SciPost Submission Page
maria goes NIFTy: Gaussian Process-Based Reconstruction and Denoising of Simulated (Sub-)Millimetre Single-Dish Telescope Data
by Jonas Würzinger, Joshiwa van Marrewijk, Thomas W. Morris, Richard Fuchs, Tony Mroczkowski, Lukas Heinrich
Submission summary
| Authors (as registered SciPost users): | Jonas Würzinger |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.01600v1 (pdf) |
| Code repository: | https://github.com/jwuerzinger/CMB_denoising/tree/main |
| Date submitted: | Sept. 16, 2025, 11:19 a.m. |
| Submitted by: | Jonas Würzinger |
| Submitted to: | SciPost Physics Proceedings |
| Proceedings issue: | The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025) |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approach: | Computational |
Abstract
(Sub-)millimetre single-dish telescopes feature faster mapping speeds and access larger spatial scales than their interferometric counterparts. However, atmospheric fluctuations tend to dominate their signals and complicate recovery of the astronomical sky. Here we develop a framework for Gaussian process-based sky reconstruction and separation of the atmospheric emission from the astronomical signal based on Numerical Information Field Theory (\texttt{NIFTy}). To validate this novel approach, we use the \textit{maria} software to generate synthetic time-ordered observational data mimicking the MUSTANG-2 bolometric array. This approach leads to significantly improved sky reconstructions versus traditional methods.
Current status:
Awaiting resubmission
Reports on this Submission
Strengths
- The scientific content appears sound, and the methodology is clearly explained.
- The comparison with traditional mapmaking provides a clear benchmark for the success of their technique.
Weaknesses
- Mild phrasing/clarification improvements as requested below
Report
The authors present a framework for reconstructing sky maps from single-dish (sub-)millimetre telescope data. By combining the maria simulator with the NIFTy Bayesian inference framework, they demonstrate a method to reconstruct atmospheric emission and astronomical signals. The approach is validated against synthetic data mimicking MUSTANG-2 observations, showing a significant reduction in residuals compared to a standard Maximum Likelihood method (minkasi here).
I believe that the contribution is well-suited for these proceedings. The scientific content appears sound, and the methodology is clearly explained. The comparison with traditional mapmaking provides a clear benchmark for the success of their technique. I recommend publication after the authors address a few very minor presentation points listed below.
I believe that the contribution is well-suited for these proceedings. The scientific content appears sound, and the methodology is clearly explained. The comparison with traditional mapmaking provides a clear benchmark for the success of their technique. I recommend publication after the authors address a few very minor presentation points listed below.
Requested changes
- Typo on Page 2: In the second paragraph of the intro there is a repeated word: "...generates location-specific weather including including turbulence..."
- "extended faint constant astronomical signals" should have some commas separating the descriptors
- Figure 1: It would possibly be nice for the reader if the true map was shown for comparison here as well as Fig. 2. It would also be nice to have a panel showing the mean reconstruction and variance estimates as described in the text below the figure.
- Performance above 1 Hz: The authors comment that the noise component (n) is not included in the NIFTy model, and that the map-making degrades above 1 Hz where it is the dominant contribution. Is it possible to add a very short comment on how feasible it would be to include this component?
- Discussion around Eqs. (1 - 3): This is the only portion of the proceedings where I would recommend mild changes to presentation given the wide readership. In particular, I would recommend additional clarity on the following: i) the definition and statistics of n - what is this modelling? How is it distributed? ii) are there references for the beam pattern chosen? iii) Why is a 2-dimensional/1-dimensional CFM appropriate (again, brief comments)? iv) What exactly are the model parameters in \xi? This is the point that I think would be most useful to clarify (e.g. where has the time dependence gone?) v) Clarification on Eq. (3) - I am interpreting this as the p.d.f of a multivariate normal distribution for the noise n = [d - R(s,t) - a(t)] with covariance matrix N. Is this correct?
- "allows to" -> allows us to
- References: I believe Ref. [4] is now published, just from an inspire search, can this be updated?
Recommendation
Publish (easily meets expectations and criteria for this Journal; among top 50%)