Embed2Scale (Earth Observation & Weather Data Federation with AI Embeddings) is a European research project developing new artificial-intelligence approaches to make the rapidly growing volumes of Earth observation and weather data easier to access, exchange and analyse.

Modern satellite and environmental datasets provide unprecedented information about the Earth system, but their size increasingly limits how efficiently they can be transferred and processed. Embed2Scale addresses this "data gravity" problem using AI-based neural compression and learned representations, or embeddings. These methods can compress complex geospatial datasets into much smaller representations while retaining information needed for downstream scientific and operational applications.

The project is developing AI compressors capable of reducing data volumes by up to 1000 times, enabling faster data exchange, decentralised applications and large-scale environmental analyses. Embed2Scale tests these technologies across four application areas: maritime awareness, aboveground biomass estimation, climate and air-pollution prediction, and crop-stress and early-yield detection.

 

Our Climate Processes Group contributes to the climate and air-pollution application of Embed2Scale, exploring how AI-based representations of Earth observation and weather data can support atmospheric and climate science.

Our work focuses on assessing whether compressed AI embeddings retain the information needed to characterise important properties of the atmosphere and their spatial and temporal variability. By connecting emerging machine-learning approaches with our expertise in aerosols, clouds, atmospheric composition and climate, we help test the scientific value of these new representations for environmental applications.

A particular objective is to explore how information contained in large and heterogeneous geospatial datasets can be represented more efficiently while remaining useful for atmospheric analysis and prediction. This provides a demanding scientific test of the Embed2Scale approach: successful compression needs not only to reduce data volumes but also to preserve the subtle environmental signals required to understand variability in climate and air quality.

The work contributes to the broader goal of making the rapidly expanding volume of Earth observation data more accessible for environmental science, while enabling analysis at spatial and temporal scales that would otherwise be computationally challenging.

Embed2Scale (Earth Observation & Weather Data Federation with AI Embeddings) is funded by the European Union’s Horizon Europe programme under Grant Agreement No. 101131841. The contribution of our Climate Processes Group is funded through the UK Government’s Horizon Europe funding guarantee by Innovate UK, part of UK Research and Innovation, under award No. 10113603.