Learning Kilometer-Scale Weather Prediction with Global-Regional Alignment
ScaleCast aligns pretrained global weather models with kilometer-scale regional forecasts without retraining per driver.
ScaleCast aligns pretrained global weather forecasts with kilometer-scale regional prediction across mismatched grids. Using ERA5 analyses on a 0.25-degree grid and CERRA reanalysis at 5.5 km, one model supports Pangu-Weather, GraphCast, and HRES without driver-specific retraining. Fine-tuning on HRRR at 3 km shows adaptation to another regional domain. Windstorm cases improve cyclone positioning and core-pressure estimates, and HadISD station comparisons show closer agreement on local temperature and humidity changes.
- One trained model accepts Pangu-Weather, GraphCast, and HRES drivers.
- Experiments use ERA5 at 0.25 degrees and CERRA at 5.5 km.
- HRRR fine-tuning adapts the framework to 3 km resolution.
- Windstorm cases improve cyclone position and core-pressure estimates.
- Organizations
- ECMWF
- AI models
- ScaleCastPangu-WeatherGraphCast
From the original article
1 min read at arxiv.orgKilometer-scale regional weather forecasting is essential for local weather warnings and weather-sensitive decisions. Existing data-driven approaches often rely on numerical forecasts for large-scale guidance or require additional training of global forecasting components.
By Guowen Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.