Rivers are essential to the Earth's water cycle and deeply impact many human societies and ecosystems, yet they are currently monitored poorly at the global scale. In-situ gauging stations are distributed sparsely and heterogeneously and do not cover much of the world, whereas remotely sensed images are spatially and temporally dense and available globally. Remotely sensed multispectral images, such as the ones acquired by Landsat missions, are available to enable the analysis and surveying of rivers using suitable algorithms. However, existing algorithms are limited in ways that restrict the coverage of the produced results and that prevent the automated analysis of river networks at large scales over short periods of time. Ideally, river maps should be as “live” as possible, e.g., computed quickly and continuously as new Earth imaging data becomes available. Towards advancing progress on this problem, we describe an automated river analysis and mapping engine, RivaMap, that enables the computation of large-scale hydrography data sets from remotely sensed data in a short period of time. RivaMap facilitates water resource management by providing tools to delineate rivers and to estimate their width. As a practical application of RivaMap, we present a continental-scale centerline and width data set of North American rivers, that is automatically computed on Landsat data. We validate our mapping engine by comparing the RivaMap-generated data to a similar data set, NARWidth, and also to in-situ measurements. Our experimental results show that RivaMap is able to efficiently and accurately extract rivers from remotely sensed images at large scales. The outcomes of this research, the software, and the computed exemplary data set are publicly available.