Skip to main content

Signal Correction to Reveal other Earths

Periodic Reporting for period 1 - SCORE (Signal Correction to Reveal other Earths)

Reporting period: 2020-01-01 to 2021-06-30

Searching for life signatures on other planets is one of the key endeavours of astrophysics, and today,
we are in a unique position to make this possible. The TESS satellite has detected interesting Earth candidates
and PLATO in a few years from now will detect real Earth-twins orbiting bright stars, which will allow follow-up
studies with JWST and ELTs to characterize the atmosphere of those exoplanets.

However, TESS will only measure the radius of the detected Earth-twins, which is not enough to interpret the
spectroscopic features in their atmospheres. The mass is also required, and it can be obtained using the
radial-velocity (RV) technique, which measures the gravitational influence of an exoplanet on its host star.

To measure the mass of the Earth-twins that TESS will detect, the community have built incredible RV
instruments that can reach a RV precision of 0.25 m/s (ESPRESSO commissioning). Such an extreme precision
is required to measure the tiny signature of an Earth-twin, however, this is without considering the perturbing
signals induced by its host star, by Earth’s atmosphere and by instrumental noise. Indeed, we know that these
perturbing signals mask completely the signal induced by an Earth-twin, and now that the RV
instruments have the sensitivity to detect such planets, it is urgent to develop novel methods for
mitigating the different perturbing signals.

Understanding the different perturbing signals is extremely challenging and require incredible data. The PI
have built two telescopes that feed Sun-light into the best RV instruments. The obtained data are of
exceptional quality, and the goal of SCORE is to analyse them, explore novel promising methods for
mitigating the different perturbing signals and find the tiny signatures of Earth and Venus. This will
open the way towards the mass-measurement of Earth-twins, which is essential in the quest for finding
life elsewhere, but also to understand planetary formation and dynamics. SCORE will therefore benefit
the entire exoplanet community.
Techniques performed so far to probe and correct for perturbing stellar activity signal have been based on the radial-velocity (RV) and a few other timeseries.
However, to obtain a precise radial-velocity for a star, one need first to obtain a very high-resolution spectrum (Resolution > 100'000). One of the main ideas
of the SCORE project is to look for stellar signatures directly in the spectral time-series, providing much more information than the RV and a few other timeseries.

Before analysing high-resolution spectra data to search for stellar signatures, ones need to extract at best those spectra from the raw images obtained by high-resolution spectrographs (i.e. HARPS, HARPS-N, ESPRESSO).
Different instrumental effects have to be accounted for, to obtain spectra that are not sensitive to the instrument. The PI of the SCORE project developed new data reduction software for HARPS and HARPS-N
to mitigate at best all the different instrumental signals. Published in Dumusque et al. 2021 (https://ui.adsabs.harvard.edu/abs/2021A%26A...648A.103D/abstract) the new data reduction applied to the HARPS-N
solar RV data show an improvement of up to 20% in RV precision. This new data reduction allows us to obtain more precise high-resolution spectra, which is needed to search for stellar signature.

After mitigating at best the different instrumental signals, we started looking for stellar signals in time series of the high-resolution spectra. We published in Cretignier et al. 2020 and 2021 https://ui.adsabs.harvard.edu/abs/2021arXiv210607301C/abstract
a new method called RASSINE and YARARA to help analysing time series of spectra and look for systematics. We looked at how each spectral line behave as a function of time, if group of spectral lines
(grouped by their physical properties or by their position on the instrument detector) were correlated with stellar signal proxies (activity indexes, moments of the cross-correlation function).
The conclusion of this work is that there is still a lot of instrumental signals that persist and needs to be corrected for. This is done by YARARA, and after mitigating their impact, we start to see
stellar signatures that we can correct for. This work is in its beginning, and we really want to continue in this direction, but we are already able to improve the precision of RV by ~20 % when mitigating
the different signal that we see at the spectral level.

High-resolution spectra contain so much information that it is difficult to explore all the different stellar signature that could exist. We therefore investigated the use of Machine Learning algorithms to help us in the process.
Although ML algorithms work like black boxes, if something interesting comes out of them, we can then try to track down the origin of the behaviour. We therefore tried to use Principal Component Analysis at each spectral line
level, or at the spectrum level itself, but without a lot of success so far. We also investigate the power of the SCALPEL method (Collier-Cameron et al. 2021, https://ui.adsabs.harvard.edu/abs/2021MNRAS.505.1699C/abstract)
in mitigating stellar signal. Although the original paper presents a revolutionising technique, preliminary tests on the HARPS-N solar dataset are not so convincing. The outcome of this research for the moment is that we needed
a realistic simulation of stellar signal at the spectrum level. Postdoc Yinan Zhao started from an existing code (SOAP 2.0 Dumusque et al. 2014, https://ui.adsabs.harvard.edu/abs/2014ApJ...796..132D/abstract) and recoded it
to be able to use GPUs, which allows a gain in speed of two magnitudes. Generating 100 spectra affected by stellar signal now takes dozens of seconds. This new code, that will soon be published in open-source, will allow us to
test further ML algorithms on data that we fully understand (only affected by stellar signals and no other perturbing signals).

Even-though the solar RV data obtained from HARPS-N and HARPS can still be improved, with tools like YARARA (see above), PhD student Khaled Al Moulla started to compare the data obtained by the two instruments. He first cleaned
the data from all known systematics (e.g. https://ui.adsabs.harvard.edu/abs/2019MNRAS.487.1082C/abstract). Analysing the cleaned data in the Fourier space allows us to get an unprecedented view on the different types of stellar
signal seen on the Sun. It is possible to estimate precisely the timescale of the different components, and this could help in the future to mitigate these stellar signals using different analysis techniques. This study also put into light
the instrumental limitation of the obtained RVs. Showing that HARPS-N and HARPS are limiting the RV precision is a very strong argument to push for a solar telescope on the more precise ESPRESSO spectrograph. A paper is foreseen
on this specific subject.
The state of the art in the field of stellar signal mitigation in RV consist in going away for the RV time series and looking for extra information in the high-resolution
spectra available behind each RV measurement. In this sense, the SCORE project is really focusing on the state of the art, and is trying to push it further by investigating
new stellar signal proxies and developing clever ways of analysing spectral time series to disentangle stellar and planetary signals.
Spectral cleaning using YARARA