Galaxy Zoo DECaLS: Volunteer and Machine Morphology Catalogs

Galaxy Zoo DECaLS provides detailed visual morphology classifications for approximately 314,000 galaxies, combining citizen scientist volunteer labels with deep learning model predictions. The dataset includes multiple classification campaigns, with a later GZD-5 campaign using a modified schema to better detect galaxy mergers and weak bars. It contains both raw volunteer responses and automated posteriors modeled as Dirichlet distributions, offering a resource for studying galaxy evolution and benchmarking machine learning methods in astronomy.

Use Cases

• Training supervised machine learning models for galaxy morphology classification using volunteer-provided labels.
• Comparing the performance and biases of human volunteer classifications against automated deep learning predictions.
• Studying the prevalence of specific morphological features like bars, mergers, or spiral arms across a large sample of galaxies.
• Analyzing the consistency of citizen science classifications, especially for galaxies with only a few volunteer responses in the GZD-5 campaign.

Strengths

• Combines human intelligence from over 314,000 galaxy classifications by volunteers with automated deep learning predictions.
• Includes a modified classification schema in the GZD-5 campaign specifically designed to improve detection of mergers and weak bars.
• Provides full posterior distributions (Dirichlet concentrations) for model predictions, not just summary statistics.

Limitations

• Key metadata such as exact row counts, column names, and the last update date are not provided in the available sources.
• The GZD-5 campaign includes many galaxies with only approximately five volunteer responses, which may limit statistical confidence for those entries.
• The availability of the associated galaxy image files is noted as pending integration with the repository platform.

Provenance

Source

Data released with the paper 'Galaxy Zoo DECaLS: Detailed Visual Morphology Measurements from Volunteers and Deep Learning for 314,000 Galaxies' by Mike Walmsley (University of Oxford).

Collection Method

Classifications gathered via the Galaxy Zoo citizen science platform (www.galaxyzoo.org) and derived from a trained deep learning model.

Time Range

Covers the GZD-1, GZD-2, and GZD-5 classification campaigns.

Geography

Observations are from the DECaLS (Dark Energy Camera Legacy Survey) sky survey.