2025ApJ...994...76L

We present GDCluster, a fully automated algorithm for decomposing a spectral-line data cube of interstellar gas into coherent structures. Assuming a multi-Gaussian nature of observed spectra, GDCluster employs and augments the derivative spectroscopy technique for precise parameter estimation, incorporates spatial-continuity constraints during spectral fitting, and extends these constraints to spatial clustering. This approach effectively resolves velocity blending structures in position-position-velocity (PPV) space-which is particularly critical for ubiquitous H I spectra where emissions from multiple phases are severely blended. Applied to the all-sky HI4PI data, a 10° × 10° CRAFTS survey region, and a 45° × 10° MWISP survey region, GDCluster extracts 45,299, 2247, and 47,119 structures in H I and CO (1-0), respectively. Comparative analyses demonstrate GDCluster's superiority over DBSCAN for separating overlapping spectra with complex velocity components.