Predicting roadway flood susceptibility using crowdsourced traffic data

Abstract:

Water ponding and pluvial flash flooding (PFF) on roadways can pose a significant risk to drivers, with rapid water level rises that can sweep away people and vehicles with little warning. Meanwhile, climate change, growing urbanization, augmenting imperviousness, and aging stormwater infrastructure make these events more frequent. Using a physics-based model to predict pluvial flooding at road segment scale requires notable terrain simplifications, which brings uncertainty and unobserved variables, such as blockage of stormwater inlets, into the results, especially in highly urbanized areas where micro topographic features typically govern the actual flow dynamics. This study evaluates the potential for flood observations collected from Waze–a community-based navigation app–to estimate susceptibility of the road network to PFF at road segment scale. Highly localized surface depressions and their catchment descriptors are derived from a 1-meter-resolution bare-earth digital elevation model (BE-DEM). Next, we investigated the correlation of the Waze flood reports with any well-known flood observations and models, such as the National Flood Hazard Layer (NFHL), high water marks, and low water crossing data inventories. This descriptive analysis showed that the highest correlation of flood reports exists with local surface depressions rather than river flooding. Accordingly, two data-driven models, Negative Binomial regression and Random Forest regression, are implemented to predict the frequency of flooding in classified storm events (light, moderate, and severe) as a proxy for flood susceptibility in every road segment located on surface depressions. Applying the models in two watersheds within the City of Dallas showed that the Negative Binomial regression model performed the best, with predictive errors of 6% for extreme storms, 3% for moderate storms, and 1% for light storms. The results for the entire city will be presented at the conference.