SAMI: Sydney–AAO Multi-object IFS

• SAMI is a fibre-fed multi-object integral field spectrograph combining the multiplex advantage of MOS with the spatial resolution of IFUs using 13 hexabundles.
• It facilitates simultaneous spatially resolved spectroscopy of multiple galaxies to study kinematics, metallicity gradients, and star-formation processes.
• The SAMI Galaxy Survey has produced comprehensive datasets, including spectral cubes and 2D maps for over 3000 galaxies, driving new insights into galaxy evolution.

SAMI is the Sydney–AAO Multi-object Integral field spectrograph: a fibre-fed, multi-object integral-field unit on the 3.9 m Anglo-Australian Telescope that combines the multi-object spectrograph with the spatial multiplex advantage of an integral field spectrograph. Its defining innovation is the deployment of 13 imaging fibre bundles (“hexabundles”) over a 1-degree diameter field, allowing simultaneous spatially resolved spectroscopy of multiple galaxies and establishing the instrumental basis for the SAMI Galaxy Survey (Croom et al., 2011). In the survey era, the same platform was used to build a mass- and environment-spanning nearby-galaxy sample, culminating in a final public release of 3068 unique galaxies with blue and red spectral cubes, binned products, aperture spectra, and two-dimensional maps of emission-line and stellar properties (Croom et al., 2021).

1. Concept and scientific rationale

SAMI was developed in response to a long-standing asymmetry in extragalactic spectroscopy. Over the preceding decades, multi-object spectroscopy surveys such as 2dFGRS and SDSS had built samples of $10^5$–$10^6$ galaxies, while single-object integral-field spectrographs such as SAURON and PMAS had demonstrated the value of spatially resolved spectroscopy for gas and stellar kinematics, metallicity and age gradients, star-formation maps, and outflows. The bottleneck was that monolithic IFUs could observe one galaxy at a time, rendering statistically large resolved surveys prohibitively expensive. SAMI was designed as the first instrument to combine the object multiplex of a MOS with the spatial multiplex of an IFS, so that multiple galaxies could be observed simultaneously, each with two-dimensional coverage (Croom et al., 2011).

That design goal was not only instrumental but explicitly survey-driven. Early survey planning specified a three-year program targeting 3400 galaxies across the GAMA fields and eight clusters, with the aim of covering a broad range in stellar mass and environment, from isolated field galaxies through groups to clusters of $10^{15}\,M_\odot$ (Brough, 2014). This positioned SAMI between small, high-information single-object IFU studies and very large single-aperture redshift surveys, and made it possible to frame questions about angular momentum, metallicity gradients, quenching, turbulence, feedback, and environmental dependence in a statistically resolved way rather than through aperture averages alone.

A recurrent misconception is that SAMI should be understood only as an instrument. In the literature, however, “SAMI” also denotes an integrated observational program: hardware, plate configuration, reduction pipeline, target selection, public data releases, and the scientific analyses built from those components. This broader usage is evident across the technical, survey, and science papers associated with the project (Allen et al., 2014).

2. Instrument architecture and observing configuration

At the core of SAMI are 13 “hexabundles,” each a lightly fused close-packed array of 61 circular, multimode fibres. In the prototype description, each fibre had a 105 $\mu$m core and 0.22 NA, with cladding etched down to 5 $\mu$m within a $\sim 30$ mm fused region, producing a 75 percent fill-factor; later survey descriptions quote a 73 percent filling factor for the survey instrument. On sky, each fibre core subtends 1.6 arcsec, and each hexabundle subtends approximately a 15 arcsec diameter region. The bundles are deployable across a 1-degree prime-focus field at the AAT, behind the $f/3.4$ triplet corrector [(Croom et al., 2011); (Brough, 2014)].

The optical train couples the fibre bundles to the AAOmega double-beam spectrograph. In the standard survey configuration, the blue arm used the 580V grating over $3700$–$5700$ Å at $R\approx 1730$, while the red arm used the 1000R grating over $10^6$0–$10^6$1 Å at $10^6$2. In the broader AAOmega configuration space, SAMI could access approximately $10^6$3–$10^6$4 Å at $10^6$5–$10^6$6, depending on the chosen grating and central wavelength. The spectral resolving power was written explicitly as

$10^6$7

This configuration was chosen to capture the principal stellar absorption features in the blue and the H$10^6$8-centered nebular diagnostic set in the red (Croom et al., 2011).

Operationally, the prototype and survey phases differed in how the 13 bundles were allocated. The prototype system allowed 13 hexabundles to be deployed over the field. In the survey configuration, 12 hexabundles typically observed galaxies and one observed a simultaneous flux-calibrator star, with 26 dedicated sky fibres also mounted on the plate (Brough, 2014). Survey observations generally used a 7-point dither pattern of $10^6$9 s to fill the incomplete fibre fill factor, while early commissioning also experimented with pointed and nine-point dithered observations (Fogarty et al., 2012).

Plate manufacture and field setup were non-trivial because SAMI used a plug-plate system at prime focus. The automated configuration work describes a 24 cm steel plate carrying stacked observation configurations, with science plates containing three fields sharing a common central guide-star hole, and calibration plates carrying up to eight fields. The automation stack used a thin Java controller layer, a C++ configuration and optimisation layer, and visual verification through Aladin. This operational infrastructure was part of making high-throughput, multi-field observing practical at the required positional precision (Lorente et al., 2012).

3. Survey design, target selection, and data releases

The SAMI Galaxy Survey was built around stepped, volume-limited selection in stellar mass and redshift. The primary GAMA-region sample covered $10^{15}\,M_\odot$0 and $10^{15}\,M_\odot$1, and used a colour-plus-absolute-magnitude proxy for stellar mass: $10^{15}\,M_\odot$2 Four primary mass/redshift bins defined 90 percent of the sample, and lower-priority filler samples extended coverage both to lower masses and slightly higher redshift. The survey footprint comprised the three GAMA equatorial fields G09, G12, and G15, totaling 144 deg$10^{15}\,M_\odot$3, with extensive ancillary data from GALEX, SDSS, KiDS, VIKING, WISE, Herschel-ATLAS, GMRT, FIRST/NVSS, ALFALFA, and GAMA spectroscopy (Brough, 2014).

Because the GAMA fields do not sample the highest-density environments well, SAMI added an eight-cluster program. A dedicated cluster redshift survey with 2dF/AAOmega established cluster membership, dynamical properties, and the cluster target catalogues. Across the eight low-redshift clusters, the combined redshift catalogue contained 21,257 reliable redshift measurements and 2899 confirmed cluster member galaxies, with spectroscopic completeness of $10^{15}\,M_\odot$4 for $10^{15}\,M_\odot$5 and $10^{15}\,M_\odot$6. The resulting cluster sample spanned $10^{15}\,M_\odot$7–$10^{15}\,M_\odot$8 and included both relatively relaxed systems and clusters with strong indications of merger-related substructure (Owers et al., 2017).

Public data release began early. The Early Data Release provided fully calibrated blue and red datacubes for 107 galaxies drawn from the GAMA regions, together with variance, weight, and covariance extensions and ancillary catalogue information from GAMA (Allen et al., 2014). The third and final release, DR3, expanded this to 3068 unique galaxies, including for the first time the cluster sample of 888 unique galaxies. For each galaxy, DR3 includes two primary spectral cubes, three spatially binned cubes per arm, standardized aperture spectra, and complete two-dimensional maps from parameterized fitting to emission-line and absorption-line spectral data (Croom et al., 2021).

The distinction between planned sample size and final public sample is important. Early survey papers refer to an intended sample of $10^{15}\,M_\odot$9 galaxies (Brough, 2014), whereas DR3 documents the released sample of 3068 unique galaxies (Croom et al., 2021). Both figures are part of the record, but they correspond to different stages of the program rather than to inconsistent statements about a single dataset.

4. Reduction methodology, cube construction, and calibrated products

The reduction chain begins with 2dfdr, which performs bias subtraction, flat-fielding, fibre tracing, wavelength calibration, extraction to row-stacked spectra, fibre throughput correction, and sky subtraction. Flux calibration is tied to primary spectrophotometric standards, while simultaneous secondary standards support telluric correction and field-by-field transmission tracking. In later pipeline versions, sky subtraction was improved by a simple median combination of sky fibres and PCA removal of residuals around bright lines, and telluric correction across the red arm used molecfit [(Allen et al., 2014); (Croom et al., 2021)].

Datacube construction is a central technical component of SAMI because the raw fibre sampling is sparse and irregular. The “Cubism and covariance” methodology adopted a drizzle-like resampling from the circular fibre footprints to a regular 0.5 arcsec spaxel grid, preserving total flux, minimizing further PSF smoothing, and retaining a weight map for relative exposure. To recover resolution, the effective fibre “drops” were shrunk by a factor $\mu$0, and seven dithered exposures were combined after registration. The same work also introduced compressed covariance tracking: rather than storing a full covariance matrix for the entire cube, the pipeline stored a reduced covariance representation that retained 90 percent of the covariance information while incurring only a modest increase in data volume (Sharp et al., 2014).

The resulting survey products are unusually rich for a multiplexed IFS program. DR3 provides, per galaxy, blue and red cubes on a $\mu$1 grid of 0.5 arcsec spaxels and 2048 spectral pixels, adaptive Voronoi bins to median $\mu$2 Å$\mu$3, annular and sector-binned cubes, circular apertures of 1.4, 2, 3, and 4 arcsec diameter, a physical 3 kpc aperture, and elliptical $\mu$4 apertures from both Sérsic and MGE fits. Two-dimensional maps include emission-line fluxes, Balmer-decrement extinction, BPT ionization classification, gas velocity and velocity dispersion, multi-Gaussian H$\mu$5 decomposition, stellar $\mu$6, $\mu$7, $\mu$8, $\mu$9, kinematic asymmetry, stellar rotation-axis position angle, $\mu$0, Lick indices, and SSP-equivalent ages and abundances (Croom et al., 2021).

Several later science analyses depend explicitly on this infrastructure. Emission-line studies use LZIFU to model and subtract the continuum and to fit up to three Gaussian components per strong line on a spaxel-by-spaxel basis (Poetrodjojo et al., 2018). Stellar kinematic work uses pPXF with optimal templates derived from the MILES library, validated against ATLAS$\mu$1D galaxies observed with a single hexabundle (Fogarty et al., 2015).

5. Performance characteristics and observational limitations

Instrumental performance evolved across the prototype, early survey, and upgraded survey phases. Prototype laboratory tests of a single hexabundle core yielded 96 percent throughput at 625 nm and 89 percent at 435 nm within $\mu$2, but once spliced to the 42 m ribbonized run and slit block the central core throughput dropped to $\mu$3 in the red and $\mu$4 in the blue because ribbonizing and cabling introduced focal-ratio degradation. On-sky standard-star measurements confirmed system throughput in the blue arm of 26–46 percent and in the red arm of 30–40 percent, peaking near 7000 Å (Croom et al., 2011). Later instrument documentation reports that new cable packing limited FRD losses to $\mu$5 in $\mu$6 and $\mu$7 in $\mu$8, whereas the original ribbonised cable had up to 50 percent FRD loss in the blue; after upgrades, overall end-to-end throughput was approximately 30 percent in the blue and 40 percent in the red (Brough, 2014).

Sky subtraction and flux calibration were characterized quantitatively. In the prototype, median sky-residual continuum across $\mu$9–$\sim 30$0 Å was $\sim 30$1 RMS of the sky level, sufficient for absorption-line work to $\sim 30$2 mag arcsec$\sim 30$3 with 2–3 hr exposures (Croom et al., 2011). For the Early Data Release pipeline, median sky continuum residuals were 1.2 percent in the blue and 0.9 percent in the red, with emission-line residuals of 0.8 percent and 0.9 percent respectively. Relative flux calibration uncertainty was 4.1 percent systematic plus 4.3 percent statistical, and atmospheric dispersion was removed with an accuracy of 0.09 arcsec, less than one-fifth of a spaxel (Allen et al., 2014).

Spatial resolution was limited primarily by seeing and fibre sampling rather than by the spectrograph itself. Typical survey PSF FWHM in final cubes was 2.1 arcsec, with a reported range of $\sim 30$4–$\sim 30$5 arcsec in early data (Allen et al., 2014). The datacube-construction paper showed that subcritical sampling and incomplete fill factor produced only a 10 percent degradation in the final image resolution recovered, with measured broadening of only $\sim 30$6 arcsec in the blue and $\sim 30$7 arcsec in the red for calibration-star mosaics (Sharp et al., 2014). That limitation has direct astrophysical consequences: in the metallicity and ionization study, H II regions were not resolved at the SAMI spatial resolution, so maps were constructed on a spaxel-by-spaxel basis rather than by explicit H II-region segmentation (Poetrodjojo et al., 2018).

Spectral resolution likewise varied by arm and application. The prototype reported blue-arm FWHM $\sim 30$8 Å and red-arm FWHM $\sim 30$9 Å, corresponding to $f/3.4$0 km s$f/3.4$1 and $f/3.4$2 km s$f/3.4$3, and noted typical H$f/3.4$4 velocity-field map precision of $f/3.4$5 km s$f/3.4$6 per spaxel for $f/3.4$7 discs (Croom et al., 2011). Later DR3 documentation gives instrumental dispersions of $f/3.4$8 km s$f/3.4$9 in the blue and $3700$0 km s$3700$1 in the red (Croom et al., 2021). The red arm’s relatively high resolution is one reason SAMI became particularly effective for ionized-gas kinematics and H$3700$2-based studies.

6. Scientific results, interpretation, and legacy

SAMI’s scientific output spans stellar dynamics, ionized-gas physics, chemical cartography, star-formation regulation, environmental dependence, and galaxy structure. In the Pilot Survey of 106 galaxies in Abell 85, 168, and 2399, stellar velocity and dispersion maps were used to compute the specific stellar angular momentum proxy

$3700$3

showing a trend between $3700$4 and galaxy concentration such that late-type galaxies are less concentrated, higher-angular-momentum systems, while fast-rotating early types are more concentrated and lower in angular momentum. The same study measured kinematic misalignment,

$3700$5

and found that 83 percent of fast rotators were aligned, whereas only 38 percent of slow rotators were aligned, consistent with fast rotators behaving like oblate spheroids and slow rotators being more likely mildly triaxial (Fogarty et al., 2015).

Resolved nebular diagnostics were another major outcome. The DR1 metallicity and ionization mapping study of 25 star-forming face-on spirals used an iterative $3700$6–$3700$7 scheme to solve jointly for gas-phase metallicity and ionization parameter $3700$8, finding typical metallicity gradients of $3700$9 to $5700$0 dex $5700$1, an average gradient of $5700$2 dex $5700$3, and a typical ionization-parameter range $5700$4. Only two galaxies showed radial gradients in $5700$5, and no significant overall correlation was found between ionization parameter and SFR, sSFR, or metallicity. A central methodological conclusion was that $5700$6 dex within individual galaxies is large enough to bias diagnostics that assume a fixed ionization parameter (Poetrodjojo et al., 2018).

SAMI also produced resolved star-formation maps at scale. The $5700$7 H$5700$8-based SFR maps released in DR1 were used to examine the global and resolved star-forming main sequence. Galaxies further below the main sequence were more likely to have flatter star-formation profiles. Early-type galaxies separated into an on-sequence population with centrally concentrated star formation similar to late types and a quenched population with strongly reduced star formation, most strikingly in the nuclear regions. Across morphologies, denser environments were associated with decreased specific star-formation rate from the outside in, supporting an environmental contribution to quenching such as ram-pressure stripping or galaxy interactions (Medling et al., 2018).

Ionized-gas kinematics provided additional constraints on ISM driving mechanisms. In eight strictly selected pure star-forming galaxies, SAMI showed a flat distribution of ionized-gas velocity dispersion as a function of SFR surface density on sub-kpc scales, with observed $5700$9 at low $R\approx 1730$0 typically a factor of $R\approx 1730$1–2 above simple feedback-driven expectations. The analysis concluded that star-formation feedback alone is insufficient in these systems and that gravity, galactic shear, and/or magnetorotational instability may also drive turbulence (Zhou et al., 2017).

Early commissioning already demonstrated the survey’s sensitivity to extranuclear phenomena. In ESO 185-G031, SAMI identified extended minor-axis emission with ionization and kinematic properties consistent with a large-scale galactic wind. Off-plane spaxels crossed into the AGN/shock-excited regime in BPT space, H$R\approx 1730$2 residuals reached $R\approx 1730$3 km s$R\approx 1730$4, and Magellan MMTF imaging showed H$R\approx 1730$5 filaments extending up to 11 arcsec above and below the disk. The result was significant partly because conventional single-fibre spectra would have been dominated by nuclear emission and would likely have missed the extranuclear wind signature (Fogarty et al., 2012). A different early survey discovery, in the dwarf galaxy GAMA J141103.98-003242.3, was a luminous unresolved H II complex contributing $R\approx 1730$6 percent of the galaxy’s H$R\approx 1730$7 luminosity, with $R\approx 1730$8, lower than the rest of the galaxy by $R\approx 1730$9 dex, and a smoothly varying ionized-gas velocity field consistent with co-rotation rather than interaction-driven disturbance (Richards et al., 2014).

Later work used the statistical depth of the full survey to revisit several debated questions. A logistic-regression analysis of fast versus slow rotators found that a model using stellar mass, star-formation rate, $10^6$00-band half-light radius, and an ellipticity threshold variable captured the probability of slow rotation, and that once those quantities were fixed, the local environmental surface density $10^6$01 gave no further information about whether a galaxy is a slow rotator (Vaughan et al., 2024). In a separate study of spin–neighbour coherence, the luminosity-weighted mean velocity offset between SAMI galaxies and neighbours within 1 Mpc was $10^6$02 km s$10^6$03 for absolute-luminosity weighting, but the appearance of similarly significant negative signals at larger separations led to the conclusion that the modest $10^6$04–$10^6$05 signals may result from cosmic variance or coincidental scatter rather than a robust universal coupling (Mai et al., 2022).

SAMI data also constrained structural evolution. Comparisons between SAMI galaxies and equilibrium models showed that five gigayears of disc fading produced $10^6$06 and $10^6$07, substantially smaller than the observed differences between star-forming spirals and passive lenticulars. Even after allowing for progenitor bias using EAGLE size evolution, the conclusion was that intrinsic dynamical evolution must be important in the transition from star-forming discs to passive discs (Croom et al., 2021). In early-type galaxies on the Fundamental Plane, SAMI found coefficients $10^6$08, $10^6$09, and $10^6$10, with FP residuals most strongly correlated with luminosity-weighted SSP age at $10^6$11 significance. Stellar-population mass-to-light variations accounted for most of the FP scatter but not the full tilt away from the virial plane (D'Eugenio et al., 2021).

The larger significance of SAMI lies in demonstrating that multiplexed IFS with hexabundles is technically feasible on survey scales and scientifically productive across a wide range of galaxy-evolution problems. The prototype paper explicitly framed higher-multiplex, wider-field descendants as a route to surveys of $10^6$12 to $10^6$13 galaxies (Croom et al., 2011), while later methodological work showed that SAMI-trained models can be exported to other surveys and that future programs such as Hector, WEAVE, and 4MOST can use SAMI-based post-stratification or target selection for slow rotators (Vaughan et al., 2024). In that sense, SAMI is both a specific instrument-survey system and a proof of concept for large, statistically resolved spectroscopy of nearby galaxies.