Deep Learning and Remote-Sensed Observations Reveal Global Underestimation of River Obstructions

River obstructions are a subject of global concern due to their impact on river connectivity and aquatic ecosystems. However, detecting and quantifying these structures, especially small and undocumented ones, remains a major challenge due to limitations in existing data sets and detection methods. This study focuses on improving the global detection of river obstructions and revealing their spatial distribution patterns. We developed a deep-learning-based detection framework combined with manual validation, resulting in the Deep Learning-Global River Obstructions Database, which comprises 50,061 river obstructions identified globally. This represents a 64% increase over previous estimates, which were based solely on manual identification. Spatial analyses reveal strong correlations between obstruction density and factors such as Gross Domestic Product, agricultural expansion, urbanization, and river morphology. By enhancing the precision and comprehensiveness of river obstruction data, our open-source data set provides a solid foundation for accurate assessment of global river connectivity, basin-to-continental-scale hydrological modeling, and impact assessments.