Testing masking effectiveness using multi-line image cubes based on COSMOS2020 for [CII] line intensity mapping at $z_{[CII]} > 3.5$

Technical Summary: Testing Masking Effectiveness Using Multi-Line Image Cubes Based on COSMOS2020 for [CII] Line Intensity Mapping at $z_{[CII]} > 3.5$

Problem Statement
Line Intensity Mapping (LIM) offers a unique capability to probe the three-dimensional distribution of galaxies and the intergalactic medium during the Epoch of Reionization (EoR) and post-EoR eras without resolving individual sources. A primary target for future LIM experiments, such as the EoR-Spec module on the Fred Young Submillimeter Telescope (FYST), is the ionized carbon fine-structure line ([CII]) at $z \approx 3.5 - 8.2$. However, LIM observations in the submillimeter regime are heavily contaminated by foreground emission, particularly from Carbon Monoxide (CO) rotational transitions originating from lower-redshift galaxies ($z < 2$). Unlike traditional surveys, LIM cannot inherently distinguish between the target [CII] signal and contaminant CO emission within the same frequency channels. Previous simulations often lacked realistic foreground contamination or did not fully test the efficacy of masking techniques required to recover the [CII] power spectrum. This work addresses the challenge of recovering [CII] signals in the presence of significant CO contamination and instrumental noise, specifically testing whether masking strategies can effectively isolate the target signal in the E-COSMOS field.

Methodology
The authors extended the empirical simulation framework established by Clarke et al. (2024, C24) to create multi-line intensity cubes for the EoR-Spec frequency range (205–420 GHz) and two hypothetical lower-frequency extensions (90 and 150 GHz).

1. Data and Sample Construction:

   • The study utilized the COSMOS2020 galaxy catalogue (version 4.1.11) covering a 1.44 deg$^2$ area (E-COSMOS).
   • To address the intrinsic incompleteness of the catalogue, the authors generated two sample types: a "lower limit" sample based strictly on observed galaxies (FARMER LP) and an "extrapolated" sample. The extrapolation involved duplicating galaxies to match mass completeness statistics derived from CANDELS and fitting Schechter functions to the stellar mass function, thereby adding faint and bright galaxies to better represent the true sky population.
   • Bulk properties (Stellar Mass, SFR, $L_{IR}$) were used to assign line luminosities.

2. Line Luminosity Modeling:

   • [CII]: Three models were applied to bracket the uncertainty: a conservative lower limit based on local dwarf galaxies (De Looze et al. 2014), a middle-ground model based on SFR (Schaerer et al. 2020), and an upper limit based on high-redshift ALPINE-ALMA galaxies (C24_m3).
   • CO: Models for the $J=1-0$ transition were derived from $L_{IR}$ relations (Sargent et al. 2014; Li et al. 2016). Other CO transitions ($J=2-1$ up to $J=13-12$) were derived using Spectral Line Energy Distribution (SLED) templates, mixing bright (NGC 253) and faint (Milky Way) galaxy profiles based on the galaxy's position in the main sequence.
   • Cube Construction: Luminosities were converted to intensities and inserted into 3D voxels. The cubes were convolved with the instrument beam (2D Gaussian in map space, 1D Lorentzian in frequency space) and sub-gridded by $3\times3\times3$ to facilitate masking.

3. Noise Modeling:

   • The study incorporated both white noise and correlated noise (1/f noise) based on Time-Ordered Astrophysics Scalable Tools (TOAST) simulations for EoR-Spec. The noise cubes were scaled to represent the full 2000-hour observing period.

4. Masking Techniques:

   • Targeted Masking: Utilized the COSMOS2020 catalogue as a foreground mask. Bright galaxies known to emit CO were identified, and voxels around them were masked. The authors tested various mask radii (0.5$\sigma$ to 2.5$\sigma$) and iteratively determined an optimal masking order based on an efficiency parameter (CO removal per unit volume masked).
   • Blind Masking: Did not rely on an external catalogue. Instead, it identified local maxima in the total intensity cubes across multiple frequency bands. Voxels were paired if their spatial separation and frequency difference corresponded to a single galaxy emitting in two different lines (e.g., different CO transitions or CO and [CII]). This technique was tested with and without the hypothetical 90/150 GHz bands.

Key Results

• CO vs. [CII] Signal Strength:

  • Above 300 GHz ($z_{[CII]} < 5$), the predicted [CII] and CO intensities are comparable (within $\sim0.5$ dex), with CO dominating at lower frequencies.
  • Below 300 GHz, CO emission exceeds [CII] by more than 1 dex in non-extrapolated models, making recovery via masking extremely difficult without additional frequency coverage.
  • The CO power spectra generally exceed the predicted atmospheric sensitivity limits for $k > 0.5$ Mpc$^{-1}$, but fainter models fall below the noise floor at low $k$ modes.

• Targeted Masking Efficacy:

  • Success: Targeted masking successfully recovered [CII] power spectra above 300 GHz when using wide masks ($\ge 1.5\sigma$). In ideal conditions (complete foreground catalogue), the recovered [CII] signal was distinct from the residual CO.
  • Limitations: At 280 GHz, even maximum masking failed to recover [CII] due to the overwhelming CO signal.
  • Overmasking: The study identified a specific risk of "overmasking" (accidentally removing [CII] signal) when masking the CO $J=5-4$ transition at 410 GHz with bright galaxy masks. This suggests specific transitions must be excluded from masking strategies in the E-COSMOS field.
  • Catalogue Completeness: If the foreground catalogue is incomplete (missing faint extrapolated galaxies), targeted masking becomes significantly less effective, as unmasked CO sources remain.

• Blind Masking Efficacy:

  • Blind masking proved superior when the foreground catalogue is incomplete, as it does not rely on prior knowledge of galaxy locations.
  • The technique is most effective when utilizing the hypothetical 90 and 150 GHz bands, which allow for more robust voxel pairing across frequency. Without these bands, blind masking is less efficient but still outperforms targeted masking with an incomplete catalogue.
  • Blind masking is highly sensitive to noise; at high frequencies (>300 GHz), noise spikes can be mistaken for line emission, reducing the technique's effectiveness by up to 0.5 dex.

• Noise Impact:

  • Combined white and correlated noise hinders [CII] recovery above 300 GHz. Achieving a Signal-to-Noise (S/N) ratio $>5$ for the clustering signal would require approximately five times the currently planned observing time (or improved instrument sensitivity).
  • While CO can be recovered below 300 GHz, [CII] recovery in this regime remains unlikely without cross-correlation techniques due to the dominance of CO and noise.

Significance and Claims
The authors claim that their work provides a realistic, empirically grounded assessment of the feasibility of [CII] LIM with the EoR-Spec instrument. The study demonstrates that:

1. Feasibility: Recovering [CII] via masking is feasible for frequencies above 300 GHz, provided that wide masking radii are used and the foreground catalogue is sufficiently complete.
2. Necessity of Multi-Band Data: The inclusion of lower frequency bands (90/150 GHz) or cross-instrument observations is critical for the success of blind masking and for cleaning CO contamination effectively.
3. Methodological Validation: The empirical approach using COSMOS2020, combined with extrapolation and SLED templates, yields CO estimates consistent with ALMA, VLA, and NOEMA observations, validating the simulation framework.
4. Operational Guidance: The paper identifies specific operational constraints, such as the need to avoid masking the CO $J=5-4$ transition in E-COSMOS to prevent overmasking, and highlights that current noise levels will likely require extended observing times or advanced noise mitigation (e.g., PCA cleaning) to achieve robust detections.

The authors conclude that while contaminant mitigation via targeted and blind masking is a viable strategy for EoR-Spec, the success of the experiment depends heavily on the completeness of foreground catalogues, the availability of multi-frequency data, and the management of instrumental noise. They position this work as a foundation for future cross-correlation studies and more advanced deconfusion techniques.