Use of the WeatherGenerator foundation model for atmospheric reconstruction over Scandinavia

WeatherGenerator is a pan-European initiative that integrates advanced machine learning architectures with high-performance computing to build an open, kilometer-scale foundation model of the coupled Earth system. The project is structured into four thematic areas; the application presented here represents one of the twenty-two developed within Theme 3 by project partners.

The Norwegian Meteorological Institute (MET Norway) plans to use the foundation model to reconstruct several decades of atmospheric variables over Scandinavia, ideally covering the most recent 40 years. The main goal is to apply the WeatherGenerator framework for climate monitoring purposes. The reconstructed datasets comprise near-surface variables as well as atmospheric variables across multiple pressure levels. This approach combines diverse data sources (including in situ observations, reanalysis datasets, and numerical model outputs) by exploiting the foundation model to produce coherent, high-resolution fields for both weather and climate monitoring. A target spatial resolution of a few kms is achieved through data fusion methods and a dedicated tail network trained to generate gridded analyses at this scale.

This contribution aims to present MET Norway’s results from the initial phase of the WeatherGenerator project. The emphasis is therefore on preliminary findings, outlining the methodological framework and highlighting the potential of these approaches for high-resolution climate monitoring.

------------------------------------------

Note: The WeatherGenerator project (grant agreement No101187947) is funded by the European Union. Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the Commission. Neither the European Union nor the granting authority can be held responsible for them.