Research
Projects

My project aims to develop machine-learning models that can identify chemical species and atmospheric properties by analyzing the observed spectra from an exoplanet. Studying exoplanet atmospheres helps us understand the atmosphere’s composition, weather, planet formation and the potential for habitability.

This project was done under Machine Learning for Science (ML4Sci) as a Google Summer of Code (GSoC) 24' contributor.

ML4Sci is an open-source organization that brings together modern machine learning techniques and applies them to cutting edge STEM problems and GSoC is a global, online program focused on bringing new contributors into open source software development.

I am working on the development of a flexible planetary atmosphere model which solves the radiative transfer equation to generate thermal emission spectra of exoplanetary atmospheres. The package includes multiple opacity calculation methods – line-by-line, bicubic sampling, and the correlated-k approach. I have also coupled a Bayesian retriever to constrain the parameters by analyzing the observed spectra from various instruments. The different components of the package have been benchmarked and it has been successfully validated against the established radiative transfer codes.

I am currently preparing the documentation for the package and it will soon be available along with the manuscripts currently in preparation. The accompanying manuscript will detail this tool’s mathematical foundations, numerical methods, and potential applications in exoplanetary science. The goal for the next part of my thesis is to implement a radiative-convective thermochemical equilibrium module in the package. I will also explore using the model to study directly imaged planets.

Self-consistent models solves the radiative-convective thermochemical equation to get the PT profiles and chemical abundances. These are used to generate forward models for comparing with the observed data. It is computationally expensive and takes a lot of time to converge, hence making it challenging to run nested sampling using on-the-fly generated RCTE forward models. This led me to explore and develop a grid retrieval algorithm that implements nested sampling by interpolating on a grid of precomputed self-consistent forward models. The grid retrieval method directly interpolates the spectra between the grid points while performing atmospheric on observed data. This requires less computational resources and significantly reduces analysis time while maintaining accuracy on the predicted parameters.