Key points are not available for this paper at this time.
Advancements in Earth Observation, coupled with the swift progress in big data analysis and access to distributed computing and storage, open up exciting possibilities for the development of Digital Twins of the Earth. These Digital Twins hold the potential to transform disaster preparedness, allowing us to foresee extreme events and assess the effectiveness of various policy measures. Within this framework, we propose here a specialized Digital Twin dedicated to flood disasters. Its primary goal is to enhance flood resilience by introducing an innovative inundation forecasting service that provides early warnings and enhances preparedness. To ensure the product aligns with user needs, a multi-tiered strategy for collecting user requirements was implemented. Key features identified by users include hourly flood depth predictions, updated daily, with a 72-hour lead time. The integration of local data and models for impact analysis at local scales was also recognized as crucial. The chosen pilot studies for this project focus on the winter 2020 storms in the Severn Catchment, UK and the summer 2021 storm in the Alzette Catchment, Luxembourg. Both events were observed by the Copernicus Sentinel-1 mission. To meet user requirements, the study aims to incorporate existing state-of-the-art global and regional near-real-time flood monitoring and forecasting products, namely GloFAS (Global Flood Awareness System) and GFM (Global Flood Monitor). The Digital Twin thus consists of four key elements: Numerical Weather Prediction (NWP) model, based on ECMWF or French/German weather service forecasts; Land surface model and rainfall-runoff model, i.e. GloFAS HTESSEL or LARSIM; Hydrodynamic model, with LISFLOOD-FP model for both catchments; Flood impact assessment model, based on KONTUR population dataset and OpenStreetMap. By integrating the GFM and GloFAS products through data assimilation, the Digital Twin is capable of short-term as well as medium-range daily inundation forecasting, reducing predictive uncertainties. The data assimilation strategy is flexible and accommodates various global- and local-scale models and resolutions. Its implementation involves particle filtering enabling weighted combinations of pre-computed flood depth maps based on LISFLOOD-FP, aligned with flood extent maps observed by GFM, providing a more accurate representation of the real world. Not only this strategy is spatiotemporally transferable but it is also adaptable to new test sites without extensive retraining or reconfiguration. The outcomes of this proof-of-concept study can lay the groundwork for future research in the field, contributing to closing the global flood protection gap.
Nguyen et al. (Sat,) studied this question.