Reproduce “WeatherDreamer: Learning a World Model from Multimodal Earth 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
Kate Duffy, Elise Penn, Tristan Ballard, August Posch, Thomas J. Vandal
WeatherDreamer learns a recurrent assimilation cycle from observations across 18 observing systems, without reanalysis training or prescribed sensor-error models. It carries a learned state between hourly steps and updates it with observations. For 10 m wind at 534 withheld stations, its 1-to-12-hour CRPS was 1.09 m s⁻¹, below GEFS and IFS ENS; results vary, and upper-air departures remain worse than GSI.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.