This study assesses the change of the seasonal runoff characteristics in 98 catchments in central Europe between the reference period of 1981–2010, and in the near future (2011–2040), mid future (2041–2070) and far future (2071–2099). Therefore, a large ensemble of 50 hydrological simulations featuring the model WaSiM-ETH driven by a 50-member ensemble of the Canadian Regional Climate Model, version 5 (CRCM5) under the emission scenario Representative Concentration Pathway (RCP 8.5) is analyzed. A hierarchical cluster analysis is applied to group the runoff characteristics into six flow regime classes. In the study area, (glacio-)nival, nival (transition), nivo-pluvial and three different pluvial classes are identified. We find that the characteristics of all six regime groups are severely affected by climate change in terms of the amplitude and timing of the monthly peaks and sinks. According to our simulations, the monthly peak of nival regimes will occur earlier in the season and the relative importance of rainfall increases towards the future. Pluvial regimes will become less balanced with higher normalized monthly discharge during January to March and a strong decrease during May to October. In comparison to the reference period, 8% of catchments will shift to another regime class until 2011–2040, whereas until 2041–2070 and 2071–2099, 23% and 43% will shift to another class, respectively.

This study introduces a holistic approach for the hydrological modelling of peak flows for the major Bavarian river basins, referred to as Hydrological Bavaria. This approach, intended to develop a robust modelling framework to support water resources management under climate change conditions, comprises a regionalized parameterization of the water balance simulation model (WaSiM) for 98 catchments in high temporal (3 h) and spatial (500 m) resolution using spatially coherent information and an automatized calibration (dynamically dimensioned search–simulated annealing, DDS-SA) for storage components. The performance of the model was examined using common metrics (Nash & Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE)). The simulations provided the means for the calculation of a level of trust (LOT) by comparing observed and simulated high flows with a five, ten, and 20-year return period. These estimates were derived by the Generalized Pareto Distribution (GPD) applying the peak over threshold (POT) sampling method. Results show that the model overall performs well with regard to the selected objective measures, but also exhibits regional disparities mainly due to the availability of meteorological inputs or water management data. For most catchments, the LOT shows moderate to high confidence in the estimation of return periods with the hydrological model. Therefore, we consider the holistic modelling approach applicable for climate change impact studies concerned with dynamic alterations in peak flows.

​​​​​​​Severe floods with extreme return periods of 100 years and beyond have been observed in several large rivers in Bavaria in the last 3 decades. Flood protection structures are typically designed based on a 100-year event, relying on statistical extrapolations of relatively short observation time series while ignoring potential temporal non-stationarity. However, future precipitation projections indicate an increase in the frequency and intensity of extreme rainfall events, as well as a shift in seasonality. This study aims to examine the impact of climate change on the 100-year flood (HF₁₀₀) events of 98 hydrometric gauges within hydrological Bavaria. A hydrological climate change impact (CCI) modeling chain consisting of a regional Single Model Initial-condition Large Ensemble (SMILE) and a single hydrological model was created. The 50 equally probable members of the Canadian Regional Climate Model version 5 large ensemble (CRCM5-LE) were used to drive the hydrological model WaSiM (Water balance Simulation Model) to create a hydro-SMILE. As a result, a database of 1500 model years (50 members × 30 years) per investigated time period was established for extreme value analysis (EVA) to illustrate the benefit of the hydro-SMILE approach for a robust estimation of HF₁₀₀ based on annual maxima (AM) and to examine the CCI on the frequency and magnitude of HF₁₀₀ in different discharge regimes under a strong-emission scenario (RCP8.5). The results demonstrate that the hydro-SMILE approach provides a clear advantage for a robust estimation of HF₁₀₀ using the empirical probability of 1500 AM compared to its estimation using the generalized extreme value (GEV) distribution of 1000 samples of typically available time series sizes of 30, 100, and 200 years. Thereby, by applying the hydro-SMILE framework, the uncertainty from statistical estimation can be reduced. The study highlights the added value of using hydrological SMILEs to project future flood return levels. The CCI of HF₁₀₀ varies for different flow regimes, with snowmelt-driven catchments experiencing severe increases in frequency and magnitude, leading to unseen extremes that impact the distribution. Pluvial regimes show a lower intensification or even decline. The dynamics of HF₁₀₀ driving mechanisms depict a decline in snowmelt-driven events in favor of rainfall-driven events, an increase in events driven by convective rainfall, and almost no change in the ratio between single-driver and compound events towards the end of the century.

Severe floods with extreme return periods of 100 years and beyond have been observed in several large rivers in Bavaria in the last three decades. Flood protection structures are typically designed based on a 100-year event, relying on statistical extrapolations of relatively short observation time series while ignoring potential temporal non-stationarity. However, future precipitation projections indicate an increase in the frequency and intensity of extreme rainfall events, as well as a shift in seasonality. This study aims to examine the impact of climate change on the 100-year flood (HF₁₀₀) events on 98 hydrometric gauges within the Hydrological Bavaria. A hydrological climate change impact (CCI) modelling chain consisting of a regional single model initial condition large ensemble (SMILE) and a single hydrological model was created. The 50 equally probable members of the CRCM5-LE were used to drive the hydrological model WaSiM to create a hydro-SMILE. As a result, a database of 1,500 model years (50 members x 30 years) per investigated time period was established for extreme value analysis (EVA) to illustrate the benefit of the hydro-SMILE approach for a robust estimation of the HF₁₀₀ based on annual maxima (AM), and to examine the CCI on the frequency and intensity of HF₁₀₀ in different discharge regimes under a strong emission scenario (RCP8.5). The results demonstrate that the hydro-SMILE approach provides a clear advantage for a robust estimation of the HF₁₀₀ using empirical probability on 1,500 AM compared to its estimation using the generalized extreme value (GEV) distribution on 1,000 samples of typically available time series size of 30, 100, and 200 years. Thereby, by applying the hydro-SMILE framework the uncertainty from statistical estimation can be reduced. The CCI on the HF₁₀₀ varies for different flow regimes, with snowmelt-driven catchments experiencing severe increases in frequency and intensity, leading to unseen extremes that impact the distribution. Pluvial regimes show a lower intensification or even decline. The study highlights the added value of using hydrological SMILEs to project future flood return levels.

Heavy rainfall and resulting flash flood events have been in the focus of research and the public in recent years. The relevance of the topic will become more prominent with increasing temperatures due to climate change. Extreme rainfall events in Germany like 2014 in Münster (North Rhine-Westphalia) or 2016 in Simbach am Inn (Bavaria) and Braunsbach (Baden-Wurttemberg) have also raised public awareness. Hydrodynamic models for the simulation of fluvial events have been developed for a long time and are often used. However, the question arises to what extent these methods can be used for pluvial events. Hydrodynamic models allowing precipitation input are therefore well suited for the simulation of pluvial events, as they can display flow paths, depths, and velocities in high resolution. Nevertheless, defining precipitation without infiltration leads to an overestimation of the surface runoff. For this problem, an improved event simulation can be achieved by nesting hydrological processes into the hydrodynamic simulation procedure. In this study, we are using TELEMAC-2D as a hydrodynamic model because it uses precipitation in a spatially and temporally distributed manner and can be used very well by high-performance computing. LARSIM (Large Area Runoff Simulation Model) and WaSiM (Water Flow and Balance Simulation Model) are used as hydrological models. The methodology for simulating flash floods can be divided into two important processes: runoff generation and runoff concentration. These are divided according to the strength of the respective model types: Runoff generation: SCS-CN value method (TELEMAC-2D), Green Ampt method (LARSIM), layer-resolving Richards method (WaSiM) Runoff concentration: Strickler roughness approach (TELEMAC-2D), Kalinin-Miljukov method (LARSIM), flow time index method (WaSiM) In this study, we examine three different types of couplings: (1) The runoff concentration is calculated using the hydrodynamic model, the runoff generation is carried out using the CN value method. (2) The runoff generation in the entire catchment is calculated using the hydrological processes (LARSIM/WaSiM). The runoff concentration is still generated by the hydrodynamic model. (3) The runoff concentration in the upper catchment area is also calculated using hydrological methods, only the urban area is calculated hydrodynamically. We compare the different coupling types with each other using some real flash flood events. The results are presented with the aim to identify which approach is necessary for a good representation of the flash flood event. This depends mainly on the local conditions in the catchment area (e.g.  culverts, land use) and the rainfall event (e.g. rainfall intensity and duration). The findings from this study will be transferred to unobserved catchments in the further course.