Urban Environment

The case study addresses urban resilience, mobility, and climate adaptation by developing tailored FMs to detect environmental changes, monitor climate stress, and assess risks using Fiber-Optics Distributed Acoustic Sensing (FO-DAS). It combines multimodal data (satellite Earth observation from Sentinel-1/-2/-3, Landsat, and EnMAP, UAV imagery, street-level photogrammetry, cadastral data, volunteered sources like OpenStreetMap, in-situ weather and air quality, and FO-DAS time series) curated into FAIR, analysis-ready datasets for model development.

Urban FMs build on Vision–Language Models (VLMs) optimised to geolocate, interpret, and describe urban scenes, to track road conditions, infrastructure degradation, and climate stress indicators such as heat islands or vegetation decline. Such FMs are obtained either by fine-tuning and retraining state-of-the-art foundation model architectures using use-case specific data and tasks.

The outputs are explainable, transferable, and usable by researchers and city planners. Validation takes place in European cities (Munich, Leipzig, Grenoble, Strasbourg) complemented by additional benchmarking sites.

Over the course of the project, the pilot will progress from prototype models released as open source (M18) to an operational AI FM pipeline integrated with EOSC-ready services (M36). User feedback from urban planners, civil protection agencies, and citizen associations is envisaged through workshops and Advisory Board interactions, to illustrate benefits and guide service alignment with operational needs.