The European Space Agency (ESA) is these days developing a new unified Zarr-based file format for Sentinel (1, 2, and 3) mission products under the Earth Observation Processing Framework (EOPF) initiative. EOPF Zarr enables scalable, cloud-native access to Earth Observation data and represents a significant shift in how EO products are distributed by ESA.

This talk introduces the EOPF Zarr format: its design, current status, and practical capabilities. We'll also cover the growing ecosystem of open-source tools being built around it, from plugins for GDAL, xarray, QGIS, R, and Julia, to interactive browser-based exploration of Sentinel imagery with no downloads or preprocessing required.

Whether you're new to cloud-based EO workflows or want to learn about the new format, this session will give you a clear picture of where EOPF stands today and how to get started with it.

Building analysis-ready Earth observation products starts well before any algorithm runs. Source data need to be accessible, complete, up to date. That sounds obvious, but doing it reliably across multiple satellite missions while backfilling years of historical archives is not an easy task.

This talk is about how we built that foundation. The starting point is a simple Argo CronWorkflow that queries a STAC API and downloads one day of data to S3. Nothing impressive, but Argo already gives you things a cron job doesn't: built-in retries, a web UI showing exactly which step failed, and the full log. Your Python script doesn't change, you're just not the (only) one watching it anymore.

This talk follows what happened when we scaled this up across various satellite products. Each problem we ran into pushed us to add something: fan-out parallelism when sequential backfills were taking days, STAC as a logbook of what had already been ingested and what is missing, and eventually an observability layer when we needed to understand periods of higher error rate.

The combination of autonomous backfill and automated monitoring creates a system that self-corrects at two levels: individual failed items are retried via STAC gap detection, while systemic issues surface in daily reports for human intervention.
All the tools are open source: Argo Workflows, STAC API, Python, Kubernetes, CI pipelines Attendees will leave with a concrete understanding of what Argo Workflows gives you at each stage of complexity, from replacing a cron job to running a system you can trust "unsupervised".