Get Started
Home
Topics
Search
Library
Research questionHow can probabilistic weather downscaling recover site-specific conditions from coarse grids despite unresolved terrain and land-surface effects?Coarse weather analyses and forecasts cannot represent the terrain and land-surface properties that systematically shape conditions at individual sites. Downscaling must infer these local departures while retaining probabilistic predictions across locations and times.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.Earth observation embeddings are effective sub-grid descriptors for probabilistic weather downscalingThe evidence concerns a ConvCNP using compressed 10 m Earth observation embeddings alongside approximately 25 km atmospheric fields from ERA5 reanalysis, with additional results using Aurora forecasts. It reports tests across five climatically diverse regions, with stations held out in space and time and experiments involving newly deployed station networks; reported targets include 2 m temperature and 10 m wind speed.research paper · Sep 3, 2026
Related questions
How can AI weather models produce high-resolution global forecasts without relying solely on analysis data?How can we select the best probabilistic weather forecast model for a specific downstream decision?How can we improve weather forecasts two to six weeks ahead despite rapidly degrading skill?How can cloud-cover forecasts extend beyond nowcasting despite imperfect NWP initialization while preserving satellite spatial detail?