ThousandWorlds: Global Climate Model Simulations for Rocky Exoplanets

Approximately 1,700 global climate model simulation outputs for rocky exoplanets in or near the habitable zone, described in the ThousandWorlds benchmark paper. The dataset was authored by Edward Stevenson and last updated on 2026-04-28. It includes three predefined splits of increasing complexity and two evaluation protocols for machine learning model comparison.

Use Cases

• Train machine learning emulators for climate models based on the 8 continuous planet properties and GCM label.
• Benchmark model performance against inter-GCM error using the provided 'shared-planets' evaluation protocol.
• Analyze atmospheric patterns on exoplanets using the gridded fields of temperature, humidity, winds, clouds, and radiative fluxes.
• Test spectral methods on climate data using the provided spherical harmonic coefficients and inverse-SHT weights.

Strengths

• Contains approximately 1,700 simulation outputs, providing a substantial corpus for analysis.
• Includes outputs from 5 different global climate models (GCMs), allowing for model intercomparison.
• Offers three predefined dataset splits of increasing complexity, facilitating structured evaluation.
• Provides gridded atmospheric variables on a 32×64×10 latitude-longitude-pressure grid for detailed spatial analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Some simulations in the full dataset contain missing fields represented as NaNs, requiring handling.

Provenance

Source

Harvard Dataverse

Collection Method

Outputs from global climate model (GCM) simulations.

Freshness

Last updated 2026-04-28 14:55:47; freshness should be verified.