Exoplanet Atmosphere Characterization

Exoplanet Atmosphere Characterisation plays a vital role in understanding chemical compositions, weather patterns and habitability of the exoplanet. Traditional retrieval models take hours per planet so using machine learning for this is a perfect alternative. I have simulated a physically realistic spectral dataset of atmospheres under regimes like equilibrium transmission, disequilibrium transmission, non-isothermal emission, and photochemical emission with instrument specific noise injection. Then two models: disequilibrium-aware GRU model and Variational Autoencoder for Pressure Temperature Profiles are used to predict various atmospheric parameters. Additionally, to detect biosignatures we can use an unsupervised convolutional Autoencoder trained on normal equilibrium spectra to flag anomalous molecular signatures. Finally, I plan on validating all my models on real JWST observational data proving that these models are worthy of real world applications.