Cosmoglobe DR2. VI. Disentangling hot and cold thermal dust emission with Planck HFI

TL;DR

This work tackles the challenge of disentangling thermal dust emission by introducing a four-component, high-resolution model that separates hot and cold dust from Gaia-traced nearby dust and Ha-correlated dust. Hot and cold dust amplitudes are allowed to vary per pixel while the nearby and Ha templates provide fixed spatial morphology with global spectral parameters, calibrated against Planck HFI and FIRAS data. The approach yields a robust decomposition that correlates hot dust with the FIRAS C II 158 μm map and cold dust with HI, capturing over 98% of the dust variance across frequencies and offering a more efficient, physically motivated alternative to traditional three-parameter MBB fits. This model underpins the Cosmoglobe DR2 reanalysis of COBE-DIRBE data, enabling improved foreground mitigation for CMB studies and related infrared analyses.

Abstract

We present a four-component high-resolution model of thermal dust emission for microwave and sub-mm frequencies derived from Planck HFI, WHAM and Gaia. The resulting high-resolution model derived here forms the basis for the thermal dust model employed in the Cosmoglobe DR2 reanalysis of COBE-DIRBE. The four dust components are called ``cold dust'', ``hot dust'', ``nearby dust'', and ``Ha correlated dust'', respectively, and trace different physical environments. The spatial distributions of the nearby dust and Ha dust components are defined by the Edenhofer et al. Gaia 3D extinction model and the WHAM survey, respectively, while the hot and cold dust components are fit freely pixel-by-pixel to the Planck HFI data. We use a global parameter grid search coupled to an amplitude map Gibbs sampler to fit this model to Planck HFI data. In agreement with the companion low-resolution analysis, we find that the hot dust component is strongly correlated with the FIRAS Cii map, while the cold dust component is strongly correlated with the HI4PI Hi map. Despite its fewer degrees of freedom per pixel compared to the Planck 2015 legacy dust model, we find that this new model performs competitively in terms of overall residuals, capturing over 98% of the full-sky dust variance for all channels. When fitting a spatially varying 3-parameter MBB model to the new dust model with isotropic SEDs, we find very similar spatial distributions to those of the official Planck analysis, and this new model thus represents an economical decomposition of previously published spatially varying spectral parameter maps. We conclude that this new model represents both a statistically more efficient summary of thermal dust in the microwave and far-infrared regimes and a physically more realistic decomposition of the sky compared to the traditional 3-parameter MBB model. (abridged)

Figures (5)

• Figure 1: All masks used for different stages of the analysis. HFI: Mask used during the HFI Gibbs sampling routine for the monopoles. FIRAS: Mask used during the FIRAS Gibbs sampling routine for the monopoles. $\mathsf{G_{857}}$: Mask used during the 857 GHz gain ($\mathsf{G_{857}}$) Gibbs sampling routine. Grid search: Mask used to compare the $\chi^2$ in the grid search and when calculating correlations and summary statistics.
• Figure 2: Contour plot of relevant statistical tests and $\chi^2$ (in red) for (clockwise from top left) cold dust, hot dust, H$\alpha$ dust and nearby dust. The black line shows the fiducial point used for all parameters and grey grids mark the unphysical regions of the parameter space. Maps of a subset of the grid points can be found in \ref{['app:GridSearchTests']}.
• Figure 3: $\chi^2$ (in red) as a function of nearby (top) and H$\alpha$ correlated (bottom) dust amplitude scale. Green and blue curves show correlations with key companion components. Note that the range of the scale of the $\chi^2$ axes are very small.
• Figure 4: DR2 Cold and hot dust components (top panels) as compared with relevant tracers @series HiHiHi and @series CiiCiiCii (bottom panels).
• Figure 5: Spectral energy density (SED) for the four dust components across the HFI bands. Note that the H$\alpha$ dust is a negative amplitude since it is an extinction component, but the absolute value is plotted for clarity. The lower bound is set by a 70% $f_\mathrm{sky}$ and the upper bound by a 94% $f_\mathrm{sky}$. The grey vertical lines are the locations of the six Planck HFI bands.