Earth Observation (EO) has become one of the most demanding and scientifically important application domains for modern machine learning. The rapid emergence of large-scale EO foundation models, multimodal satellite archives, self-supervised learning, and scientific AI workflows has transformed how EO data are analysed. At the same time, the increasing complexity of models, datasets, evaluation protocols, and downstream tasks has made automation a necessity rather than a convenience.

AutoML provides many of the methodological foundations required to address these challenges, including automated model selection, hyperparameter optimization, meta-learning, neural architecture search, workflow optimization, and empirical performance modelling. However, AutoML systems can only reason about tasks, datasets, and models if those artifacts are described in a consistent, machine-readable, and interoperable way. This is the role of open science infrastructure: FAIR data principles, controlled vocabularies, ontologies, standardised metadata, and benchmark repositories that accumulate empirical knowledge across tasks and studies.

Recent developments in AI assistants and agentic systems suggest that future EO systems may automatically design, execute, and evaluate complete analysis pipelines. Such systems require rigorous benchmark suites, reproducible evaluation methodologies, rich metadata, and accumulated empirical knowledge before they can become scientifically reliable. Open science infrastructure is precisely what provides this foundation.

This workshop brings together the AutoML, Earth Observation, and open science communities to discuss the next generation of automated EO machine learning systems.