Clumpiness: Time-domain classification of red-giant evolutionary states

TL;DR

This paper introduces Clumpiness, a new supervised classification method that uses time-series summary statistics and Gaia data to accurately determine the evolutionary states of red-giant stars from various space mission datasets.

Contribution

The paper presents Clumpiness, a novel classification tool that achieves over 91% accuracy in identifying red-giant evolutionary states from short and long-duration space-based observations.

Findings

Achieved ~91% classification accuracy with Kepler data.

Maintained >91% accuracy with short datasets like TESS and K2.

Enabled fast, reliable classification of large stellar datasets.

Abstract

Long, high-quality time-series data provided by previous space-missions such as CoRoT and $\mathit{Kepler}$ have made it possible to derive the evolutionary state of red-giant stars, i.e. whether the stars are hydrogen-shell burning around an inert helium core or helium-core burning, from their individual oscillation modes. We utilise data from the $\mathit{Kepler}$ mission to develop a tool to classify the evolutionary state for the large number of stars being observed in the current era of K2, TESS and for the future PLATO mission. These missions provide new challenges for evolutionary state classification given the large number of stars being observed and the shorter observing duration of the data. We propose a new method, $\mathtt{Clumpiness}$, based upon a supervised classification scheme that uses "summary statistics" of the time series, combined with distance information from the Gaia mission to predict the evolutionary state. Applying this to red giants in the APOKASC catalogue, we obtain a classification accuracy of ~91% for the full 4 years of $\mathit{Kepler}$ data, for those stars that are either only hydrogen-shell burning or also helium-core burning. We also applied the method to shorter $\mathit{Kepler}$ datasets, mimicking CoRoT, K2 and TESS achieving an accuracy >91% even for the 27 day time series. This work paves the way towards fast, reliable classification of vast amounts of relatively short-time-span data with a few, well-engineered features.