A flood prediction framework: integrating seamless predictions into urban hydrological modeling

In recent times, the escalating occurrences of intense precipitation and flooding have exposed substantial socio-economic repercussions, with projections indicating a further rise in their impact due to climate change. Addressing this issue necessitates timely warnings for actions like neighborhood evacuations. However, issuing such warnings poses a dual challenge. On the one hand, it demands accurate forecasts, a difficult task given the heterogeneous nature of rainfall. On the other hand, modeling hydrological processes tied to flood prediction in urban and valley settings proves arduous due to their nonlinear characteristics. Additionally, the accuracy and lead time of forecasted precipitation significantly influence hydrological models, making it challenging for a warning system to generate reliable predictions of flooding events.

This study introduces a comprehensive flood prediction framework that combines: 1) a probabilistic seamless prediction model spanning up to 12 hours, achieved by blending 48 ensemble members from radar-based nowcasting and numerical weather prediction (NWP) ALARO/AROME models, and 2) a distributed hydrodynamic model tailored for urban flood prediction. The primary objective is to evaluate the framework's efficacy in predicting catchment responses, accounting for inherent uncertainties within the models.

For illustrative purposes, rainfall rate estimates are derived from the rain-gauge adjusted radar product managed by the Royal Meteorological Institute of Belgium (RMI). The blended forecast product is sourced from the open-source pysteps community, customized by the RMI for operational use. The hydrodynamic model for flood prediction is implemented through the InfoWorks ICM software, configured to simulate flooding at street level in the city of Antwerp, Belgium. Case studies involve impactful events that led to flooding in major cities within the Flanders area.

Initial findings indicate that, for a rapidly evolving convective storm, precipitation forecasts remained reliable up to 180 minutes in advance, while the flood forecast model predicted flooding levels 2 hours in advance. This analysis is anticipated to underscore the advantages and limitations of an integrated probabilistic approach to flood prediction at urban scales, emphasizing the necessary compatibilities among rainfall products and their representation of uncertainties. The insights gained from this study will contribute to the development of data-driven urban flood prediction models in Belgium for real-time hydrological forecasting.