HDMClouds: A hierarchical decomposition of molecular clouds based on Gaussian mixtures

The identification and characterization of independent entities within molecular clouds is a key challenge for astronomical data analysis. The ever-increasing volume, resolution and sensitivity of observations demand automatic routines to identify and deblend candidate entities to be analysed. Additionally, the intrinsically hierarchical nature of molecular gas distributions demands an automatic identification of the nesting relations between these entities. We propose a novel approach for decomposing molecular clouds in two steps: first we fit the data to a Gaussian mixture with many components, then reconstruct the cloud using a hierarchical model using a Gaussian-mixture reduction algorithm. We use a continuous-space representation, because it is well suited for disentangling coupled entities of emission compared with pixel-based ones, and build a tree structure to represent the hierarchical connections between mixture components. This allows us to select different groups of components in the tree without additional computational effort, including overlapping substructures. We assess our proposal quantitatively and qualitatively using data from the Atacama Large Millimeter Array (ALMA) science verification archive, as well as synthetic data. We also compare the results from some state-of-the-art clump identification algorithms. The experiments and comparisons show that our approach is an effective way to inspect and represent the hierarchical structure of molecular clouds.