Improving TESS satellite measurements of exoplanet WASP19b

Summer 2019. Used TESS data repository, Python (notably emcee Markov-Chain Monte Carlo Python package)

Under the supervision of Dr. Björn Benneke at the University of Montreal's Institution for Research on Exoplanets, I improved the transit and eclipse analysis of WASP19b, a hot Jupiter orbiting around a star 815 lightyears away. This exoplanet offers a wonderful test case to develop new lightcurve analysis, as its very short orbital period ensures frequent transits/eclipses.

A transit occurs when the exoplanet crosses in front of the host star, temporarily blocking its light. This causes a large drop in the light we measure coming from this star. An eclipse occurs when the exoplanet crosses behind its star. An eclipse causes a small drop in the star's lightcurve.

WASP19b's host star's lightcurve. In red: detected transits. Data taken by MIT's TESS satellite.
The same lightcurve, but de-trended. Notice that the transits still vary in depth.

The traditional way to measure transits and eclipses is to process the entire lightcurve at once using Markov Chain Monte-Carlo walkers. Because of the sheer amount of data, a variety of characteristics can be measured (size, weight, etc.). These contribute to a single joint measurement for all transits, and a single joint measurement for all eclipses. However, as we can see in the de-trended lightcurve, each event is different.

Results of the joint fit for transit and eclipse depths.

It was unclear to what extent these joint measurements were accurate

To check their accuracy, I had to measure individual transit/eclipse depths. Unfortunately, there is not enough data in the these individual portions of the lightcurve to inform the MCMC walkers. To solve this, I used the joint fit's MCMC chains' covariance matrix to inform my individual fit's initial guess. That way, my walkers had enough information to measure individual transit/eclipse depth.

Visual depiction of the covariance (dependency) between the lightcurve's 8 parameters. Calculated using the joint fit's MCMC chains.
Checking that the parametric relationships given by the joint fit's covariance matrix match satellite data.

Using this informed prior distribution, I was able to individually fit all 29 transits and 29 eclipses. Happily, their weighted means were consistent with joint fit results. They also gave us new information about the exoplanet's reflectivity and orbit shape that was lost in the joint fit.

Results of the individual transit fits compared to the joint fit.

Such data analysis tricks are really relevant when working with smaller, cheaper satellites such as TESS. Developing tools to compensate for their noisiness is key, as these satellites are continuously sending enormous amounts of freely available data.