Trace4EO presented at the 5th ECMWF–ESA Machine Learning Workshop

Trace4EO was presented at the 5th ECMWF–ESA Machine Learning Workshop in Bologna, an event focused on the use of machine learning for Earth Observation, Earth system modelling and prediction.

The workshop provided an opportunity to discuss one of the key challenges addressed by Trace4EO: how to strengthen trust in increasingly complex EO and AI workflows. As EO data volumes grow and machine-learning methods are applied across more stages of data processing and analysis, users need reliable ways to understand and verify how highly derived outputs are produced.

A recurring topic during the discussions was the need to go beyond explainability alone. In workflows involving multiple input assets, processing steps, model versions and fine-tuning stages, trust depends not only on understanding model behaviour, but also on being able to verify the full history of data, models and derived products.

The Trace4EO Solution

Trace4EO addresses this challenge by developing an open-source provenance framework for EO data and AI assets. The solution establishes a verified Chain of Trust based on three core elements:

Keyless identity signing: Trace4EO uses Sigstore to support identity-bound, short-lived certificates, reducing reliance on long-lived private keys and strengthening the integrity of the signing process.
Immutable transparency logging: Processing events are recorded in Rekor, creating an independently verifiable audit trail that supports transparency across the workflow.
Asset-level provenance: Input data, models and derived outputs are protected through cryptographic fingerprints, enabling the ancestry of EO products and AI models to be traced across processing and fine-tuning stages.

The discussions in Bologna confirmed the relevance of verifiable provenance for the EO community, particularly as AI-based methods become more widely used in both research and operational workflows. By providing traceable and independently verifiable histories for data and models, Trace4EO supports more transparent, reproducible and trustworthy Earth Observation outputs.