ensemble statistics demonstrate that HETDEX redshifts are unbiased, with an average offset of 0.0026 and a scatter of σ = 0.22. The normalized median absolute deviation is consistent with zero bias (NMAD = 0.0003), and the catastrophic outlier fraction (defined by ∣Δz∣/(1 + z) > 0.2) is 5.9%.

Zoom In Zoom Out Reset image size

As noted above, the DESI–HETDEX program specifically targeted faint emission-line galaxies near the classification boundary between Lyα and [O ii] and, therefore, consists primarily of objects with g  >  23. By design, there are no false positives or neighbor confusion. In contrast, the heterogeneous collection of object classes in the COSMOS spectroscopic compilation is biased toward brighter sources. Despite these differences, the catastrophic outlier fractions are nearly identical in the two samples: 5.97% for COSMOS and 6.03% for DESI–HETDEX.

In the COSMOS comparison, catastrophic outliers can arise from incorrect counterpart associations, including occasional HETDEX false positives or nearby on-sky neighbors within the matching radius. In contrast, the DESI–HETDEX sample is constructed from emission-line detections identified independently by both surveys, so catastrophic failures primarily result from incorrect emission-line identification. The most common case is confusion between Lyα and [O ii], which produces the two distinct outlier tracks visible in Figure 8. Overall, these results demonstrate that HETDEX redshifts are robust across the survey footprint, with only a modest tail of failures dominated by low-S/N detections or occasional misclassifications.

3.4.1. Source Classification

HETDEX sources are broken into five source_types. Source counts by field are provided in Table 1, and the magnitude distributions for each source type are shown in the histograms in Figure 9.

1. 1.  
   star—Galactic stellar sources; classified by Diagnose and assigned z_hetdex = 0.
2. 2.  
   agn—Active galactic nuclei; identified via the HETDEX AGN catalog or Diagnose; redshift range: 0 < z_hetdex < 4.6.
3. 3.  
   lae—Lyα emitters classified by ELiXer; redshift range: 1.88 < z_hetdex < 3.52.
4. 4.  
   oii—Low-redshift galaxies with detected [O ii] emission (λ3727 Å), classified via Diagnose or ELiXer; redshift range: 0 < z_hetdex < 0.48.
5. 5.  
   lzg—Low-redshift galaxies without detected emission lines classified by Diagnose using continuum template fits; redshift range: 0 < z_hetdex < 0.5.

Zoom In Zoom Out Reset image size

The HETDEX emission-line detection algorithm produces PSF-weighted line-flux values by assuming the object being measured is a point source. However, many nearby [O ii] line emitters are extended, resulting in underestimated measurements of the objects’ fluxes. In addition, the emission-line detection pipeline contains an upper limit on the continuum value, so very bright emission-line galaxies are completely missing from the emission-line database, although they are found in the HETDEX continuum catalog and are assigned a redshift. Without an associated emission line, they are initially classified as a low-redshift galaxy (an “lzg”). Figure 10 illustrates this issue. The two leftmost sources are only found in the continuum search method with no reported [O ii] line flux; in the third galaxy, the HETDEX-report line flux severely underestimates the total [O ii] emission from the galaxy.

Zoom In Zoom Out Reset image size

For these resolved galaxies, we supplement the HETDEX line-flux measurements with estimates of galaxies’ total line emission. A major strength of the wide-IFU (dithered) coverage provided by HETDEX is that the observations automatically produce an emission-line map of resolved galaxies. However, due to the grid-like distribution of IFUs in the HET’s focal plane, many of these systems have incomplete coverage, as their light extends beyond their IFU’s limits. We deal with this issue by using shape information from ancillary imaging observations either provided by HSC or DECam Legacy Survey (DECALs; A. Dey et al. 2019).

Isophotal elliptical apertures based on broadband imaging are included in HETDEX’s ELiXer classification tool (D. Davis et al. 2023a). The program applies Source Extraction and Photometry (SEP; K. Barbary 2016) to all available broadband imaging at the location of each HETDEX detection. We use the ELiXer catalog selected=True option (indicating this was the imaging counterpart of choice) and preferentially choose r-band measurements over g-band measurements to define each galaxy’s elliptical aperture. In general, the r-band imaging has a fainter limiting magnitude, and better image quality.

Elliptical parameters for each low-z galaxy are provided in HPSC2 (columns described in Table 5) under the columns major, minor, and theta. Additionally, the aperture center and the measured continuum aperture magnitude are in the expanded Detection Information Table in Appendix B in the columns ra_aper, dec_aper, mag_aper, and mag_aper_err.

Table 5. Source Observation Table Column Descriptions

Name	Description
source_name	HETDEX IAU designation (e.g., HETDEX J123449.19+511733.7).
source_id	HETDEX Source Identifier.
shotid	Integer observation ID: int(date+obsid).
ifuslot	String identifier of IFU location in focal plane.
RA	R.A. for source (ICRS deg).
DEC	Decl. for source (ICRS deg).
RA_det	Representative detectid R.A. (ICRS J2000 deg)
DEC_det	Representative detectid decl. (ICRS J2000 deg)
gmag	SDSS g magnitude measured in the HETDEX spectrum.
Av	Applied extinction correction in the V band.
z_hetdex	HETDEX spectroscopic redshift.
z_hetdex_src	Source/method for HETDEX spectroscopic redshift.
z_hetdex_conf	Confidence (0–1) in HETDEX spectroscopic redshift source.
source_type	One of star, lae, agn, lzg, oii.
detectid	Detection ID of representative detection (where selected_det = True in the Detection Info Table).
field	Survey field: dex-fall, dex-spring, cosmos, goods-n, nep, or ssa22.
flux	Extinction-corrected line flux in 10−17 erg s−1 cm−2 at wave. −999.0 for continuum sources but is correctly assigned in flux_lya or flux_oii.
flux_err	Markov Chain Monte Carlo (MCMC) error in extinction-corrected line flux.
flux_aper	Extinction-corrected [O ii] line flux in 10−17 erg s−1 cm−2 measured in the resolved galaxy aperture.
flux_aper_err	Uncertainty in flux_aper.
flag_aper	Aperture-flux usage flag for lum_oii: 1 = use flux_aper; 0 = use PSF line flux flux.
major	Major axis (arcseconds) of the resolved [O ii] aperture ellipse (from imaging).
minor	Minor axis (arcseconds) of the resolved [O ii] aperture ellipse (from imaging).
theta	Position angle (degrees) of the aperture ellipse.
logL_lya	Lyα luminosity erg s−1 from extinction-corrected flux.
logL_lya_err	Uncertainty in logL_lya.
logL_oii	[O ii] luminosity erg s−1 from flux if flag_aper=0 or from flux_aper if flag_aper=1.
logL_oii_err	Uncertainty in logL_oii.
flux_lya	Lyα flux in 10−17 erg s−1 cm−2 from extinction-corrected flux.
flux_lya_err	Uncertainty in flux_lya.
flux_oii	[O ii] flux in 10−17 erg s−1 cm−2 from flux if flag_aper=0 or flux_aper if flag_aper=1.
flux_oii_err	Uncertainty in flux_oii.
sn	Signal-to-noise ratio for line emission. Note that continuum sources (det_type==“cont”) will have sn = −999.0.
det_type	Detection type of the main detectid for the source: “line” or “cont.”
apcor	Aperture correction applied to the spectrum at 4500 Å.
p_conf	LAE/faint [O ii] emitter confidence score from RF classifier. 1 = high confidence, 0 = low confidence.
p_cnn	LAE/faint [O ii] emitter confidence score based on CNN classifier. 1 = high confidence, 0 = low confidence.

Note. Bad values are −999.0 for floating-point columns and n/a for string columns. This table is available in its entirety in machine-readable form at https://hetdex.org/data-results/ and on Zenodo.

Download table as:  ASCIITypeset image

For every galaxy (both “lzg” and “oii”) at z_hetdex < 0.5, we first construct a sky-subtracted narrowband image centered on the observed [O ii] wavelength (given by z_hetdex). The cube is collapsed spectrally over a ± 15 Å window to form the line-flux map. To remove continuum emission, we create two additional 50 Å wide collapsed images placed ± 10 Å outside the line-map limits. The average of these two continuum images is subtracted from the narrowband image, producing a continuum-subtracted [O ii] emission-line-flux map.

Because many resolved galaxies extend beyond the HETDEX IFU footprint, only part of the elliptical aperture defined from broadband imaging is actually covered by IFU data. After generating the continuum-subtracted map, we therefore measure the [O ii] flux within the portion of the aperture that has fiber coverage, compute the fraction of the aperture that is sampled, and scale the measured flux by the inverse of this coverage fraction. This yields an aperture-corrected estimate of each galaxy’s total [O ii] flux assuming the flux distribution is smooth and evenly distributed within the aperture. In practice, emission-line regions can exhibit radial structure and clumpy morphology, so this correction should be regarded as an approximate estimate of the total flux. The aperture-corrected fluxes and their uncertainties are provided in the catalog under the columns flux_aper and flux_aper_err, respectively.

A significant number of low-z galaxies exhibit a positive [O ii] line-flux value within the resolved galaxy aperture, despite an absence of an [O ii] flux detection in the initial pipeline database. While the redshift of these systems remains unchanged, their classification is revised from “lzg” to “oii.” Using this method, 32,031 sources in the HPSC2 catalog have been reclassified as [O ii] emitters from an initial classification as a low-z passive galaxy.

For the two main emission-line datasets, we include columns for their continuum-subtracted, extinction-corrected line fluxes (flux_oii, flux_lya) and corresponding luminosities (logL_oii, logL_lya), along with their respective uncertainties marked by the _err subscript. This information is only relevant to sources with [O ii] and Lyα emission lines and pertains exclusively to objects categorized with source_type as “oii,” “lae,” or “agn.” Only the AGN with Lyα-emission within the HETDEX spectral range will have reported flux values. For accessing other emission-line-flux data, users are directed to Appendix B.

The native source detection pipeline as described in K. Gebhardt et al. (2021) is designed to deliver continuum-subtracted line fluxes under the assumption that the source is unresolved. However, many HETDEX sources are resolved. Moreover, an additional shortcoming in the HETDEX detection procedure is that an upper bound on continuum counts is set during the search for emission lines. As a result, the line flux emitted at the central location of a bright [O ii] emitter or AGN is not reported.

As outlined in Section 3.5, the issue is addressed for [O ii] emitters by fitting spatially resolved line fluxes for each system with z_hetdex < 0.5. When the aperture flux, denoted as flux_aper, exceeds the pipeline-reported flux, this spatially integrated flux measurement value is assigned to flux_oii, and the flag_aper is set to 1. Conversely, if the HETDEX pipeline’s estimate of the [O ii] surpasses the aperture flux, the pipeline flux is preserved in flux_oii, and flag_aper is set to 0. If no [O ii] flux is associated with the source, the flag_aper is assigned a value of −1, and the source is not to be classified as an “oii.”

For LAEs, it is generally acceptable to assume that their morphology resembles that of a point source, especially considering the relatively poor image quality delivered by the HET. Research focused on quantifying surface-brightness profiles and determining isophotal radii suggests that for a subset of LAEs with an S/N > 6 (E. Mentuch Cooper et al. 2026), an exponential model with a scale length of ∼10 kpc is a more accurate representation of LAE morphology. This improved model increases the reported Lyα flux by a factor of ∼1.3 over the point-source calculation as reported in E. Mentuch Cooper et al. (2026). However, we have chosen not to implement any corrections to adjust for this discrepancy. We also have opted to not provide spatially resolved flux values for extended Lyα emitters; these measurements can be found in E. Mentuch Cooper et al. (2026).

As mentioned above, some AGNs are often bright enough so that their continuum counts are above the maximum allowed for an emission-line search. In these objects, the Lyα line flux is measured by subtracting the surrounding continuum and fitting a multicomponent Gaussian model to the emission line (C. Liu et al. 2025). These line-flux values are reported in flux_lya in HPSC2. For AGNs whose Lyα emission lies outside the spectral range of the VIRUS spectrographs, no flux is reported. Readers that desire more advanced line-flux values are referred to the AGN catalogs created from HETDEX data release HDR4 (C. Liu et al. 2025).

3.6.1. Extinction Correction due to Dust Attenuation

ML and AI tools provide efficient methods for data classification, and the HETDEX collaboration has explored several such approaches. Unsupervised learning has proven valuable for identifying supernovae (J. Vinkó et al. 2023) and AGNs (V. Tardugno Poleo et al. 2023) within the rich HETDEX dataset, while also offering interactive methods for artifact removal and sample cleaning.

For HPSC2, ML/AI methods have been most successful in identifying and removing artifacts from the raw detection database and improving the low-S/N emission-line sample. In this section, we describe three approaches that have proven particularly effective and that are incorporated into HPSC2 and the PDR1 data cube mask model.

3.7.1. Rapid Automatic Image Categorization

Every emission-line and continuum detection in the HETDEX catalog is passed through a classification workflow using the RAIC Labs.⁴⁰

For each emission-line detection, we generate a narrowband cutout, 20″ on a side, with the wavelength dimension collapsed over ±2σ around the fitted central wavelength of the line, where σ is taken from the Gaussian profile fit produced by the HETDEX pipeline. No continuum subtraction is performed. For continuum detections, cutouts of the same size are produced by collapsing the spectrum over the rest-frame wavelength range 4000–5000 Å. In both cases, the detection of interest is centered within the cutout. These narrowband cutouts are uploaded to the RAIC Labs classification platform, which projects the image set into a low-dimensional feature space where unsupervised learning naturally groups detections with similar morphology. The tool provides an interactive GUI in which users can visually explore this feature space and interactively choose selections to assign labels to groups of sources simultaneously. This approach enables efficient labeling of large numbers of detections with consistent morphological classifications.

The resulting label sets are shown in Figure 11. For emission-line detections, the labeled sets are point source, extended object, streak, and meteor. For continuum detections, the categories include saturated object, calibration issue, star, bad dither, bad fiber, satellite, galaxy, streak, low counts, and faint star. Each category is built up to include at least 1000 example labels.

Zoom In Zoom Out Reset image size

Once a sufficient number of labels has been collected via the RAIC Labs interactive GUI, the system trains its internal classification model to apply those labels to previously unlabeled detections. Each source in the HETDEX catalog (emission-line or continuum) is passed through the relevant classification model, and is assigned a likelihood score, set between 0.0 and 1.0, for each label for each detection. This process means that any HETDEX detection can be assigned multiple labels. Those that are above the threshold classification score of 0.5 are assigned that label.

The labels for “meteor” and “satellite” are used to help identify contaminants in the catalog and create the masking for each category, respectively (see Sections 2.3.7 and 2.3.11.) From the continuum labels, we flag the labels “calibissue,” “baddither,” and “badfiber” from the catalog when their label score is >0.3. From the emission-line labels, we use the flag and remove any detections with the “streak” label and a score >0.35.

From our internal quality-controlled catalog 113,954 detections are flagged by RAIC. We do not include these in any public catalogs. While many of these artifacts are found through our other automated flagging procedures, 41,002 are only found using the RAICLabs AI classification platform; this demonstrates the success of this software as a tool for efficiently labeling and classifying HETDEX detections. These detections are masked in the data cubes under the BADDET bitmask described in Section 2.3.14.

3.7.2. Citizen Science + Unsupervised Learning

3.7.3. CNN Classifier: p_cnn

3.7.4. Random Forest Classifier: p_conf

We train a single, compact random forest (RF) classifier to distinguish high-confidence LAEs from low-confidence detections in the COSMOS Legacy field. This field benefits from multiple HETDEX observations, which allow for confirmation of genuine LAEs, and provides a set of high-confidence LAEs suitable for use in supervised machine learning methods.

Training is restricted to COSMOS LAE emission-line detections and excludes any objects previously classified as AGNs. It is further limited to faint objects with g_HETDEX > 22. Classification labels are binary, with 1 denoting a high-confidence LAE candidate and 0 denoting a low-confidence LAE candidate. It is important to note that verifying that an HETDEX detection is real is much more straightforward than proving the candidate is noise, unless the object exhibits characteristics of an obvious artifact. Because HETDEX is dominated by low-S/N emission lines, uncertainties in the noise model lead to detections with questionable validity.

High-confidence classifications originate from several sources. HETDEX observed some regions of the COSMOS Legacy field multiple times; any LAE detected in a second observation beyond its original detection is considered a validated source. We also include objects confirmed through positional and redshift and/or spectral matches to other spectroscopic programs (via the compilation by A. A. Khostovan et al. 2026), or deep narrowband surveys. Included in the latter category are SILVERRUSH (z = 2.2 and 3.3; S. Kikuta et al. 2023), ODIN (z = 2.4 and 3.1; N. M. Firestone et al. 2024), and the SC4K narrow- and medium-band surveys (z ∼ 2.2–3.4; D. Sobral et al. 2018). To increase sample size and better match the distribution of HETDEX LAEs, we also include sources classified as “Real” in the citizen science project DEE (L. R. House et al. 2023), with a confidence score of p_DEE ≥ 0.8. In total, 2935 sources are assigned real. The 4582 low-confidence sources come exclusively from DEE with p_DEE ≤ 0.1.

We explore a number of catalog columns as possible feature vectors for our LAE training set. Ultimately, we find that a compact feature set performs well, relying only on parameters measured directly from the detection pipeline:

where wave is the wavelength of the line, sn is the pipeline’s signal-to-noise, chi2 is the reduced χ² value of the Gaussian line-fits to the possible feature, linewidth is the standard deviation for the line width Gaussian, and continuum is the pipeline’s fitted value for the continuum underlying the emission line.

We use scikit-learn’s RandomForestClassifier (version 1.5.2) with standard settings (F. Pedregosa et al. 2011), including n_estimators=100, criterion=‘gini’, and max_features=‘sqrt’. The classifier’s output is a probability-like score, p_conf, assigned to each detection in our catalog. As with the CNN classifier, p_conf is computed for both the full LAE sample and for a subset of fainter [O ii] emitters. Our Random Forest model achieves an accuracy of 0.89 and an area under the receiver operating characteristic curve (AUC) of 0.95 when trained on COSMOS repeat detections and ancillary spectroscopic matches. It reliably separates high-confidence LAEs from low-confidence detections, with most predictive power supplied by the sn (0.303) and continuum (0.286) features. The remaining features—wave, linewidth, and chi2—provide smaller but meaningful contributions, with relative importances of 0.189, 0.115, and 0.107, respectively.

To determine the optimal thresholds that maximize the LAE sample size and minimize false-positive contamination, we take a systematic evaluation approach that is demonstrated in Figure 12. This figure shows how the confirmation rate of a subset of LAEs from the COSMOS Legacy field varies with a number of different selection criteria thresholds. As stated above, it is difficult to prove with certainty that a low-S/N emission-line detection is not real unless it is visually verified as an artifact: when the completeness fraction is below 50%, it can take a prohibitively large number of observations to say with >90% confidence that a candidate flagged by the HETDEX pipeline is a false detection. Conversely, it is relatively straightforward to prove an object is real. If the object is found in more than one HETDEX observation, or it has already been cataloged by an independent spectroscopic or narrowband survey, then there is no question that the object exists. Thus, instead of plotting the false-positive rate on the y-axis, we plot the confirmation rate, which can be considered an upper limit on the inverse of the false-positive rate.

Zoom In Zoom Out Reset image size

The key to measuring the confirmation rate is to select a sample in which confirmation is possible. While the entire COSMOS field has a large number of spectroscopic redshift sources, provided by the compilation by A. A. Khostovan et al. (2026), and narrowband sources that overlap with the HETDEX LAE population, the majority of HETDEX objects will not be found in ancillary studies. Conventional spectroscopic surveys, which require a broadband detection to identify candidates for follow-up redshift measurements, generally target brighter galaxies (with consequently larger stellar masses) than the average HETDEX LAE. Moreover, while the physical properties of the LAEs identified in deep narrowband surveys, such as SILVERRUSH, ODIN, and SC4K, are similar to those of HETDEX LAEs, these programs target very narrow redshift slices of the Universe. The parameter space explored by HETDEX is quite unlike any other LAE survey except for the small coverage () provided by the MUSE COSMOS sample (i.e., T. Urrutia et al. 2019). Operating with longer exposure times and reaching deeper flux sensitivity, this program is too limited in size (and redshift overlap) to provide useful comparison.

The most effective method to confirm an HETDEX LAE is by using HETDEX itself. Unfortunately, due to the large variation in observing conditions, the flux sensitivity of any given observation can be vastly better than another. One repeat observation alone is generally not enough to confirm. Most HETDEX LAEs straddle the 50% completeness limit in a typical HETDEX observation and in that regime, it takes at least five independent observations to excluded the reality of a detection with 95% confidence. By design, we limit the COSMOS confirmation sample to those observations that contain at least eight overlap apertures. This results in a small set of 414 LAEs that have the potential to be confirmed by repeat HETDEX observations as well as ancillary spectroscopic and narrowband surveys. Future dedicated HETDEX validation observations at the HET/VIRUS (obtained by PI Laurel Weiss in 2025) will expand this statistical baseline.

For any threshold on a classifier score, we define the confirmed LAE fraction as

where the denominator is the number of LAEs in the confirmation sample whose classifier scores exceed the threshold (either a selection-cut based on S/N, p_cnn, or p_conf or a combination of these), and the numerator is the subset of those LAEs that are confirmed by repeat HETDEX observation or by ancillary COSMOS catalogs in that same sample.

In Figure 12, the solid dark-blue curve shows the confirmation fraction for different p_conf thresholds, described in Section 3.7.4, as a function of the total number of LAEs included in the full HETDEX catalog sample above the indicated threshold; this is annotated with text for clarity. As expected, for all metrics, we see a trend where higher thresholds correspond to an increased confirmed LAE fraction (and consequently less contamination from false positives), indicating that the use of more stringent criteria results in a more reliable selection of LAEs. However, this comes at the cost of decreasing the overall sample size, as fewer candidates meet the stricter selection criteria. For example, a threshold of p_conf = 0.5 increases the confirmation fraction from 62% to 80% but reduces the sample of LAE candidates from 1.6 million candidates to 600,000 LAE candidates. We therefore adopt p_conf = 0.5 as a practical compromise between confirmation fraction and sample size, as higher thresholds yield only modest additional improvements in confirmation while rapidly reducing the number of LAE candidates available for statistical studies.

Variation based on thresholds for the CNN classifier, p_cnn, described in Section 3.7.3, is shown in the dashed purple line. Confirmation of the sample increases as the threshold for p_cnn is raised, but the sample size decreases significantly. As the curve plateaus at p_cnn ∼ 0.3–0.7, the confirmation rate does not indicate a single obvious threshold from this metric alone. In S. Mukae et al. (2026), additional criteria are explored, including the recovery of known LAEs in the DESI–HETDEX validation sample (M. Landriau et al. 2025) and ancillary COSMOS catalogs. These analyses show that adopting a CNN threshold of p_cnn = 0.5 provides a practical balance between completeness and contamination, recovering the majority of confirmed LAEs while maintaining a catalog size close to the survey’s target LAE number density. Because the CNN score uncertainties are typically ∼0.2, sources with scores in the range ∼0.3–0.7 are intrinsically ambiguous, making a threshold near 0.5 a natural dividing point between high- and low-confidence classifications.

To compare against alternative selection criteria, we also compute (i) a combined RF + fixed value CNN sequence requiring a fixed p_cnn = 0.5 alongside varying p_conf (light blue), (ii) pure S/N cuts at S/N ∈ {4.8, 5.2, 5.5, 6.0, 10.0}, and (iii) pure CNN cuts at thresholds from 0 to 1.

It is crucial to strike a balance between maximizing the confirmation rate and preserving a sufficiently large sample for statistical analysis. In this release, we adopt thresholds of p_conf ≥ 0.5 and p_cnn ≥ 0.5 to provide a robust sample while maintaining a sufficiently large catalog for statistical analyses. The choice of p_conf = 0.5 reflects a balance between confirmation fraction and sample size, as higher thresholds yield only modest improvements in confirmation while rapidly reducing the number of LAE candidates. The CNN threshold of p_cnn = 0.5 is motivated by the analysis of S. Mukae et al. (2026), which shows that this value provides a practical balance between completeness and contamination while preserving the survey’s target LAE number density. These values reduce the LAE sample from 1,632,604 to 426,654, and the [O ii] sample from 559,544 to 491,411.

Future work will involve exploring alternative selection methods and thresholds to optimize these criteria. For users who are interested in exploring a less stringent selection criteria, we provide the full suite of LAE candidates in the Detection Information Table in Appendix B. Candidates at lower p_conf and p_cnn values are included. This flexibility allows future users to tailor their analyses based on specific scientific goals.

The file ifu-index.fits (also available in .h5/.txt format) is a master look-up table for each IFU data cube observation made by HETDEX.

Table 6 describes the columns from the IFU index table. Each row in the table represents a single IFU observation, with the parent observation labeled as shotid. An HETDEX observation will have between 14 and 78 IFU observations, identified by column ifuslot. The shotid/ifuslot combination provides the unique identifier of an HETDEX IFU data cube.

Table 6. IFU Metadata Schema

Column Name	Description	Data Type	Units
shotid	Shot ID integer representing a shot: DATEOBS	int64	⋯
ifuslot	IFU slot number (e.g., 046, 063, etc.)	str	⋯
ra_cen	R.A. of IFU center (ICRS J2000)	float32	deg
dec_cen	Decl. of IFU center (ICRS J2000)	float32	deg
flag	Fraction of IFU with usable data (1 = fully usable, 0 = fully flagged)	float32	⋯
flag_badamp	Fraction of IFU unaffected by bad amplifier(s)	float32	⋯
flag_badfib	Fraction of IFU unaffected by bad fiber(s)	float32	⋯
flag_meteor	Fraction of IFU unaffected by meteor trail contamination	float32	⋯
flag_satellite	Fraction of IFU unaffected by satellite trail contamination	float32	⋯
flag_shot	Fraction of IFU unaffected by shot-level flagging	float32	⋯
flag_throughput	Fraction of IFU unaffected by low throughput	float32	⋯
field	Field identifier (e.g., “dex-spring”)	str	⋯
objid	Object ID (string identifier)	str	⋯
date	Date of the observation (YYYYMMDD)	int32	⋯
obsid	Observation index	int32	⋯
ra_shot	R.A. of shot center (ICRS J2000)	float64	deg
dec_shot	Decl. of shot center (ICRS J2000)	float64	deg
paa	Position angle of focal plane	float64	deg
n_ifu	Number of active IFUs in the shot	int32	⋯
fwhm_virus	Seeing FWHM from VIRUS	float32	arcsec
fwhm_virus_err	Error in seeing measurement	float32	arcsec
response_4540	Effective normalized system throughput at 4540 Å in 360 s	float32	⋯
ambtemp	Ambient temperature	float32	°C
datevobs	Observation date string	str	⋯
dewpoint	Dew point	float32	°C
exptime	Exposure time	float32	s
humidity	Relative humidity	float32	%
mjd	Modified Julian Date	float32	d
nstars_fit_fwhm	Number of stars used for seeing estimate	int32	⋯
obsind	Observation index (unused)	int32	⋯
pressure	Barometric pressure	float32	Torr
structaz	Telescope azimuth	float32	deg
time	Start time of exposure	str	UTC
trajcdec	TRAJDEC from telescope header	float32	deg
trajcpa	TRAJPA (position angle) from telescope header	float32	deg
trajcra	TRAJRA from telescope header	float32	deg

Notes.

^aThe pa column is the HET tracker rotation angle. The position angle east of north on the celestial sphere is given by .This table is available in its entirety in machine-readable form at https://hetdex.org/data-results/ and on Zenodo.

Download table as:  ASCIITypeset image

The central sky coordinate of each IFU is listed in Table 6. Users can search for data cubes that overlap within an arcminute of this coordinate and then check the WCS of the data cube for exact coverage. An example query is given in Section 4.5.1. Observational quality assessment values for seeing (FWHM), and transparency (response_4540) are tabulated to allow for querying of desired shots. We include the fwhm_virus value, which is produced by the HETDEX pipeline based on VIRUS observations.

For each IFU, the fraction of the data cube flagged due to bad amplifiers, bad fibers, meteors, and satellites is reported in the aptly named columns. Of the full set of IFU data cubes, 91.6% of the data remain unflagged, meaning that 8.4% of the data cubes are flagged at all wavelengths. Breaking this down by source of contamination, 7.8% of the total data cube volume are flagged due to bad amplifiers, 0.2% due to bad fibers, 0.12% due to satellites, and 0.04% due to meteors. Additional wavelength-dependent masking can occur; this mask is contained in the BITMASK HDU.

The creation of the primary catalog delivered in this release is the HETDEX Public Source Catalog 2 (HPSC2), which was described in Section 3. Here we describe its data model. HPSC2 provides one row per source observation and summarizes the key physical and observational properties for each object. HPSC2 includes the source position, redshift, physical size when relevant, and the dust- and aperture-corrected [O ii] or Lyα fluxes and luminosities. It contains all necessary information for most scientific analyses. To support more specialized use cases, we also provide the full Detection Information Table, described in Appendix B, which lists every individual HETDEX line emission and continuum detection that passes the quality checks and line-parameter criteria in Sections 2.3 and 3.2. This expanded table contains per-detection measurements and metadata for all HETDEX observations of a source and can be used to trace each HPSC2 entry back to its constituent detections. While the detailed Detection Information Table enables advanced analyses, HPSC2 provides a streamlined and user-friendly summary appropriate for most users.

An HETDEX source, identified by source_id, is a collection of all detections at the same on-sky position combined through the detection grouping method described in Section 3.3. If the source is observed more than once, its source_name will be the same, but the source_id will be different, as will the reported catalog measurements. For each source_id row, we report a single representative detection identifier, detectid, which may be matched to the Detection Information Table for each source observation in the detectid column; this column corresponds to the detection member with the brightest (i.e., smallest) g_HETDEX value for all sources that are not LAEs. For LAEs, the highest-S/N Lyα line detection is the selected representative detectid. A user may search the Detection Information Table for this representative detectid by the selecting the column selected_det==True.

For sources identified as lae or oii, we provide the objects’ Lyα or [O ii] line-flux in the columns flux_lya and flux_oii, along with the corresponding errors on these quantities. As discussed in Section 3.5, for each low-z galaxy, an aperture [O ii] line flux is measured: flux_aper at z_hetdex. This flux is assigned as the source’s flux_oii if it is a positive value and the major axis of the galaxy (as measured on broadband imaging) is >2″; otherwise, the measured line flux comes from the native pipeline flux value obtained from the line fit to the extracted spectrum of the brightest detectid in the source group. Line fluxes and associated errors are converted to intrinsic [O ii] and Lyα line luminosities (denoted logL_lya and logL_oii) using our best measured redshift, z_hetdex, and the cosmological model defined by Planck Collaboration et al. (2020).

The following files are included in this release:

1. 1.  
   HETDEX Public Source Catalog 2 (HPSC2, columns described in Table 5): hetdex_sc2_vX.X.dat/.fits. This table consists of one row per source observation. For each source observation, it provides the source’s J2000 equatorial coordinates, and redshift (z_hetdex). Every source is classified into one of the following source_type options: lae, oii, agn, lzg, or star as described in Section 3.4. For sources with either Lyα or [O ii] line emission, the table provides the optimal measurement for the dust-corrected, aperture-corrected flux, and luminosity in flux_lya, flux_oii, logL_lya, and logL_oii.
2. 2.  
   HPSC2 spectra in FITS format: hetdex_sc2_spec_vX.X.fits. Each row in HPSC2 provides the corresponding 1D extracted spectra in an FITS file format consisting of four Header Data Units (HDUs). Multiple HDUs are included as listed in Table 7. The primary HDU is empty. HDU1:INFO contains a copy of the main HPSC2 catalog. At the same row index for each source in this table, HDU2:SPEC and HDU3:SPEC_ERR contain the aperture-corrected, 1D PSF-weighted extinction-corrected spectra and their associated uncertainties in , computed according to the procedure outlined in Section 3.6. The final HDU4:WAVELENGTH is a 1036 array corresponding to the spectral dimension in angstroms. All spectra have the same spectral range from 3470–5540 Å in steps of 2 Å.
3. 3.  
   Supplemental Detection Information Table (columns described in Table 11): hetdex_sc2_detinfo_vX.X.dat/.fits. This table contains specific information for every quality-controlled line emission and continuum detection. Every emission-line detection row contains all parameter information, including the emission line’s observed wavelength, its fitted parameters, and measured flux. If the observed wavelength corresponds to a commonly found spectral species⁴² at redshift z_hetdex, the species is indicated in the column line_id. There are also several columns related to imaging counterpart matches, redshift assignments, and emission-line classification as found by ELiXer (D. Davis et al. 2023a). Also included are a number of columns containing details about the specific observation, the instrument, and the detection grouping parameters. A full column description is provided in Table 11. Detailed information concerning this catalog is provided in Appendix B.

Table 7. Format of the HETDEX Public Source Catalog 2 (HPSC2, Columns Described in Table 5) Spectra FITS File

HDU Name	Type	Dimensions	Description
0:PRIMARY	PrimaryHDU	...	...
1:INFO	BinTableHDU	7367R × 27C	Source information for each catalog source, one row per source observation.
2:SPEC	ImageHDU	(1036 × 1,107,763)	Extinction- and aperture-corrected, PSF-weighted 1D spectrum at the source’s representative detectid at position RA_det, DEC_det in units of .
3:SPEC_ERR	ImageHDU	(1036 × 1,107,763)	Uncertainty in SPEC.
4:WAVELENGTH	ImageHDU	(1036,)	Wavelength array from 3470–5540 Å in 2 Å bins.

Note. This file contains the full HPSC2 table (Table 5) in HDU1 (also available in a simple .dat ASCII format). The spectrum provided is a 1D PSF-extracted spectrum at the representative detectid location. Different spectral extractions can be obtained from HETDEX data cubes directly.

Download table as:  ASCIITypeset image

An essential diagnostic for HETDEX detections is provided by the ELiXer reports that are produced for every detection in the quality-controlled HETDEX catalog. A full description of an ELiXer detection report is provided in the Appendix of D. Davis et al. (2023a). These reports provide a summary of ancillary imaging data at the location of the detection. Small cutouts of the amplifier array containing the four highest weighted fibers that contribute to the PSF-extraction of the detection. Also included is a full plot of the HETDEX PSF-extracted spectrum, associated error with an expanded view of the spectral region of the detection. These reports are used to visually validate HETDEX detections. We include in this release 3,296,101 ELiXer reports for all detections in full supplemental HETDEX catalog described in Appendix B. This includes every detection in the main HPSC2 catalog.

They can be accessed individually through the HETDEX data mount at https://web.corral.tacc.utexas.edu/hetdex/HETDEX/pdr/pdr1/detect/elixer/. The files are organized based on the first five digits of their detectid into directories to allow for easier navigation. For example, detectid=4022202460 is located at https://web.corral.tacc.utexas.edu/hetdex/HETDEX/pdr/pdr1/detect/elixer/40222/4022202460.jpg.

There are two primary ways a user can access HETDEX public data: (1) use the IFU index table to determine dexcubes of interest and then download these cubes locally for personal analysis, or (2) utilize a public JupyterHub access point. In this section we outline these two points of access and present some basic data access examples including making line-flux maps at a spectral region of interest and doing 1D spectral extractions.

4.4.1. Remote Access

For users without direct HETDEX JupyterHub access, data cubes may be downloaded manually. First, you will need to download the IFU index file from https://web.corral.tacc.utexas.edu/hetdex/HETDEX/pdr/pdr1/ifu-index.fits. Then files can be downloaded through a terminal using wget:

where url = http://web.corral.tacc.utexas.edu/hetdex/HETDEX/pdr/pdr1/datacubes/20190405020/dex_cube_20190405020_034.fits, for example. The options here ensure the same file structure as is located on the host directory system so that only minor edits of the base path are require in the tutorial notebooks.

The script can be easily adapted to download many cubes by looping over lists of shotid and ifuslot values. For examples of downloading data cubes, including batch parallel downloading options, see https://github.com/HETDEX/dexcube/blob/main/notebooks/02-DownloadingCubes.ipynb and https://github.com/HETDEX/dexcube/blob/main/notebooks/12-BatchDownloads-ForRemoteUsers.ipynb.

4.4.2. JupyterHub Access

A public JupyterHub is available for all public users through the Texas Advanced Computing Center at https://jupyter.tacc.cloud/. The JupyterHub provides access to the full data release and is the most efficient way to perform bulk actions on the survey data. For example, one can perform searches on a catalog of coordinates and extract spectra, create narrowband like images and run other analysis packages. We also include an interactive widget, CubeWidget, to explore HETDEX IFU cubes. An example of CubeWidget is shown in Figure 14.

Zoom In Zoom Out Reset image size

A TACC account is required for access, but users do not need to be associated with a project allocation. More details on JupyterHub access can be found at https://hetdex.org/data-results/. No persistent storage is offered to public users. Tutorial notebooks can be run and edited, but when the user’s server is shut down, the server will reboot back to the original container. It is strongly recommended you should download your work and any data products that you produce.

The best approach for a user to become familiar with the data model is to run the notebooks found on the dexcube repository.⁴³ These tutorials introduce the data model and provide simple examples of querying the survey and downloading data cubes of interest for remote users. Users will generally want to do two things with HETDEX data cubes: (1) extract spectra at a list of catalog coordinates and (2) create line-flux maps at specific wavelength and coordinate locations. In this section we demonstrate, using Python, how to query, access and interpolate HETDEX data cubes.

4.5.1. Coordinate Query Example

The following Python snippet demonstrates how to query the HETDEX survey for data cubes near a given sky coordinate. This example searches for IFU observations within 37″ of a Lyα nebula at z = 2.53.

4.5.2. Spectral Extractions

Spectra can be extracted from HETDEX data cubes by applying a circular aperture at a given sky position and summing the flux across spatial pixels within the aperture. This process is the most direct way to obtain 1D spectra from the 3D cubes, and it can be adapted depending on the science case. A simple example is shown in Figure 15. Additional examples are provided on the dexcube repository.⁴⁴

Zoom In Zoom Out Reset image size

For more robust use cases, we provide a dedicated spectral extraction script, get_spectra.⁴⁵ This tool performs proper aperture corrections and allows for a variety of masking options. For example, in Figure 15, we removed the BADCAL mask flag so that the AGN’s Lyα-emission line at λ = 3563 Å could be recovered, even though it overlaps with a region that is typically masked. Users are encouraged to adapt the example code to their own needs and apply masking strategies appropriate for their targets.

Finally, while spectra can be extracted directly from the full data cube, efficiency can often be improved by slicing the cube along the spectral axis to limit the wavelength range of interest. This reduces memory load and speeds up the extraction when only a narrow spectral region is required.

4.5.3. Collapsing Cubes to Make Line-flux Maps

To visualize emission at a known wavelength, data cubes can be collapsed along the spectral axis across a small wavelength window around the targeted line. This process produces a pseudonarrowband “line-flux” image. A similar procedure can be performed on an adjacent/wide-pass set of wavelengths to build a “continuum” image, which can be scaled and subtracted to remove underlying continuum. During this interpolate process, masked pixels (bitmasks or zeros) should be set to “NaN” before summation so they don’t bias the collapse.

Below is an example of collapsing in a ±Δλ window around Lyα at a known observed wavelength. In this case, we collapse ± 10 Å at λ = 4295.9 Å. No continuum subtraction is done here, but an example is available online.⁴⁶

In Figure 16, we extend this simple example by collapsing the data cube into narrowband and continuum images and then visualizing their spatial distribution across multiple epochs for the same changing-look AGN shown in Figure 15. For each observation covering the AGN, we build (i) a line-flux map by collapsing the cube within ±50 Å of Lyα at 3563 Å and (ii) a continuum map from 3800–5200 Å. Square cutouts centered on the AGN are extracted with consistent physical size, but the orientation here is in the IFU plane, so they are not matched in WCS. Lyα narrowband images are overlaid as contours on top of the gray-scale continuum maps made from the HETDEX data cubes. This approach highlights the combined spatial and spectral advantage offered by HETDEX observations.

Zoom In Zoom Out Reset image size

During its first year (2017 January 1 to 2018 January 1), VIRUS operated with fewer than 20 IFUs. The data quality was poor, with about half of the amplifiers flagged as bad, and detectors were plagued by pox and charge traps. Over time, IFU units were replaced, and new generations of detectors proved to be more stable and clean of artifacts. As a result, early catalog sources and IFU data cubes are of subpar quality. Early observations can be identified using the shotid variable. Values beginning with 2017 correspond to early data and should be used with caution. The shotid appears in the naming structure of each data cube and is included in all source catalogs. We do not recommend the use of LAE samples from this era without manual inspection. Users should be cautious; these data are included mainly for follow-up studies of external samples.

HETDEX observations are very heterogeneous. The sensitivity of each spaxel varies considerably depending on wavelength, detector/amplifier, and observing conditions. The flux sensitivity of HETDEX is quantified through extensive source injection simulations calibrated against the survey noise properties (K. Gebhardt et al. 2021). In these simulations, artificial emission lines are added directly into the fiber spectra across the full wavelength range of VIRUS, and recovery rates are measured using the same detection algorithms applied to the real data.

The survey reaches ∼50% completeness at a line flux of 1.1 × 10⁻¹⁶ erg s⁻¹ cm⁻², depending on wavelength and seeing conditions, based on simulations. Sensitivity decreases toward the blue end of the spectrograph, but remains relatively uniform redward of λ ≃ 4700 Å.

We do not include completeness corrections, as they need to be appropriately adapted and tested with the interpolated data cube model presented in PDR1. As a result, PDR1 is not suited for direct luminosity function or number density measurements of faint line emitters without additional modeling. While PDR1 provides the essential raw data products to do so, the completeness models needed to translate catalog detections into population statistics are beyond the scope of the release.

Users may attempt their own source detection algorithms on the released data cubes; however, this should be done with caution. The primary targets of HETDEX, LAEs, are extremely faint emission-line sources that typically occupy only a couple spatial and spectral resolution elements in the VIRUS IFU data. As a result, noise fluctuations and instrumental artifacts can occasionally mimic emission-line detections.

We have made a concerted effort to provide systematic masking of nonastronomical signals through the mask model described in Section 2.3. In addition, some detections were removed during catalog construction using automated machine learning classifiers and manual vetting procedures that operate at the detection level rather than directly within the cube masks. To improve reproducibility for users performing independent analyses, we introduce an additional bitmask flag (see Section 2.3.14), BADDET, which masks the locations of detections that were rejected by these procedures but were not captured by the previous noise model. This mask is applied as a 5 × 5 × 5 pixel cube (corresponding to 25 × 25 × 10 Å) centered on the emission-line detection and is included in the MASK array HDU of the data cube FITS files.

Despite these masking efforts, some artifacts may still remain and may be difficult to identify using the data cubes alone. Intermediate data processing products, particularly the 2D spectral images used in S. Mukae et al. (2026), provide additional diagnostic power for distinguishing real emission lines from artifacts. Users developing independent detection algorithms should therefore consider using the provided mask extensions and, where possible, auxiliary data products to minimize contamination from nonastronomical features.

Source detection is performed on the internal fiber database using a PSF-extracted grid search, with noise estimated by propagating fiber uncertainties into PSF-weighted spectral extractions. All detections are assumed to be point sources. Due to the interpolated nature of the public IFU data cubes, one-to-one recovery is not expected, even with masking applied.

To quantify agreement between independent detections on the public data cubes and the released catalog, we re-measured the S/N for 1000 faint emission-line sources drawn from the expanded source catalog (see Appendix B), including those below the primary selection threshold (p_conf < 0.5 or p_cnn < 0.5). Using the same PSF-weighted Moffat extraction (β = 3.0, FWHM from the IFU index), and S/N estimator as the HETDEX pipeline (K. Gebhardt et al. 2021), we find a median recovered S/N of 0.75  ±  0.23 relative to catalog values.

Agreement improves at higher significance: sources with S/N ≳ 6 approach unity with low scatter, while dispersion increases near the detection threshold (S/N ∼ 5). The ∼25% deficit arises from correlated noise introduced by spatial interpolation onto the grid, unlike the uncorrelated noise in native fiber spectra. For high-confidence detections (p_conf > 0.5 and p_cnn > 0.5), 95% of sources with S/N ≥ 6 are recovered at S/N > 4, decreasing to 81% for S/N ≥ 5.5 and 5% near S/N ∼ 5.

Flux calibration in HETDEX is performed independently within each IFU. While this approach works well for detecting the low-S/N emission lines that are the primary targets of HETDEX, it can lead to inaccurate flux measurements near bright or extended sources such as large galaxies, bright stars, or the occasional planetary nebula that would benefit from larger-scale modeling of the sky.

To address this, we also generate a full-focal-plane sky estimate using the entire VIRUS array for each observation. In practice, however, the current implementation of the full-frame sky model introduces low-level spectral features that degrade the default flux calibration. For this reason, fiber spectra calibrated with the full-frame sky model are not included in this release, although this may be revisited in future data products. Despite these limitations, M. Lujan Niemeyer et al. (2026) successfully used full-frame sky-subtracted data to perform Lyα intensity mapping by statistically removing contaminating components.

Because of these issues, we caution users against performing intensity-mapping analyses (e.g., M. Lujan Niemeyer et al. 2026) or absolute sky-brightness measurements (L. H. Weiss et al. 2025) directly from the public PDR1 data cubes. The local sky subtraction provided here is optimized for source detection but is not suitable for analyses requiring a globally uniform sky model. Users studying very extended sources should also note that the local sky subtraction can suppress low surface-brightness emission on spatial scales comparable to the IFU footprint.

One of the biggest strengths of HETDEX is its potential for studying the properties of faint targets via spectral stacking. By combining thousands to hundreds of thousands of spectra, the faint spectral continuum signal from LAEs is boosted revealing details about its stellar population (D. Davis et al. 2023b) and signatures of neutral hydrogen in the circumgalactic medium (L. H. Weiss et al. 2024; M. Mirza Khanlari et al. 2025).

When stacking HETDEX spectra, we account for faint sky residuals by constructing a correction spectrum from stacks of 200 randomly selected empty-sky apertures per field (D. Davis et al. 2023b). This residual stack, typically <1% of the sky level in individual spectra, is subtracted from source stacks to remove small calibration offsets that only become significant when combining large numbers of spectra. This measured offset, which is applied in published HETDEX stacking analyses, such as D. Davis et al. (2023b), L. H. Weiss et al. (2024), and M. Mirza Khanlari et al. (2025), is not included in the HETDEX data cubes.

Emission-line detections are identified through a two-stage grid search of the data cubes. An initial coarse search is performed in spatial steps and 8 Å spectral steps using a Gaussian line profile with the instrumental resolution (σ = 1.7 Å). Continuum emission is subtracted locally, and candidate lines with S/N > 4 and χ² < 3 are retained. A refined raster with spatial steps and an unconstrained σ then optimizes the fit, with duplicates within 3″ and 3 Å merged. Continuum detections are identified by searching for fibers with more than 50 counts in either a blue (3700–3900 Å) or red (5100–5300 Å) window (corresponding to g ∼ 22.5). Candidate positions are refined with a PSF raster fit, and spectra are extracted at the best-fit location.

Each row in the catalogs is assigned a unique integer detectid. A single astronomical source may have multiple detection entries if it is spatially extended, has multiple emission lines, or is bright enough to be detected in via its line and continuum emission. Additionally, the same emission line observed in separate exposures will receive independent detectid identifiers. Users should therefore treat the detection catalogs as a superset of measurements from which higher-level source catalogs can be constructed.

The catalogs are stored in HDF5 format and can be accessed directly with PyTables (F. Alted et al. 2002). Each internal data release has its own set of detections for the date range, as outlined in Table 8. To streamline querying, a master index file detect_index_hdr5.h5 can first be queried either by coordinates, detectid, or other variables (see Table 9) to determine the detection’s survey and detection type (line versus cont). This query indicates which survey HDF5 file to open (e.g., detect/detect_{survey}.h5 for line detections, or cont/cont_{survey}.h5 for continuum detections). The matching rows in the /Detections and /Spectra tables are then read to obtain sky coordinates, shotid/ifuslot, the central wavelength for line detections, and the pipeline 1D spectrum and errors.

For emission-line detections, cw stores the fitted line center in angstroms. The pipeline spectrum (column spec1d, more information in Table 10) contains PSF-weighted flux in units of 10⁻¹⁷ erg s⁻¹ cm⁻² in 2 Å bins (users should divide by 2 to obtain conventional flux density units of 10⁻¹⁷ erg s⁻¹ cm⁻² Å⁻¹, as is provided in the 1D spectra provided in HPSC2). A calibration error is present in the raw H5 detection spectra. Output spectra should be normalized by the spectrum “/pdr1/detect/wdcor.txt” as shown in the example above.

Table 10. Column Descriptions in the Raw Detections HDF5 Table Groups

Table	Column	Description
Detections	detectid	Unique integer detection ID
	inputid	Input extraction ID (DATEvOBS_inputname)
	shotid	Shot identifier (YYYYMMDDnnn)
	date	Observation date (YYYYMMDD)
	obsid	Observation number on that date
	ra, dec	R.A. and decl. (ICRS J2000 deg)
	wave, wave_err	Line central wavelength (Å) and uncertainty
	flux, flux_err	Line flux 10−17 erg s−1 cm−2 and uncertainty
	linewidth, linewidth_err	Gaussian σ (Å) and uncertainty
	continuum, continuum_err	Continuum flux density and uncertainty
	sn, sn_err	Emission-line signal-to-noise ratio and uncertainty
	chi2, chi2_err	χ2 of line fit and uncertainty
	fiber_id	Brightest fiber identifier
	multiframe	specid/ifuslot/ifuid/amp identifier
	fibnum	Fiber number
	expnum	Dither exposure number
	x_raw, y_raw	CCD coordinates of detection
	x_ifu, y_ifu	IFU position (arcseconds)
	specid, ifuslot, ifuid, amp	Instrument identifiers
	weight	Weight of brightest fiber
	apcor	Aperture correction factor
	sn_cen	S/N summed across 7 pixels at line center
	flux_noise_1sigma	1σ noise estimate on flux
	sn_3fib, sn_3fib_cen	S/N for three-fiber extractions
Spectra	detectid	Unique integer detection ID (links to Detections)
	wave1d	Wavelength array (Å)
	spec1d, spec1d_err	Flux-calibrated, PSF-weighted, aperture-corrected spectrum and error in 10−17 erg s−1 cm−2 in 2 Å bins
	counts1d, counts_err	Counts and uncertainty
	apsum_counts, apsum_counts_err	PSF-weighted counts and uncertainty
	spec1d_nc, spec1d_nc_err	Spectrum without aperture correction and uncertainty
	apcor	Applied aperture correction (wavelength dependent)
Fibers	detectid	Unique integer detection ID (links to Detections)
	ra, dec	Fiber R.A., decl. (ICRS J2000 deg)
	fiber_id	Fiber identifier string
	x_ifu, y_ifu	Fiber position in IFU (arcseconds)
	multiframe	specid/ifuslot/ifuid/amp identifier
	fibnum	Fiber number
	expnum	Dither exposure number
	distance	Distance to requested R.A., decl. (arcseconds)
	wavein	Input wavelength for extraction
	timestamp	Exposure time stamp
	date, obsid	Observation date and number
	flag	Fiber flag
	weight	Weight applied in PSF sum
	ADC	Five-element array of ADC values
	specid, ifuslot, ifuid, amp	Instrument identifiers
	x_raw, y_raw	CCD coordinates of fiber

Download table as:  ASCIITypeset image

Continuum catalogs (cont_hdr*) have the same structure, although many line-specific columns (e.g., wave, flux, sn) are zero.