Where should your Earth observation AI run?

Earth observation AI is moving from experiments to services that public authorities, civil protection and infrastructure operators depend on. That raises questions that a model card does not answer. Where does the model run? Whose law applies to the data and the results? Who could switch the service off, and could you move it somewhere else?

This course gives you a structured way to answer those questions. It explains the European rules that shape where EO AI may run, from GDPR transfers and the US CLOUD Act to the AI Act and the Commission's Cloud Sovereignty Framework. It compares four deployment patterns: public cloud next to the data, a sovereign EU cloud, an on-premises cluster, and air-gapped or edge deployment. And it shows you how to size the hardware for a national inference run from three simple estimates.

In two hands-on notebooks you run Prithvi-EO-2.0, an open EO foundation model, with no internet connection, export it to ONNX, quantise it, measure the trade-offs, verify its signature, and score deployment options for sovereignty. The same notebooks also run, unchanged, on a Linux virtual machine in a European cloud.

8 video lessons · 52 min of video · 2 hands-on notebooks · 8 quizzes · capstone

Legal content is factual, dated "as of September 2026", and is not legal advice.

What you will learn

By the end of the course you can:

  • Distinguish data residency, data sovereignty, operational and technological sovereignty, and recognise sovereignty-washing in a vendor offer

  • Explain which EU and third-country rules shape where an EO AI workload may run

  • Score a deployment option with the SEAL levels and eight objectives of the Commission's Cloud Sovereignty Framework

  • Compare four deployment patterns for EO AI and choose one for a given set of requirements

  • Size the hardware for an EO inference job from memory, throughput and storage estimates

  • Compare renting and owning with the prices that apply when you take the course

  • Run an open EO foundation model fully offline on a laptop CPU or a cloud virtual machine

  • Export it to ONNX, quantise it to int8 and measure speed, memory and output similarity

  • Apply supply-chain controls: checksums, model signing, safe weight formats, mirrors and logging

Who is this course aimed at?

  • EO and GIS practitioners who deploy or commission AI models and want to run them on infrastructure they control.
  • Managers and public-sector decision-makers who must justify where an EO AI service runs, and need a clear view of the rules and the trade-offs without becoming infrastructure engineers.
  • Cloud and IT teams supporting geospatial units, who want to understand the specific needs of EO workloads.

Hands-on details. No background in infrastructure or machine-learning operations is assumed. For the hands-on notebooks you need a laptop with about 8 GB of RAM and basic Python. You need no account, no API key and no GPU. The notebooks run offline from the course materials, on a laptop or on a Linux virtual machine.

Course curriculum

  1. 01

    • 1.1 Why sovereignty matters for EO AI

    • Quiz 1.1

    • 1.2 Data, jurisdiction and cross-border access

    • Quiz 1.2

    • 1.3 The AI Act and Europe's sovereignty frameworks

    • Quiz 1.3
  2. 02

    • 2.1 Deployment architectures for EO AI

    • Quiz 2.1

    • 2.2 Hardware and sizing

    • Quiz 2.2

    • Course materials: notebooks, data and setup guide
  3. 03

    • 3.1 Running open EO models offline

    • Quiz 3.1

    • 3.2 Securing the AI supply chain

    • Quiz 3.2

    • Hands-on 1: Offline inference benchmark

    • Hands-on 2: Sizing and sovereignty assessment
  4. 04

    • 4.1 Decision framework and wrap-up

    • Quiz 4.1

    • Capstone: a two-page deployment brief

Your instructor

Dr. Maria Makaronidou is a Forest Engineer and Environmental Scientist who earned her PhD in Environmental Science from Lancaster University (2020), focusing on the local climate effects of solar parks. She also holds an Integrated Master in Forestry and Natural Environment from the Aristotle University of Thessaloniki (AUTh), specialising in forest infrastructure and ArcGIS. Currently, she serves as an AI-Learning Design and Strategy Consultant at GOVADI, developing geospatial courses for the geo.university platform. Her extensive academic career includes serving as an Associate Lecturer and Tutor at Lancaster University (2015–2019), where she led along with class lectures, field research in Mediterranean ecosystems and workshops on energy metrics. From 2025 to the present, she continues her teaching as an Adjunct Lecturer at AUTh, instructing courses on Snow Science, Hydraulics, and Environmental Geotechnical works. Simultaneously, she is a Scientific Associate at the Athena Research Center, managing European projects like SciLake and EOSC-EVERSE. Her expertise spans Agrivoltaics, Environmental Impact Assessments (EIA), and the application of AI and Big Data in environmental configurations. Further, Maria is a member of the EOSC expert group and the Geotechnical Chamber of Greece. Her research is widely published, covering microclimatic variability and the ecological footprint of infrastructure.

Maria Makaronidou

Senior Instructor

Frequently asked questions

  • Do I need a background in infrastructure or machine learning?

    No. The course explains the legal, architectural and hardware concepts from first principles. For the notebooks, reading and running Python in a Jupyter notebook is enough.

  • Do I need a GPU, a cloud account or an API key?

    No. Both notebooks run offline on a laptop CPU with about 8 GB of RAM. They also run unchanged on a Linux virtual machine in a European cloud, if you prefer.

  • Is this legal advice?

    No. The legal content is a factual overview as of September 2026. For a decision about a specific service, consult your legal or data protection team.

  • Why are there no prices in the videos?

    Cloud and hardware prices change quickly, so the course teaches the method instead. In Hands-on 2 you enter the prices that apply on the day you take the course.

  • How long does the course take?

    About 52 minutes of video, plus 4 to 6 hours for the notebooks, quizzes and capstone.

  • Was AI used to produce this course?

    Yes. The narration uses a synthetic voice generated with ElevenLabs, and the slides, scripts, quizzes and notebooks were drafted with AI assistance and reviewed by the instructor. The hands-on results shown in the videos come from real runs of the notebooks.

  • Do I get a certificate?

    Yes, a certificate of completion once all lessons, quizzes and the capstone are done.

Use of AI tools

Transparency notice regarding the course content creation

The academic concepts and slides for this course were developed by our lecturer. The narration was co-authored using AI tools and subsequently edited. The narration is provided by a synthetic AI voice.