Development of a coastal inundation compound modelling framework

Environmental hazards represent a major socio-economic challenge where floods events are the most impactful in terms of global population affected (UNDRR, 2020). Coastal areas are exposed to multiple met-ocean extreme events which can occur separately or combined. Storm surges associated with wind waves, heavy rainfall and tides can lead to catastrophic inundation events associated with breakdown of structures, food and water insecurities and loss of lives. Additionally, climate changes are associated with two coastal risk factors: a) an increase of extreme events (Schiermeier, 2011; Vitousek et al., 2017) and b) an increase of sea level rise (IPCC, 2018).

Different approaches exist to flood modelling (Vousdoukas et al.,2016; Dottori, Martina and Figueiredo, 2018), varying by complexity and accuracy. Simple hydrological models, which operate by integrating the 2D shallow water equation in a flood-plain, offer a good trade-off between computational demand and good skills in simulating real coastal flood events (Smith, Bates and Hayes, 2012). Since accurate inundation modelling is of great importance for risk prevention and management of coastal areas, a system that can be reallocated and calibrated for different regions is a forefront of the research topic.

As a first case study, the flood event of February 2015 in Emilia-Romagna Region (Italy) was selected. The event was characterized by a combination of heavy rain, waves and tides which leads to one of the highest water levels ever recorded in the area (Perini et al., 2015). The model was run with different Digital Elevation Models and forced with water levels provided by Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA) station. The results were compared with observational data of inundation maps.  A broad agreement was found between inundation maps produced by the model and observational data, though with significant local discrepancies. The main differences between model and observations can be ascribed mainly to DEM’s local uncertainty. Work is in progress to include the different types of forcings and to elaborate machine-learning based protocols of calibration to locally improve the model skill, by a) optimizing the mean elevation of the DEM using the modelled and observed flooded areas and b) best-fitting Manning coefficients over the DEM using land use data.