Documentation

Changes between DR2 and DR3

As one of the more mature products of the SAMI galaxy survey, the DR3 Emission Line Data Products have fewer changes from the previous data releases. Most changes instituted are about including or excluding data that experience has shown us to be useful or less trustworthy. Such changes include:

• We now include emission lines affected by proximity to sky-lines. Due to the varying redshifts of the galaxies within the SAMI galaxy survey, some strong emission lines lie in a wavelength region near to a strong sky line. Previously these were not fit, however, the line could still be clearly seen in the spectral cube. We now fit for these lines, but have flagged the data to indicate that this line may be untrustworthy.
• We only include the 1 component fits for Aperture data but include higher order Balmer lines and some weaker emission lines. Due to the way that LZCOMP was trained we found that it was unable to process correctly the spectra from the aperture bins. Therefore we release only the 1-component fits. However, given the higher S/N in the binned spectra, we now also fit for $H\gamma$, $H\delta$, and $H\epsilon$, as well as the weaker forbidden lines [OIII]4363 and [NeIII]3869.
• We now also specifically mask spaxels on the outer parts of the original (and some adaptively-binned) maps that, due to the reduction and cubing process, had significant missing pixels in the spectral data preventing a good fit to the underlying continuum.

1 Spectral Analysis

The main goal of LZIFU (LaZy-IFU, Ho et al. 2016) is to extract emission line maps and ionised gas kinematic maps from SAMI data cubes. We achieved this in two steps. On a spaxel-to-spaxel basis, we first model and subtract the continuum, then fit emission lines to the continuum-free spectrum. We briefly introduce the procedures below. A full description of the pipeline LZIFU is presented in Ho et al.2016. Further details of LZIFUs application to SAMI DR3 can be found in the DR3 paper (Croom et al. submitted).

1.1 Continuum Fitting

1.1.1 Unbinned cubes

For the unbinned, "default" cubes, we follow the methodology in DR2 to better fit low S/N continuum and use the fits to the Voronoi-binned data, which has S/N$\ \sim10$ in the continuum, to constrain the number of templates that are then used to fit the individual spaxels that lie within the Voronoi bin. This process is fully described in Owers et al 2019, but summarising: we first fit the spectrum in each Voronoi-bin using pPXF (penalized Pixel Fitting; Cappelari & Emsellem 1994) with a subset of the MILES simple stellar population (SSP) spectral library (Vazdekis et al. 2010) that contains four metallicities ([M/H] = -0.71, -0.40, 0.00, 0.22) and 13 logarithmically-spaced ages ranging from 0.0063-15.85 Gyrs. Following Cid Fernandes et al. (2013), the MILES SSPs are supplemented with younger SSP templates drawn from Gonzalez Delgado et al. (2005) with metallicities [M/H] = -0.71, -0.40, 0.00 and ages 0.001-0.025 Gyr. During the fitting, emission line templates are included for the Balmer lines, as well as strong forbidden lines. Importantly, this simultaneous fitting of emission and absorption components allows the regions surrounding the age-sensitive Balmer absorption lines to be included in the continuum fits. The stellar kinematics are not fitted for during this process, instead we use those derived by the SAMI stellar kinematics pipeline, included as part of this release.

The subset of SSP templates that have non-zero weights assigned in the fits to the Voronoi binned spectra are then used during the fitting of each spaxel contained within the region defined by the Voronoi bin. Again, emission lines are fitted simultaneously, and the stellar kinematics are fixed to those measured by our stellar kinematics pipeline, while allowing pPXF to re-determine the optimal template weights only for spaxels where the S/N>5. For spaxels with S/N<5, the weights determined during the Voronoi binned fitting are used to produce a single optimal template, while the stellar kinematics are fixed to those derived from the Voronoi-binned data. This helps to guard against poor fits due to low S/N. In all of the pPXF fitting described above, we include a 12th order multiplicative polynomial. This continuum fit is then used in LZIFU to subtract the continuum and measure the final line fluxes for the 1- to 3-components fits. Overall this method produces similar line fluxes to those found in DR1, but with better constraints in spaxels with low-S/N continua, and some systematic offsets in galaxies with significant Balmer absorption features.

1.1.2 Binned data and aperture spectra

For binned data and aperture spectra we follow the same continuum fitting process as used in SAMI Data Release 1. LZIFU first convolves the blue and red spectra to a common spectral resolution (i.e., the blue resolution) and then stitches the two spectra together. The merged spectrum is sent to the penalized pixel-fitting routine (pPXF; Cappellari & Emsellem 2004) for modeling the continuum. Only channels not contaminated by strong sky lines or optical emission lines are included in the fitting process. In addition to bad channels rejected by the data reduction pipeline, we mask out the vicinity ($\rm\pm10$ Å) of strong sky lines (at 5577, 6360, and 7340 Å), and those at 40 Å (full-width) around common emission lines ([OII]3726,29, H$\delta$, H$\gamma$, [OIII]4363, H$\beta$, [OIII]4959, 5007, [OI]6300, [NII]6548,83, H$\alpha$, and [SII]6716,31). Only channels between 3700 Å and $6950\times (1+z)$ Å are considered due to the limited spectral coverage of the stellar templates.

As with the unbinned cubes, we adopt the MILES SSP templates with the Salpeter IMF of four metallicities ([M/H] = $-0.71$, $-0.40$, 0 and $+0.22$) and 13 ages (equally spaced logarithmically from 63.1 Myr to 15.8Gyr) from Vazdekis et al. (2010), supplemented with the younger SSP templates drawn from Gonzalez Delgado et al. (2005) . Additive Legendre Polynomials (orders 2 to 10) are also included to fit simultaneously with the stellar templates. To reject potential bad channels, we iterate the continuum fit while imposing a robust-sigma clipping at the 5$\sigma$ level. Note that this is different from what the standard CLEAN keyword in pPXF does.

These templates are at a slightly lower spectral resolution than the red arm of our spectra; therefore in low stellar velocity dispersion galaxies ($\sigma<$30 km s$^{-1}$), we may incorrectly subtract the H$\alpha$ absorption. However, most of these low stellar velocity dispersion galaxies have strong emission lines with high equivalent widths, so this systematic error is unlikely to have a big effect.

1.2 Emission Line Fitting - Single and Multiple Components

After subtracting the continuum, LZIFU models emission lines in the residual spectrum as Gaussians and performs a bounded-value nonlinear least-squares fit using the Levenberg-Marquardt least-squares method implemented in IDL (MPFIT; Markwardt 2009). Only channels ±20 Å around line centres, as inferred from the galaxy redshift (LZIFU input redshift), are included in the fitting process. Each galaxy is fit three times, assuming that the line is composed of 1, 2, or 3 velocity components having a Gaussian shape. All the lines are fit simultaneously and each kinematic component is constrained to share the same velocity and velocity dispersion. In total, we fit 11 strong optical lines simultaneously, [OII]3726,29, Hβ, [OIII]4959,5007, [OI]6300, [NII]6548,83, Hα, and [SII]6716,31. The [OIII]4959 and [NII]6548 lines are both constrained to be one-third in amplitude (and flux) of [OIII]5007 and [NII]6583, respectively, due to the physics of their emission. We do not produce line maps for these two lines. In the aperture data, we also fit for $H\gamma$, $H\delta$, and $H\epsilon$, as well as the weaker forbidden lines [OIII]4363 and [NeIII]3869 due to the higher S/N found in the binned spectra.

Some spaxels at the edge of the fibre bundles have a significant fraction of the channels masked due to issues arising from the cubing routine or other reduction steps. Such spectral drop-outs prevent the continuum from being correctly fit and in some cases actually mask the emission-lines as well. Spaxels where a significant ($>500$) fraction of the spectral channels in either the blue or red cubes are masked are flagged in our routine and are masked (set to NaN) in our final output data.

As a final step, the LZIFU fits with 1-, 2-, and 3-components for each galaxy is run through a trained Neural Network called LZCOMP (with the network and training described in Hampton et al. 2017) that decides what number of components best reproduces the emission line profiles in each spaxel. This then returns a final "recommended" component map, where the flux of each emission line is a sum of up to 3 Gaussian components with corresponding velocity and velocity dispersion maps for each component. As an additional constraint, we require that all components have a signal-to-noise ratio of at least 5 in H$\alpha$. If this condition is not met, we reduce the number of components by 1. 3 components becomes 2 components and 2 components becomes 1 component. Generally only spaxels with sufficiently high S/N in the emission lines (S/N > 20) are able to be decomposed in this manner, with single Gaussian fits sufficient to reproduce most faint lines.

2 Data Quality

2.1 Errors in the line fits

One sigma errors of line fluxes (except for Balmer lines, see below), velocities, and velocity dispersions are taken straight from MPFIT, which uses the Levenberg-Marquardt technique to solve the maximum likelihood problem, and the Jacobian matrix of the line model to estimate the covariance matrix for parameters. We have tested the robustness of MPFIT through Monte Carlo simulations. In the absence of any systematic errors, such as those from continuum subtraction or those from variance of the data, the one sigma errors returned by LZIFU (from MPFIT) are robust. We discuss in the following sections some of the possible systematic errors we have identified that you should be aware of, some of which have been incorporated into the line flux error extensions.

2.2 Continuum subtraction

When fitting the continuum it is assumed that the spectral templates and polynomial are a perfect representation for the observed continuum within the noise. However, as discussed in the DR3 paper and many other spectral fitting works, there may be systematic errors associated with the templates in either the broad continuum shape or specific absorption features. At high Equivalent Widths (EW) or low S/N the systematic errors associated with the subtraction of the continuum have little impact on the emission lines with the fit error a reasonable representation. However, at low EW and high S/N, any systematic errors associated with the continuum subtraction can dominate, with the Balmer lines particularly susceptible due to the underlying Balmer absorption features. While we have done our best to account for this when fitting the continuum as described above, we caution users to be aware of this issue in the low EW regime.

2.3 Non-zero covariance

When a line is fit with more than one component, the fluxes of different components are highly correlated. Typically the correlation coefficients are negative because the total flux, the sum of all components, remains approximately the same. This means when summing the fluxes of two components together, the combined error is not the quadrature sum of the errors. The error of the sum should be taken directly from the first slice in MS2D, which takes the covariances into account.

2.4 Errors of [OII]3726,31 doublets

In almost all lines, the wavelength separations between different lines are much larger than the spectral resolutions such that fluxes between different lines are independent of each other, i.e. the covariances are zero. This is not true for the [OII]3726,31 doublets. The doublets are marginally resolved by the blue spectrograph and therefore non-zero covariance can be important. Currently, the covariances are not reported and therefore directly summing the flux errors (of the two lines) in quadrature would overestimate the errors.
In some cases where signal-to-noise ratio is poor, LZIFU would use only one Gaussian to model one of the lines and report zero flux for the other. In this case, the error on the line with zero flux would be NaN (parameter touches boundaries) and the error on the non-zero line will be a more representative estimate.
Because the doublets are only marginally resolved, and the lines are usually at the very blue parts of the spectrograph, we highly recommend that you use only the summed flux for any analysis.

3 Released Data Products

The Emission Line Data Products for SAMI DR3 includes maps of total emission line fluxes for the strong emission lines ([OII]3726,29, Hβ, [OIII],5007, [OI]6300, [NII]6583, Hα, and [SII]6716,31), with maps of the associated gas velocities and gas velocity dispersions. We release maps in two different versions tied to the fitting method: as 1-component or "1-comp" maps, where each line in each spaxel is fit by a single Gaussian profile with a velocity and velocity dispersion tied between lines, and; recommended component or "recom-comp" maps, where each line in each spaxel is fit by up to 3 velocity components with tied velocities and velocity dispersions and the number of components determined by a neural network. The fitting process can be found above. These maps are released for both the original (0.5") resolution data and all binning schemes.

One exception to this is the annular-binned data. Due to the rotational profiles of many galaxies, many of the annular bins appear to have double-peaked or "horned" profiles, making a single Gaussian profile unsuitable. Therefore for the annular-binned maps we only release a 2-component or "2-comp" fit, obtained by fitting two Gaussians to each line in each spaxel or spectrum, accounting better for the apparent profile.

With the recommended-component maps, the only line for which we are releasing individual fluxes for multiple components is Hα. Hα emission line flux maps are arrays of dimension 50×50×(N+1), where the (N+1) slices are: [total H$\alpha$ flux (ie the sum of all components), the flux in component 1, ..., the flux in component N], where N is the recommended number of components. Due to the strong degeneracies involved in decomposing emission lines (especially with tied velocity structures and the low spectral resolution in the blue) other weaker lines may have strong uncertainties in the individual components, so only the total flux (along with appropriate covariance-included errors) is released for the other lines, in a 50×50 map. Multi-component fits are also included for the gas kinematic measurements. The format of these data are 50×50×(N+1) arrays, where the (N+1) slices are: [NaN, V or $\sigma$ of component 1, ...V or σ of component N]. The first (zeroth) slice is included as NaN in order to preserve matching with the Halpha emission line map format.

3.1 Kinematics maps

Line-of-sight velocity and velocity dispersion maps (and relative errors maps) obtained by fitting simultaneously 11 emission lines with a single Gaussian component. The velocities are calculated with respect to the heliocentric redshift as measured by the GAMA survey and are given in units of km/s. Each file has two extensions:

PRIMARY | ionised gas velocity map (1-comp are 50×50 maps , recom comp are 50×50×4 cubes with a NaN map, and velocities of the 1st, 2nd and 3rd component if it exists)

V_ERR | error on the ionised gas velocity

and

PRIMARY | ionised gas velocity dispersion map (1-comp are 50×50 maps , recom comp are 50×50×4 cubes with a NaN map, and velocity dispersions of the 1st, 2nd and 3rd component if it exists)

VDISP_ERR | error on the ionised gas velocity dispersion

3.2 Line flux maps

2D image obtained by summing all the flux associated to a given emission line in each spaxel (and relative error maps). Units are $10^{-16} \mathrm{ergs}\ \mathrm{s}^{-1} \mathrm{cm}^{-2}$. Each line flux map has two extensions:

PRIMARY | line flux map (50×50 maps with pixels 0.5", except for H$\alpha$ in the recom-comp maps which is a 50×50×4 cube with the total flux and flux in the 1st, 2nd and 3rd component if it exists).
{LINE}_ERR | error on the line flux

4 Value-Added Products

In addition to the line maps we also release as part of SAMI DR3 value-added products where we have applied standard conversions to create additional maps for: the gas-phase extinction determined from the decrement of the Hydrogen emission lines, a mask to distinguish spaxels dominated by star-formation (as compared to strongly regions affected by AGN, shocks from winds & supernovae, diffuse ionised gas, etc.), and star-formation (surface density) rates that account for the extinction and star formation mask, as well as the luminosity distances determined from the flow-corrected redshifts and a standard $\Lambda$CDM cosmology. Below we describe the routines used to create these maps for those who wish to create there own, and full details can be found in Medling et al. (2018).

4.1 Extinction Maps

Extinction maps are calculated using the Balmer decrement and the Cardelli et al. (1989) extinction law.
For each spaxel:

$$balmerdec = \frac{H\alpha}{H\beta}$$

$$balmerdecerr = balmerdec * [(H\alpha_{err}/H\alpha)^2 + (H\beta_{err}/H\beta)^2]^{0.5}$$

$$attencorr = (\frac{balmerdec}{2.86})^{2.36}$$

$$attencorrerr = |\frac{attencorr * 2.36 * balmerdecerr}{balmerdec}|$$

These maps will be in units of attenuation correction factor — such that you can multiply this map by the Halpha cube to obtain de-extincted Halpha cubes. Note that, when the Balmer decrement is less than 2.86, no correction will be applied (attenuation correction factor = 1., error = 0.).

Additionally, we have set to NaN the correction and error for spaxels with Halpha flux > 40 ($\times 10^{-16} \mathrm{erg}\ \mathrm{s}^{-1} \mathrm{cm}^{-2}$) and Balmer decrement > 10. These numbers were chosen to eliminate spurious Halpha fits to the edges of the fibre bundles. Errors are given as 1-sigma uncertainties on the attenuation correction.

Each file has two extensions:

PRIMARY | attenuation correction map
EXTINCT_CORR_ERR | error on the attenuation correction

4.2 Star Formation Masks

We classify each spaxel using (when possible) [OIII]/Hbeta, [NII]/Halpha, [SII]/Halpha, and [OI]/Halpha flux ratios to determine whether the emission lines are dominated by photoionization from HII regions or other sources like AGN or shocks, using the BPT/VO87 diagnostic diagrams and dividing lines from Kewley et al. (2006). We only classify spaxels with ratios that have a signal-to-noise ratio of at least 5. Emission is classified as star-forming in a given diagnostic if:

$$\log([OIII]/H\beta) < (0.61 / (\log([NII]/H\alpha)-0.05) + 1.3$$ (Kauffmann et al. 2003)

$$\log([OIII]/H\beta) < (0.72 / (\log([SII]/H\alpha)-0.32) + 1.3$$ (Kewley et al. 2001)

$$\log([OIII]/H\beta) < (0.73 / (\log([OI]/H\alpha) +0.59) + 1.33$$ (Kewley et al. 2001)

We additionally add a likely classification of "star-forming" to spaxels with $\log([NII]/H\alpha) <-0.4$ without an [OIII] detection and to spaxels with Halpha detections but no [N II], [S II], [O I], or [O III] detections.

Each spaxel in the map is 1 (when star formation dominates the emission in all available line ratios) or 0 (when other ionization mechanisms dominate), so it may be simply multiplied by the Halpha flux map to produce "Halpha from star formation" maps.

Each file has a single extension:

SFMask | Star formation mask map

4.3 Star Formation Rate Maps

Star formation rate (SFR) maps (in $M_\odot \mathrm{yr}^{−1}$) are obtained from extinction-corrected H$\alpha$ maps. As mentioned above, extinction-corrected Hα maps are derived by multiplying the attenuation correction factor map (computed from smoothed data) by the observed (i.e., not smoothed) Hα map. The relevant Star Formation Masks have been applied to zero out Hα emission that may be contaminated by AGN, LINER, or shock emission. When calculating the distances, we are using the flow-corrected redshifts z_tonry from the SAMI-matched GAMA catalog for the GAMA regions and the cluster redshift from Owers et al. 2017. for cluster galaxies. We assume $H_0=70$, $\Omega_m=0.3$ and $\Omega_\Lambda=0.7$.

$$SFR=L(Hα)×(7.9/1.53)×10^{−42} [M_\odot \mathrm{yr}^{−1}]$$ following Kennicutt (1998) with conversion to a Chabrier Initial Mass Function (Chabrier 2003). L(Hα) is the Hα luminosity in each spaxel corrected for internal extinction via the Balmer decrement.

Star formation rate density and error maps (in units of $M_\odot \mathrm{yr}^{−1} \mathrm{kpc}^{−2}$) are also provided. Errors are given as 1-sigma uncertainties on the SFR. Each file has two extensions:

PRIMARY | Star formation rate map

SFR_ERR | Error on the star formation rate

and

PRIMARY | Star formation rate surface density map

SFRSurfDensity_ERR | Error on the star formation rate surface density

We emphasise that these SFR maps are not appropriate for calculating a global (total) SFR, for two reasons: a) These maps use a clean sample of star-forming spaxels, not a complete one. Thus, it is possible (and likely) that some regions of star formation may be excluded, particularly in ambiguous cases, and b) Measurements of derived quantities will have more reliable measurements if the spectra are summed and then emission lines refit, rather than summing together many low-S/N fits. In general, the aperture spectra measurements in the Gas1compApertures and GasRecomcomApertures tables should be used for global SFR measurements.

SFR maps retain the same dimensions as the Hα emission line maps. Note that this means the input 50×50 extinction maps and star formation masks are applied to all components of the multicomponent Hα maps.

5 Aperture Tables

We also provide the emission line data products (ie line fluxes, ionised gas velocity & velocity dispersion, and star formation related quantities) for the set of 7 different fixed apertures centred on the galaxy described in core data products (1".4, 2", 3", 4" and 3 kpc diameter circular apertures and two elliptical apertures with semi-major axis of radius Re from a Sersic fit and ReMGE from and MGE fit, respectively). We provide only a 1-component measurement for all apertures, as our neural net, LZCOMP, was not trained on such data and unable to deal with the spectra. These measurements are contained in the table Gas1compApertures. Each row contains all measurements for a single galaxy. The columns have a structure {Measurement}_{Aperture}, e.g. HALPHA_3_ARCSECOND. Additional columns contain some general galaxy properties applicable to all apertures. For details see the SAMI DR3 schema browser.