Star-galaxy separation strategies for WISE-2MASS all-sky infrared galaxy catalogs (1401.0156v2)

Abstract: We combine photometric information of the WISE and 2MASS all-sky infrared databases, and demonstrate how to produce clean and complete galaxy catalogs for future analyses. Adding 2MASS colors to WISE photometry improves star-galaxy separation efficiency substantially at the expense of loosing a small fraction of the galaxies. We find that 93% of the WISE objects within W1<15.2 mag have a 2MASS match, and that a class of supervised machine learning algorithms, Support Vector Machines (SVM), are efficient classifiers of objects in our multicolor data set. We constructed a training set from the SDSS PhotoObj table with known star-galaxy separation, and determined redshift distribution of our sample from the GAMA spectroscopic survey. Varying the combination of photometric parameters input into our algorithm we show that W1 - J is a simple and effective star-galaxy separator, capable of producing results comparable to the multi-dimensional SVM classification. We present a detailed description of our star-galaxy separation methods, and characterize the robustness of our tools in terms of contamination, completeness, and accuracy. We explore systematics of the full sky WISE-2MASS galaxy map, such as contamination from Moon glow. We show that the homogeneity of the full sky galaxy map is improved by an additional J<16.5 mag flux limit. The all-sky galaxy catalog we present in this paper covers 21,200 sq. degrees with dusty regions masked out, and has an estimated stellar contamination of 1.2% and completeness of 70.1% among 2.4 million galaxies with $z_{med}= 0.14$. WISE-2MASS galaxy maps with well controlled stellar contamination will be useful for spatial statistical analyses, including cross correlations with other cosmological random fields, such as the Cosmic Microwave Background. The same techniques also yield a statistically controlled sample of stars as well.

• The paper introduces a star-galaxy separation strategy utilizing SVM and a straightforward W1-J color criterion that rivals multi-dimensional classifiers.
• The study achieves only 1.2% stellar contamination and a 70.1% completeness rate across a sample of 2.4 million galaxies.
• The methodology enhances future cosmic surveys by improving galaxy catalog purity, thereby refining analyses of large-scale structure and galaxy evolution.

An Analysis of Star-Galaxy Separation Strategies in WISE-2MASS Infrared Galaxy Catalogs

The paper presents an exploration of strategies for distinguishing stars from galaxies within the combined photometric data of the Wide-field Infrared Survey Explorer (WISE) and the Two Micron All Sky Survey (2MASS). The goal of harnessing these extensive infrared datasets is to construct a catalog that is both clean and complete, which will facilitate future astronomical analyses by minimizing stellar contamination in the galaxy selection.

Methodology and Techniques

Central to the paper is the application of machine learning, particularly the Support Vector Machine (SVM), for classifying objects based on photometric data. The authors leveraged the rich multi-color data available from the WISE and 2MASS surveys, applying these machine learning tools to effectively distinguish between stars and galaxies. They found that about 93% of WISE objects within the magnitude limit of W1 < 15.2 mag have a corresponding 2MASS match, which provides a substantial basis for high-confidence classifications.

One of the pivotal findings is the identification of W1-WISE − J-2MASS as a potent and straightforward parameter for star-galaxy separation. This simple color-based criterion achieved comparable performance to the multi-dimensional SVM classifier, underscoring the potential for computational efficiency without a significant trade-off in accuracy.

Results and Implications

The paper highlights several key outcomes:

• The use of W1-J color as a star-galaxy separator yielded a clean catalog with an estimated stellar contamination of only 1.2% and a completeness rate of 70.1% for a substantial sample of 2.4 million galaxies.
• The SVM algorithm demonstrated an impressive classification efficacy, achieving high levels of completeness and reduced contamination.
• Practical limitations, such as contamination from the Moon's glow, were effectively handled through empirical strategies, resulting in improvements in the homogeneity of the galaxy map.

The paper signifies a step forward in refining galaxy catalogs for large-scale structure analyses, particularly in cross-correlation studies with other cosmological datasets like the Cosmic Microwave Background. The significance of such advancements lies in their potential to bolster the accuracy of cosmological conclusions drawn from these huge datasets.

Future Directions

The implications of this work extend to various applications in astrophysics and cosmology, including enhancing the precision of galaxy evolution studies and improving the constraints on cosmic density fields. In the future, these techniques could be further refined with additional datasets to probe even deeper into the universe, potentially incorporating new machine learning models or integrating supplementary catalogs.

This paper underscores the ongoing evolution of classification strategies in astronomical surveys, demonstrating the critical role of machine learning in managing and interpreting the burgeoning volumes of data characteristic of contemporary cosmic surveys. The methodologies and findings detailed in the paper lay foundational work for future research aiming to optimize the balance between catalog completeness and contamination mitigation, vital for advancing our understanding of the cosmic fabric.