Reproduce “FuXi: A cascade machine learning forecasting system for 15-day global weather forecast” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Lei Chen, Xiaohui Zhong, Feng Zhang, Yuan Cheng, Yinghui Xu, Yuan Qi, Hao Li

Researchers at Fudan University developed FuXi, a cascade machine learning system, to provide 15-day global weather forecasts. The system achieved performance comparable to the ECMWF Ensemble Mean and extended skillful forecast lead times for key atmospheric variables.

Key insights

  • The cascade ML architecture effectively mitigates error accumulation, allowing ML models to maintain accuracy for forecasts up to 15 days by optimizing individual models for specific time horizons.
  • Machine learning models can achieve global weather forecast performance comparable to the ECMWF Ensemble Mean, marking a significant advancement beyond deterministic NWP parity.
  • While ML ensembles can achieve comparable CRPS, generating consistent and reliable ensemble spread requires more sophisticated, flow-dependent perturbation strategies.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.