SEAWRD

Surrogate Emulator for Aquatic World Radius Determination - "sea-ward"

Surrogate model creator for predicting the radius of irradiated ocean worlds. For installation instructions, tutorials, and detailed documentation, start here.

Motivation

Planetary interior modelling for ocean worlds is a computationally demanding exercise, involving a lot of hydrodynamical considerations dependent on the composition and physical properties of a given exoplanet. This can take a number of minutes per-planet, which grows to be incredibly large when performing hundreds of thousands of simulations.

A cheap approximation is available in the form of surrogate models. A neural network can act as a general function learner, i.e., something that maps inputs to outputs, and so we can use pre-ran expensive simulation data to train a small neural network to reproduce the simulation's results with great accuracy in a fraction of the time.

This is what Surrogate Emulator for Aquatic World Radius Determination is for! Based on user-provided hyperparameters and data, it can train an appropriate surrogate model to be used as an approximation for the full hydrodynamical simulations, namely as a predictor of the radius of the planet.

Contributors

SEAWRD as an open-source project was first pursued as a part of Code/Astro Workshop 2026 by Group 13, with members:

The code, concepts, configuration, and set-up for the start of SEAWRD came from the following contributors, who are therefore equal authors of this project:

Attribution

Please cite the DOI if you make use of this software in your research.

Acknowledgements

The data source file "DNN_data_IOP_Aguichine2021.dat" used in the example usage Jupyter notebook is from Aguichine et al. (2021), and you are encouraged to read their paper found here:

A. Aguichine, O. Mousis, M. Deleuil, and E. Marcq, “Mass–Radius relationships for irradiated ocean planets,” The Astrophysical Journal, vol. 914, no. 2, p. 84, Jun. 2021, doi: 10.3847/1538-4357/abfa99.