Revealing Global Patterns of Hydropower Plants via Multimodal AI

Hydropower stands as the dominant source of renewable electricity worldwide, playing a pivotal role in global transitions to low-carbon energy systems and climate change mitigation. Yet, the planetary distribution of hydropower infrastructure remains poorly quantified at a global scale—a critical gap that hinders accurate assessments of energy security, freshwater resource allocation, and environmental sustainability. Current public inventories, which are largely compiled through fragmented bottom-up reporting schemes reliant on national or regional submissions, are plagued by pervasive incompleteness, inconsistent geospatial referencing, and significant lags in updates, rendering them inadequate for evidence-based global policy and conservation planning. Here, we present a multimodal artificial intelligence (AI) framework that enables the automated identification of hydropower plants from remote sensing imagery via a globally uniform, top-down methodology. Applied to 8,330,487 river segments across the globe, this framework detects 12,640 hydropower installations, 55.7% of which are unrecorded in leading contemporary public inventories. The resultant global dataset uncovers striking regional disparities and transboundary clustering in hydropower development. It further demonstrates that hydropower infrastructure impacts 56.97% of the world’s protected areas, with marked biomass loss occurring during the construction phase. Complementary hydrological analyses reveal that 29.9% of these installations have experienced declining runoff over the past two decades, while 12.0% are exposed to high flood risk. This work establishes a scalable framework for monitoring global hydropower expansion and its associated environmental and climatic risks, providing a critical foundation for evidence-based energy and conservation policy. The study releases a topdown remote sensing-based hydropower monitoring platform https://glohydro.cn.