Quantitative flood risk evaluation to improve drivers' route choice decisions during disruptive precipitation

This article describes a data-driven approach to flood risk exposure evaluation and route delineation during heavy rainfall events. We cross-referenced diverse geospatial and drainage infrastructure datasets with the street network of Hoboken to uncover the factors that increase flood risk. Elevation, slope, precipitation level, imperviousness, and distance to the drainage system's outlets were the most significant predictors to link flooding. We used the link flood risk patterns found in the data to train a reinforcement learning model that generates routes that avoid flood-prone areas. We benchmarked the route assistance model with shortest path and most reliable path algorithms, demonstrating our model has balanced path length and path reliability. We provided the flood risk model outputs at the link-level, which city authorities can use to plan road closures ahead of heavy precipitation events. The route assistance model can be used by drivers to better navigate flood-prone environments by detouring around riskier areas or canceling trips altogether.

Automated summary

This article presents a data-driven approach to assess flood risk exposure and determine routes during heavy rainfall events. By analyzing diverse geospatial and drainage infrastructure datasets, the factors contributing to increased flood risk were identified. Using a reinforcement learning model trained on the flood risk patterns in the data, routes that avoid flood-prone areas can be generated. The route assistance model was benchmarked against other algorithms and showed a balanced consideration of path length and reliability.