ThousandWorlds Benchmark Accelerates Exoplanet Climate Emulation with ML.

The quest for extraterrestrial life relies heavily on interpreting atmospheric signatures of exoplanets, which necessitates a deep understanding of their climates. Traditional Global Climate Models (GCMs) are essential for this, but their computational demands are immense, often requiring millions of core-hours and extensive expert involvement. To address this bottleneck, the ThousandWorlds benchmark has been introduced. This new, machine learning-ready dataset is specifically designed for exoclimate emulation and for broader applications in low-data, multi-simulator, parameter-to-field regression. It compiles around 1800 simulations from five different GCMs. The dataset maps eight key planetary parameters to detailed 3D atmospheric fields, including temperature, humidity, winds, clouds, and radiation. ThousandWorlds offers three progressively challenging subsets and two evaluation protocols, allowing for robust testing of machine learning methods. Initial evaluations show that Gaussian Process-based methods currently outperform off-the-shelf deep learning, indicating a unique challenge regime for ML.