EarthGAN: Can we visualize the Earth's mantle convection using a surrogate model?

TL;DR

This paper explores using a GAN-based surrogate model to visualize Earth's mantle convection data, making complex scientific simulations more accessible on standard hardware.

Contribution

It introduces a novel GAN approach to generate visualizations of mantle convection data, enabling easier analysis without high-performance computing.

Findings

Surrogate model can generate useful mantle convection visualizations

Preliminary results show promising similarity to ground-truth data

Code is publicly available for reproducibility

Abstract

Scientific simulations are often used to gain insight into foundational questions. However, many potentially useful simulation results are difficult to visualize without powerful computers. In this research, we seek to build a surrogate model, using a generative adversarial network, to allow for the visualization of the Earth's Mantle Convection data set on readily accessible hardware. We present our preliminary method and results, and all code is made publicly available. The preliminary results show that a surrogate model of the Earth's Mantle Convection data set can generate useful results. A comparison to the "ground-truth" is provided.