Gaussian process-based Bayesian imaging of bolometric (sub-)millimetre data

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: We're looking at the paper: "Gaussian process-based Bayesian imaging of bolometric (sub-)millimetre data".

Jocelyn: Here's the gist of the paper. Widefield single-dish telescopes operating at (sub-)mm wavelengths tend to feature much faster mapping speeds than radio interferometers at the same wavelengths, and are more suitable for measuring signals on larger spatial scales<ref:2610.27733#pg2>.

Vera: This paper introduces maria-nifty, a Gaussian process-based Bayesian imaging framework for reconstructing sky maps from single-dish telescope data<ref:2610.27733#pg2>.

Jocelyn: It demonstrates the method's application to real data observed with the MUSTANG-2 bolometric array on the 100-meter Green Bank Telescope for the galaxy cluster RXJ1347.5-1145<ref:2610.27733#pg3>.

Vera: The key finding is that this method cleanly separates individual astrophysical signals, even though they are superposed on the sky<ref:2610.27733#pg9>.

Paper summary: Vera: So, to recap, the paper "Gaussian process-based Bayesian imaging of bolometric (sub-)millimetre data" introduces this Gaussian process based Bayesian imaging framework for reconstructing sky maps from single dish telescope data and its main claim is that it reconstructs those maps better than other methods.

Jocelyn: They’re using a Gaussian process to model the sky as if it has some underlying spatial correlation, and then they use Bayesian inference to figure out what the actual sky map is given the noisy observations.

Vera: That’s right, they are using modular generative models to separate the astronomical signal from the atmospheric emission while providing pixel wise uncertainty quantification for the reconstructed sky maps <ref:2610.27733#pg3>.

Vera: They applied this to real data from the MUSTANG-two bolometric array on the one hundred meter Green Bank Telescope for RXJ1347 point 5-one thousand one hundred forty-five <ref:2610.27733#pg6>. That’s a concrete piece of evidence from real observation.

Vera: The paper also outlines how they built their prior models to separate the astronomical signal from atmospheric contributions and instrumental noise <ref:2610.27733#pg4>.

Jocelyn: That’s the trick, they model the atmosphere using a parametric approach where each detector is modeled as a linear combination of N B-spline basis functions a(t) = Xn j=one ξ aij Bj,k(t) <ref:2610.27733#pg13>.

Subrahmanyan: The prior models are constructed to separate the astronomical signal from atmospheric contributions and instrumental noise <ref:2610.27733#pg4>.

Vera: They assume the astronomical signal is static, and each component s i is described as a two dimensional correlated field s i ∼ G(s i, S) <ref:2610.27733#pg7>.

Jocelyn: And they use a softplus function φ(z) = log1 + e z to enforce positivity for compact sources <ref:2610.27733#pg9>.

Subrahmanyan: The main challenge in reconstructing the sky map from the observed data lies in effectively separating the astronomical signal from atmospheric contributions and instrumental noise <ref:2610.27733#pg3>.

Conclusion: Vera: So we've looked at how they used this Gaussian process thing to clean up those dusty submillimetre observations from single dishes, and now we’re wrapping up what it actually means for the sky mapping community.

Jocelyn: I think the title itself, "Gaussian process based Bayesian imaging of bolometric (sub)millimetre data," is pretty descriptive, but it really boils down to them using a statistical model that assumes some underlying smoothness in the sky—that's the Gaussian process part—and then they use Bayesian inference to figure out what's actually hidden behind all that noise and atmosphere.

Subrahmanyan: For me, the implication is about reliability. They are essentially giving us a way to reconstruct a cleaner sky map by acknowledging that real astronomical signals tend to be smoother than the noisy data we get in a single scan <ref:2610.27733#pg10>.

Vera: Exactly, because they explicitly model both the astronomical signal and the atmospheric emission separately, which means they can tell you not just what's there, but how much of it is likely noise or weather contamination.

Jocelyn: And when you look at the results on that galaxy cluster RXJ1347 point five-one thousand one hundred forty-five they managed to cleanly separate the extended cluster emission from that super compact source at its center using this approach <ref:2610.27733#pg1>.

Subrahmanyan: That’s a good demonstration of how this framework helps disentangle sources when they are piled up together on the sky, which is exactly what we need for understanding structure formation across different scales <ref:2610.27733#pg2>.

Vera: The main thing they show is that this method provides a baseline comparison against traditional maximum likelihood map-making, and it gives us those pixel wise uncertainty values, which lets us know how much faith we can put in any specific part of the final map.

Subrahmanyan: It sets a new standard for how we might handle these widefield single dish datasets when we’re trying to build cosmological models based on galaxy clustering or cluster dynamics <ref:2610.27733#pg2>.

Vera: And while they've shown it works well on this specific cluster, the next step is scaling this up to those much larger datasets that the next generation of telescopes will produce.

Jocelyn: Yeah, so we're still looking at how they might handle the sheer volume of data coming from instruments like AtLAST and how that affects their computational load.

Subrahmanyan: That’s where we need to see if this approach scales up reliably for those much larger datasets <ref:2610.27733#pg2>.