Cross-scale characteristics of extreme precipitation events under climate change

Yang, Jingkun; Poschlod, Benjamin; Ament, Felix

doi:10.5194/egusphere-2026-4111

Preprints

https://doi.org/10.5194/egusphere-2026-4111

Preprints

21 Jul 2026

| 21 Jul 2026

Cross-scale characteristics of extreme precipitation events under climate change

Jingkun Yang, Benjamin Poschlod, and Felix Ament

Abstract. Extreme precipitation is intensifying globally, with Central Europe emerging as a hotspot for accelerating hydroclimatic risks. Traditional hazard metrics often fail to capture the spatial extent and compound temporal characteristics of these events. This study employs the cross-scale Weather Extremity Index (xWEI) – a metric integrating intensity, duration, and area – to systematically analyze the 200 most extreme precipitation events over Germany. Using high-resolution (3 km) convection-permitting COSMO-CLM simulations under the RCP8.5 scenario, we compare event characteristics across historical (1971–2000), present (1990–2019), and future (2031–2060; 2071–2100) climate states. Our results reveal a profound intensification: averaged over the top 200 events and relative to the historical baseline, the xWEI is projected to increase by 27 % and 45 % until the near and far futures, respectively. This is driven by increasing peak rainfall intensity, accompanied by a fundamental structural shift in the extreme-event population of the top 200 events. The composition transitions from being dominated by short-lived (1–4 h), small-scale (< 5,000 km²) events toward more persistent (12–24 h) and spatially extensive rainfall systems. Furthermore, we demonstrate that the perceived severity of these changes depends on the statistical frame of reference. When return periods are re-calibrated ("adapted") to each specific climate period, adjusting for the dominant effect of peak intensification, they reveal a second-order intensification that remains hidden in traditional assessments. This effect, diagnosed primarily between the historical and subsequent climates, manifests as a rise in the cross-duration and cross-space characteristics: events are becoming more extreme across a broader range of durations and areas simultaneously, significantly expanding the total volume of extreme precipitation. These findings highlight the emergence of more complex, high-impact hazard structures that necessitate the multi-scale assessment capabilities of the xWEI framework.

Received: 09 Jul 2026 – Discussion started: 21 Jul 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

Jingkun Yang, Benjamin Poschlod, and Felix Ament

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-4111', Anonymous Referee #1, 07 Aug 2026
General Comments
The study compares precipitation event characteristics for COSMO-CLM simulation data across different time periods, from the past to the future under the RCP8.5 scenario. The authors use the dGEV and adapted dGEV distributions to assess extreme rainfall return periods across durations, which is then applied within the xWEI framework to quantify event extremeness and identify the scale of maximum impact relevance.
While the manuscript is well written, and the authors demonstrate a strong statistical understanding, the scientific narrative requires strengthening. Specifically, there is a gap in the fundamental understanding of the climate model products (COSMO-CLM) and hydro-climatological phenomena that underpins the post-processing statistical frameworks. Therefore, there are methodological limitations that arise, which call into question the key findings and interpretations of the results - especially regarding the shift in trend towards more persistent and spatially extensive rainfall systems. These concerns have been detailed systematically below.
Specific Comments
In Line 115, the authors mention the use of a duration specific intensity threshold that is based on absolute values using the DWD severe weather warning threshold. While this is useful as a warning system, it is less than optimal when the focus is multi-hazard hydroclimatic risks. There is a geographic bias when statistically evaluated over large regions. Some areas (like the Alps) may naturally receive more rainfall, the lowland areas receive much lower rainfall on average, but also have higher economic activity (more dense, agricultural activity, etc.), and therefore far more economic damage potential.For this reason, if the study that chooses an absolute threshold with the top 200 extreme events, it would introduce a spatial bias. The authors should consider transitioning to a grid cell specific percentile baseline (like 99^th percentile per duration per grid cell) to normalize the spatial heterogeneity or provide a rigorous statistical justification to prove that the current analysis does not skew the resulting datasets towards specific geographic sub-regions.

Please clarify what are the statistical scores (such as RMSE, Bias, MAE, FSS, SR, Taylor Diagrams) for extreme precipitation of the COSMO CPM model used, with respect to observed datasets, including, if necessary, spatial maps indicating these scores. Model biases in COSMO are predominant across sub-regions due to terrain, orographic changes, etc. Therefore, a positive bias in the Alps compared to a negative bias in the lowlands, for instance, could skew the results further towards specific sub-regions when using CPM datasets. Furthermore, the performance of the CPM model may drop significantly when considering extreme events. This should be justified in the study with respect to observed data, with the proper use of citations and/or analysis.

In Line 223, the authors indicate that each time-period is analyzed independently. However, Voit and Heistermann (2022) studied specific regions, and adapting the framework for a shifting regional analysis every time period must be justified more thoroughly. For instance, does the shift in the extremity of the events still hold if we consider grids fixed over all the time periods analyzed? Or is the changing atmospheric current shifting the locations of the storm at the subregional scale? It may be better if the authors provide a separate sensitivity analysis, by performing an analysis over a few fixed selected sub-regions across all the time periods, or this should be justified thoroughly. In addition, the authors should provide individual maps indicating the locations of the top 200 events for each of the four time-periods.

An indication that the composition of the top-ranked subset events shifts from smaller, short-lived events to a larger and more persistent events would suggest that the subset is becoming more dominated by synoptic-scale events. Such an analysis should be supported by a seasonal classification, which would clarify by how much the seasons dominated by synoptic scale events change over time, compared to the increase in convective scale events. Combining these scales into the same structural category for analysis provides an incomplete result. Because synoptic scale events are generally much larger than convective events in both duration and volume, an increase in both would skew the representation of the entire subset towards synoptic scale events when selecting only the top 200 most extreme events, due to the sheer mathematical weightage that synoptic scale events possess. The issue of extreme precipitation varying seasonally and regionally has been mentioned in previous studies (Zhong et al., 2025). Please justify.

In continuation with the previous point, the shift of the results from smaller and short-lived events to larger and persistent events is also a known bias for models called the ‘drizzle effect’ or a ‘precipitation persistence’. This effect should also manifest during extreme events. Specifically, modelled datasets often tend to stretch high high-intensity, short-duration events over a longer durations and larger spatial volume (Carlo et al., 2026; Moon et al., 2019). Unless prior validation for this specific COSMO configuration have demonstrated that this smoothing bias is not present, addressing this issue would be critical for the credibility of the results.

The use of RCP8.5 (or SSP5-8.5) may be unjustified. The scenario framework for CMIP7 has recently considered it implausible, and intends to have a high emissions scenario that is expected to result in forcings lower than SSP5-8.5 (Van Vuuren et al., 2026). Given that the novelty of this study is central CPM dataset use for climate projections, it would be beneficial instead to understanding these analyses for more realistic climate projections, such as SSP2-4.5 or SSP3-7.0, rather than SSP5-8.5 that may now be considered obsolete.

The authors have noted that the MIROC5 model is considered a median performing model, yet it is among the ‘driest’ in the CMIP5 ensemble and show a notable warm and dry summer bias. It is unclear how this, in combination with the previous point about the extreme SSP5-8.5 scenario, would compound in the case of extreme precipitation events. It would be better if the study used driving models that are median performance specifically in terms of their representation of convective precipitation. The results for an analysis of a dry-driven model subject to an implausible extreme SSP5-8.5 scenario may not give satisfactory results. Moreover, if the model shows a dry bias in summer - a season known for convective scale events - then this would further skew the composition of the results of extreme events towards synoptic scale events as highlighted in Point 4 previously.

The ‘second-order effect’ is not clearly understood from this study, and we feel the authors should elaborate on this more. Does this mean the intensification for short and medium duration of rainfall as mentioned in Line 288? Because this is in contrast with the conclusions that indicate a shift towards medium and long duration events.

In estimating the dGEV and adapted dGEV parameters in Section 3.1.2, the authors should also elaborate on any uncertainties when estimating the parameters or sensitivity to outlier events, and whether there was any evaluation on how this could affect the return period.

Technical Corrections
Line 161: Maybe it would be better to rephrase the concluding line as follows instead: “Hence, we refrain from mixing dGEV parameters from COSMO-CLM driven by MIROC5 (1971–2000) with the simulation driven by ERA5.”

References
Carlo, O., Fedele, G., Stocchi, P., Calmanti, S., Raffa, M., Mercogliano, P., and Lionello, P.: A nested time-window framework for evaluating sub-daily precipitation extremes in two convection permitting model datasets, https://doi.org/10.5194/egusphere-2026-4515, 30 July 2026.
Moon, H., Gudmundsson, L., Guillod, B. P., Venugopal, V., and Seneviratne, S. I.: Intercomparison of daily precipitation persistence in multiple global observations and climate models, Environ. Res. Lett., 14, https://doi.org/10.1088/1748-9326/ab4169, 2019.
Voit, P. and Heistermann, M.: A new index to quantify the extremeness of precipitation across scales, Nat. Hazards Earth Syst. Sci., 22, 2791–2805, https://doi.org/10.5194/nhess-22-2791-2022, 2022.
Van Vuuren, D. P., O’Neill, B. C., Tebaldi, C., Sanderson, B. M., Chini, L. P., Friedlingstein, P., Hasegawa, T., Riahi, K., Govindasamy, B., Bauer, N., Eyring, V., Fall, C. M. N., Frieler, K., Gidden, M. J., Gohar, L. K., Högner, A., Jones, A. D., Kikstra, J., King, A., Knutti, R., Kriegler, E., Lawrence, P., Lennard, C., Lowe, J., Mathison, C., Mehmood, S., Nicholls, Z., Prado, L. F., Zhang, Q., Rose, S. K., Ruane, A. C., Sandstad, M., Schleussner, C. F., Seferian, R., Sillmann, J., Smith, C., Sörensson, A. A., Panickal, S., Tachiiri, K., Vaughan, N., Vishwanathan, S. S., Yokohata, T., Zecchetto, M., and Ziehn, T.: The Scenario Model Intercomparison Project for CMIP7 (ScenarioMIP-CMIP7), Geosci. Model Dev., 19, 2627–2656, https://doi.org/10.5194/gmd-19-2627-2026, 2026.
Zhong, P., Brunner, M., Opitz, T., and Huser, R.: Spatial Modeling and Future Projection of Extreme Precipitation Extents, J. Am. Stat. Assoc., 120, 80–95, https://doi.org/10.1080/01621459.2024.2408045, 2025.
Citation: https://doi.org/10.5194/egusphere-2026-4111-RC1
RC2: 'Comment on egusphere-2026-4111', Anonymous Referee #2, 10 Aug 2026

The authors analyse the effect of climate change on the cross-scale extremity of precipitation events using the COSMO-CLM convection permitting model and the xWEI. They introduce two different analyses based on different fits of the dGEV, which aim to account for evolving rainfall characteristics. Generally, I think that it makes sense to assess extreme precipitation events across scales with metrics such as the xWEI, especially if the aim is to understand hydrological impacts. While it is a good idea to split the analysis into two different dGEV fits, the study has several shortcomings, which are listed below. In my opinion, the study, in its current version, does not meet NHESS standards.
Please see attached PDF for the detailled review

Citation: https://doi.org/10.5194/egusphere-2026-4111-RC2
RC3: 'Comment on egusphere-2026-4111', Anonymous Referee #3, 12 Aug 2026

General comments:

The manuscript provides an analysis of the characteristics of extreme precipitation in the past, present and future climate. In the study well established methods for detection and categorisation of the extremity of rainfall events are applied to COSMO-CLM simulations under the RCP8.5 scenario spanning four different 30-year periods in the past, the present, the near and the far future. For each period the 200 most extreme events according the cross-scale Weather Extremity Index (xWEI) are investigated in terms of changes in durations and spatial extent. A shift to towards larger and more persistent events is found in the present, the near and far future compared to the past.

The study fits in the scope of NHESS. It is well-written, the goal and methods are clearly explained and the results are easy to follow and well-illustrated. However, the choice of the dataset used is questionable and should be reconsidered before publishing the manuscript. Please, find my comments and suggestions below:
Specific comments:

• The authors chose COSMO-CLM simulations forced by boundary conditions from MIROC5 for three of the four datasets investigated. As the authors state themselves, MIROC5 is among the driest models in the CMIP5 ensemble. Therefore, it might not be the best choice for studying changes in extreme precipitation. Expanding the dataset to include further simulations forced by other boundary conditions would significantly strengthen the conclusions drawn in this study.

• Four time periods are investigated: 1971-2000 (past), 1990-2019 (present), 2031-2020 (near future) and 2071-2100 (far future). Three of the four periods are based on COSMO-CLM forced by boundary conditions from MIROC5, but the simulations representing the present are based on COSMO-CLM driven with ERA5. Therefore, only three of the periods are directly comparable. Especially in the comparison with adapted dGEV it is not clear to me, how much of the change, e.g., between the past and the present is related to an actual shift in the characteristics of extreme precipitation events and how much is related to the differences in the model simulations. A direct comparison of the past (1971-2000) between COSMO-CLM / MIROC5 and COSMO-CLM ERA5 would be beneficial for this study to examine the differences between the simulations.

• The study deals only with the 200 most extreme events. Could the authors also comment on the expected structural changes in precipitation events in general and set that into perspective with the extreme events?
Technical comments:

P.3, L74: There are two references for Haller et al. (2022). Which one is meant here?

P.4, L.124-125: Here it is written that the minimum size of an event must be 9 km² for durations of 3 hours or less and 3 km² times the duration for longer events. The model has a resolution of 3 km, therefore a grid cell has 9 km². Does that mean that an event with a duration of 4 hours must cover at least one grid cell or two?

P.11, L.260-261: Are the authors really refer to Figure 5e and f here? In the previous sentence they compare the historical to the present climate which would be Figure 5a and d.

Fig. 4 and 6: The blue and the black dot are hard to distinguish in the legend. I would suggest using more distinct colours or also different symbols.

Fig. 4 and 6: For both, fixed and adapted dGEV, it seems like the most extreme events are quite similar or even weaker in the future compared to the past. Is there any explanation for this behaviour?

P.16, L.400-405: The references for Lenderick et al. 2021 und 2017 are not arranged in the alphabetic order.

P.17, L432-434: Lengfeld et al. (2023) is not cited in the manuscript.

Citation: https://doi.org/10.5194/egusphere-2026-4111-RC3

Jingkun Yang, Benjamin Poschlod, and Felix Ament

Model code and software

xWEI calculation for the top 200 precipitation events over Germany based on convection-permitting climate simulations Jingkun Yang https://doi.org/10.5281/zenodo.19723990

Jingkun Yang, Benjamin Poschlod, and Felix Ament

Viewed

Total article views: 243 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
159	61	23	243	20	17

HTML: 159
PDF: 61
XML: 23
Total: 243
BibTeX: 20
EndNote: 17

Views and downloads (calculated since 21 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	98	23	9	130
Aug 2026	37	12	5	54
Sep 2026	24	26	9	59

Cumulative views and downloads (calculated since 21 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	98	23	9	130
Aug 2026	37	12	5	54
Sep 2026	24	26	9	59

Viewed (geographical distribution)

Total article views: 218 (including HTML, PDF, and XML) Thereof 218 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 22 Sep 2026

Short summary

Heavy rainfall is becoming more severe, yet common measures capture only its peak strength, not how long it lasts or how wide an area it covers. Using high-resolution climate simulations for Germany, we rank the 200 most extreme events across past, present and future. We found that future events are not only more intense but also last longer and spread over larger areas. This means future rainfall hazards will be more complex, and assessments must consider strength, duration and extent together.


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