Global hydropower infrastructure and its environmental risks mapped by multimodal AI

Hydropower is the largest source of renewable electricity globally, yet its infrastructure is poorly mapped at scale. Existing public inventories, compiled predominantly from bottom-up reporting, are often incomplete and lack consistent geolocation. Here, we develop a multimodal AI framework for the automated detection of hydropower plants from remote-sensing imagery within a globally consistent, top-down manner. Applied to 8,330,487 river segments worldwide, the framework identifies 12,640 hydropower plants, 55.7% of which are absent from state-of-the-art public inventories. The resulting dataset reveals pronounced regional disparities and transboundary clustering of hydropower development, and shows that hydropower infrastructure affects 56.97% of the world’s protected areas, with substantial biomass loss during construction. Hydrological analyses further indicate that 29.9% of plants have experienced declining runoff over the past two decades and 12.0% are exposed to high flood risk. This work provides a scalable framework for monitoring global hydropower development and associated environmental and climatic risks.