SIMLA Pipeline: Spitzer/IRS Cube Assembly

• SIMLA Pipeline is a methodology and software toolkit designed to create validated mid-infrared spectral cubes from Spitzer/IRS low-resolution mapping data.
• It employs novel partitioning of instrumental backgrounds and robust statistical cleaning to subtract time-varying zodiacal and detector artifacts.
• The pipeline uses CUBISM for 3D cube construction and integrates comprehensive validation protocols to ensure high-quality outputs for Galactic and extragalactic research.

The SIMLA pipeline is a methodology, software toolkit, and operational procedure for constructing validated, background-subtracted mid-infrared spectral cubes from Spitzer Space Telescope Infrared Spectrograph (IRS) low-resolution mapping data (5.2–38 μm, $R \sim 60$–130). Designed to deliver research-grade spatial–spectral data products for hundreds of resolved and unresolved Galactic and extragalactic targets, the pipeline embodies novel partitioning of instrumental backgrounds, robust statistical cleaning of detector artifacts, and sophisticated 3D cube assembly leveraging the CUBISM software interface. Its development and algorithmic details are codified in "SIMLA: The Spitzer Infrared Spectrograph Mapping Legacy Archive" (Donnelly et al., 13 Dec 2025).

1. Ingest and Preprocessing of Spitzer/IRS Basic Calibrated Data

The SIMLA workflow initiates by ingesting Basic Calibrated Data (BCD) produced by the IPAC pipeline (v S18.18.0). BCDs are then organized by Astronomical Observation Request (AOR), IRS module (SL1, SL2, SL3, LL1, LL2, LL3), RAMPTIME, and for the long-low (LL) modules, also by detector bias epoch. This granular grouping is essential for capturing instrumental configuration dependencies, ensuring that subsequent background modeling and calibration are performed within homogeneous observational subsets. Each BCD includes associated bad-pixel masks and uncertainty extensions which propagate through later processing.

2. Construction and Subtraction of Instrumental and Astrophysical Backgrounds

Each AOR receives a custom 2D background image constructed as the sum of three physically and instrumentally motivated components: the Zodiacal Emission Model (ZEM), a per-AOR "baseline frame" of stable (dark+ZEM residual) structure, and time-variable pixel offsets measured via "shard" analysis.

2.1 Zodiacal Emission Model (ZEM)

For every AOR $k$, the Kelsall et al. (1998) zodiacal model spectrum $Z_k(\lambda)$ is evaluated at the Spitzer ephemeris and pointing. This is differenced against a Campaign-dependent "superdark" average $\hat Z_{\rm CVZ}(\lambda)$, normalized at 10.95 μm (SL1) or 28.75 μm (LL1). The time-dependent amplitude $a(t)$ is fit from the observed residuals in the Continuous Viewing Zone (CVZ). The 2D background is then generated by projecting these differences onto the BCD geometry via an inverse wavesamp mapping $W^{-1}_{ij}$, as

$$\mathcal{Z}_{k,ij}(\lambda) \;=\; W^{-1}_{ij}\{\,Z_{k}(\lambda)-a(t)\,\hat Z_{\rm CVZ}(\lambda)\}$$

producing a synthetic zodiacal emission image per BCD.

2.2 Baseline Frame (Stable Dark + ZEM Residuals)

A large library of blank-sky ("dark AOR") BCDs, each with its ZEM estimate subtracted, forms the basis for per-pixel median background estimation. These are partitioned into four bins in modeled 12 μm sky intensity $z_d$. Per bin, the 1 σ, 5-iteration sigma-clipped median defines $s_{z',ij}$ frames. The science AOR $k$'s ZEM intensity $z_k$ is used to linearly interpolate between nearest bins: $$S_{k,ij} \;=\; \mathrm{Interp}(z_k;\{z'_L,s_{z'_L,ij}\},\{z'_H,s_{z'_H,ij}\})$$ producing the baseline frame $S_{k,ij}$ tuned to the science background conditions.

2.3 Time-variable Pixel Offsets (Shards)

Detector footprints are divided along the slit into five "shards" (strips), with edge trimming (8% for SL, 4% for LL). Shards qualifying as "dark" are selected by:

• WISE W3 intensity: $$| \langle I_{\rm W3} \rangle_{\rm shard} | < 0.1\; \mathrm{MJy/sr}$$
• BCD median: $|I_{\rm shard}| < 2\; \mathrm{MJy/sr}$ (after $\mathcal{Z}_{d,ij}$ and $S_{d,ij}$ are subtracted)
• Proximity in ZEM level: $|\Delta z| \le 10\; \mathrm{MJy/sr}$
• Proximity in epoch: $\Delta t \le 10$ days

Selected shards are stacked to a target depth (50 is default). Across this stack, a 1.5 σ, 3-iteration sigma-clipped mean is taken per pixel: $$D_{k,ij} \;=\; \frac{1}{N}\sum_{n=1}^N \left[d_{n,ij}-(\mathcal{Z}_{d,ij}+S_{d,ij})\right]$$ The final composite background for AOR $k$ is thus

$$B_{k,ij} \;=\; \mathcal{Z}_{k,ij} \;+\; S_{k,ij} \;+\; D_{k,ij}$$

This background is subtracted from every science BCD, yielding data for 3D cube assembly.

3. 3D Cube Construction and Bad Pixel Mitigation

Cube assembly is performed by CUBISM. Each background-subtracted BCD image $I^{(i)}_{pq}(\lambda)$ is mapped as detector polygons onto a world-coordinate grid (RA–Dec–$\lambda$), with wavelength-dependent distortion taken into account. For every voxel $(x, y, \lambda)$, detector–cube overlap fractions $w_{i,pq}(x,y,\lambda)$ are computed. Flux resampling is achieved through area-weighted averaging: $$F(x,y,\lambda) =\frac{\sum_{i,p,q}w_{i,pq}(x,y,\lambda)\,f_i(p,q,\lambda)}{\sum_{i,p,q}w_{i,pq}(x,y,\lambda)}$$ Variance propagates analogously.

Automatic bad-pixel detection is then implemented: if over 40% ($f_{\rm occ}=0.4$) of samples at a cube voxel are $>4\sigma$ outliers from the local median, those detector pixels are flagged and excluded. All cubes are calibrated to extended source units (slit-loss corrected). An optional stray light correction (sl_io_correct) can be applied to SL cubes.

4. Statistical Validation and Quality Assessment

Each cube undergoes several validation protocols:

• Extraction of dark-region spectra: median combines voxels whose shards passed all selection cuts; resulting surface brightness distributions are verified to be centered at $0 \pm 1$ MJy/sr.
• WISE W3 photometry cross-calibration: $5''$-radius synthetic photometry from SIMLA cubes (convolved with the W3 response) is compared to WISE background-subtracted images for $I_{\rm W3}>0.1$ MJy/sr, yielding a median ratio consistent with $1.19$.
• Noise-reduction metric: For noise-only image tests, define $$\mathcal{R} = \sigma_{\rm after}/\sigma_{\rm before}$$, with typical $\mathcal{R} \sim 0.6$–0.8.
• Sigma-clipping for outlier rejection is performed throughout (baseline: 1σ, 5 iterations; shard stacking: 1.5σ, 3 iterations).

5. Configuration, I/O Structure, and Parameter Defaults

Inputs

• BCD FITS files (all SL/LL mapping AORs), uncertainties, bad-pixel masks
• AOR metadata (WCS grid, step pattern, RAMPTIME, Campaign)
• Zodiacal model engine (Kelsall 1998, ephemeris)
• WISE W3 all-sky coadds

Outputs

• For each AOR / suborder: background-subtracted data-cube FITS (flux, variance, mask)
• Raw (no IO-correction) and IO-corrected SL cubes
• Dark-region masks and QA plots

Table: Key Pipeline Parameters

Parameter	Default/Range	Description
WISE shard cut	$|I_{\rm W3}|<0.1$ MJy/sr	Selects truly dark shards
BCD shard cut	$|I|<2$ MJy/sr	Spectral median after background
$\Delta z_{\rm max}$ (ZEM)	10 MJy/sr	Shard vs target ZEM proximity
$\Delta t_{\rm max}$	10 days	Temporal proximity for stacking
Target shard depth	50	No. of shards per stack
Baseline bin count	4	12μm intensity bins
Sigma-clip (baseline)	$1\sigma$, 5 iter	Outlier rejection in median
Sigma-clip (shard stack)	$1.5\sigma$, 3 iter	Outlier rejection in stacking
Bad-pixel flag (CUBISM)	$4\sigma$, $\geq 40\%$	Fractional occurrence threshold

6. Performance, Scope, and Future Development

Pipeline throughput is nearly linear in number of BCDs per AOR: a fully sampled SL map ($\sim50$ BCDs) is processed in approximately 10–20 minutes, while large LL maps ($\sim2000$ BCDs) can require several hours on an 8-core workstation. The current release covers only low-resolution modules (R$\sim$60–130); coverage of high-resolution (SH/LH), moving-target, and multi-AOR mosaics is reserved for future versions.

Planned extensions include enhanced bad-pixel flagging using full-mission rogue maps, automated stitching of multi-AOR mosaics, inclusion of IRS high-resolution cubes, and integration with JWST/MIRI drizzle kernels.

A documented limitation is that maps with mean surface brightness $<0.1$ MJy/sr can suffer loss of astrophysical emission due to aggressive dark-background selection. All described algorithmic steps, parameter choices, and validation procedures are traceable to (Donnelly et al., 13 Dec 2025), enabling robust reproducibility and adaptation of the SIMLA pipeline for independent Spitzer/IRS mapping studies.