Abstract
The 5th ECMWF-ESA Machine Learning Workshop (Bologna, Italy; 13-17 April 2026) convened the Earth System Observation and Prediction (ESOP) community to assess progress and new directions in the general field of the application of Machine Learning to ESOP and the specific thematic areas of Machine Learning (ML) for Digital Twins, hybrid ML-physics systems, ML for Earth observation (EO), end-to-end Data Assimilation and Prediction and high-performance computing. The 5-day workshop was held in a hybrid format at the Tecnopolo Data Manifattura (DAMA) in Bologna (IT) with an interactive online component, featuring 57 expert talks and 88 poster presentations. The workshop emphasised both the increasing scale and sophistication of ML for ESOP—including potential end-to-end workflows for future operations—and the parallel shift from CPU-centric to GPU/TPU-centric computing, with additional paradigms (quantum, edge, neuromorphic) emerging. Building on the reporting structure of the previous workshop report (ref. 1), this document summarises outcomes across the five thematic areas covering current topics and emerging trends. All the workshop presentations and recordings are available online at https://ecmwfevents.com/i/5th-ecmwf-esa-machine-learning-workshop/public/agenda .