Southern Germany's 100-year flash flood discharge expected to increase by 30 % under an RCP8.5 climate

Voit, Paul; Hundhausen, Marie; Seregina, Larisa; Feldmann, Hendrik; Ahrens, Bodo; Heistermann, Maik

doi:10.5194/egusphere-2026-1229

Preprints

https://doi.org/10.5194/egusphere-2026-1229

Preprints

08 Apr 2026

| 08 Apr 2026

Southern Germany's 100-year flash flood discharge expected to increase by 30 % under an RCP8.5 climate

Paul Voit, Marie Hundhausen, Larisa Seregina, Hendrik Feldmann, Bodo Ahrens, and Maik Heistermann

Abstract. The frequency and intensity of convective heavy rainfall events are generally assumed to increase in a warming climate. So far, however, it remained difficult to assess the corresponding changes in flash floods. This difficulty resulted from the mismatch between the coarse spatial resolution of global and regional circulation models, which do not explicitly resolve convective processes, and the small spatial extent of flash-flood-prone headwater catchments. Recently, though, the results of several convection-permitting climate models (CPMs) became available for parts of Germany. Our study presents the first attempt to utilize these high-resolution data (1 h, 3 km) for an assessment of flash-flood changes in southern Germany. Based on an ensemble of 6 CPM models, we simulated the runoff for the periods 1971–2000 (historical) and 2071–2100 under the RCP8.5 scenario for the German part of the Danube basin. We then compared the 100-year return levels of rainfall and discharge maxima between these two periods. The results indicate an increase of 100-yr flood return levels for 94 % of the Danube subcatchments with a median increase of 30 % across all subcatchments under the RCP8.5 scenario.

Received: 05 Mar 2026 – Discussion started: 08 Apr 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1724 KB)

Supplement (3659 KB)

Download & links

Paul Voit, Marie Hundhausen, Larisa Seregina, Hendrik Feldmann, Bodo Ahrens, and Maik Heistermann

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-1229', Alison Kay, 14 May 2026
Review of “Southern Germany’s 100-year flash flood discharge expected to increase by 30% under an RCP8.5 climate” by Voit et al.
This is an interesting and well-written paper using data from multiple CPMs with a flash-flood model to investigate potential future changes in such floods across southern Germany. The higher spatial and temporal resolution of data from CPMs are clearly going to be important for flood modelling, at least for some events/catchments (see for example my own recent work using hourly 5km CPM precipitation data for Britain https://onlinelibrary.wiley.com/doi/10.1002/hyp.70523).
I only have relatively minor comments:
My main issue is that I don’t really like the term ‘flash-flood’ as I feel it is ill-defined in the literature generally. Can you be precise about what you mean by this term, and are you sure that all the events you are analysing fit that definition?

It is stated in Section 2.2, and shown in Fig 1, that one of the CPMs does not include the whole drainage area. How does this affect modelling of downstream catchments - surely it means some water-input to these catchments is missing? This should be discussed.

In Section 3.2 it says that the modelling is limited to March-October, because you focus on convective rainfall and have no snow routine. But how are antecedent conditions then properly accounted for, if you don’t run the model continuously? By antecedent conditions I mainly mean soil moisture, but also is there no snowpack still to melt at the start of March in some places in this region? Can ‘flash floods’ sometimes be caused by rapid snowmelt and/or rain-on-snow, and how might this change in future?

Section 2 says that the flash-flood model has been ‘proven and tested’ previously. A brief outline of what this actually means would be helpful. E.g. presumably this has used observed driving data and compared against observed flows, but where, over what range of catchments/events, how were you judging performance?

In Section 3.3, it’s rather confusing to describe extracting annual maxima if you are only modelling Mar-Oct…

Also in Section 3.3, you say that basins where the shape parameter is outside a certain range are excluded, but the number excluded varies by CPM. This means that an inconsistent number of catchments is used for various later plots. Is there a sub-set of catchments that are well-behaved for all CPMs, which could be used subsequently?

Section 4.2: I like the pooling of annual maxima from all ensemble members – since you are looking at 100-year return period events this larger sample will be helpful (although the question is whether this could skew results in a certain direction based on members of the ensemble, since it’s an ensemble of opportunity rather designed for this purpose). Can you make it clear that ‘fitting to pooled AM’ is what you mean for the bottom plot in Fig. 2 (i.e. it’s not just taking all the points from the above 6 plots and plotting boxplots for the larger set)?

Adding a horizontal line at 1 for the plots in Fig 2 would help, since this represents the boundary between increases or decreases.

You could also note that the pooled data might be better used with the empirical distribution rather than fitted EVD, although this might be more the case with initial-condition climate ensembles than the multi-model ensemble you have here (e.g. https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019GL081967).

Line 182: what do you mean by ‘significantly though not substantially’?!

Line 189: ‘ensure’ or ‘guarantee’.

Line 194: I’m confused by the ‘results not shown’ comment. The reduced spread of ratios is in the bottom plot in Fig 2, compared to plots above? Or do you mean that you have checked the robustness of GEV fits. If the latter, could you add this info in the Supp Mat?

Alison Kay, UKCEH, May 2026
Citation: https://doi.org/10.5194/egusphere-2026-1229-RC1
RC2:
'Comment on egusphere-2026-1229', Anonymous Referee #2, 01 Aug 2026
This study uses an ensemble of convection-permitting models and a hydrological model to investigate changes in flash flood peaks in the German part of the Danube catchment. The authors conclude that, for the RCP8.5 scenario, 100-year flash flood peaks are expected to increase by 30% in the region. While the evolution of flash floods in a warming climate is a highly relevant topic for NHESS, I believe this study does not meet the criteria required for publication in the journal. I therefore recommend rejection with invitation to resubmit. My main comments are detailed below.
Emphasising the RCP8.5 scenario at the end of the century when reporting projected changes could be misleading. The study uses model runs based on this scenario. While CPM runs with 30-year time periods are only available with CMIP5 projections, I believe that emphasising the RCP8.5 scenario at the end of the century is problematic in terms of communication. This is because it has been shown to be a highly unlikely scenario (Hausfather et al., 2020). Hundhausen et al. (2023), who used four of the six model runs, based their analyses on global warming levels 2 and 3. It seems that the choice of the end of the century was driven by one of the other two CPM runs, which is not transient. However, given that this study cannot explore the role of climate sensitivity comprehensively due to the number of GCM-CPM combinations used and cannot disentangle the climate change signal from internal climate variability (see next comment), there is no need to attempt to provide an overall percentage change in flash flood hazard in the region as 'expected'. Also, the methodology used to derive the 'global delta T' values reported in Table 1 is not explained. I could not find this information in Hundhausen et al. (2023). How were these values derived? I may have overlooked some explanations in Hundhausen et al. (2023), but I believe this is a significant issue, particularly in light of the challenges associated with RCP8.5.

In light of the climate ensemble used, investigating changes in the 100-year flash flood is not adequate. While I acknowledge that the authors discuss this issue several times in the paper, I believe that the model ensemble used in this study does not allow for an analysis of changes in 100-yearly flash floods between the historical and future periods. Firstly, it is not reasonable to assume that enough extremes are sampled within 30-year periods. There is a 26% chance that at least one 100-year (or rarer) event will occur in a 30-year time series. In my opinion, looking at a change signal between two 30-year periods, even with a statistical fit, is therefore not reasonable. Pooling all climate members to improve this would mix the questions of climate sensitivity and sampling, which, in my opinion, contradicts the initial aim of providing insights into projections, as stated in the title. Secondly, several studies have shown that, in order to study extreme events such as floods and their changes due to climate warming, natural variability needs to be quantified (e.g. Aalbers et al., 2017; Brunner et al., 2021; Willkofer et al., 2024). One way to sample more extremes is to perform a spatial aggregation across the different basins by emphasising the median results, as done by the authors. However, using this method to examine 100-year floods would require proper quantification of the spatial independence of events. In my opinion, the results of this study would be more reliable if the authors had looked at return periods below 10 years.

The hydrological assumptions made in this study are not fit for purpose, and they lack evaluation and references to support them. The authors begin section 3.2 with the following statement: 'Flash floods are governed by the formation and concentration of quick runoff components. Therefore, groundwater and evapotranspiration processes can be neglected which immensely simplifies the structure of the required hydrological model”. Firstly, such a statement should be referenced. Secondly, I believe that it may be challenging to find suitable references to support this statement, as it does not align with studies on runoff generation processes. Although the SCS method coupled with a GIUH has been used to study flash floods in ungauged basins, several studies have shown that surface runoff does not necessarily control the entire flood hydrograph, but that instead subsurface processes are important even in flash flood prone areas (Roux et al., 2011; Garambois et al., 2014; Zhang et al., 2021). Furthermore, the explanations in the method section do not clarify how antecedent wetness conditions are considered, especially given that evapotranspiration and groundwater are entirely neglected. Does the model run for only five days before each storm, or continuously between May and October? Thirdly, this study does not rely on any hydrological modelling evaluation. In my opinion, the provided references do not demonstrate that the used hydrological model is fit for purpose; they only show that counterfactual simulations are useful for studying flash floods, which is not the current paper's focus. Although, flash flood observations are scarce (as mentioned by the authors), the model should be evaluated against streamflow data where available, for a few events (not necessarily extreme) and locations at least. Using relative change signals, i.e. comparing simulations between the historical and future periods, does not address this issue because the bias is likely to be non-stationary, especially if intense precipitation events trigger flash floods under conditions not observed in the historical records.

The hydrological modelling assumptions may implicitly enforce flash flood scaling in line with precipitation intensity scaling. Due to the assumptions made within the hydrological modelling framework (i.e. no evapotranspiration, a limited antecedent wetness window and no subsurface processes influencing runoff), the response of the basins to increasing precipitation may be linear by construction. The authors discuss how the flash flood scaling observed in their study aligns with the widely reported C-C scaling for precipitation intensity. Apart from the fact that such scaling should be assessed using local temperature changes rather than global ones, I think the scaling may be linear due to the modelling assumptions made. While this is a reasonable assumption, I do not think it is supported in this study. Assuming that flash floods are not modulated by any subsurface processes could lead to misunderstandings.

No evaluation is carried out to assess the ability of the CPMs to reproduce precipitation observations for the different sub-basins, and no bias correction is applied. The authors state that using a ratio (i.e. a relative change signal) accounts for potential biases in the CPM simulations implicitly. However, biases can be non-stationary, and as the authors also state, the rainfall-runoff relationship is non-linear. Therefore, biases in the CPM simulations could be translated non-linearly into flash flood projections. An evaluation of the CPM performance is required, along with clear explanations showing how this assumption is valid in the context of flash flood projections.

The definition of a flash flood is not explicitly given. From the paper, I understand that any flood occurring between May and October in any of the sub-basins is considered a flash flood. This is a strong assumption that should at least be made explicit and justified based on data or suitable references.

There is only one result plot. While I appreciate short, concise papers, I believe this falls short of the requirements of a journal such as NHESS. Many other analyses could be performed to provide further insight into the simulations conducted in this study. For example, how do the different change signals relate to the hydrological variability between large numbers of sub-basins? How do changes in precipitation intensity drive flash flood responses? Could changes in precipitation characteristics cause a shift in flash flood seasonality? Some more explicit analyses of the variability between different CPMs in generating flash floods in future could be performed. The maps in the supplementary materials could be used, at least partially, in the main manuscript to illustrate the different change ratios.

References
Aalbers, E. E., Lenderink, G., van Meijgaard, E., & van den Hurk, B. J. (2017, September). Local-scale changes in mean and heavy precipitation in Western Europe, climate change or internal variability? Climate Dynamics, 50, 4745–4766. doi:10.1007/s00382-017-3901-9
Brunner, M. I., Swain, D. L., Wood, R. R., Willkofer, F., Done, J. M., Gilleland, E., & Ludwig, R. (2021, August). An extremeness threshold determines the regional response of floods to changes in rainfall extremes. Communications Earth and Environment, 2. doi:10.1038/s43247-021-00248-x
Garambois, P. A., Larnier, K., Roux, H., Labat, D., & Dartus, D. (2014, February). Analysis of flash flood-triggering rainfall for a process-oriented hydrological model. Atmospheric Research, 137, 14–24. doi:10.1016/j.atmosres.2013.09.016
Hausfather, Z., & Peters, G. P. (2020, January). Emissions – the ‘business as usual’ story is misleading. Nature, 577, 618–620. doi:10.1038/d41586-020-00177-3
Roux, H., Labat, D., Garambois, P.-A., Maubourguet, M.-M., Chorda, J., & Dartus, D. (2011, September). A physically-based parsimonious hydrological model for flash floods in Mediterranean catchments. Natural Hazards and Earth System Sciences, 11, 2567–2582. doi:10.5194/nhess-11-2567-2011
Willkofer, F., Wood, R. R., & Ludwig, R. (2024, July). Assessing the impact of climate change on high return levels of peak flows in Bavaria applying the CRCM5 large ensemble. Hydrology and Earth System Sciences, 28, 2969–2989. doi:10.5194/hess-28-2969-2024
Zhang, G., Cui, P., Gualtieri, C., Zhang, J., Ahmed Bazai, N., Zhang, Z., . . . Lei, M. (2021, December). Stormflow generation in a humid forest watershed controlled by antecedent wetness and rainfall amounts. Journal of Hydrology, 603, 127107. doi:10.1016/j.jhydrol.2021.127107
Citation: https://doi.org/10.5194/egusphere-2026-1229-RC2

Paul Voit, Marie Hundhausen, Larisa Seregina, Hendrik Feldmann, Bodo Ahrens, and Maik Heistermann

Supplement

https://doi.org/10.5194/egusphere-2026-1229-supplement

Paul Voit, Marie Hundhausen, Larisa Seregina, Hendrik Feldmann, Bodo Ahrens, and Maik Heistermann

Viewed

Total article views: 597 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
411	157	29	597	66	26	22

HTML: 411
PDF: 157
XML: 29
Total: 597
Supplement: 66
BibTeX: 26
EndNote: 22

Views and downloads (calculated since 08 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	218	92	13	323
May 2026	120	29	8	157
Jun 2026	22	3	5	30
Jul 2026	41	21	2	64
Aug 2026	10	12	1	23

Cumulative views and downloads (calculated since 08 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	218	92	13	323
May 2026	120	29	8	157
Jun 2026	22	3	5	30
Jul 2026	41	21	2	64
Aug 2026	10	12	1	23

Viewed (geographical distribution)

Total article views: 575 (including HTML, PDF, and XML) Thereof 575 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 10 Aug 2026

Short summary

Extreme rainfall is expected to increase with a warming climate. However, it remains unclear how this will impact future floods. This study utilizes an ensemble of 6 new datasets, containing the future projected rainfall, and compares the simulated flash floods of the past (1971–2000) and future (2071–2100) under the RCP8.5 scenario for the German part of the Danube catchment. Our results indicate an increase of 100-yr flood return levels of 30 % across all subcatchments under this scenario.


Total:	0
HTML:	0
PDF:	0
XML:	0