the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Aug 2026
Observed and projected changes in extreme precipitation intensity and temporal patterns: a case study of the Misa River Basin (Central Italy)
Abstract. Extreme hydrological events such as floods are intensifying under climate change, yet uncertainties in local-scale rainfall projections impede reliable adaptation strategies. This study examines future rainfall extremes and their daily timing in the Misa River basin (Central Italy) using the Weather Research and Forecasting (WRF) and Consortium for Small-scale Modeling in Climate Mode (COSMO-CLM) models, together with rain gauge observations, comparing historical periods (1979–2008 and 1981–2008, respectively) with a future period (2039–2068) under RCP8.5. Mean-state precipitation is projected to decrease annually and in summer, while November average daily precipitation increases. In contrast, extreme precipitation shows a different pattern: both models project more frequent extremes relative to their own historical baselines, with a historical 100-year event becoming 10–20 times more frequent by mid-century. However, the models diverge when evaluated against observed historical flood events: WRF shows a broadly uniform response, with return periods for past events reduced by nearly 50 % and annual exceedance probability doubling, while COSMO shows an event-dependent response, with events remaining within the 95 % confidence bound. Sub-daily event magnitudes, including average peak intensity and nine-hour maximum accumulation, decrease in both models. Overall, both models agree that extremes intensify in a model-relative sense, but this agreement breaks down against observed benchmarks and sub-daily magnitudes, reflecting model-chain spread rather than uniform intensification. These results suggest that historically based design standards may underestimate future rainfall hazards under some projections, underscoring the need to account for model-chain uncertainty in flood risk planning across the Misa River basin.
- Preprint
(3088 KB) - Metadata XML
-
Supplement
(1700 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-4780', Anonymous Referee #1, 26 Aug 2026
-
AC1: 'Reply on RC1', Dawit Meskele, 14 Sep 2026
Reply to referee comments RC1.
Thank you for considering our study interesting and relevant to the regional and local context. We also thank the reviewer for highlighting the limitations of the study. Addressing these comments gave us the opportunity to strengthen our analysis, improve the robustness of the results, and ultimately provide a stronger manuscript. All comments have been addressed in the relevant sections point-by-point, and our responses are in "Bullet." The changes introduced in the revised manuscript are shown in "bullet italics."
1. Lines 190–240, 611–666: Historical events are reconstructed using 1.5-km ERA5-driven WRF, whereas future events come from 4-km WRF and ~2.2-km COSMO simulations. Please perform WRF-historical vs. WRF-future and COSMO-historical vs. COSMO-future comparisons using identical model configurations.- We accepted and acknowledged this comment. We clarify that our core extreme-value analysis already uses matched same-model configurations: Figure 8 compares COSMO-historical vs. COSMO-RCP8.5 (~2.2 km) and WRF-historical vs. WRF-RCP8.5 (4 km), each from continuous 30-year simulations within the same modeling chain, isolating the climate signal as recommended. The noted resolution mismatch applies specifically to the event-level structural comparison, where the eight historical benchmarks are event-scale hindcasts compared against future extremes drawn from the continuous WRF and COSMO climate runs. Restricting the historical benchmarks to only those falling within the model's 30-year historical period would exclude most of the catastrophic flood-causing events (6 of 8), while extending the continuous historical simulation to cover the full 82-year observational record at 1.5-km resolution would require a new multi-year, convection-permitting integration beyond this study's computational scope.
- We agree that this limits attribution of storm-structure differences exclusively to climate change for this comparison. Figure 8 and its matched-model GEV analysis were already part of the original manuscript; this matched-configuration comparison is distinct from the event-level structural comparison, which uses different-resolution modeling chains by design and is interpreted accordingly. Moreover, the 1.5-km hindcasts are devoted to reconstructing the observed meteorological evolution of specific events at high spatiotemporal resolution, whereas the future events emerge from independent, free-running climate trajectories; rerunning only the observed event dates at 4-km/2.2-km resolution would therefore test model sensitivity but would not constitute a paired historical–future climate experiment.
- Lines 311–331, 531–600: Please clarify how “basin-maximum precipitation” is defined.
- We thank the reviewer for this comment. Basin-maximum precipitation on a given day is taken as the highest daily precipitation value recorded among all grid (modeled) or gauge stations (observed) within the basin boundary. This approach is more appropriate for extreme hazard analysis because it captures the wettest location within the basin and avoids diluting local precipitation extremes through spatial averaging.
- We added these updates to the revised manuscript in section 2.3.2. “...Basin-maximum precipitation on a given day is taken as the highest daily precipitation value recorded among all modeled grid or gauge stations within the basin boundary. The annual maximum series used for GEV fitting was then obtained by taking the annual maximum of this daily basin-maximum time series...”
- Lines 255–295: With only five long-term gauges, IDW may strongly influence spatial patterns. Please provide the full workflow and a sensitivity analysis of interpolation assumptions.
- We acknowledged this comment. As this sensitivity analysis is the robustness of the study, we have summarized the workflow and rationale in the Methods (Section 2.3.1) of the revised manuscript, with full leave-one-out cross-validation results and the comparison of IDW against nearest-neighbor, elevation-regression, and unweighted-average interpolation provided in the Supplementary Material (Table S2, Fig. S2):
- “...Given the limited number of gauges (five stations) in the basin, the sensitivity of the IDW-based interpolation (p = 2) to interpolation assumptions was assessed through leave-one-out cross-validation and by comparison against nearest-neighbor/Thiessen assignment, unweighted spatial averaging, and elevation-regression interpolation, and their results are provided in the Supplementary Material (Table S2, Fig. S2). This sensitivity analysis indicated that basin-representative return levels varied by no more than ~6% across the IDW power parameter and by ~4% when compared against nearest-neighbor and elevation-regression interpolation, supporting IDW as appropriate for this gauge network...”
- Lines 286–299: Daily bias correction followed by redistribution using raw-model hourly fractions doesn’t independently correct hourly precipitation structure. Validate sub-daily characteristics against hourly observations where possible or clearly state this limitation when interpreting future hourly extremes.
- We thank the reviewer for this comment. We agree and have added the following clarification in Section 4.
- We have added an explicit statement of this limitation in Section 4, in the revised manuscript: “... The method of fragments preserves the raw model's hourly timing but does not independently correct sub-daily structure; in the absence of hourly gauge observations for validation, sub-daily statistics should be interpreted as inherited from the raw model rather than independently bias-corrected. More broadly, robust validation of sub-daily precipitation structure in CPMs requires high-resolution hourly observations from both rain gauge networks and gridded observational products; such data remain notably limited across much of Italy, and their expansion represents an important direction for future model evaluation in basins....”
- Lines 355–361: The choice of a 70-mm threshold and exactly eight future events need stronger justification.
- We thank the reviewer for this valuable suggestion; providing a more detailed explanation of our methodological choices strengthens the manuscript by making the rationale clearer. We have clarified the rationale for both the 70-mm threshold and the selection of eight future events in Section 2.3.2 of the revised manuscript:
- "...A minimum daily precipitation threshold of 70 mm for the basin peak was adopted, corresponding to the lowest daily peak recorded among the eight historical flood-causing events (71.4 mm, which has a return period of 1.8 years), rounded down to a conservative value. The threshold was applied independently to each future model output over the projection period, identifying 19 exceedance events in COSMO and 24 in WRF. The eight future events with the highest basin-peak precipitation from each model were selected for comparison with the eight benchmark events, thereby providing equal-sized samples for a consistent comparison of their rainfall characteristics. This fixed-rank selection was performed independently for each model and was intended solely to identify comparable upper-tail events for event-based analysis, rather than to characterize projected exceedance frequency, which is assessed separately using the complete exceedance populations and the GEV-based analysis..."
- Lines 778–799: The sensitivity analysis shows that some 9-h windows capture only 40–60% of daily precipitation. Please complement the 9-h analysis with full-event or 24-h rainfall characteristics, especially when discussing flood impacts.
- We thank the reviewer for this comment. In Section 3.2.3, we have added concrete hourly intensity evidence for events of 40–60% of daily, which has lower 9-hour core capture (e.g., Historical E06, COSMO E04, WRF E01, etc.), showing their secondary peaks and full-day structure (Fig. S8), which complements the 9-hour analysis with full-event, 24-hour characteristics. We have also added pointers to this 24-hour characterization at the two points in the mass-curve discussion (Sect. 3.2.2). These updates appear in Sections 3.2.2 and 3.2.3.
- Lines 531–600: Table 3 contains extremely wide confidence intervals. Therefore, precise values such as 182→97 or 182→860 years should not be overemphasized.
- We thank the reviewer for this comment. We agree that the wide confidence intervals warrant caution in interpreting the point estimates. We have revised the text so that every return period value is explicitly framed as a "central estimate" rather than a precise value, and we direct the reader to the full 95% confidence intervals in Table 3 at each relevant point in the discussion.
- These updates are made in Section 3.1.2 (ii) in the revised manuscript.
- Lines 819–859 and Abstract: The models strongly disagree for observed benchmark events, sub-daily comparisons contain model-chain uncertainty, and no hydrological model is used. Please distinguish clearly between projected precipitation hazard and actual flood risk.
- To clearly state model-chain uncertainty and distinguish projected precipitation hazard from flood risk, we replace "flood risk" terminology with "precipitation/rainfall hazard" throughout, since no hydrological and/or hydraulic simulation was done in this manuscript, but it’s suggested for future/other works.
- We clearly made these changes in the revised abstract and relevant body sections.
- Please check consistency in terminology, including “CPM,” “RCM,” “extreme event,” “benchmark event,” and “historical event,” as these terms are sometimes used interchangeably.
- We acknowledge highlighting this terminology issue. We have clarified and standardized the terminology throughout the manuscript. “RCM” refers broadly to regional climate models, while “CPM” is used for the convection-permitting simulations (finer resolutions). “Benchmark events” refers specifically to the eight documented flood events, whereas “extreme events” refers to precipitation values that are extremely relative to a defined statistical distribution or threshold. The term “historical event” has been avoided where it does not denote a distinct methodological category.
Citation: https://doi.org/10.5194/egusphere-2026-4780-AC1
-
AC1: 'Reply on RC1', Dawit Meskele, 14 Sep 2026
-
RC2: 'Comment on egusphere-2026-4780', Anonymous Referee #2, 13 Oct 2026
This manuscript examines projected changes in precipitation climatology, daily extremes, and intra-storm rainfall patterns in the Misa River basin using rain-gauge observations, two convection-permitting climate simulations, and event-specific WRF hindcasts. The topic is relevant to Natural Hazards and Earth System Sciences, particularly given the basin’s exposure to damaging rainfall events. Connecting future precipitation distributions with recognizable historical benchmarks is potentially useful for communicating local rainfall hazards.
The manuscript also makes a welcome effort to distinguish model-relative changes from comparisons against observed events and explicitly acknowledges substantial uncertainty. However, several central results are not yet sufficiently supported:
1. Section 2.3.2. There are two related concerns. First, Eq. (2) appears to combine incompatible GEV shape-parameter conventions. The authors should identify the convention and provide the fitted parameters. Second, Tables 2 and 3 cannot represent the same future distributions as currently presented. For example, Table 2 gives COSMO future return levels of 165.57 mm and 179.01 mm for 200 and 500 years, respectively. A threshold of 170.0 mm must therefore have a return period between those values, not 79.3 years, if the same fitted distribution is used.2. Lines 329–343: Table 3 compares a distribution fitted to observations from 1951–2023 with future model distributions for 2039–2068. The models are calibrated against substantially shorter historical windows ending in 2008. Consequently, the comparison combines projected changes within each model chain and residual differences between modelled and observed extremes. Differences in the periods、spatial sampling and representation may also alter the results. Thus, a larger future COSMO return period than the observed historical estimate does not, by itself, establish that climate change makes that rainfall magnitude less frequent. It may also reflect a remaining baseline mismatch. Use a common historical window as a sensitivity test, while retaining the longer observational record as valuable additional context may help. Additionally, the statements that severe historical thresholds become less frequent under COSMO should be revised unless that conclusion is demonstrated relative to COSMO’s own historical distribution.
3. “Basin maximum,” “basin average,” “maximum hourly intensity,” and “maximum cumulative rainfall” refer to different quantities, but their construction is not sufficiently clear.
4. Section 3.1.1: The present description suggests that corrected series at five station locations are interpolated to reconstruct the full model domain. If so, the resulting spatial field is substantially determined by those five series and the interpolation weights. This would require qualification of the claim that the method preserves the model’s physically coherent spatial gradients.5. The historical event sample comes from 1.5 km ERA5-driven WRF hindcasts, whereas future events come from 4 km EC-Earth-driven WRF and approximately 2.2 km CMCC-CM-driven COSMO. These comparisons change resolution, forcing, model configuration, and climate period simultaneously. The sampling design also needs revision. Eight documented flood-producing events collected over many decades are not statistically equivalent to eight rainfall-ranked events selected from a 30-year simulation. Equal sample size does not ensure comparable rarity, seasonality, or selection criteria.
Minor comments:
1. Titled “Observed and projected changes” implies an analysis of changes within the observational record. But the observations are mainly used for calibration and benchmarking. Either add an appropriate observational change analysis or revise the title and objectives to match the work performed.
2. Provide actual initialization times, run lengths, and retained analysis intervals.
3. State the observation-day boundary, treatment of storms crossing midnight, and whether “daily extremes” mean calendar-day totals or rolling 24-hour maxima.
Therefore, I recommend major revision requiring substantial reanalysis.Citation: https://doi.org/10.5194/egusphere-2026-4780-RC2 -
AC2: 'Reply on RC2', Dawit Meskele, 13 Nov 2026
Reply to referee comments for RC2.
This manuscript examines projected changes in precipitation climatology, daily extremes, and intra-storm rainfall patterns in the Misa River basin using rain-gauge observations, two convention-permitting climate simulations, and event-specific WRF hindcasts. The topic is relevant to Natural Hazards and Earth System Sciences, particularly given the basin exposure to damaging rainfall events. Connecting future precipitation distributions with recognizable historical benchmarks is potentially useful for communicating local rainfall hazards.
The manuscript also makes a welcome effort to distinguish model-relative changes from comparisons against observed events and explicitly acknowledges substantial uncertainty. However, several central results are not yet sufficiently supported:- We thank the reviewer for considering our study interesting and relevant to the regional and local context. We also appreciate the reviewer’s constructive comments on the study’s limitations, which have helped us strengthen the analysis, improve the robustness of the results, and enhance the overall quality of the manuscript. All comments have been addressed point by point. For clarity, the reviewer’s comments are presented in numbered format, followed by our responses in bullet points. Corresponding revisions have also been incorporated into the manuscript, as appropriate.
- Section 2.3.2. There are two related concerns. First, Eq. (2) appears to combine incompatible GEV shape-parameter conventions. The authors should identify the convention and provide the fitted parameters. Second, Tables 2 and 3 cannot represent the same future distributions as currently presented. For example, Table 2 gives COSMO future return levels of 165.57 mm and 179.01 mm for 200 and 500 years, respectively. A threshold of 170.0 mm must therefore have a return period between those values, not 79.3 years, if the same fitted distribution is used.
- We thank the reviewer for identifying this issue and apologize for the confusion in the notation of the discussion version. First, regarding the GEV Convention & Fitted Parameters, We use the Coles (2001) convention, for which ξ > 0 denotes a heavy-tailed Fréchet distribution, ξ = 0 the Gumbel limit, and ξ < 0 a Weibull-type distribution with a finite upper bound in the updated manuscript in Section 2.3.2, and in addition, the fitted GEV parameters for all observational baselines and model runs have been added to Table S5 in the Supplementary section. Second, Harmonization of Table 2 and Table 3: The reviewer is entirely correct that a 170.0 mm threshold must fall between the 200-year (165.57 mm), and 500-year (179.01 mm) return levels. The previous value of 3 years was an artifact from an earlier draft; we apologize for the issue. We have unified the analysis results so that all tables and figures derive from identical fitted distributions in section 3.1.3. Under the corrected distribution, the future return period for the 170.0 mm event under COSMO RCP8.5 is 269 years [95% CI: 30 to >1000 yr], which strictly obeys monotonicity with Table 2. Table 3, Figure 9, and the accompanying text have been updated in the new version of the manuscript in section 3.1.3.
- Lines 329–343: Table 3 compares a distribution fitted to observations from 1951–2023 with future model distributions for 2039–2068. The models are calibrated against substantially shorter historical windows ending in 2008. Consequently, the comparison combines projected changes within each model chain and residual differences between modelled and observed extremes. Differences in periods of spatial sampling, and representation may also alter the results. Thus, a larger future COSMO return period than the observed historical estimate does not, by itself, establish that climate change makes that rainfall magnitude less frequent. It may also reflect a remaining baseline mismatch. Use a common historical window as a sensitivity test, while retaining the longer observational record as valuable additional context may help. Additionally, the statements that severe historical thresholds become less frequent under COSMO should be revised unless that conclusion is demonstrated relative to COSMO’s own historical distribution
- We thank the reviewer for this important observation. We agree that a direct comparison between the long observational record (1951–2023) and the future model distributions (2039–2068) combines the projected climate signal with differences in sampling period, spatial representation, and model bias. The available historical simulations cover 1979–2008 for WRF and 1981–2008 for COSMO. We have therefore revised the interpretation so that projected changes are assessed within each modelling chain, comparing their historical data periods with 2039–2068 for COSMO. The observational record is retained as an independent benchmark for the magnitude and historical context of the documented events, rather than as the baseline from which the climate-change signal is inferred. Table 3 and the related text have been revised accordingly, and statements that interpreted differences between the observed historical distribution and future COSMO distribution as a direct climate-change signal have been removed or rephrased.
- “Basin maximum,” “basin average,” “maximum hourly intensity,” and “maximum cumulative rainfall” refer to different quantities, but their construction is not sufficiently clear.
- We thank the reviewer for this comment. In response, we have revised the relevant sections of the manuscript to clarify the meaning and calculation of the four-rainfall metrics (basin maximum rainfall, basin average rainfall, maximum hourly intensity, and maximum cumulative rainfall). The corresponding clarifications have been incorporated directly into the main text at the relevant locations in the revised manuscript, ensuring that the distinction between these spatial and temporal metrics is clearer.
- Section 3.1.1: The present description suggests that the corrected series at five station locations are interpolated to reconstruct the full model domain. If so, the resulting spatial field is substantially determined by those five series and the interpolation weights. This would require qualification of the claim that the method preserves the model’s physically coherent spatial gradients.
- We thank the reviewer for this perceptive observation. We revised the description in Section 3.1.1 to better clarify the role and limitations of gauge-anchored spatial interpolation. We clarified the robustness of catchment-scale integrated totals and extreme return levels based on the sensitivity analysis (Table S2 and Fig. S2). These sensitivity tests indicate that the basin-scale metrics used here are comparatively insensitive to the interpolation methods tested, while local spatial gradients remain affected by the sparse gauge network and the interpolation scheme.
- The revised paragraph in the manuscript reads: "...Quantile-based coupled with spatial interpolation effectively resolves localized seasonal amplitude discrepancies across the basin, bringing both models into close agreement with the observed baseline. While reconstructing continuous gridded fields from a sparse five-station network inevitably smooths sub-grid convective variability and reflects the distance-weighting geometry of the gauges, sensitivity evaluations (Table S2, Fig. S2) confirm that catchment-scale integrated totals and extreme return levels remain robust (varying by less than ~6%). By accounting for interpolation smoothing, this approach improves the agreement with observations across the catchment and reduces systematic distributional errors while preserving the temporal dynamics of the climate projections"
- The historical event sample comes from 1.5 km ERA5-driven WRF hindcasts, whereas future events come from 4 km EC-Earth-driven WRF and approximately 2.2 km CMCC-CM-driven COSMO. These comparisons change resolution, forcing, model configuration, and climate period simultaneously. The sampling design also needs revision. Eight documented flood-producing events collected over many decades are not statistically equivalent to eight rainfall-ranked events selected from a 30-year simulation. Equal sample size does not ensure comparable rarity, seasonality, or selection criteria.
- We thank the reviewer for this important comment. We acknowledge that the historical and future simulations differ in spatial resolution, driving data, and regional climate model configuration. Consequently, differences between the simulations cannot be attributed exclusively to climate change, because the effects of these modelling differences cannot be isolated within the present experimental design. Regarding event selection, the eight historical benchmark events were documented daily extreme-precipitation events associated with flooding caused over a 70-year period, whereas the eight future events were selected from the upper tail of simulated daily precipitation extremes of 30-year periods for each model.
- We acknowledge that these samples were selected using different criteria and that equal sample sizes do not ensure comparable rarity, return periods, or seasonality. The purpose of the event-based analysis is to examine the hourly rainfall characteristics of selected historical and future extremes, including hourly intensity, cumulative rainfall, and temporal rainfall patterns within a nine-hour window. We therefore interpret this analysis as a comparison of event characteristics, rather than a statistically equivalent comparison of historical and future event populations. The distribution-based analysis of precipitation extremes and return levels is considered separately, and projected changes are assessed primarily by comparing historical and future simulations within each modelling chain. The event-based analysis is retained to investigate the hourly rainfall characteristics of selected historical and future extremes, rather than to establish statistically equivalent event samples. We have clarified this distinction and the limitations arising from the different event-selection criteria in the revised manuscript in Section 4. Accordingly, the results are interpreted as indicative of potential changes in rainfall temporal structure, rather than as a statistically exhaustive assessment of changes in the event of rarity or frequency.
Minor comments:
- Titled “Observed and projected changes” implies an analysis of changes within the observational record. But the observations are mainly used for calibration and benchmarking. Either add an appropriate observational change analysis or revise the title and objectives to match the work performed.
- We thank the reviewer for the observation. To align the title more precisely within the scope of the study, we have revised the title to: “Projected changes in extreme precipitation and temporal patterns in the Misa River basin (Central Italy): A multi-model evaluation against observed benchmarks”. In addition, we have also harmonized the abstract and stated research objectives in the Introduction section to reflect this focus.
- Provide actual initialization times, run lengths, and retained analysis intervals.
- We thank the reviewer for this comment. Each historical event was simulated as an individual 48-hour WRF run initialized at 00 UTC on the day preceding the event day, using ERA5 for initial and hourly lateral boundary conditions. The first 6 hours were treated as spin-up, and the 00-24 UTC interval of the event day was retained for the event analysis. These model setups are clearly written in section 2.2.2 for the hindcasted event, whereas the WRF model simulation for the claimed change scenario for section 2.2.3 was found in Napoli et al. (2023).
- State the observation-day boundary, treatment of storms crossing midnight, and whether “daily extremes” mean calendar-day totals or rolling 24-hour maxima.
- We thank the reviewer for this comment. We have revised the relevant text (Section 2.3.2) to explicitly clarify the observation-day boundary (00:00–24:00 Local Time), the treatment of storms crossing midnight, and the definition of daily extremes. We now specify that daily rainfall totals are based on fixed calendar days rather than rolling 24-hour maxima, consistently applied to both rain gauge observations and regional climate model outputs (WRF and COSMO-CLM).
Citation: https://doi.org/10.5194/egusphere-2026-4780-AC2
-
AC2: 'Reply on RC2', Dawit Meskele, 13 Nov 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 225 | 147 | 92 | 464 | 101 | 80 | 81 |
- HTML: 225
- PDF: 147
- XML: 92
- Total: 464
- Supplement: 101
- BibTeX: 80
- EndNote: 81
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript investigates historical and projected changes in extreme precipitation in the Misa River Basin. The topic has regional relevance and may provide useful information for local flood-risk assessment and climate adaptation. However, several methodological and statistical issues limit the robustness of the results. Therefore, substantial revisions are required before the manuscript can be considered for publication.
1. Lines 190–240, 611–666: Historical events are reconstructed using 1.5-km ERA5-driven WRF, whereas future events come from 4-km WRF and ~2.2-km COSMO simulations. Please perform WRF-historical vs. WRF-future and COSMO-historical vs. COSMO-future comparisons using identical model configurations.
2. Lines 311–331, 531–600: Please clarify how “basin-maximum precipitation” is defined.
3. Lines 255–295: With only five long-term gauges, IDW may strongly influence spatial patterns. Please provide the full workflow and a sensitivity analysis to interpolation assumptions.
4. Lines 286–299: Daily bias correction followed by redistribution using raw-model hourly fractions does not independently correct hourly precipitation structure. Validate sub-daily characteristics against hourly observations where possible or clearly state this limitation when interpreting future hourly extremes.
5. Lines 355–361: The choice of a 70-mm threshold and exactly eight future events needs stronger justification.
6. Lines 778–799: The sensitivity analysis shows that some 9-h windows capture only 40–60% of daily precipitation. Please complement the 9-h analysis with full-event or 24-h rainfall characteristics, especially when discussing flood impacts.
7. Lines 531–600: Table 3 contains extremely wide confidence intervals. Therefore, precise values such as 182→97 or 182→860 years should not be overemphasized.
8. Lines 819–859 and Abstract: The models strongly disagree for observed benchmark events, sub-daily comparisons contain model-chain uncertainty, and no hydrological model is used. Please distinguish clearly between projected precipitation hazard and actual flood risk.
9. Please check consistency in terminology, including “CPM,” “RCM,” “extreme event,” “benchmark event,” and “historical event,” as these terms are sometimes used interchangeably.