The 3-D data cubes that constitute the final MaNGA data products are constructed from a few tens to a few thousands of individual spectra that have been combined onto a regular grid. This page will explain the structure of the raw data (spectra dispersed onto individual CCDs), intermediate data processed through the 2-D (per-exposure) stage of the data reduction pipeline, and final data products processed through the 3-D (per plate) stage of the reduction pipeline. Additionally, we describe the preimaging data drawn from the NSA reprocessing of the original SDSS imaging survey; this preimaging data is used for the astrometric alignment of the MaNGA spectral imaging.
The default storage format for MaNGA data (images and cubes) is multi-extension FITS (gzipped to save space). The zeroth extension of such files is blank except for the global header. All extensions of such files are labelled with the EXTNAME header keyword so that they can be read by extension name instead of extension number.
(see details here)
Aside from some new keywords, the format is the same as for BOSS raw data. These keywords include:
- Dither position (MGDPOS, values of N,S,E,C). C means no dither; N,S,E are the 3 allowed dither values for the MaNGA dither triangle. (For more details see the MaNGA Observing Strategy paper, Law et al. 2015).
- Dither location (MGDRA, MGDDEC), i.e. offset of the pointing from the nominal center in arcsec of RA and DEC. These are the values actually used by the DRP in determining the astrometric solution.
- PLATETYP=’MANGA’ or ‘APOGEE-2&MANGA’, as appropriate (i.e., whether the plate contains only MaNGA holes, or both MaNGA and APOGEE holes for co-observing)
- SRVYMODE=’MaNGA stare’, ‘MaNGA dither’, or ‘APOGEE-lead’ as appropriate. Most galaxy plates are ‘MaNGA dither’, all-sky plates are ‘MaNGA star’, and bright-time stellar library plates are ‘APOGEE-lead’.
2-D Reduction Pipeline Output
Each exposure through the MaNGA instrument is processed separately through the 2-D Data Reduction Pipeline (DRP) up to and including flux calibration and combination of spectra from individual cameras across the dichroic break.
The structure of these files is loosely based on the BOSS spFrame type files.
That is, they are row-stacked spectra (RSS), two dimensional arrays in which each row corresponds to an individual one-dimensional spectrum.
The following files can (and should) be read using the routine ml_mgframeread.pro?.
List of files:
3-D Reduction Pipeline Output
Once a given plate is complete (ie. all of the required sets of dithers are collected), the 3D stage of the DRP extracts the relevant rows for each IFU from the mgCFrame files, computes the astrometric solution of each, and combines the exposures into row-stacked spectra and data cubes.
These output files follow the naming convention [plate]-[ifudesign], which uniquely identifies a given galaxy observation. Note, however, that if a galaxy were to be reobserved on a different plate it would have a different ‘plateifu’ identifier. ‘plateifu’ thus uniquely identifies a set of observations, while ‘mangaid’ uniquely identifies an astronomical target.
In this section NFIBER is taken to mean the number of fibers in a given IFU (e.g., 19, 37, 61, 91 or 127), and NEXP is the number of exposures.
There are two kinds of output files: row-stacked spectra (RSS) that contain each individual spectrum stacked atop each other into into a 2d format, and a data cube that combines individual spectra together into a rectified 3d data cube. Note that RSS files and cubes are provided for all MaNGA dark time targets, and thus include both galaxies and spectrophotometric standard stars.
MaNGA adopts the HDUCLASS FITS header extension keyword structure (see https://heasarc.gsfc.nasa.gov/docs/heasarc/ofwg/docs/ofwg_recomm/r8.html) to indicate the type of information contained in the science, error, and mask (i.e., data quality, or DQ) extensions (and the relationship between those extensions). We define HDUCLASS SDSS.
The science extension has HDUCLAS1=IMAGE (for RSS files) or CUBE (for data cubes) and HDUCLAS2=DATA. The ERRDATA and QUALDATA keywords in this extension header point to the error and DQ extensions.
The error extension has HDUCLAS1=IMAGE or CUBE and HDUCLAS2=ERROR. Valid HDUCLAS3 entries are MSE, RMSE, INVMSE, INVRMSE. MaNGA uses INVMSE (i.e., we provide inverse variance). The SCIDATA and QUALDATA keywords in this header point to the science and DQ extensions.
The DQ extension has HDUCLAS1=IMAGE or CUBE and HDUCLAS2=QUALITY. MaNGA uses HDUCLAS3=FLAG64BIT, indicating that the DQ extension should be interpreted as a bitmask with up to 64 independent bits available. The SCIDATA and ERRDATA keywords in this header point to the science and error extensions.
In all cases, the dimensionality of the ERROR and QUALITY extensions matches that of the DATA extension.
These are the row-stacked, flux-calibrated fiber spectra for a given galaxy across all exposures. The “LOGRSS” file has logarithmic wavelength sampling from log10(lambda/Angstroms)=3.5589 to 4.0151 (NWAVE=4563 spectral elements). The “LINRSS” file has linear wavelength sampling from 3622.0 to 10353.0 Angstroms (NWAVE=6732 spectral elements). Both files contain one row for each fiber, for a total of NFIBER*NEXP rows.
The detailed data models for the LOGRSS and LINRSS files can be found here. In brief, they contain extensions for the flux (in units of 10-17 erg/s/cm2/Angstrom/fiber), the inverse variance, the pixel mask, the spectral line spread function for each fiber, the wavelength vector, the median spectral resolution as a function of wavelength for the fibers in this IFU, the standard deviation of spectral resolution as a function of wavelength for the fibers in this IFU, a binary table describing the individual exposures that make up the file, and arrays of the effective X and Y positions in arcsec of each fiber (as a function of wavelength) relative to the IFU center.
These are the final 3d data cubes for a given galaxy that combine all fiber spectra across all exposures as described here.
The LOGCUBE data cube has logarithmic wavelength sampling from log10(lambda/Angstroms)=3.5589 to 4.0151 (NWAVE=4563 spectral elements), and 0.5 arcsec spatial pixels (spaxels) for a total size of NX x NY x NWAVE pixels; LINCUBE is the same, except it has linear wavelength sampling from from 3622.0 to 10353.0 Angstroms (NWAVE=6732 spectral elements).
The detailed data models for the LOGCUBE and LINCUBE files can be found here). Similar to the RSS files, they contain extensions for the flux (now in units of 10-17 erg/s/cm2/Angstrom/spaxel), inverse variance, pixel mask, wavelength vector, median spectral resolution and standard deviation thereof, and a binary table describing the individual exposures that make up the file. Additionally, they also included reconstructed broadband ‘griz’ images created from the spectral data cube, and estimates of the reconstructed point source profiles in each of the ‘griz’ bands.
SDSS Legacy Imaging
MaNGA requires that broadband imaging data for target galaxies be available for the DRP to use to register astrometric solutions against and to estimate achieved S/N depth. We use the NASA Sloan Atlas (NSA) re-reduction of the SDSS imaging data, and we provide FITS postage stamps of this re-reduction for each MaNGA galaxy released as part of DR13.
The detailed data model for the preimaging data can be found here. In brief, they are multi-extensions FITS files giving the flux, inverse variance, and psf in each of the four ‘griz’ bands.