Machine Learning for Earth Observation powered by Supercomputers

Course Description

This course offers an advanced exploration of machine learning and high-performance computing (HPC) techniques tailored to satellite remote sensing and Earth observation (EO). Starting with the historical evolution of Earth observation—from early aerial photography to today’s high-throughput satellite constellations such as Sentinel and Landsat—the curriculum situates modern EO within the context of an unprecedented “big data” era, driven by terabytes of imagery captured daily.

Building on foundational concepts in traditional pixel-based analysis, students will progress to contemporary approaches that leverage the deep learning revolution and the emergence of AI foundation models. These models employ self-supervised learning and Transformer-based architectures to extract structure and meaning from vast amounts of unlabeled data at scale. The course emphasizes how these advanced machine learning paradigms are reshaping EO data analytics beyond classical handcrafted methods.

A core theme of the semester is the intersection of large-scale machine learning and HPC. Students will examine how modern supercomputers enable the training and deployment of Geospatial AI foundation models for Earth observation. This includes discussions on distributed training, parallel algorithm design, and performance optimization on HPC infrastructures. The course also covers benchmarking and performance evaluation, introducing students to standardized evaluation frameworks and metrics to assess model scalability, efficiency, and accuracy in real-world Earth observation applications.

To integrate theory and practice, students complete a semester-long capstone project. Through hands-on development with Python APIs and open models, students will design, implement, and evaluate machine learning solutions on real EO datasets. Projects reinforce the application of theoretical concepts to practical problems, preparing students for research or industry roles at the forefront of AI for EO (see the official page of the course).

Learning Outcomes

Upon completion of the course, the student should:

• Understand and contextualize core concepts inEarth observation and machine learning, including the evolution of remotesensing data, characteristics of big EO datasets, and the foundationalprinciples behind modern Geospatial AI models (e.g., self-supervised learning,Transformer architectures).  
• Apply machine learning and deep learningtechniques to real Earth observation problems, demonstrating proficiency inmodel selection, training, validation, and interpretation for tasks such asland-cover classification, change detection, and environmental monitoring.  
• Leverage high-performance computing resources todesign, scale, and optimize large machine learning workflows, includingdistributed training and fine-tuning of models across multiple GPUs, andevaluate performance using standardized benchmarks.
• Develop and deliver a practical, project-basedsolution that integrates remote sensing satellite data, Python-based APIs, andopen models, showing the ability to translate theoretical knowledge intooperational Earth observation applications with documented results.

Course Material

https://gabcav.github.io/ML4EO-with-Supercomputers/