IoW_20230321 - Gaia

Image of the Week

GaiaVari: a citizen science project to help Gaia variability classificaton

Figure 1. Field Guide of the GaiaVari citizen science project with on the top left corner the light curve (brightness change over time), on the top right corner the colour-magnitude diagram indicating the observations of the source. On the bottom left corner the folded light curve for a given period is shown, and next to it, on the bottom right, the position of the object in the Milky Way with respect to the Solar System (both with a face-on view and an edge-on view). Source: GaiaVari Project

In June 2022, the Gaia consortium published the most extensive catalogue of variable sources over the entire sky as part of Gaia Data Release 3 (DR3). Around 10.5 million variable sources, with classifications into 24 different variability types, were published. The variability catalogue was obtained using machine learning methods in a supervised classification scheme based on training datasets compiled from the literature.

Gaia’s variability processing and analysis includes several modules dedicated to different tasks, as described in Eyer et al. (2022). These modules include the computation of statistical parameters, the detection of variability, the characterisation of light curves, the classification of variable objects, and specific object studies (SOS) that confirm or reclassify the identifications from previous stages and refine the description of variability of specific types.

A semi-supervised classification approach was employed, firstly training multi-stage random forest classifiers with sources of known types in the literature, followed by a preliminary classification of the Gaia data. Then in a second training phase, a selection of the first classification results was carried out to improve the representation of some classes, after which the improved classifiers were applied to the Gaia data. Dedicated validation classifiers were used to reduce the level of contamination in the published results.

The large number of sources and the sparse sampling of Gaia time series, however, do not allow for a “perfect” classification of all sources. Validation tests performed on the published data by comparing the results obtained with external catalogues suggest that for a small, but significant number of variables, Gaia classification does not match their classification in other catalogues of variable stars.

Moreover, for a number of variable stars, the estimated main parameters differ from the parameters provided in these external catalogues. In some cases, the reason for the discrepancy is the small number of epochs used to calculate the periods. In others, the reason is unknown.

There are also light curves hampered by the presence of outliers which were not flagged / removed by the automated pipelines leading also to wrong variability classifications. On top of that, some sources may display secular trends, experience outbursts or display anomalous behaviour which may have resulted in misclassifications, as they may have escaped proper characterization by the automated software.

Unfortunately, the scrutiny of individual sources is beyond the scope of the Gaia consortium's activities, and this is why GaiaVari was developed, a citizen science project funded by ESA through the ESAC Space Science Faculty. In collaboration with Gaia Coordination Unit 7 (CU7) team in Geneva, GaiaVari offers the opportunity to conduct a more targeted analysis on individual sources to the worldwide citizen science community. This is expected to complement the classifying efforts of the Gaia consortium.

The GaiaVari project will be deployed within the Zooniverse environment, the biggest platform gathering citizen scientists from all over the world. It will provide revised classifications of thousands of variable stars from Gaia’s data release 3 for selected subsets of different types.

The information provided by the volunteers participating to this classification effort will be used to critically review the mismatches found with the classifications provided in Gaia DR3. Also, it could lead to discover sources showing peculiar variability behaviour and flag them for further follow-up and analysis, some of which may result in science to be published in one or more papers, of which (a subset of) the citizen scientists are expected to be co-authors. By analyzing Gaia data, citizen scientists can help train the classification toward the next Gaia Data Release 4, planned not before the end of 2025, and improve the automated classification process.

In a first wave of classifications, launched on 21 March 2023, citizen scientists will be requested to analyse the information collected by Gaia on a subset of 10,000 variable stars. The first wave is mostly composed of relatively bright sources classified by the automated processing, such as RR Lyrae, Cepheids, Long Period Variable Stars and Eclipsing Binaries, with the addition of a small percentage of other types of variable stars.

Based on the information provided, namely their epoch photometry, folded light curve and derived variability period, the location in the colour-magnitude diagramme and situation in the Galaxy, plus the background information contained in a dedicated tutorial, citizen scientists are asked to pick one category for the class of variability. Their contributions will be compared to machine learning predictions in order to help train and improve the AI algorithms used by the Gaia Consortium.

In the coming weeks or months, subsequent waves containing additional sources, and perhaps new variability classes, will be put in the hands of the citizen scientists for their analysis. Registering on Zooniverse is open to all and can be done by everyone interested in participating in this citizen science project. To join the scientific adventure of GaiaVari, please visit the project’s page, linked below.

IoW_20230321 - Gaia

Image of the Week

GaiaVari: a citizen science project to help Gaia variability classificaton

Access GaiaVari Citizen Science Project here

Image of the Week Archive