Source-linked AI summary
Spectral Mapping Reconstruction of Extended Sources
J. D. T. Smith, L. Armus, D. A. Dale, H. Roussel, K. Sheth, B. A. Buckalew, T. H. Jarrett, G. Helou, R. C. Kennicutt
TL;DR
Extended-source three-dimensional spectroscopy is usually performed with dedicated integral field spectrographs, but detector tradeoffs and long-slit observing challenges motivate an alternative. The paper reconstructs full spectral cubes from rastered single-slit observations using a flux-conserving algorithm, addresses extended-source calibration, and implements the method in Cubism for Spitzer IRS maps. Cubism has been used for individual maps, ultra-deep surveys, and large-area mosaics.
Problem
Dedicated integral field designs face detector tradeoffs, while reconstructing cubes from overlapping slit spectra requires handling mixed spatial-spectral information, unknown relative offsets, and flux conservation.
Method
The paper reconstructs full spectral cubes from rastered single-slit observations using pseudo-rectangles and δ pixel-overlap weights, with extended-source flux calibration and the Cubism implementation.
Results
Cubism has been used for maps containing ∼9000 spectral data frames, ultra-deep blind surveys, and large-area mosaics covering hundreds of square arcminutes.
Takeaways & Limitations
Spectral mapping with a fixed single-slit instrument provides sensitive, spatially resolved infrared spectroscopy while retaining efficient point-source follow-up.
Takeaways & Limitations
The reconstruction introduces adjacent-pixel noise correlations, although filled maps often minimize them through high pixel redundancy.
Abstract
from arXiv · showhide
Three dimensional spectroscopy of extended sources is typically performed with dedicated integral field spectrographs. We describe a method of reconstructing full spectral cubes, with two spatial and one spectral dimension, from rastered spectral mapping observations employing a single slit in a traditional slit spectrograph. When the background and image characteristics are stable, as is often achieved in space, the use of traditional long slits for integral field spectroscopy can substantially reduce instrument complexity over dedicated integral field designs, without loss of mapping efficiency -- particularly compelling when a long slit mode for single unresolved source followup is separately required. We detail a custom flux-conserving cube reconstruction algorithm, discuss issues of extended source flux calibration, and describe CUBISM, a tool which implements these methods for spectral maps obtained with ther Spitzer Space Telescope's Infrared Spectrograph.
1. INTRODUCTION
Traditional integral field spectrographs trade field of view, spectral resolution, and wavelength coverage, while long slits can be inefficient for unresolved sources and difficult to use on extended sources from the ground. The paper proposes reconstructing full spectral cubes from rastered single-slit observations, leveraging stable space-based backgrounds and image characteristics.
- Instrument tradeoffs: Three-dimensional spectrograph designs must trade field of view, spectral resolution, and instantaneous wavelength coverage within finite detector area.Under identical throughput, detector performance, and an unchanging PSF, designs using the same fractional detector area have identical spectral mapping efficiency.
- Instrument tradeoffs: Integral field spectrographs capture an entire field simultaneously, mitigating sky, transmission, and PSF variations but allocating detector area to unused background around unresolved sources.This makes them relatively inefficient for single unresolved-source observations.
- Ground-based limitations: Long-slit mapping of extended sources from the ground is challenged by varying sky brightness, transmission, and PSF across potentially long integration sequences.The difficulty is especially acute for sources with surface brightness comparable to or below the sky.
- Space-based opportunity: Space-based observations minimize these long-slit limitations because astrophysical backgrounds are typically smooth and unvarying while image characteristics remain fixed.The Spitzer IRS uses fixed single slits and no moving parts across 5–38 µm.
- Proposed approach: The paper describes reconstructing full spectral cubes from scanned single-slit observations and presents a custom algorithm, flux-calibration methods, and the Cubism implementation for Spitzer IRS maps.The approach targets two spatial dimensions plus one spectral dimension.
2. SPECTRAL CUBE RECONSTRUCTION
The section develops a flux-conserving reconstruction approach for spectral cubes from overlapping slit spectra while addressing resampling noise and the spatial–spectral degeneracy of undersampled data.
- Resampling Noise: Resampling curved spectral orders can introduce periodic noise of up to 20% even when the order center is known precisely.The effect depends on extraction width and the rate at which the order center crosses detector-row boundaries.
- Resampling Noise: This resampling noise reflects irrecoverably lost information about flux distribution along the slit, rather than detector properties or PSF shape.It persists at the few-percent level near typical sampling of 2 pixels per FWHM and is worse for undersampled PSFs.
- Algorithm Design: The reconstruction avoids interpolation during order straightening and raster remapping to mitigate additional resampling noise, although undersampling cannot generally be recovered without prior slit-profile knowledge.The method instead handles curved orders and cube placement through geometric overlap calculations.
- Spectral–Spatial Degeneracy: Because slit pixels mix spatial and spectral adjacency, the adopted assumption places adjacent dispersion-direction pixels into adjacent wavelength planes, leaving spatial information to the mapping observations.Alternative assumptions distribute information alongside wavelength planes or use a varying spatial–spectral mixing angle.
- Algorithm Design: The algorithm represents each wavelength with a tilted pseudo-rectangle and computes exact overlap weights between input pixels and output cube pixels.Accurate order, slit-center, and output-grid positioning is required because errors degrade spectral or spatial resolution and increase adjacent-pixel noise correlation.
- Algorithm Design: Pixel redundancy minimizes noise correlations in practice, while the polygon-clipping method conserves surface brightness and tracks individual input contributions for discrepant-pixel rejection.Filled maps may have a dozen or more input pixels contributing to one output pixel.
3. EXTENDED SOURCE FLUX CALIBRATION
Extended-source flux calibration with slit spectrographs requires correcting point-source-based references for aperture and slit losses. The slit-loss correction depends on wavelength and on the source’s spatial flux distribution, while beam-profile validation is degenerate with the slit’s effective solid angle.
- Point-source spectrophotometric references are used because uniform, beam-filling reference sources with known intensity are generally unavailable.
- Aperture corrections are required because photometric apertures encircle only part of a reference star’s total flux.
- The aperture-loss correction function estimates wavelength-dependent losses and corrects normally extracted reference-star spectra toward infinite-aperture fluxes.
- Slit throughput varies appreciably with wavelength in diffraction-limited spectrographs because telescope diffraction broadens the PSF.
- The slit-loss correction function is the wavelength-dependent fraction of centered point-source flux admitted by the slit; without it, extended-source fluxes are overestimated.
- Synthetic-photometry tests cannot separately determine slit-loss corrections and effective slit solid angle because scaling one can be compensated by changing the inferred beam.
4. CUBISM
Cubism implements the paper’s cube-reconstruction methods for Spitzer IRS spectral maps. It combines two-dimensional spectral images into three-dimensional cubes and calibrates its effective pixel solid angle against imaging photometry.
- Cubism combines IRS mapping spectral images into cubes with two spatial dimensions and one spectral dimension.
- The tool is implemented in IDL and includes cube-analysis tools alongside the reconstruction functionality.
- Cubism uses pipeline-processed two-dimensional BCD spectral images by default and applies the twin-clipping reprojection algorithm.
- Its effective solid angle per pixel is calibrated by comparing integrated spectrophotometry with corrected photometry from several instruments in bright SINGS galaxies.
- The calibrated solid-angle terms agree with measured slit widths within 10%.
4.2. Cube Construction
Cubism constructs a cube grid around the mapped region and places spectra using reconstructed slit positions. It processes each IRS sub-slit independently and supports contribution tracking, accelerated clipping, and simulated-data validation.
- The output cube grid bounds the mapped region, with default pixel sizes matching module sampling from 1.′′85 to 5.′′08.
- Spectra are offset according to slit-center positions, with telescope-attitude reconstruction producing more stable results than requested positions.
- Each of the six IRS sub-slits is treated independently, and one cube can contain data from only one sub-slit.
- Cube accounts store the fractional contributions of input pixels to each output pixel and enable statistical outlier rejection.
- The polygon-clipping core is written in C for speed, with an IDL-native fallback that incurs a significant speed penalty.
- Cubism was extensively tested on simulated spectral maps to validate reconstruction and flux conservation.
4.3. Uncertainty
Cubism propagates input pixel uncertainties into full error cubes for extracted spectra and line or continuum maps. These estimates capture statistical ramp uncertainties but omit later calibration and processing errors.
- Full error cubes are generated by standard propagation of input BCD-level uncertainties through the reconstruction equation.
- The resulting uncertainties support error estimates for extracted spectra and constructed line or continuum maps.
- The estimates include statistical ramp uncertainties but neglect errors from calibration and later processing steps.
4.4. Background
Background frames are subtracted from each input spectral record at the two-dimensional level to remove astrophysical foregrounds and backgrounds and mitigate time-varying detector artifacts.
- Near-in-time background frames are subtracted from each input data record before cube reconstruction.The frames may come from the observations, dedicated offset observations, or other suitable sources when direct backgrounds are unavailable.
4.5. Bad Pixels
Cubism combines instrument and user-defined bad-pixel information, with optional automatic detection controlled by sigma-trimming and minimum-contribution parameters.
- Bad pixel masks combine permanent nonresponsive pixels, exposure flags, and user-marked global or record-specific exclusions.The instrument flags include saturation and flat-field difficulties reported in the input frames.
- Automatic bad-pixel detection uses rσ as the sigma-trim threshold and fmin as the minimum contributing fraction.Contributing pixel residuals are tested against an estimated deviation derived from variance or median absolute deviation.
- Outlier testing can use residuals computed with or without subtracted background for each output cube pixel.
4.6. Components
Cubism provides project management, visualization, extraction, and map-making components that operate on shared spectral-mapping data and cube products.
- 4.6. Components: The Cubism interface has a project window, a multipurpose viewer, and a spectrum viewer with map-creation capabilities.These components communicate through a shared scriptable object-oriented data container.
- 4.6.1. Project Window: The project pane manages records, calibration inputs, bad pixels, background records, cube-build parameters, and cube construction.Projects can be loaded, saved, or recovered, and records can be added, removed, disabled, combined, sorted, or examined.
- 4.6.2. CubeView: CubeView displays individual records, record stacks, spectral cubes, maps, and visualization images through context-dependent tools.It supports background toggling, pseudo-rectangle editing, bad-pixel editing, cube-plane navigation, and extraction.
- 4.6.2. CubeView: CubeView can show input stacks with overlays and bad pixels, slit-position visualizations, or cube-derived stacks with extracted regions.
- 4.6.2. CubeView: Multiple CubeView instances can operate simultaneously, with each project communicating with its own viewer tools.
- 4.6.3. CubeSpec and Map Creation: Cube extractions are displayed in CubeSpec, while maps can average foreground and continuum wavelength regions using weighted or uniform continuum estimates.The tool also supports saved map sets, redshift shifts, line fitting, and filter-curve weighting.
4.7. Outputs
Cubism exports cubes, maps, and spectral extractions in standard FITS and IPAC Table formats, retaining spectral coordinates, extraction metadata, and WCS information.
- Full spectral cubes are exported as FITS with wavelengths encoded as a lookup table following FITS spectral-coordinate standards.Maps are written as standard two-dimensional FITS images with their foreground and background wavelength ranges recorded.
- Spectral extractions are exported in IPAC Table format with labels, units, and coordinates describing the extraction rectangle.
- All output products include WCS coordinate systems with estimated accuracy of 1′′.
5. OPTIMIZED SPECTRAL MAPPING
High pixel redundancy is central to producing high-quality spectral cubes, while careful raster stepping prevents gaps and uneven photometric response.
- 5. OPTIMIZED SPECTRAL MAPPING: Pixel-sampling redundancy primarily determines spectral-cube quality and supports robust statistical rejection of bad pixels.It also reduces the impact of slit-throughput or detector-response variations and noise correlation between neighboring output pixels.
- 5. OPTIMIZED SPECTRAL MAPPING: At least 4 different detector pixels should sample each sky position for high final cube quality with the IRS.
- 5. OPTIMIZED SPECTRAL MAPPING: Stepping by one-half the slit width in the dispersion direction is critical because wider spacing risks gaps or uneven photometric response.
6. SUMMARY
The paper presents single-slit raster mapping as an efficient route to integral-field spectroscopy in stable observing regimes, supported by a reconstruction algorithm and its Cubism implementation.
- 6. SUMMARY: Rastering a single long slit can efficiently provide integral-field spectroscopy for extended sources when backgrounds and image characteristics are stable.The approach can reduce instrument complexity and size when a long-slit mode is already needed for isolated point-source followup.
- 6. SUMMARY: A resampling-noise-minimizing, two-pass polygon-clipping algorithm reconstructs spectral cubes with two spatial and one spectral dimension.
- 6. SUMMARY: A Cassiopeia A map combines three continuum-subtracted lines, [Ne ii], [Ar iii], and [S iv], while retaining a full 5–38 µm spectrum at each position.
- 6. SUMMARY: Cubism implements the reconstruction algorithm for Spitzer IRS spectral maps and has been used for individual maps, ultra-deep surveys, and large-area mosaics.The reported applications include maps with ∼9000 individual spectral data frames and mosaics covering hundreds of square arcminutes.