Predicting combined tidal and pluvial flood inundation using a machine learning surrogate model

• DOI: 10.1016/j.ejrh.2022.101087
• Publication Date: 2022-04-22T12:42:06Z
• Authors: Faria T. Zahura; Jonathan L. Goodall
• Keywords: Pluvial; Surrogate model

Flooding increases in recent years, particular for coastal communities facing sea level rise, have brought renewed attention to real-time, street-scale flood forecasting. Such models using conventional physics-based modeling approaches are often unrealistic real-time decision support use cases due their long model runtime. Machine learning offers an alternative strategy whereby a surrogate can be trained mimic relationships present within the and, after training, run seconds rather than hours. This study used Random Forest (RF) algorithm emulate 1D/2D simulating surface water depths urban watershed Norfolk, Virginia. Environmental features from selected set of pluvial and tidal events topographic information roadway were input variables train model. Results show potential predict extent depth both events. Furthermore, differentiate between flooding locations dominated by or impacted mechanisms. Flood reports mobile app Waze validation 90% agreement with Finally, feature importance methods investigated interpret performance RF understand contribution different physical localized flooding.