Energy-dependent gamma-ray morphology estimation tool in Gammapy

TL;DR

This paper introduces a new tool within Gammapy for quantifying energy-dependent gamma-ray source morphology, using a full forward-folding approach and likelihood comparison, demonstrated on real and simulated data.

Contribution

A novel energy-dependent morphology estimation tool in Gammapy that employs a 3D likelihood approach and hypothesis testing for gamma-ray sources.

Findings

Significant variability detected in real H.E.S.S. data (9.8σ).

The tool successfully identifies energy-dependent morphology in simulated CTA data (9.7σ).

Applicable to sources of various sizes, demonstrating broad utility.

Abstract

An understanding of the energy dependence of gamma-ray sources can yield important information on the underlying emission mechanisms. However, despite the detection of energy-dependent morphologies in many TeV sources, we lack a proper quantification of such measurements. We introduce an estimation tool within the Gammapy landscape, an open-source Python package for the analysis of gamma-ray data, for quantifying the energy-dependent morphology of a gamma-ray source. The proposed method fits the spatial morphology in a global fit across all energy slices (null hypothesis) and compares this to separate fits for each energy slice (alternative hypothesis). These are modelled using forward-folding methods, and the significance of the variability is quantified by comparing the test statistics of the two hypotheses. We present a general tool for probing changes in the spatial morphology with energy, employing a full forward-folding approach with a 3D likelihood. We present its usage on a real dataset from H.E.S.S. and on a simulated dataset to quantify the significance of the energy dependence for sources of different sizes. In the first example, which utilises a subset of data from HESSJ1825-137, we observe extended emission at lower energies that becomes more compact at higher energies. The tool indicates a very significant variability (9.8{\sigma}) in the case of the largely extended emission. In the second example, a source with a smaller extent (~0.1{\deg}), simulated using the CTAO response, shows the tool can still provide a statistically significant variation (9.7{\sigma}) on small scales.