Researchers have developed WeatherDreamer, a novel world model that learns weather forecasting directly from multimodal Earth observations without relying on traditional data assimilation systems. This model can infer a latent state, carry it forward with learned dynamics, and correct it with new observations, even predicting measurements from systems it wasn't trained on. WeatherDreamer's ensembles outperform NOAA's Global Ensemble Forecast System (GEFS) for temperature, dew point, and wind up to 12 hours ahead, and show competitive skill in inferring precipitation and cloud-top height. The system achieves this by learning assimilation and forecasting together, processing an hourly global analysis in approximately 1.5 minutes on a single GPU. AI
IMPACT This research demonstrates a new approach to weather modeling that could lead to more efficient and accurate forecasting systems.
RANK_REASON The cluster contains a research paper detailing a new machine learning model for weather forecasting. [lever_c_demoted from research: ic=1 ai=1.0]
- European Centre for Medium-Range Weather Forecasts
- Global Ensemble Forecast System (GEFS T1534) evaluation for tropical cyclone prediction over the North Indian Ocean
- Gridpoint Statistical Interpolation
- National Oceanic and Atmospheric Administration
- WeatherDreamer
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →