Geomorphologists have long debated the relative importance of disturbance magnitude, duration and frequency in shaping landscapes. For river-channel adjustment during floods, some argue that cumulative flood ‘power’, rather than magnitude or duration, matters most. However, studies of flood-induced river-channel change often draw upon small datasets. Here, we combine Sentinel-2 imagery with flow data from laterally-active rivers to address this question using a larger dataset. We apply automated algorithms in Google Earth Engine to map rivers and detect their lateral shifting; we generate a large dataset to quantify channel change during 160 floods across New Zealand, Russia, and South America. Widening during these floods is best explained by their duration and cumulative hydrograph. We use a random forest regression model to predict flood-induced channel widening, with potential applications for hazard management. Ultimately, better global data on sediment supply and caliber would help us to understand flood-driven change to river planforms.