Radar-based Fine-scale Rainfall Estimation and Probabilistic Nowcasting of Urban Flooding

Urban flooding is a highly relevant topic for recent years due to its social-economic impact. Growing urbanisation and climate change amplify the risk, bringing new challenges to the management of urban flooding. Supplementary to infrastructural and planning measures, urban flood forecasting that could warn citizens to take precaution measures, offers immediate mitigation effects.Urban flood forecasting is, however, difficult, as urban flooding is a highly uncertain phenomenon. Hence, probabilistic approaches are recommended. Traditional ways of probabilistic forecasting of flooding combine stochastic rainfall forecasts with deterministic urban drainage models. They often suffer from the lack of computational efficiency and the lack of representation of the multiple sources of uncertainty, including rainfall, model structures and components, and infrastructural failures. Therefore, this thesis was designed to contribute to the development of an optimised urban flood forecasting methodology. Given the limited lead times involved in urban flood forecasting, the term "nowcasting" was used.Because rainfall is the driving force of the urban flooding, the first aim of the thesis was to provide a better radar rainfall observation strategy to improve the rainfall input used in urban drainage models. To achieve this, an advanced X-band polarimetric radar was installed for a pilot study in the city of Gent, Belgium. Its capability and its added value were studied in comparison with other existing radar technologies and techniques. The impact study revealed that polarimetric measurements could improve the radar quantitative rainfall estimation. Although, the spatial and temporal resolutions of the current operational C-band radar network of the Royal Meteorological Institute of Belgium (RMI) of 5 min / 1 km were found to be sufficient for urban flood modelling. Dual-polarised X-band radar is recommended for enhancing the radar network at locations where the coverage is insufficient and for large cities.The second aim of this thesis was to improve the current urban flood models in order to achieve fast probabilistic flood predictions. For this purpose, a new hybrid modelling approach was developed. Fast surrogate models were built based on hydrodynamic models for two cases, namely the cities of Antwerp and Gent, Belgium. Results show that the models could provide probabilistic flood predictions at manhole locations with satisfactory efficiency and accuracy. The approach was further adapted to apply at a coarser scale, which was tested for two suburban catchments in Antwerp, Belgium. Results show that consideration of crowdsourced flooding records and incorporation of water levels in the river downstream reduce the model uncertainty significantly.Finally, the thesis made clear which steps and approaches are needed to set up an urban flood nowcasting system strategically. It is encouraged to enhance the radar products for places with high socioeconomic importance and to use a hierarchical modelling regime to accommodate the different levels of prediction needs.

Jaar van publicatie:2020