Flooding is a significant global hazard, affecting billions and exacerbated by climate change and inadequate investment. While advancements in hydro-meteorological monitoring and forecasting have been made, translating these into reduced socio-economic costs remains a challenge. The 2021 Western European floods highlighted this gap, despite advanced FEWS. This paper addresses the limitations of current FEWS by developing a high-resolution, impact-based system that incorporates near-real-time inundation and impact forecasts and their associated uncertainties. This approach aims to provide valuable insights for improved disaster preparedness and tailored emergency actions by providing information to local authorities for better risk-based decision-making. The system extends the traditional forecasting chain by integrating high-resolution hydrodynamic and impact forecasting, offering crucial lead-time maps and impact indicators to support timely responses.

Existing state-of-the-art FEWS, such as GloFAS and EFAS, often rely on interpolating pre-calculated flood hazard maps, which lack real-time dynamics and spatial consistency. These systems often fall short of the expectations of regional flood managers needing high spatio-temporal resolution forecasts. While probabilistic forecasts are increasingly recognized as valuable, particularly for rare events, challenges remain in propagating uncertainties throughout the forecasting chain and representing uncertainty in impact indicators in a way that is useful for decision-making. Existing operational systems often face computational limitations when generating high-resolution inundation and impact forecasts in real-time, especially with high-fidelity hydrodynamic models. Several studies have explored simplified methods and model emulation to improve computational efficiency, but these approaches may struggle with diverse flood scenarios or complex interactions among drivers, and they may inaccurately simulate unprecedented extremes.

The proposed system integrates a chain of models: ICON-D2-EPS (numerical weather prediction), mHM (mesoscale hydrologic model), and RIM2D (high-resolution hydrodynamic model). ICON-D2-EPS provides ensemble precipitation forecasts, which are input to mHM to generate streamflow and water level forecasts. These forecasts are then used to drive RIM2D, creating high-resolution (10-meter) inundation maps. The system incorporates uncertainty quantification throughout the chain. The study utilized the 2021 Ahr River flood in Germany as a case study to hindcast the event. The ICON_D2_EPS ensemble predictions, initialized every 3 hours, were used to force mHM to predict water levels. The probabilistic forecasts and exceedance probabilities of warning levels were analyzed, noting discrepancies between the forecasts and observed water levels, largely due to uncertainties in precipitation forecasts. The system was then extended with RIM2D to provide high-resolution inundation maps including lead-time estimates. Lead-time maps, considering forecast persistence, were generated for the ensemble median and maximum to estimate the time available for emergency actions. These were validated against Copernicus Emergency Management Service (CEMS) Rapid Mapping products. The system also quantifies impacts on buildings and infrastructure, benchmarking results against CEMS data. Finally, the study examines how probabilistic information can be effectively communicated to support flood management decisions, addressing challenges in communicating uncertainties and the need for context-specific decision-making protocols.

The hindcast of the 2021 Ahr River flood demonstrated the feasibility of near-real-time impact forecasting. The ensemble precipitation and water level forecasts showed considerable variation, highlighting the uncertainties in the forecasting chain. Although the ensemble median water level forecasts underestimated the actual flood peak, the probability of exceeding the 100-year flood threshold (HQ100) was greater than 50% for forecasts issued within 17 hours of the peak. At 11 hours before the peak, this probability surged to 90%. A comparison with official forecasts showed similarities but also differences attributable to various factors, including precipitation estimation methods and model parameters. High-resolution lead-time maps generated by RIM2D provided valuable information on the time available for response, ranging from 6 to 30 hours for the ensemble median and 24 to 48 hours for the maximum. The maximum ensemble member showed better alignment with the actual flood extent than the ensemble median, emphasizing the importance of considering the full range of uncertainty. The impact forecasting accurately predicted the number of buildings, roads, and railways affected, although with some overestimation and underestimation depending on the forecast initialization and ensemble statistic used. The uncertainty analysis provided visual representations to show flood managers how the inundated areas are likely to be above specific thresholds and provides a perspective for unusual events. Overall, the proposed system offers a significant improvement over traditional FEWS by providing high-resolution, impact-based forecasts that account for uncertainties and provide decision-makers with richer, more timely information.

The study successfully demonstrates a high-resolution, impact-based FEWS capable of providing timely and valuable information for disaster preparedness and response. The integration of high-resolution hydrodynamic modeling and impact forecasting significantly improves the information available to decision-makers compared to traditional FEWS based on pre-calculated hazard maps or gauge-based warnings. The results highlight the critical role of uncertainty quantification and communication in decision-making. The findings emphasize the need for a shift towards impact-based warnings, which provide more actionable insights than hazard-based warnings. The use of ensemble forecasting effectively captures the inherent uncertainties in weather and hydrological predictions, providing a more realistic representation of potential flood impacts. The method allows for a more dynamic and responsive early warning system. However, the study also highlighted the complexities and challenges of implementing such systems at a national scale, including data availability, computational power, and the need for further research to improve accuracy in predicting rare, extreme events.

This study demonstrates a proof-of-concept for a high-resolution, impact-based flood early warning system. The system successfully integrated advanced numerical weather prediction, hydrological, and hydrodynamic models to generate near real-time inundation and impact forecasts, including uncertainties. The hindcast of the 2021 Ahr River flood showed the system’s potential to provide valuable lead time and actionable insights. Future work should focus on operationalizing the system at a larger scale, improving the accuracy of rare event predictions, and developing strategies for effectively communicating uncertainties to decision-makers.

The study was a hindcast of a single, extreme flood event. While the results are promising, further testing with a larger number and diversity of flood events is necessary to fully validate the system. The accuracy of the forecasts is dependent on the quality of the input data, particularly high-resolution meteorological data. The computational demands of high-resolution hydrodynamic modeling remain a challenge, although the use of GPUs mitigated this to some extent. Effective communication of uncertainties to end-users is critical and requires further investigation. Finally, the study focused on a specific geographic location with readily available data; extending to other regions may require further data collection and model adaptation.