Best of both worlds: Fusing hyperspectral data from two generations of spectro-imagers for X-ray astrophysics

¹ Université Paris-Saclay, Université Paris Cité, CEA, CNRS, AIM, 91191 Gif-sur-Yvette, France
² FSLAC IRL 2009, CNRS/IAC, La Laguna, Tenerife, Spain

^★ Corresponding author; julia.lascar@cea.fr

Received: 22 July 2024
Accepted: 25 November 2024

Abstract

Context. With the recent launch of the X-Ray Imaging and Spectroscopy Mission (XRISM) and the advent of microcalorimeter detectors, X-ray astrophysics is entering in a new era of spatially resolved high-resolution spectroscopy. But while this new generation of X-ray telescopes offers much finer spectral resolution than the previous one (e.g. XMM-Newton, Chandra), these instruments also have coarser spatial resolutions, leading to problematic cross-pixel contamination. This issue is currently a critical limitation for the study of extended sources, such as galaxy clusters or supernova remnants (SNRs).

Aims. To increase the scientific output of XRISM’s hyperspectral data, we propose that it be fused with XMM-Newton data, and seek to obtain a cube with the best spatial and spectral resolution of both generations. This is the aim of hyperspectral fusion. In this article, we explore the potential and limitations of hyperspectral fusion for X-ray astrophysics.

Methods. We implemented an algorithm that jointly deconvolves the spatial response of XRISM and the spectral response of XMM- Newton. To do so, we construct a forward model adapted for instrumental systematic degradations and Poisson noise, and then tackle hyperspectral fusion as a regularized inverse problem. We test three methods of regularization: spectral low-rank approximation with a spatial Sobolev regularization; spectral low-rank approximation with a 2D wavelet sparsity constraint; and a 2D–1D wavelet sparsity constraint.

Results. We test our method on toy models constructed from hydrodynamic simulations of SNRs. We find that our method reconstructs the ground truth well even when the toy model is complex. For the regularization term, we find that while the low-rank approximation works well as a spectral denoiser in models with less spectral variability, it introduces a bias in models with more spectral variability, in which case the 2D-1D wavelet sparsity regularization works best. Following the present proof of concept, we aim to apply this method to real X-ray astrophysical data in the near future.