Astrophysical facilities now routinely observe molecular gas clouds over huge areas of the sky, in some cases creating maps with millions of pixels, each with hundreds to thousands of independent velocity measurements. Analysing such vast datasets is a major technological challenge and key barrier to further understanding ISM dynamics. Well-established techniques such as moment analysis and position-velocity diagrams are simple and quick to implement. However, they either rely on averaging in the spectral domain or integrating over spatial dimensions. As a result, these techniques do not provide a full description of the gas kinematics.
The ScousePy package provides a method by which complex spectroscopic data can be fitted in a systematic way. ScousePy decomposes each spectrum into a set of modelled emission features: a technique known as spectral decomposition. This method overcomes the previous limitations because it quickly and efficiently yields a description of all prominent emission features observed in spectroscopic data.
For more information, including tutorials on how to use ScousePy, head over to here.
All tutorial data can be found here.
Agglomerative Clustering for ORganising Nested Structures
Reference: Henshaw et al. 2019
acorns is an n-dimensional unsupervised machine-learning algorithm designed for the clustering of spectroscopic position-position-velocity data. acorns is based on a technique known as hierarchical agglomerative clustering, the primary function of which is to generate a hierarchical system of clusters within discrete data. HAC methods fall into two main categories: ‘bottom-up’ or ‘top-down’. acorns follows the bottom-up approach in that each singleton data point begins its life as a ‘cluster’. Traditionally, clusters merge until only a single cluster remains (containing all data). The output of this technique is often visualized graphically as a dendrogram. Although acorns was designed with the analysis of discrete spectroscopic PPV data in mind (rather than uniformly spaced data cubes), clustering can be performed in n-dimensions, and the algorithm can be readily applied using information in addition to PPV measurements.
Additional Tools
A collection of useful, and some less useful WIP, astronomy modules can be found under my astrojh repo on GitHub.