Reproduce “Data driven weather forecasts trained and initialised directly from observations” 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
Anthony McNally, Christian Lessig, Peter Lean, Eulalie Boucher, Mihai Alexe, Ewan Pinnington, Matthew Chantry, Simon Lang and 6 more
Researchers at the European Centre for Medium-Range Weather Forecasts (ECMWF) developed an observation-based forecasting system that directly predicts future weather observations and parameters from past raw observational data. The deep learning model successfully forecasted satellite radiances and inferred surface weather conditions, demonstrating a path to bypass traditional numerical weather prediction components like data assimilation.
Key insights
- A neural network can learn to predict future observations directly from past raw observational data, bypassing the need for physics-based models or data assimilation.
- The transformer-based architecture effectively handles irregularly distributed observational data in space and time by leveraging attention mechanisms.
- The model can infer unobserved or indirectly measured weather parameters, such as surface winds, solely from satellite radiance data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.