Published November 9, 2026 | Version 1.0.0

Results dataset for: An AI-Based OSSE Framework to Evaluate Learned Physical Consistency: A Multi-Model Study Using FourCastNet, Aurora, Prithvi WxC, and GraphCast

  • 1. University of Maryland, Baltimore County, Baltimore, MD, USA
  • 2. Earth Sciences Division, NASA Ames Research Center, Moffett Field, CA, USA

Description

Derived results (RMSE time series and tables, regime-stratified (cloud, terrain, latitude) statistics, spatial bias fields, spillover analyses and figures) from truth-insertion Observing System Simulation Experiments with four AI weather models: FourCastNet v2, Aurora, Prithvi WxC and GraphCast. Each model is run for 60 days (240 x 6 h steps); from day 30, selected fields (T, U/V, Z) are replaced with reanalysis truth (ERA5; MERRA-2 for Prithvi WxC) globally or within masks, with and without correlated observation-error perturbations.

Supports the manuscript An AI-Based OSSE Framework to Evaluate Learned Physical Consistency: A Multi-Model Study Using FourCastNet, Aurora, Prithvi WxC, and GraphCast (Diggikar, Halem, and Raman; submitted to Journal of Geophysical Research: Machine Learning and Computation). See README.md for file structure, conventions and units. Raw model forecasts are not included and are available from the corresponding author on request.

Files

Aurora_results.zip

Files (3.0 GB)

Name Size
md5:8ca986aa9a04ba02505a22e7c8605340
1.1 GB Preview Download
md5:c97ea94806a2e4ebd7ffb0c9c5f3387e
398.5 MB Preview Download
md5:c0f3ad0389a13a5cb74b97c95521bd87
1.0 GB Preview Download
md5:a23ba999fc46ec25a8d25e6ef5812ae8
392.0 MB Preview Download
md5:e476cd72e53351f37880a5e35f270376
89.4 MB Preview Download
md5:1098ae39fc2a290bac893d26d04dd0f7
4.3 kB Preview Download

Additional details

Related works

Is supplemented by
Software: 10.5281/zenodo.24092590 (DOI)
Software: 10.5281/zenodo.24092592 (DOI)
Software: 10.5281/zenodo.24092594 (DOI)
Software: 10.5281/zenodo.24092596 (DOI)
Software: 10.5281/zenodo.24092598 (DOI)