Detecting galaxies in a large H{\sc i}~spectral cube

TL;DR

This paper presents a parallel Python pipeline for efficient detection of galaxies in large H I spectral cubes, optimizing source finding parameters and proposing alternative methods for improved robustness.

Contribution

The authors develop an MPI-based Python pipeline that processes large H I spectral data cubes in parallel and analyze the impact of source finder parameters on detection efficiency.

Findings

Detection efficiency varies with SoFiA-2 parameters.

Optimal kernel sizes are identified for different flux bins.

Single kernel choice is insufficient for all galaxy types.

Abstract

The upcoming Square Kilometer Array (SKA) is expected to produce humongous amount of data for undertaking H{\sc i}~science. We have developed an MPI-based {\sc Python} pipeline to deal with the large data efficiently with the present computational resources. Our pipeline divides such large H{\sc i}~21-cm spectral cubes into several small cubelets, and then processes them in parallel using publicly available H{\sc i}~source finder {\sc SoFiA-$2$}. The pipeline also takes care of sources at the boundaries of the cubelets and also filters out false and redundant detections. By comapring with the true source catalog, we find that the detection efficiency depends on the {\sc SoFiA-$2$} parameters such as the smoothing kernel size, linking length and threshold values. We find the optimal kernel size for all flux bins to be between $3$ to $5$ pixels and $7$ to $15$ pixels, respectively in the spatial and frequency directions. Comparing the recovered source parameters with the original values, we find that the output of {\sc SoFiA-$2$} is highly dependent on kernel sizes and a single choice of kernel is not sufficient for all types of H{\sc i}~galaxies. We also propose use of alternative methods to {\sc SoFiA-$2$} which can be used in our pipeline to find sources more robustly.