Abstract

De-Trending Time Series Data for Variability Surveys

Dae-Won Kim (Harvard-Smithsonian Center for Astrophysics)

Pavlos Protopapas(1, 2), Charles Alcock(1), Yong-Ik Byun(3) 1 : Harvard-Smithsonian Center for Astrophysics (CfA), USA 2 : Initiative in Innovative Computing (IIC), Harvard Univ., USA 3 : Yonsei Univ., Seoul, Korea

We present an algorithm for the removal of trends in time series data. Trends in time series could be caused by various systematic and random noise sources such as cloud passages, changes of airmass or CCD noise. Those trends dilute the intrinsic signals of stars and thus have to be removed in order to recover the intrinsic signals. We determine the trends from subsets of stars (template stars) that are highly correlated among themselves. These subsets of stars were selected based on a hierarchical tree clustering algorithm. A bottom-up merging algorithm based on the normality test is developed to identify the subsets of template stars which we call clusters. After identification of clusters, we determine the trends per cluster by weighted sum of normalized light curves. We then use a multiple linear regression method to de-trend all individual light curves based on these determined trends. Experimental results with synthetic light curves which contain artificial trends and events are presented. Developed algorithm can be applied to not only astronomical time series but also all kind of time series which show trends. We'll applied our algorithm to the whole light curves set in Time Series Center at IIC as well.

Mode of presentation: poster