Quantifying River Widths of North America from Satellite Imagery

River width is a fundamental predictor variable in many hydrologic, geomorphic, and biogeochemical models, yet current large-scale models rely on theoretical hydraulic geometry relationships that do not fully capture natural variability in river form. Here we present the first high-resolution dataset of long-term mean width of North American rivers wider than 30 m. The dataset contains 7.93 million georeferenced width measurements derived from Landsat TM and ETM+ imagery that were acquired when rivers were most likely to be at mean discharge. We built the dataset by developing an automated procedure that selects and downloads raw imagery, creates cloud-free normalized difference water index images, histogram balances and mosaics them together, and produces a water mask using a dynamic water-land threshold technique. We then visually inspected and corrected the mask for errors and used RivWidth software to calculate river width at each river centerline pixel. We validated our dataset using >1000 United States Geological Survey and Water Survey of Canada in situ gauge station measurements. Error analysis shows a robust relationship between the remotely sensed widths and in situ gauge measurements with an r ² = 0.86 (Spearman's = 0.81) and a mean absolute error of 27.5 m. We find that North American river widths lie on logarithmic frequency curve with some notable exceptions at widths <100 m. This dataset can be used to improve our understanding of the water, carbon, and nitrogen cycles, as well as large-scale landscape evolution models. Our results also allow for the characterization of the extent of rivers likely to be observable by the planned Surface Water and Ocean Topography (SWOT) satellite mission.