This is the first in a planned series of Featured Project pages, which aim to take a
more in-depth view of a current DiRAC project, learning more along the way about the science
questions being addressed and how DiRAC facilities and resources are helping. We’ll also meet
the Principal Investigator and the research team at first hand, to hear a little more about their
motivations and stories.
Exoplanet studies will get a huge boost with the planned launch of the ARIEL mission in 2029, which will study the atmospheres of some 1000 remote worlds in our galactic neighbourhood. Ingo Waldmann’s team based at the Centre for Space Exochemistry Data (CSED) at University College London plans to study the atmospheric chemistry of gaseous planets, to learn about their formation, subsequent development, and even climate. They use DiRAC resources to model the light spectra observed each time the planet’s disk passes in front of its star. By mapping atmospheric models across a range of planetary and stellar parameters, the project will provide a basis for detailed population studies providing a holistic context to the in-depth atmospheric models being developed in preparation for ARIEL’s launch.
Notable results to date include the discovery of
water vapour in the atmosphere of a “super-Earth” K2-18b, 110ly distant in the constellation of Leo, and the identification of the hottest planet known to date – KELT-9b – where atmospheric temperatures in excess of 4000K permit the presence of metallic components such as iron and titanium.
The group’s research is founded on detailed simulation using codes such as TauREx running on the Data Intensive service (Skylake and Wilkes2) at Cambridge, which have to undergo constant improvement in both performance and complexity as the quality and quantity of observational data increases. However, with the advent of ARIEL the anticipated increased data throughput demands have driven the development of artificial intelligence-based inference methods. The researchers reach out to the AI community via an annual ARIEL Machine Learning Data Challenge, born out of the necessity to prepare analysis tools ready for population-level studies. A challenge related to a specific aspect of the mission is issued, and solutions solicited from talents in both industry and academia, regardless of subject discipline, seniority, or geographical location. Last year’s challenge hosted by NeurIPs, one of the most prestigious AI conferences, concerned measuring molecular abundance from Ariel observation, one of the key goals of the space mission. Participants were asked to provide solutions which are fast while remaining faithful to those obtained with conventional techniques.
Dr. Gordon Yip studied Astrophysics at UCL both as undergrad and postgrad.
“Before my PhD I had never heard of Machine Learning, and to be honest I literally needed to google it at that time. Four years down the line, I am now actively using Machine Learning in my research.”
He is currently a postdoctoral researcher at UCL, working with Dr Ingo Waldmann to develop ML techniques specifically for problems within the field of exoplanetary science. He coordinates the ML Working group within the ARIEL Consortium.
“Organising and managing the Data Challenge is almost my daily job now, but I am also involved in a few projects, such as speeding up the chemical network, looking at the interpretability of neural networks and developing alternative approaches to speed up our data analysis.“
Dr. Quentin Changeat joined the Space Telescope Science Institute (STScI) in Baltimore in September 2022 as a European Space Agency Research Fellow, but remains affiliated to UCL in an honorary position. STScI manages both Hubble (HST) and James Webb (JWST) Space telescopes for NASA, so nowadays his research is focussed mostly here; however, he intends to continue his work on European exoplanet missions, in particular ARIEL, inspired by the fundamental questions:
K2-18b image: ESA/Hubble, M. Kornmesser, CC BY 4.0, via Wikimedia Commons