‹ Back to space software jobs

Job profile 5 of 8 · Example: ESA Φ-lab, Frascati

AI and Earth observation: analysing satellite images before they land

Satellites deliver more images than people can look at: the EU's Copernicus programme alone sends 16 terabytes to Earth every day6. Artificial intelligence sorts and interprets this data, on the ground and already directly on board20.

ESA · FrascatiCarinthia: VillachNeural networks
See your way inAll 8 job profiles

Free, no sign-up, every statement with a source.

Lake Wörthersee in Carinthia, imaged by Sentinel-2 (2026)Image: European Union, Copernicus Sentinel-2 · attribution

01 · Research

What is researched here: AI that thinks on board a satellite

ESA's Φ-lab was founded in spring 2018 at the ESRIN centre in Frascati near Rome. Around 40 people from 12 nations work there, and it has hosted more than 80 visiting researchers23.

In 2020, Φ-sat-1 was the first experiment in which a neural network analysed images on board: it detects clouds so that only usable images go to Earth20,22. The next generation is meant to detect ships, monitor forests and turn satellite images into street maps, among other things20.

In Villach, Silicon Austria Labs researches virtual sensors using AI in the SUNRISE project, for rocket engines24.

Satellite image of Lake Wörthersee with drawn cloudsDiagram: cloud mask
Own animation of the Φ-sat-1 principle: detect clouds (marked red), then remove them. The clouds are drawn; the image is real.

02 · End product

The end product: maps and measurements everyone can use

The Vienna company EOX combined Sentinel-2 images into an almost cloud-free map of the whole Earth, made of more than 80 trillion pixels65.

Φ-sat-1, the size of a shoebox, showed that filtering in space means sending only usable data and saving transmission22.

ESA's technical centre ESTEC in Noordwijk, NetherlandsImage: ESA · CC BY-SA 3.0 IGO

03 · Daily work

What the work looks like: 4 tasks with satellite images

  1. Classify images: detect clouds, ships or forests automatically.
  2. Find anomalies: report changes a person would miss.
  3. Create maps: turn a satellite image into a street map.
  4. Reduce data volume: sort on board so that less needs to be transmitted to Earth.

Tasks based on the planned applications of Φ-sat-220.

Keywords from the sources

Neural networks (CNN)CopernicusSentinel-2AI for Earth observation

04 · Your path

Your way in: 4 steps

  1. HTL, computer science

    In Astro Pi your Python program collects data on the ISS, for example about life on Earth. Register by 22 February 202755.

  2. Bachelor

    Robotics and Artificial Intelligence at the University of Klagenfurt, in English35.

  3. Master

    Artificial Intelligence and Cybersecurity in Klagenfurt62 or “Earth System from Space” in Graz36.

  4. Entry

    As an ESA Graduate Trainee1; the Φ-lab also hosts visiting researchers23.

05 · Who is hiring?

More employers for AI and Earth observation

All 140 jobs and routesWrite a cover letter
‹ Back to space software jobs
Buy Teddy a Coffee Back to top