Estudio de los efectos sistem\'aticos de SOPHIE+ con algoritmos de aprendizaje autom\'atico

TL;DR

This paper introduces a machine learning-based method to correct instrumental drift in the SOPHIE+ spectrograph, improving precision in radial velocity measurements for exoplanet searches without relying solely on standard star monitoring.

Contribution

It presents a novel supervised machine learning approach to predict and correct the zero point drift of SOPHIE+ using environmental and instrumental data.

Findings

Achieved 1.47 m s$^{-1}$ prediction precision for instrumental drift.

Developed a dataset with 645 observations and over 120 features.

Potential to reduce dependence on standard star monitoring for drift correction.

Abstract

SOPHIE+ is a echelle spectrograph located in Haute-Provence Observatory, France. It can reach a precision of near 1 m s$^{-1}$ by simultaneus calibration. However, the zero point shows a low frequency drift of a few m s$^{-1}$ that must be corrected to achieve the needed precision for the current exoplanet search programs. To this end, four radial velocity standard stars are monitored regularly to measure the instrumental drift. In this work, we propose a new way to correct the instrumental drift of instruments like SOPHIE+. We use supervised machine learning techniques to predict the zero point drift with environmental, instrumental and observational features as input. A dataset with 645 observations and more than 120 features was built. We explored various algorithms and achieved a precision of 1.47 m s$^{-1}$ precision on the predictions of the instrumental drift. These techniques have the potential of allowing a method of correction without the need of monitoring standard stars and also can give us knowledge about the instrument that could be used to improve its stability and precision.