Graph-based interpretation of the molecular interstellar medium segmentation

We present a generalization of the giant molecular cloud identification problem based on cluster analysis. The method we designed, SCIMES (Spectral Clustering for Interstellar Molecular Emission Segmentation) considers the dendrogram of emission in the broader framework of graph theory and utilizes spectral clustering to find discrete regions with similar emission properties. For Galactic molecular cloud structures, we show that the characteristic volume and/or integrated CO luminosity are useful criteria to define the clustering, yielding emission structures that closely reproduce `by-eye' identification results. SCIMES performs best on well-resolved, high-resolution data, making it complementary to other available algorithms. Using <sup>12</sup>CO(1-0) data for the Orion-Monoceros complex, we demonstrate that SCIMES provides robust results against changes of the dendrogram-construction parameters, noise realizations and degraded resolution. By comparing SCIMES with other cloud decomposition approaches, we show that our method is able to identify all canonical clouds of the Orion-Monoceros region, avoiding the overdivision within high-resolution survey data that represents a common limitation of several decomposition algorithms. The Orion-Monoceros objects exhibit hierarchies and size-line width relationships typical to the turbulent gas in molecular clouds, although `the Scissors' region deviates from this common description. SCIMES represents a significant step forward in moving away from pixel-based cloud segmentation towards a more physical-oriented approach, where virtually all properties of the ISM can be used for the segmentation of discrete objects.