Large Scale GIS-Based 2D Hydraulic Modelling: Improving the Analysis of Flood Dynamics with the Use of Remote Sensing and Volunteered Geographic Information

Population growth driving the human pressure in riverine areas, mostly in developing countries, together with the sea level rise due to climate change is causing an intensification of flood-related damages and fatalities. As a result, territorial planning for managing flood risk and flood-prone areas and non-structural measures (e.g. early warning systems) for flood forecasting are usually developed, principally adopting hydrologic and hydraulic modelling. Numerical models require a large amount of data for model calibration, validation towards flood dynamics understanding and inundation map updating. Data Assimilation (DA) methods are useful tools for improving flood forecasting models and reducing their uncertainties. This work investigates the integration of hydro-geomorphic models, traditional data (static stage gages) and novel data sources, such as remotely sensed images and Crowdsourced data (Volunteering Geographic Information or VGI), for observation-driven improvements of hydro-modelling tools. The Tiber river basin, the second largest basin in Italy, was selected as case study with a focus domain on the approximately 120 km channel upstream of Rome for its strategic importance in the protection of the historical city centre and the coastal urbanized zone. Hydro-geomorphic models are used both as forcing inputs and for delineating the computational domain of a quasi-2D hydraulic model that represents the core of the water level forecasting model within the Data Assimilation framework. Specifically, a parsimonious hydrological modelling algorithm was implemented, calibrated and validated for calculating the flow hydrographs of the ungauged small basins contributing to the study area. Furthermore, to delineate the boundaries computational domain of the hydraulic model for the Data Assimilation application, a DEM-based hydro-geomorphic floodplain delineation algorithm adapted from literature was tested with different DEMs and considering also its parametrization varying the stream orders. Results obtained by the geomorphic algorithm also provided reasonable ranges of the scaling law parameters, originally calibrated from in situ surveys, and here adapted for a DEM-based approach, paving the way for larger scale expeditious flood prone area mapping, that can be consider as a secondary aim of the proposed research. The delineation of the computational domain with this methodology is aimed to avoid the inclusion of hillslope areas, improving the computational efficiency of the Data Assimilation method. The adopted DA methodology is the Ensemble Kalman Filter (EnKF) that requires multiple simulations for representing the uncertainties related to the model and the observations errors. New approaches were proposed for integrating, as observations in the DA method, traditional static sensors, and simultaneously remotely sensed images and VGI data. Despite the static sensor have already been adopted in literature as observations in a DA framework, some new technical measures were necessary for integrating them in Quasi-2D hydraulic model. As auxiliary analysis for the application of the DA methodology, water extension mapping from multispectral images was investigated for selected flood events and a methodology taking into account the ensemble of the hydraulic simulations for deriving the water surface elevation from the satellite image was developed. The assimilation of satellite images resulted to be effective, since the whole computational domain is interested by the water levels correction, although the improvement of the model performance persisted for only some hours of simulation. Despite the scarce availability of VGI data for real flood events in the study area, their usefulness have been investigated considering the uncertainties related to their reliability mostly in terms of accuracy and time allocation. Results show the potential of new data for improving the performance of the flood model, partially overcoming the limitations and the potential scarce availability of the traditional sensors. Finally, the simultaneous integration of all the three types of observations gave promising results, improving the performance of the model compared to the ones obtained assimilating only Satellite images or VGI observations. Future work is needed to test satellite images but mostly the VGI data component because of the limited availability of these data and the not well known error related to their reliability. Furthermore, computational time for an ensemble of 2D hydraulic model simulations is still quite onerous. However, these limitations can be overcome soon by the increasing availability of Satellite remote sensed and VGI data and the considerable growth of the computational power of processors.