the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Apr 2026
First insights into CMIP6-based hydrological projections for Central European rivers – using a small ensemble of convection-permitting climate simulations for +2 and +3 °C global warming levels
Abstract. A warmer climate affects the hydrological regimes of rivers. It is essential to quantify these changes in order to evaluate vulnerabilities, future risks, and to develop effective adaptation strategies. To gain insights into the regional impacts of the latest generation of the Coupled Model Intercomparison Project Phase 6 (CMIP6), the NUKLEUS CMIP6 ensemble (five members, three GCMs coupled to two convection-permitting RCMs) was used for first hydrological simulations for 53 German subcatchments of the Rhine, Elbe, Danube, Weser and Ems rivers with a water balance model (LARSIM-ME).
Within this preliminary ensemble, results for a 2 °C and 3 °C global warming level (GWL) show generally decreasing mean (MQ) and high flows (MHQ) in western Germany (Rhine, Weser, Ems), while in the eastern catchments (Elbe, upper Danube) high flows are projected to increase, compared to the reference period 1961–1990. Further, decreases generally display for low flow indicators (MNQ) – especially for GWL 3 °C – except for heavily snow-affected catchments.
Although the results resemble features of previously observed hydrological change in those catchments (no major flood events in the Rhine River for 30 years with a MHQ decrease compared to the previous 30-years-period, very dry conditions in the last decade on a national level, significant regime changes in the Alpine region), they should be treated with caution. The east-west gradient, which manifests in the MQ and MHQ response to 2 °C or 3 °C warming, has not been present in discharge projections of the prior CMIP5 generation. In addition, the ensemble used is comparatively small and includes two RCMs (CCLM and ICON) that are quite similar with regard to the parameterization of the precipitation processes. Nevertheless, a strong influence of the different GCMs was evident and the new phenomena of decreasing mean and high flows in western Germany could be a new climate change signal originating from the GCMs. In future analysis, change signals should be reassessed with a wider ensemble.
- Preprint
(3834 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (extended)
- RC1: 'Comment on egusphere-2026-1630', Anonymous Referee #1, 11 May 2026 reply
-
RC2: 'Comment on egusphere-2026-1630', Anonymous Referee #2, 01 Jun 2026
reply
General assessment
The submitted manuscript presents hydrological projections over 53 gauges in German rivers using the NUKLEUS CMIP6 ensemble (consisting of 3 GCMs and 2 RCMs,) and the LARSIM-ME water balance model. The paper addresses relevant scientific questions within the scope of HESS and can be potentially useful as an early hydrological application of convection-permitting CMIP6-based regional climate simulations for Germany with potential value to provide first insights into these new datasets.
I believe the current version can still benefit from clearer positioning of its scientific contribution by explaining explicitly how the findings should be considered by water managers and adaptation planners. The authors are honest about the limits of the study, especially regarding the small ensemble size. I acknowledge this caution, which is important, but the manuscript can benefit from further discussion of uncertainty and decision relevance. The scientific approach and applied methods are valid, but in some parts they need better elaboration, especially regarding the spatial resolution of hydrological modelling used in this study.
Major comments
1. At present, the methodological chain follows a well-established impact-modelling structure: climate model output, regional downscaling, bias correction, hydrological modelling, and analysis of flow indicators. As the individual components are commonly used in climate change related impact studies, the contribution is mainly dataset-oriented rather than methodological. The title already uses “first insights” but the introduction and discussion should more directly explain what this preliminary analysis can and cannot tell us. A possible framing could be added in the introduction to tighten the novelty. For example, “the authors could add a sentence clarifying that given the small ensemble size, the results represent early signals from a new dataset rather than a robust uncertainty-constrained projection.”
2. The manuscript refers to NUKLEUS data as approximately 3 km and 1 hour (line 84), then remapped to the 5 × 5 km HYRAS/LARSIM grid (lines 112 and 142). As the manuscript emphasizes the use of convection-permitting climate simulations, the added value of this resolution is not clearly demonstrated.
This raises an important question: how much of the original hourly, approximately 3 km convection-permitting information is retained in the final hydrological simulations? I can understand that direct comparison with non-convection-permitting or statistically downscaled CMIP6 products might not be feasible but this limitation should be discussed explicitly.
3. Following the previous comment, "convection-permitting" is central to the title and motivation of the manuscript; therefore, readers would expect an explicit discussion of its relevance for Germany. The introduction states that the NUKLEUS ensemble provides high-resolution regional climate information, but it does not explain which German hydrological processes are expected to benefit from 3 km convection-permitting simulations. This clarification is especially important because the selected hydrological indicators are calculated from daily discharge simulations. The manuscript should explain how the benefits of hourly 3 km convection-permitting rainfall are expected to influence these daily hydrological indicators derived from the 5 km setup.
4. The manuscript cites the work of Kreienkamp et al. (2020) that used statistical-empirical downscaled CMIP6 data for Germany but it does not clearly discuss why the authors chose the small convection-permitting dynamical ensemble instead. This methodological selection decision needs to be discussed with respect to computational costs and the added value of the 3 km dynamically solved resolution, together with concerns raised in Comments 2 and 3. For example, why did the authors not use a direct 5 km statistically downscaled product? This is especially relevant because the authors apply bias correction to the dynamically downscaled data. The manuscript should therefore explain more clearly what is gained by using convection-permitting dynamical downscaling rather than available statistical or statistical-empirical downscaling products.
5. The manuscript states: "The extent to which the novel coupling of GCM to convection-permitting RCM could be responsible for the emergence of a west–east gradient should be further investigated" (lines 289–290). However, the present analysis does not isolate the contribution of convection-permitting regionalization from the influence of the driving GCMs.
It remains difficult to assess how much of the reported hydrological signal is specifically attributable to convection-permitting resolution, given that there is no direct comparison with non-convection-permitting RCMs, statistically downscaled products, or raw GCM-driven hydrological simulations. This distinction is particularly important because the manuscript also states that the GCM influence is strong and that the RCMs only slightly modulate the change signals (lines 346–349). I suggest that the authors make this interpretation more explicit and clearly distinguish between using a convection-permitting dataset and demonstrating its specific added value for the hydrological projections.
6. Uncertainty attribution, quantification, and propagation need a dedicated section in the discussion. The main results are presented as ensemble medians (e.g., Figures 6 and E1) from only five GCM–RCM combinations, while the spread across the model chains is not shown.
Given the limited set of five GCM–RCM model chains, the authors could still provide a transparent visualization of the available spread as an example. This would allow readers to assess whether the projected changes are spatially coherent and robust, or whether they are strongly influenced by individual GCM–RCM combinations. A transparent visualization of uncertainty would substantially improve interpretation and would make the study easier to compare with future analyses using larger ensembles.
In particular, it would be useful to know whether all model chains support the reported east–west gradient in MHQ changes, or whether this pattern is dominated by one or two GCM–RCM combinations (following comment #5).
In addition, the uncertainty discussion currently focuses mainly on the climate ensemble. However, hydrological-model uncertainty is also important and should be discussed more explicitly. Understandably, the study uses a single hydrological model for a national application, but it means that uncertainty related to hydrological model structure and parameterization cannot be quantified. I suggest adding a dedicated paragraph on this limitation. Useful points to discuss include parameter uncertainty, structural uncertainty, and the possibility that low-flow and high-flow indicators may respond differently to hydrological model structure.
7. The manuscript states that the first GCM realization, r1i1p1f1, was applied for all experiments (lines 83-84). This should be mentioned in the uncertainty discussion because using a single realization means that internal climate variability is not systematically sampled.
8. I would like to emphasize a concern related to the justification of the 5 × 5 km hydrological modelling setup. Since the study is motivated by approximately 3 km convection-permitting climate simulations, readers need to understand why the final hydrological simulations are performed at 5 × 5 km and a daily time step.
Given that higher-resolution distributed hydrological modelling systems are available for Germany, the authors should clarify why the 5 × 5 km LARSIM-ME setup was selected, and whether this choice was driven solely by operational relevance or model availability, or whether there are scientific reasons related to calibration quality, consistency with the HYRAS observational grid, or the focus on larger river basins. This clarification would help readers understand the actual resolution and process representation of the modelling chain, and how it relates to the convection-permitting motivation of the study.
9. The manuscript would benefit from a more explicit comparison with recent large-ensemble hydrological projection work for Germany and Central Europe. In particular, the study by Chandrasekar et al. (2026) in Climatic Change appears highly relevant because it addresses overlapping questions for German river basins using a much larger ensemble, and is therefore directly relevant for interpreting the present manuscript, which uses only five GCM–RCM combinations.
Differences between the two studies may arise from ensemble size, climate model selection, hydrological model structure, spatial resolution, bias correction, and indicator definitions. The authors should discuss these differences explicitly and clarify what the present study contributes beyond what is already available from larger-ensemble analyses.
10. The manuscript notes that the 1961–1990 reference period already corresponds to approximately +0.42 °C above pre-industrial conditions (Lines 86-89). This is important and should be made more visible in the results and discussion, because all reported changes are relative to an already warmed baseline.
11. The abstract and introduction state that the study aims to support the regional adaptation strategies, impact assessments, and adaptation planning (lines 33-34). This is an important motivation, but the discussion does not yet fully translate the hydrological indicators into implications for decision-making. It would strengthen the manuscript if the authors added a short subsection explaining what these indicators may imply for practical water management, for instance, under the EU Water Framework Directive or Germany's National Water Strategy.
The manuscript also identifies an east–west gradient in MQ and MHQ response, which the authors describe as a potentially new signal compared with CMIP5-generation projections. This is interesting, but its practical meaning remains unclear. I suggest that the authors explain whether this gradient could affect regional risk prioritisation. A useful addition would be a short paragraph linking the +2 °C and +3 °C warming-level results to current adaptation planning horizons and policy frameworks.
Minor comments
- Figures need to have a latitude and longitude coordinate grid to help readers locate the study catchments precisely.
- Figure 2 would benefit from additional information of subcatchment sizes in the caption. It’s not clear which river systems are shown and their approximate spatial extent. As it stands, readers unfamiliar with German river geography may find it difficult to interpret the spatial patterns discussed in the results.
- The conclusions do not address policy implications in the current format. Given that the abstract and introduction explicitly motivate the study with reference to adaptation planning, the conclusions should include at least a short paragraph discussing what the results mean for water management practice.
Reference:
Chandrasekar, et al (2026). Increased high flows in Germany under 1.5 K, 2.0 K and 3.0 K warming based on a large regional climate model ensemble. Climatic Change, 179, 94. https://doi.org/10.1007/s10584-026-04165-w
Citation: https://doi.org/10.5194/egusphere-2026-1630-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 260 | 114 | 23 | 397 | 14 | 20 |
- HTML: 260
- PDF: 114
- XML: 23
- Total: 397
- BibTeX: 14
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study uses an ensemble of convection-permitting models and a distributed hydrological model to present hydrological change signals for large parts of Germany. The authors present their findings for low, mean, and high flows, as well as for two global warming levels. These results reveal regional variations in the direction of hydrological changes that do not necessarily align with previous findings in the region. Overall, I believe this study is relevant to HESS readers because it uses recent, cutting-edge climate projections. However, major revisions are required before the paper can be considered for publication (see my comments below).
Major comments:
1) The reasons for publishing this work now are unclear to me. In line 50, the authors state that “A larger ensemble covering more GCMs and RCMs (EURO-CORDEX) is expected in spring 2026”, which suggests that this will happen in the coming weeks, if not already. However, because the ensemble is incomplete, the authors conclude in lines 309–312 that “A potential explanation for the stronger changes in discharge patterns as part of the present analysis, may be the relatively small model ensemble used within the present study (three GCM combined with two RCM) and the use of the two RCMs ICON and CCLM which are quite similar with regard to the parameterizations of the precipitation processes. Due to the limited bandwidth of the ensemble, the uncertainty can only be estimated with insufficient accuracy”. I wonder what will happen once the entire ensemble is available. Will the authors publish another article simply updating the results of the present study? I am not sure I understand the added value of such a strategy. While I understand the difficulties related to obtaining timely climate projections, I thought that one of the main motivations for this study was to report on new hydrological projections for Germany and their relationship to existing studies (CMIP6 vs. CMIP5; CPMs vs. RCMs). To address this issue, I believe the focus of the study should shift towards investigating hydrological sensitivity to climate change using convection-permitting models, global warming levels, and CMIP6 projections compared to previous modelling frameworks. In my opinion, this would be more interesting than comparing hydrological signals within an incomplete model ensemble (this is how I interpret the two research questions). The next two comments are related to how to achieve this.
2) In my opinion, using an ensemble of convection-permitting models at different GWLs and with 30-year time slices to drive a hydrological model across a large sample of catchments is quite novel. This approach provides a valuable opportunity to understand how flood magnitude changes in response to intensifying rainfall. It addresses the limitations of current studies which use regional climate models with coarser resolutions (e.g. Muelchi et al., 2021) or time slices that are too short (e.g. 10 years; Poncet et al., 2025), and which use the RCP8.5 scenario at the end of the century. This scenario is known to be quite unrealistic (Hausfather and Peters, 2020). However, the present study has a limitation in that the model was run at the daily time step, which limits the added value of using CPMs because CPMs mostly improve rainfall simulations at sub-daily time steps compared to RCMs (e.g. Akbary et al., 2026). While I understand that running the model at the hourly time step would be time-consuming, I believe it would make this work highly novel.
3) A comparison with the previous generation of hydrological projections for Germany is only made 'remotely' in the discussion section. Due to differences in the reference period and model ensemble, this comparison provides limited new insight into potential hydrological changes in the region. The authors cannot attribute the changes to the climate scenario or the higher-resolution climate model, due to a lack of comparability. A way in which the present study could make a stronger contribution to literature on hydrological changes would be to compare the previous generation of hydrological projections with those used in this study. This would require selecting model runs from the previous ensemble driven by at least the same GCM. As the previous runs are likely to have originated from a GCM-RCM combination (without a CPM), the reference period can probably be easily aligned with that used in this study. Identifying the period corresponding to 2 and 3°C global warming would be possible based on global GCM runs. However, reframing the study in this way would require access to the time series of the previous generation of hydrological projections for Germany, which the authors may not have.
4) Fifty-three gauges were used to analyse hydrological changes. However, as these gauges are mainly located along the main river stretches (see Figure 2), the hydrological variability studied here is limited. Since the model is distributed and appears to have been evaluated across a wider range of catchment areas (Figure 2), it would be interesting to report on the hydrological changes across a more diverse range of catchments.
5) Much of this study would benefit from clearer writing. I highlight a few examples below. I strongly advise the authors to use more precise language and consider the structure of each paragraph and section to clarify the key messages.
- Line 35: “Improvements with regards to earlier generations, featuring improved resolution (Haarsma et al., 2016), a broader array of models, and more sophisticated simulations of climate processes”. I do not understand this sentence.
- Line 43: “Regional climate models (RCMs) are used to refine the data from coarse-resolution GCMs into finer resolutions that better”. The link with the previous sentence is not clear.
- Lines 107-112: “The two additional variables… via bilinear interpolation”. This paragraph is confusing. It is not clear why the subject is reanalysis data here. It seems to be used for bias correction which is the topic of the next section. Some sentences are also not well structured.
- Section 2.3.3: What is the purpose of this section?
- Lines 269-281: “as the ensemble was only recently completed… in greater decreases of the discharges”. This section would benefit from some restructuring.
- Lines 293-297: “However… (under- or overestimation, increase or decrease)”. This paragraph on the limitations of bias correction was very difficult to follow, with only one citation to back up several claims.
In summary, I believe the topic is relevant to HESS, but the study needs reframing to make a stronger contribution. I suggest three potential ways to achieve this: Use the entire ensemble when available; reframe the paper around the sensitivity of flood peaks to rainfall intensification at the sub-daily scale, highlighting the added value of using CPMs over RCMs; or use the CMIP5 runs to enable better comparison and attribute the changes to the different components of the hydro-climatic modelling chain. Additionally, much of the manuscript requires clearer language and better sentence and section structures.
Minor comments:
- Title: “…Central European rivers..”. It think it would be more accurate to talk about hydrological Germany.
- Figure 0: Shouldn’t this be Figure 1 with a caption?
- Line 16: “(no major flood events in the Rhine River for 30 years with a MHQ decrease compared to the previous 30-years-period, very dry conditions in the last decade on a national level, significant regime changes in the Alpine region)”. I do not see how the fact that there have been no major floods in the Rhine for 30 years aligns with potential decreasing trends in flooding. Floods are subject to large interannual variability, so such a statement cannot support a proper trend analysis. I would suggest removing this from the abstract (and anywhere else in the text), unless there is strong evidence and peer-reviewed research to support this claim.
- Line 20: “RCMs”. I think that you could use the term “CPM” here.
- Section 2.2: Was any seasonal correction involved? It would also be useful to comment on whether these methods preserve trends or not, as this is different from making a non-stationary assumption. You mention this in the discussion section, but I think it's important to mention it earlier on.
- Section 2.3.2. From the description, it is unclear whether dams are taken into account in the modelling process. While I understand that the calibration and regionalisation process was designed to minimise the impact of dams, it would be helpful to explicitly state whether the model is run with a full 'natural' setup.
- Line 188: “To illustrate the hydrological impacts of the CMIP6 data…”. I would say “To illustrate the impact of using the CMIP6 data on hydrological projections…”
- Title of section 3.1: “Validation of reference periods”. The validation process here is not about the reference periods but about the simulations/projections during the reference period.
- Figure 6: o understand the changes better, I think it would be interesting to see maps showing changes in precipitation, potential evapotranspiration and soil moisture.
- Figure 7, x-axis title: remove “5 model combinations” as it is highlighted by the different colours.
- Line 261: “Floods in the Rhine catchment have been more moderate in the last 30 years, with a MHQ decrease compared to the previous 30-years period. » A reference is needed here.
- Line 290: “Especially in the model combinations coupled with ICON, there appears to be a west-east gradient in precipitation, since the bias correction in these model combinations tended to correct the western part upwards and the eastern part downwards. » This is not clear to me. Do the raw projections also show this spatial pattern?
- Line 292: “as a relative complex procedure”. I would remove this. I think quantile mapping is a widely used procedure that cannot be considered complex. On the other hand, understanding the effect of bias correction on hydrological projections can sometimes be difficult.
- Discussion section: I think that adding subsections would help to structure the argumentation.
- Figure A1: I would suggest expressing the precipitation bias per year, as a bias of 2,500 mm may seem quite large at first. Relative bias is also commonly used for precipitation in other studies.
References
Akbary, R., Dallan, E., Astagneau, P. C., Wood, R. R., Marra, F., Brunner, M. I., and Borga, M.: Scale-dependent biases in Alpine sub-daily areal precipitation extremes: added value of convection permitting models, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2026-1079, 2026.
Zeke Hausfather and Glen P. Peters: Emissions – the ‘business as usual’ story is misleading, Nature, 577, 618–620, https://doi.org/10.1038/d41586-020-00177-3, 2020.
Muelchi, R., Rössler, O., Schwanbeck, J., Weingartner, R., and Martius, O.: River runoff in Switzerland in a changing climate – changes in moderate extremes and their seasonality, Hydrol. Earth Syst. Sci., 25, 3577–3594, https://doi.org/10.5194/hess-25-3577-2021, 2021
Poncet, N., Tramblay, Y., Lucas-Picher, P., Thirel, G., and Caillaud, C: Projections of extreme rainfall and floods in Mediterranean basins from an ensemble of convection-permitting models. Climatic Change 178, 141. https://doi.org/10.1007/s10584-025-03983-8, 2025