Biases in retrieving planetary signals in the presence of quasi-periodic stellar activity (1908.02217v1)

Abstract: Gaussian process regression is a widespread tool used to mitigate stellar correlated noise in radial velocity time series. It is particularly useful to search for and determine the properties of signals induced by small-size, low-mass planets ($R_p<4R_{\rm \oplus}$, $m_p<10M_{\rm \oplus}$). By using extensive simulations based on a quasi-periodic representation of the stellar activity component, we investigate the ability in retrieving the planetary parameters in 16 different realistic scenarios. We analyse systems composed by one planet and host stars having different levels of activity, focusing on the challenging case represented by low-mass planets, with Doppler semi-amplitudes in the range 1-3 $m s^{-1}$. We consider many different configurations for the quasi-periodic stellar activity component, as well as different combinations of the observing epochs. We use commonly-employed analysis tools to search for and characterize the planetary signals in the datasets. The goal of our injection-recovery statistical analysis is twofold. First, we focus on the problem of planet mass determination. Then, we analyse in a statistical way periodograms obtained with three different algorithms, in order to explore some of their general properties, as the completeness and reliability in retrieving the injected planetary and stellar activity signals with low false alarm probabilities. This work is intended to provide some understanding of the biases introduced in the planet parameters inferred from the analysis of radial velocity time series that contain correlated signals due to stellar activity. It also aims to motivate the use and encourage the improvement of extensive simulations for planning spectroscopic follow-up observations.