Finding flares in Kepler data using machine-learning tools

Abstract

Listen

Translate article

Context. Archives of long photometric surveys, such as the Kepler database, are a great basis for studying flares. However, identifying the flares is a complex task; it is easily done in the case of single-target observations by visual inspection, but is nearly impossible for several year-long time series for several thousand targets. Although automated methods for this task exist, several problems are difficult (or impossible) to overcome with traditional fitting and analysis approaches. Aims. We introduce a code for identifying and analyzing flares based on machine-learning methods, which are intrinsically adept at handling such data sets. Methods. We used the RANSAC (RANdom SAmple Consensus) algorithm to model light curves, as it yields robust fits even in the case of several outliers, such as flares. The light curves were divided into search windows, approximately on the order of the stellar rotation period. This search window was shifted over the data set, and a voting system was used to keep false positives to a minimum: only those flare candidate points were kept that were identified as a flare in several windows. Results. The code was tested on short-cadence K2 observations of TRAPPIST-1 and on long-cadence Kepler data of KIC 1722506. The detected flare events and flare energies are consistent with earlier results from manual inspections.