Increasing the fidelity of hyperlocal simulations of urban pluvial flooding through street flooding observations

Urban pluvial flooding is a highly impactful natural hazard whose understanding remains limited by the scarcity of observations. Here, we demonstrated that continuous, spatially distributed, street-level flood depth measurements provide critical information to increase the fidelity of pluvial flooding simulations. We applied the LISFLOOD-FP two-dimensional, rain-on-grid hydrodynamic model to two dense urban basins of 10.9 and 0.8 km2, respectively, in New York City (NYC), where eight sensors from the FloodNet network recorded widespread flooding during three intense storms. We first provided insights into the generation of model domain and net precipitation forcings at the hyperlocal resolution of 1 m, required to quantify flood hazards at the pedestrian and vehicle scale and to support decision-making. We then focused on one event in the larger basin and assessed the performance of three modeling scenarios under the common condition of limited information about the underground sewer network. We found that neglecting the sewer or simplifying its effect by reducing the precipitation rates severely overestimated the observed water depths. In contrast, simulations based on runoff removal at the stormwater inlets reproduced the observed hydrographs remarkably well after calibration of a single coefficient in the outflow relationship against the sensor data. This calibrated approach proved robust, maintaining high performance in the smaller basin across all three events. As street-level flood observations become increasingly available, the proposed methods could help identify the most accurate strategies to model pluvial flooding in diverse urban landscapes under varying levels of data availability.