Earth Observation & Weather Data Federation with AI Embeddings

Project Information

Embed2Scale

Grant agreement ID: 101131841

DOI

10.3030/101131841

EC signature date 11 December 2023

Start date 1 January 2024

End date 31 December 2026

Funded under

• Digital, Industry and Space

Total cost

€ 2 618 047,50

EU contribution

€ 2 618 047,50

Coordinated by

MARTEL INNOVATE BV
Netherlands

Periodic Reporting for period 1 - Embed2Scale (Earth Observation & Weather Data Federation with AI Embeddings)

Reporting period: 2024-01-01 to 2025-06-30

Summary of the context and overall objectives of the project

Embed2Scale (E2S) successfully delved into the world of neural compression (NC) for Earth observation (EO) and weather data (WD). The petabyte-scale of spatio-temporal data generated by EO&WD from multiple sensors and hydrodynamic simulation requires a tool to comprehend the information contents impossible to digest by human inspection. With the aid of self-supervised artificial neural networks (Foundation Models) that generate embeddings, we opened the door to efficiently access geo-information.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

* WP 1:
- Draft extension of openEO API to support neural compression endpoints.
- Analysing possible impact of transforming PAZ SAR data into ARD format.

* WP 2:
- Published an IEEE GRSM review on neural compression for Earth observation, catalyzing new use‐case development.
- Launched NeuCo-Bench at CVPR 2025 as an open-source benchmarking framework, engaging 50+ international teams.
- Open-sourced up to 7000× lossy compression methods on multi-sensor imagery.

* WP 3:
- Designed an end-to-end workflow and federated deployment pipeline for model training, compression, and inference.

* WP 4:
- Co-developed four prototypes in biodiversity, cloud characterization, precision agriculture, and maritime awareness.
- Released SSL4EO-S12-downstream datasets on HuggingFace
- Provided some PAZ SAR and AIS sample data to train vessel detection.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

- First dedicated comprehensive review linking lossy neural compression to advance real-world Earth-observation tasks.
- NeuCo-Bench: the inaugural, task-centric EO compression benchmark standardizing metrics and datasets including setup of the community "Earth2Vec" with 32 members (and counting) to define embedding interoperability and best practices.
- SSL4EO-S12-downstream: the first public suite coupling compression evaluation with 12 diverse EO tasks on which the "Earth2Vec" community will build on.
- Pioneered an openEO API extension for on-demand neural compression, enabling seamless integration in EO workflows across data centers.