Abstract Quantifying gas transfer velocity in rivers, a key step in characterizing riverine ecosystems, carbon cycling, and greenhouse gas emissions, typically relies on empirical or semi‐empirical models. Although fluid mechanics theory predicts a direct influence of the depth‐to‐bed‐roughness‐size ratio (relative submergence) on gas transfer velocity, the submergence is rarely reported in gas exchange data sets, and empirical models do not include it as an explanatory parameter. We derived a new approach based on hydraulic resistance equations to reconstruct the missing submergence information and correct gas transfer velocity models using common geometric and hydraulic quantities. We demonstrate a low‐submergence bias in the calibration data of widely used empirical gas transfer velocity models, which explains their behavior and large disagreement at high submergence. We estimated the corrected carbon dioxide gas exchange fluxes for global rivers, and found that emissions from large, high‐submergence (typically lowland) rivers are significantly lower and more uncertain than previously estimated.