Mixture models to account for outliers and instrument systematics in time series.

DÍAZ, R. F.; KUPERMAN, M.; ALMENARA, J. M.; SÉGRANSAN, D.; UDRY, S.

Malargüe

Congreso; 60ª Reunión Anual de la Asociación Argentina de Astronomía; 2017

A common assumption when analysing radial velocities time series for exoplanet detection and characterisation is that the velocity uncertainties are independent and normally distributed. Recent years have seen the evolution of this model towards a model including covariance between the data points. This was shown to improve the description of the data affected by stellar activity signatures. In real life, however, time series are often contaminated by other sources of ?noise? as well (human errors at the telescope, instrument systematics, etc.), which appear as outliers from the assumed sampling distribution. Identifying these data points by eye is usually tricky and specially so when the amplitude of the sought-for signals are at the noise level. Here, we show that mixture models can be used to account for outliers and instrument systematics in radial velocity data sets. This represents a step forward in model realism and is therefore supposed to produce more robust results and planet detections. At the same time, mixture models have the potential to teach us about the instrument and the observational procedure. We first perform simulations to better understand the effect of mixture models on the inferred planet parameters and then apply them on real HARPS data.