An Integrated Machine Learning Model for Real-time Pluvial-tidal Flood Prediction in Coastal Communities: Case Study for Norfolk, VA, USA

摘要: Detailed simulation of urban flooding is most often achieved using physics-based models. Although advancement in computational power has allowed for speeding up these models, street-scale, real-time flood forecasting in urban environments using physics-based models is still infeasible due to the models long runtime. To address this problem, this study presents a Machine Learning surrogate model that uses a Random Forest algorithm to emulate urban-coastal flood simulation output generated by a 1D/2D physics-based model, TUFLOW. The surrogate model was trained and tested using flood depth time series simulated by the TUFLOW model for historic pluvial and tidal flood events. The input variables include several environmental and topographic attributes that are used to explain the characteristics of the events responsible for flooding. The surrogate model matched the extent and time series depths of …