Generative Data Assimilation of Sparse Weather Station Observations at Kilometer Scales

Journal article

Plain Language Summary: Weather forecasts rely on our knowledge of the full state of the atmosphere in the present. Weather stations provide measurements at some (sparse) locations. The atmospheric variables need to be filled in with models that transform the point measurements to a full state (a map). Usually, such models are numerical weather models encoding the physical laws of the atmosphere. They are expensive and slow to run, limiting real‐time updates to forecasts. However, recently the Machine Learning (ML) community has presented great advances in similar tasks like filling in missing parts of photographs, and even generating entire videos from a few words. This motivates our use of an ML model trained on km‐scale weather data and guided by sparse point measurements to fill in wind and rain maps on the same scale (3 km) as state‐of‐the‐art conventional models.

Authors: Manshausen, P; Cohen, Y; Harrington, P; Pathak, J; Pritchard, M

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Wiley
• Journal: Journal of Advances in Modeling Earth Systems
• Volume: 17
• Issue: 10
• Article number: e2024MS004505
• Publication date: 2025-10-28
• Acceptance date: 2025-09-16
• DOI: 10.1029/2024ms004505
• EISSN: 1942-2466
• ISSN: 1942-2466
• Language: English
• Keywords: generative AI; atmospheric state estimation; data assimilation; diffusion model; weather stations