Flooding impacts are on the rise globally, and concentrated in urban areas. Currently, there are no operational systems to forecast flooding at spatial resolutions that can facilitate emergency preparedness and response actions mitigating flood impacts. We present a framework for real‐time flood modeling and uncertainty quantification that combines the physics of fluid motion with advances in probabilistic methods. The framework overcomes the prohibitive computational demands of high‐fidelity modeling in real‐time by using a probabilistic learning method relying on surrogate models that are trained prior to a flood event. This shifts the overwhelming burden of computation to the trivial problem of data storage, and enables forecasting of both flood hazard and its uncertainty at scales that are vital for time‐critical decision‐making before and during extreme events. The framework has the potential to improve flood prediction and analysis and can be extended to other hazard assessments requiring intense high‐fidelity computations in real‐time.

Interactions between spatial and temporal variability of rainfall and catchment characteristics strongly influence hydrological response. In urban areas, where runoff generation is fast due to high imperviousness degree, it is especially relevant to capture the high spatiotemporal rainfall variability. Significant progress has been made in the development of spatially distributed rainfall measurements and of distributed hydrological models, to represent the variability of catchment's characteristics. Interactions between rainfall and basin scales on hydrological response sensitivity, however, needs deeper investigation. A previous study investigated the hydrological response in the small urbanized catchment of Cranbrook (8 km², London, UK) and proposed three dimensionless “scale factors” to identify if the available rainfall resolution is sufficient to properly predict hydrological response. We aim to verify the applicability of these scale factors to larger scales, with a distinct physiographic setting, in Little Sugar Creek (111 km², Charlotte, USA), to identify the required rainfall resolution and to predict model performance. Twenty‐eight events were selected from a weather radar data set from the National Weather Radar Network, with a resolution of 1 km² and 15 min. Rainfall data were aggregated to coarser resolutions and used as input for a distributed hydrological model. Results show that scale factors and associated thresholds are generally applicable for characterization of urban flood response to rainfall across spatiotemporal scales. Additionally, application of scale factors in observation‐based analysis supports identification of event characteristics that are poorly captured and critical improvements that need to be made before the model can benefit from high‐resolution rainfall.