Xiaoxiang Guan

Flood risk management has traditionally treated fluvial and pluvial flooding in isolation, for example, by producing inundation maps separately for each flood type. Here, we develop an integrated modeling framework capable of capturing the interplay of both flood types when they co‐occur. The framework consists of the non‐stationary weather generator nsRWG, a disaggregation method, the hydrological model mHM, the hydrodynamic model RIM2D and the damage model FLEMO _flash . It is applied to the Ahr Valley in western Germany, which was severely affected by the July 2021 flood disaster. Using this model chain, we generate 7,200 years of spatially consistent hourly climate and streamflow data (100 realizations of the period 1950–2021) that are translated into inundation and flood damage time series. Within this data set, spatially and temporally consistent 208 compound pluvial and fluvial flood (CPFF) events are identified. Their inundation characteristics and damage values are compared to those obtained under the conventional approach when both flood types are simulated in isolation. Our results demonstrate that CPFF events consistently produce greater hazard and damage than isolated flood types that contribute to a specific CPFF event. Pluvial contributions primarily expand inundation extents, while fluvial processes increase inundation depths, together amplifying overall flood impact. Compared to fluvial‐only flooding, CPFF events increase damages by up to 1,000% for frequent river flooding (return periods less than 20 years) and up to 20% for rare events (with return period exceeding 500 years) for the study area. Relative to pluvial‐only flooding, CPFF events increase damages by up to 300% for frequent events and up to 100% for events with return periods exceeding 50 years. Compound events involving moderate return periods (e.g., a 10‐year fluvial flood with a 7‐year pluvial flood) can cause damages comparable to rare single‐driver events (e.g., 50‐year fluvial or pluvial floods). However, the sum of single‐type flood hazard indicators (maximum inundation depths and areas) and corresponding damages are systematically higher than hazard and damage of the resulting CPFF events due to temporal asynchrony and spatial overlapping effects. These findings highlight the limitations of conventional flood hazard and risk assessments that analyze pluvial and fluvial events separately. The proposed framework provides a robust, integrated methodology for flood risk assessment that accounts for compounding mechanisms and offers valuable insights for improving flood risk management.

Lassen sich Ereignisse wie das Hochwasser im Juli 2021 im Voraus zuverlässig abschätzen? Hätte es schlimmer kommen können? Müssen wir aufgrund des Klimawandels mit noch größeren Ereignissen rechnen? Um solche Fragen zu beantworten, werden Modelle benötigt, die Extremereignisse realitätsnah beschreiben. Am Beispiel der Ahr werden innovative Ansätze zur Hochwassermodellierung vorgestellt. Die Modelle zeigen, dass schon eine leichte Verschiebung des Regenfeldes nach Osten die Abflussspitzen in einigen Zuflüssen der Ahr um bis zu 160 % erhöht hätte. Aufgrund des Klimawandels ist künftig mit höheren Extremniederschlägen und Hochwassern zu rechnen. Für drei der vier betrachteten Szenarien fallen diese Zunahmen moderat aus. Im Worst‐Case‐Szenario sind die Zunahmen jedoch besorgniserregend, z. B. mit einer Zunahme des 100‐jährlichen Abflusses um 43 % am Pegel Altenahr. Die Modelle sind eine wichtige Grundlage für die Bemessung von Schutzmaßnahmen und für ein verbessertes Katastrophenmanagement.

We assess the ability of a regional weather generator to represent the extremity of heavy-precipitation events (HPEs) across spatial and temporal scales. To this end, we implement a multi-site non-stationary regional weather generator (nsRWG) for the area of Germany and generate 100 sets of synthetic daily precipitation data spanning 72 years. The weather extremity index (WEI) and its recent cross-scale modification (xWEI) are applied to quantify the cross-scale extremity of synthetic and observed HPEs and to compare their distributions. The results show that the nsRWG excels in replicating the extremity patterns for almost all seven durations (ranging from 1 to 7 d) considered. The frequency of small-scale 1 d rainfall is, however, slightly overestimated. The nsRWG aptly reproduces the potential influential areas of HPEs, whether of a short or long duration. It is capable of generating precipitation events mirroring the extremity patterns observed during past disaster-causing HPEs in Germany, while simultaneously accommodating their variations. This study demonstrates the potential of the nsRWG for simulating HPE-related hazards and assessing flood risks.

After a flood disaster, the question often arises: what could have happened if the event had gone differently? For example, what would be the effects of a flood if the path of a pressure system and the precipitation field had taken a different trajectory? In this paper, we use alternative scenarios of precipitation footprints shifted in space, the so-called “spatial counterfactuals” to generate plausible but unprecedented events. We explore the spatial counterfactuals of the deadly July 2021 flood in the Ahr Valley, Germany. We drive the mesoscale Hydrologic Model (mHM) of the Ahr catchment with precipitation fields of this event systematically shifted in space. The resulting discharge is used as a boundary condition for the high-resolution two-dimensional hydrodynamic model RIM2D calibrated and validated for this area. We simulate changes in peak flows, hydrograph volumes, and maximum inundation extents and depths, as well as affected assets, and compare them to the simulations of the actual event. We show that even a slight shift in the precipitation field by 15–25 km eastward, which does not seem implausible due to orographic conditions, causes an increase in peak flows at gauge Altenahr of about 32 % and of up to 160 % at the individual tributaries. Also, significantly larger flood volumes of more than 25 % can be expected due to this precipitation shift. This results in significantly larger inundation extents and maximum depths at a number of analysed focus areas. For example, in the focus area around Altenahr, the increase in mean and maximum depths of up to 1.25 and 1.75 m, respectively, is simulated. The presented results should encourage flood risk managers as well as the general public to meet precautionary measures for extreme and unprecedented events.

Predicting unprecedented floods is essential for disaster risk reduction and climate adaptation but remains a challenge for both hydrological and deep learning models. This study evaluates three hydrological models, a Long Short-Term Memory (LSTM) network, and three hybrid models in simulating extreme floods in more than 400 catchments in Central Europe. The hybrid models integrate hydrological process variables with meteorological inputs to enhance runoff simulations. Results show that the LSTM model outperforms traditional hydrological models, while hybrid models further reduce runoff simulation errors. However, all models tend to underestimate peak discharges, with over 50 % underestimation for unprecedented floods. LSTM-based models exhibit extrapolation limits, likely due to structural and statistical constraints. To improve extrapolation to rare events, future work should integrate physical principles into deep learning, including differentiable hydrological models, physics-guided loss functions, and synthetic extreme event generation. Additionally, regional modeling approaches, such as entity-aware LSTMs, could improve predictions by leveraging spatial hydrological similarities. Combining data-driven learning with physical reasoning will be key to improving flood simulations beyond observed extremes.

Predicting unprecedented floods is essential for disaster risk reduction and climate adaptation but remains a challenge for both hydrological and deep learning models. This study evaluates three hydrological models, a Long Short-Term Memory (LSTM) network, and three hybrid models in simulating extreme floods in more than 400 catchments in Central Europe. The hybrid models integrate hydrological process variables with meteorological inputs to enhance runoff simulations. Results show that the LSTM model outperforms traditional hydrological models, while hybrid models further reduce runoff simulation errors. However, all models tend to underestimate peak discharges, with over 50 % underestimation for unprecedented floods. LSTM-based models exhibit extrapolation limits, likely due to structural and statistical constraints. To improve extrapolation to rare events, future work should integrate physical principles into deep learning, including differentiable hydrological models, physics-guided loss functions, and synthetic extreme event generation. Additionally, regional modeling approaches, such as entity-aware LSTMs, could improve predictions by leveraging spatial hydrological similarities. Combining data-driven learning with physical reasoning will be key to improving flood simulations beyond observed extremes.

Predicting unprecedented floods is essential for disaster risk reduction and climate adaptation but remains a challenge for both hydrological and deep learning models. This study evaluates three hydrological models, a Long Short-Term Memory (LSTM) network, and three hybrid models in simulating extreme floods in more than 400 catchments in Central Europe. The hybrid models integrate hydrological process variables with meteorological inputs to enhance runoff simulations. Results show that the LSTM model outperforms traditional hydrological models, while hybrid models further reduce runoff simulation errors. However, all models tend to underestimate peak discharges, with over 50 % underestimation for unprecedented floods. LSTM-based models exhibit extrapolation limits, likely due to structural and statistical constraints. To improve extrapolation to rare events, future work should integrate physical principles into deep learning, including differentiable hydrological models, physics-guided loss functions, and synthetic extreme event generation. Additionally, regional modeling approaches, such as entity-aware LSTMs, could improve predictions by leveraging spatial hydrological similarities. Combining data-driven learning with physical reasoning will be key to improving flood simulations beyond observed extremes.

We assess the ability of a regional weather generator to represent the extremity of heavy precipitation events (HPEs) across spatial and temporal scales. To this end, we implement the multi-site non-stationary Regional Weather Generator (nsRWG) for the area of Germany and generate 100 sets of synthetic daily precipitation data spanning 72 years. The weather extremity index (WEI) and its recent cross-scale modification (xWEI) are applied to quantify the cross-scale extremity of synthetic and observed HPEs and to compare their distributions. The results show that the nsRWG excels in replicating the extremity patterns for almost all 7 durations (ranging from 1 to 7 days) considered. The frequency of small-scale 1-day rainfalls is however slightly overestimated. nsRWG aptly reproduces the potential influential areas of HPEs, whether of short or long duration. It is capable of generating precipitation events mirroring the extremity patterns observed during past disaster-causing HPEs in Germany, while simultaneously accommodating their variations. This study demonstrates the potential of the nsRWG for simulating HPE-related hazard and assessing flood risks.

After a flood disaster, the question often arises: “What could have happened if the event had gone differently?” For example, what would be the effects of a flood if the path of a pressure system and the precipitation field had taken a different trajectory? In this paper, we use alternative scenarios of precipitation footprints shifted in space, the so-called “spatial counterfactuals” to generate plausible but unprecedented events. We explore the spatial counterfactuals of the deadly July 2021 flood in the Ahr Valley, Germany. We drive a hydrological model of the Ahr catchment with precipitation fields of this event systematically shifted in space. The resulting discharge is used as a boundary condition for a high-resolution two-dimensional hydrodynamic model. We simulate changes in peak flows, hydrograph volumes, maximum inundation extent and depths and affected assets and compare them to the simulations of the actual event. We show that even a slight shift of the precipitation field by 15–25 km eastwards, which does not seem implausible due to orographic conditions, causes an increase in peak flows at the gauge Altenahr of about 32 % and of up to 160 % at the individual tributaries. Also, significantly larger flood volumes of more than 25 % can be expected due to this precipitation shift. This results in significantly larger inundation extents and maximum depths at a number of analyzed focus areas. For example, in the focus area around Altenahr, the increase of mean and maximum depth of up to 1.25 m and 1.75 m, respectively, is simulated. The presented results should encourage flood risk managers as well as the general public to meet precautionary measures for extreme and unprecedented events.