AI Is Being Trained to Predict Climate Damage to France’s Heritage Sites
French conservation scientists are building an artificial intelligence model that could help forecast how climate change will affect cultural heritage, from cathedral stonework to archaeological remains. The project, led by Ann Bourgès at the French Ministry of Culture’s Centre for Research and Restoration of France’s Museums, is designed to give restorers and archaeologists a more precise way to measure deterioration and anticipate what may happen next.
The idea took shape after discussions within the European Commission four years ago, when Bourgès and colleagues concluded that climate impact on heritage needed to be quantified rather than described in broad terms. In 2022, Bourgès and two other researchers launched two doctoral projects and recruited Adèle Cormier and David Roqui, both now in the final year of their PhDs.
The team is working across multiple institutions and disciplines, drawing on conservation, geoscience, heritage studies, engineering, and computer science. Their methodology is being tested at three French sites chosen for their range of materials, climates, and historical contexts. One is the octagonal sandstone base of the spire of Strasbourg’s 13th-century cathedral, a Rayonnant Gothic landmark exposed to severe inland winters and intense summers. Another is the Bibracte archaeological site near Autun in Burgundy, a fortified Gallic settlement first excavated in the late 19th century. A third coastal site will eventually be added.
Cormier has spent two years collecting detailed climatic and material deterioration data at two of the sites. Roqui’s work focuses on multimodal learning, teaching the AI to process not only measurements but also photographs, audio, and other forms of evidence. Bourgès says the system is being trained to recognize features such as fissures and then compare images over time to assess how much damage has advanced.
That process has exposed two major challenges. The first is scale: climate change is global, but each heritage site has its own microclimate, both outdoors and inside the structure. The second is data compatibility. Temperature, humidity, and carbon dioxide can be monitored with specialist devices and satellite data, but the instruments and methods are not standardized. Even assessing a crack or crumble can be subjective, since descriptions depend on perception, language, and a strict taxonomy. Photographs add another layer of complexity because light and angle can alter what is visible.
The project also uses thermal infrared imaging to detect water ingress and salt accumulation, expanding the range of evidence the model can read. The broader goal is to create an open-source methodology that will be shared through the national Espadon website, offering a framework others can adapt. France already has a national plan for monitoring climate change’s effects on its treasures, and this project suggests how conservation may increasingly rely on data science as much as traditional expertise.
