A blind method to detrend instrumental systematics in exoplanetary light-curves

TL;DR

This paper evaluates a non-parametric, blind ICA-based method for removing instrumental systematics from exoplanet light-curves, demonstrating its effectiveness and objectivity over traditional parametric approaches using simulated data.

Contribution

It investigates the limits and advantages of pixel-ICA for detrending exoplanet light-curves, showing it performs as well or better than existing methods without prior instrument modeling.

Findings

pixel-ICA performs as well or better than other methods on simulated data

pixel-ICA offers higher objectivity due to its statistical foundation

current infrared space instruments can achieve photometric precision of 10^-4

Abstract

The study of the atmospheres of transiting exoplanets requires a photometric precision, and repeatability, of one part in $\sim 10^4$. This is beyond the original calibration plans of current observatories, hence the necessity to disentangle the instrumental systematics from the astrophysical signals in raw datasets. Most methods used in the literature are based on an approximate instrument model. The choice of parameters of the model and their functional forms can sometimes be subjective, causing controversies in the literature. Recently, Morello et al. (2014, 2015) have developed a non-parametric detrending method that gave coherent and repeatable results when applied to Spitzer/IRAC datasets that were debated in the literature. Said method is based on Independent Component Analysis (ICA) of individual pixel time-series, hereafter "pixel-ICA". The main purpose of this paper is to investigate the limits and advantages of pixel-ICA on a series of simulated datasets with different instrument properties, and a range of jitter timescales and shapes, non-stationarity, sudden change points, etc. The performances of pixel-ICA are compared against the ones of other methods, in particular polynomial centroid division (PCD), and pixel-level decorrelation (PLD) method (Deming et al. 2014). We find that in simulated cases pixel-ICA performs as well or better than other methods, and it also guarantees a higher degree of objectivity, because of its purely statistical foundation with no prior information on the instrument systematics. The results of this paper, together with previous analyses of Spitzer/IRAC datasets, suggest that photometric precision and repeatability of one part in $10^4$ can be achieved with current infrared space instruments.