The LVM Data Reduction Pipeline (DRP) takes raw exposures from our LVM-I system of 3 spectrographs, with 3 spectral channels each, fed by fibers from 4 telescopes (1 science, 2 sky, and 1 spectrophotometric) to produce a flux calibrated and sky subtracted row-stacked spectra (RSS) file. The release version of this pipeline is 1.2.0, available on github.
The LVM reduction package consists of two pipelines: a calibrations pipeline and a science pipeline. The calibrations pipeline takes the raw calibration frames to produce the necessary calibration products to reduce a science frame. On the other hand, the science pipeline takes raw science frames to produce science-ready fully calibrated and sky-subtracted spectra for each science fiber. We provide an overview of the reduction steps below.
Overview
One of the defining characteristics of the LVM-I is its stability, which stems from the fact that the fiber bundle is fixed. This, along with the telescope design, translates into LVM being a highly efficient survey. From the DRP standing, this also means a low-cadence for the sequence of calibrations exposures. Currently, under normal operations, a full calibration sequence is taken every 3–4 weeks. Calibrations can be triggered due to external perturbations to the instrument, like a seismic or thermal event, or due to bad weather, which results in down time when no science exposures take place. The calibrations pipeline is triggered when a new set of calibrations is available. Each calibration run yields fiber profile parameters, wavelength and LSF models and fiber flat fields. These calibration products are eventually used to reduce each science exposure.
The science pipeline performs the following steps to produce each sky subtracted, flux calibrated RSS file (‘lvmSFrame‘):
In the following we will describe the main reduction steps carried by the science pipeline.
2D reduction
Preprocessing of any individual science exposure includes overscan and bias subtraction, application of the pixel flat frame, and conversion to electrons using the gain values. Any quasi-static CCD artifacts (e.g., hot pixels, bad columns, etc.) are added to a pixel mask extension. Given our 900s exposure times and that the majority of observations are not dithered, we use L. A. Cosmic, the Laplacian CR rejection algorithm, to detect and flag cosmic rays. We also model and subtract stray-light.
Extraction of 1D spectra
Our 1D spectra are then extracted, assuming a physically motivated profile calculated as the convolution of a collapsed Tophat function—this captures the shape of the fiber projected onto the CCD—and a Gaussian profile—this captures the spectrograph PSF spatial variations. Before the extraction, the pipeline accounts for sub-pixel thermal shifts—typically ~0.1 pixels within a single night—by cross-matching a subset of columns against the fiber model. Afterwards, the extraction happens on all 9 cameras. At this point we conveniently stack all 3 spectrograph spectra, which yields 3 RSS files corresponding to 3 observed spectral channels (b, r, z), all containing information about all fibers in the system.
Wavelength calibration
In addition to the wavelength calibration obtained from our arc lamp images, associated with our master calibration frames, the LVM pipeline accounts for potential thermal shifts in the wavelength model. For this, we measure the Gaussian centroids of strong and isolated sky lines within the science data. This is done separately for each arm of the spectrograph, using 1–4 lines per spectrograph. Typical wavelength corrections are ~0.01 Angstroms within a single night. Finally, all spectra are sampled to a common wavelength grid, with a wavelength sampling of 0.5 Angstroms.
Heliocentric velocity corrections are calculated based on the observation time and sky position. These are added to the metadata associated with the file in the FITS header, but are not applied to the spectra.
Astrometric solution
Images from the acquisition and guide cameras, that bracket the science fiber bundle, are used to derive the final astrometric solution associated with the science field. Our astrometric accuracy is typically <1″ RMS. The astrometric solution to the central fiber of each telescope is stored in the primary header in degree units as {TEL}RA, {TEL}DEC, where TEL is any of: SCI (science), SKYE (sky east) or SKYW (sky west). This astrometric solution is used to determine the location of each individual fiber in the system. All individual astrometric positions for each fiber are stored in the position table, available in the SLITMAP extension of the RSS files.
Flux calibration
The LVM Observing Strategy is designed to provide co-temporal observations of 12 sequentially observed F-type stars during the course of one science exposure. By observing a large number of stars, we aim to mitigate any effects due to errors in the spectral typing, following the strategy previously pursued by MaNGA. These 12 stars are acquired using the spectrophotometric telescope, but fed through the same spectrograph system, with 4 stars exposed on to each spectrograph. These spectra are therefore calibrated and reduced identically to the science spectra, following the steps described above. These stars were selected based on their magnitude, isolation (to dominate our 35.3″ spaxels) and Apsis parameters from Gaia DR3.
Our preliminary method of determining the spectral sensitivity of the instrument was based on a comparison of our observed spectra with the Gaia XP spectra (STD method). However the Gaia spectra have significantly lower spectral resolution and cannot be used for telluric corrections. To overcome these limitations, the in 1.2.0 the pipeline implements a new flux calibration method based on high-resolution synthetic stellar spectra (MOD method). An additional backup sensitivity function is also calculated from an average instrumental sensitivity calculated across ~2 year of survey data (SCI method) and using bright, relatively isolated science field stars to calculate the zero point in synthetic pass-bands in all three brz channels.
Once these spectral sensitivity functions has been determined, the pipeline chooses the calibration method based on the availability of good quality standard stars: if more than 8 standard stars have good signal-to-noise levels, the default template-based MOD method is used to flux-calibrate all fibers. Otherwise the SCI method will be used. The STD is still propagated but never used and it is kept for backwards compatibility. Finally, the data from the three spectrograph channels is combined to produce a single spectrum per fiber covering our full 3600–9800 Angstrom wavelength range.
Sky subtraction
The LVM Observing Strategy is also designed to provide co-temporal observations of two sky fields. Since many of the science cases involve very spatially extended sources (e.g., Milky Way emission), we use a novel strategy to observe and model the sky in order to not subtract possible science signal along with the sky background. This is done by monitoring in parallel in two separate locations on the sky. One location is selected from a grid of 768 positions to be relatively dark region within 10 degrees of the science field, and is intended to provide constraints on the sky continuum level. The second location is selected to be very dark, particularly in Milky Way foreground emission, in order to provide constraints on the time-variable sky line features, including geocoronal emission.
These two fields are observed with small fiber bundles of ~60 fibers each, which are fed through the same spectrograph system, distributed approximately uniformly across each of the spectrographs. These spectra are therefore calibrated and reduced identically to the science spectra, following the steps described above.
Calibrated sky spectra are then analyzed separately for their constraints on the sky continuum and sky lines, and extrapolated to the location of science field. The current version of the pipeline does not implement geocoronal measurements, therefore this emission is not being removed from the science emission.
Final data product
Our sky subtracted, flux calibrated RSS files resulting from the LVM DRP are designated with the SFrame naming convention. The typical filenames follow the format lvmSFrame-{expnum}.fits, where {expnum} is the 8-digit zero-padded exposure number for a given observation. The intermediate DRP data products will not be part of this data release, but for reference these are the lvmFrame (after flat fielding), lvmFFrame (after flux calibration), and lvmCFrame (after the spectral channels are combined).
Within a single lvmSFrame file, we include different extensions. For each spectrum, we provide the flux measurements FLUX, the associated errors (reported as inverse variances IVAR), and binary masks to identify data issues MASK (True if bad pixel). All spectra are on a fixed wavelength grid that we provide, WAVE, but in the original per-fiber line spread function LSF. Also included are the subtracted sky spectra and error, SKY and SKY_IVAR, respectively. We provide each version of the sensitivity curve (FLUXCAL_MOD, FLUXCAL_SCI, FLUXCAL_STD), along with a position table associating fiber locations to right ascension and declination coordinates, SLITMAP. A full data model for our SFrame files is provided here: LVM SFrame data model.
A description of the current LVM DRP caveats can be found at LVM caveats.