meer21cm: an Analysis Pipeline and Comprehensive Toolkit for HI Intensity Mapping

Technical Summary: meer21cm: An Analysis Pipeline and Comprehensive Toolkit for HI Intensity Mapping

Problem Statement
Neutral hydrogen (HI) intensity mapping has emerged as a powerful probe of the post-reionization Universe, yet it faces unique analytical challenges distinct from conventional galaxy clustering. The HI signal is intrinsically weak and orders of magnitude fainter than radio foregrounds, necessitating aggressive component separation techniques (e.g., Principal Component Analysis) that inadvertently remove cosmological signal. Furthermore, single-dish 21 cm experiments exhibit highly anisotropic observational effects, including beam chromaticity, redshift-space distortions, and non-uniform survey geometries. These factors complicate power spectrum estimation, requiring consistent modeling of observational effects, transfer function corrections, and mock simulations that strictly adhere to survey specifications. Existing pipelines often lack the modularity, survey-oriented design, and rigorous consistency required to scale analysis to the large volumes anticipated for the Square Kilometre Array Observatory (SKAO) and its precursor, the MeerKAT Large Area Synoptic Survey (MeerKLASS).

Methodology
The authors present meer21cm, a comprehensive Python package designed as the core analysis pipeline for MeerKLASS observations. The methodology is built on a modular, survey-oriented architecture that prioritizes the definition of survey specifications (sky area, frequency range, telescope beam) over theoretical cubic box simulations.

Key methodological components include:

• Formalism: The pipeline models the HI brightness temperature and galaxy number density fields in 3D k-space. It explicitly incorporates observational effects such as telescope beam convolution, map-making kernels, gridding, and foreground removal operators. The power spectrum estimators for HI auto-power, galaxy auto-power, and HI-galaxy cross-power are derived with specific weight renormalizations to account for mean-centering and foreground cleaning.
• Mock Simulation: The pipeline generates log-normal HI density fields and Poisson-sampled galaxy catalogues within a Cartesian comoving grid. These are projected onto sky coordinates, convolved with a telescope beam, and contaminated with foregrounds (synchrotron, free-free, dust, AME, CMB) and thermal noise.
• Foreground Cleaning and Transfer Functions: A blind foreground cleaning method based on Principal Component Analysis (PCA) is implemented. To correct for the signal loss induced by PCA, the pipeline employs a numerical transfer function calculation. This involves injecting mock signals into the data, applying the cleaning procedure, and computing the ratio of the cleaned to uncleaned power spectra.
• Parameter Inference: The framework supports Bayesian inference using mock-based covariance matrices, incorporating Hartlap and Percival correction factors to account for the finite number of mock realizations.

Key Contributions

1. Modular and Survey-Oriented Design: Unlike traditional codes that start with cubic boxes, meer21cm defines calculations based on survey geometry. It utilizes Python class inheritance to modularize steps from data loading to parameter inference, ensuring that changes in data cuts or cleaning strategies automatically propagate through the entire analysis chain.
2. Comprehensive Toolkit: The package covers the full post-calibration data analysis workflow, including data I/O, foreground cleaning, power spectrum estimation, mock simulation, transfer function correction, and parameter inference. It handles both HI intensity maps and overlapping galaxy catalogues for multi-tracer analysis.
3. Rigorous Validation: The pipeline is validated through end-to-end tests using 512 mock realizations of a MeerKLASS UHF-band survey patch (750 deg², $0.6 < z < 0.8$). The code includes unit tests with ~100% coverage.

Results
The validation tests demonstrate the pipeline's accuracy in recovering power spectra:

• Log-normal Simulation: The simulated log-normal fields match the input theoretical models with fractional differences of $\sim \pm 1\%$ across all scales of interest.
• Observational Effects: When observational effects (beam smoothing, gridding) are included, the recovered power spectra for galaxy auto-power agree with the model within $\sim 1\%$ for $k < 0.2 \, h\text{Mpc}^{-1}$. HI auto-power and cross-power show slightly larger deviations due to beam modeling complexities but remain within $\pm 5\%$ accuracy.
• Transfer Function Corrections: After applying PCA cleaning and numerical transfer function corrections, the recovered HI auto-power agrees with the model within $0.5\sigma$ of the signal variance for $0.05 < k < 0.2 \, h\text{Mpc}^{-1}$. The cross-power recovery shows a systematic offset of $\sim 5\%$ but remains within $1\sigma$, attributed to differing weighting schemes between auto- and cross-correlations.
• Precision: For a survey comprising ten independent patches, the pipeline achieves percent-level accuracy in power spectrum estimation with deviations $\lesssim 0.5\sigma$ between mock and model spectra in the range $k \in [0.02, 0.2] \, h\text{Mpc}^{-1}$.

Significance
The paper positions meer21cm as a critical infrastructure for the current MeerKLASS campaign and a scalable foundation for future SKAO intensity mapping surveys. By ensuring that theoretical modeling strictly adheres to survey specifications and observational effects, the pipeline addresses the systematic complexities inherent in 21 cm data analysis. The authors claim that the package enables robust, consistent, and automated analysis of large-volume datasets, facilitating the transition from single-patch validations to full-survey cosmological constraints. The code is publicly available, well-documented, and designed to be extensible for future developments such as analytical covariance estimation, advanced foreground cleaning methods, and full-shape cosmological parameter inference.