Datasets generated in the study

This observational study employs an innovative deep learning framework to map river obstructions globally, overcoming the limitations of traditional detection methods by identifying various categories such as dams, low-head dams, locks, and partial dams. The primary dataset, known as DL-GROD (Deep Learning-Global River Obstructions Database), includes over 77,000 records of artificial river obstructions verified through meticulous visual inspection using Google Earth imagery. Additionally, we introduce a Random Forest (RF) model to analyze obstruction density in relation to geographic, hydrological, and socioeconomic factors. The detection point locations, DL-GROD, training data, and simulation results of the random forest model on a 7-level basin scale, curated during this research, are included in this item. The available DL-GROD on this platform consists of distinct samples delineated as square or rectangular segments on the map, rather than the entirety of the dataset.