Probabilistic flood prediction for urban sub-catchments using sewer models combined with logistic regression models

ABSTRACT This paper proposes a hybrid modelling approach for early urban flood warning and forecasting purpose. The hybrid model structure proposed combines a deterministic sewer model with a probabilistic logistic regression model. By varying the sewer model structures from complex hydrodynamic models to simple conceptual models, different hybrid models are constructed in order to accommodate different levels of available knowledge and data about the urban hydrological system. The hybrid models are optimized using crowdsourced flooding records. They can predict the probability of flooding at the urban sub-catchment scale. The proposed methodology is tested for two cases in Antwerp, Belgium. Promising results show its potential in making fast and reliable urban flood predictions. Reasonable predictions are made even by the simplest model form, indicating the method could work also for sub-catchments with limited information. It also shows that the proposed methodology allows identifying the most dominant hydrological processes explaining urban flooding.