Radar-Derived Intensity-Duration-Area-Frequency Relations for Assessing Hydrological Hazards in Complex Terrain

Rosin, Talia; Marra, Francesco; Gabella, Marco; Germann, Urs; Wolfensberger, Daniel; Morin, Efrat

doi:10.5194/egusphere-2026-1800

Preprints

https://doi.org/10.5194/egusphere-2026-1800

Preprints

18 May 2026

| 18 May 2026

Status: this preprint is open for discussion and under review for Hydrology and Earth System Sciences (HESS).

Radar-Derived Intensity-Duration-Area-Frequency Relations for Assessing Hydrological Hazards in Complex Terrain

Talia Rosin, Francesco Marra, Marco Gabella, Urs Germann, Daniel Wolfensberger, and Efrat Morin

Abstract. Extreme rainfall in complex terrain is highly variable across space and time, challenging accurate estimation of design-relevant return levels. While rain gauge networks provide precise point measurements, their sparse distribution limits their ability to characterise this fine-scale variability and assess areal precipitation amounts. Weather radar offers the spatial coverage and temporal resolution required.

In this study we leverage 9 years (2016–2024) of summer (June–August) precipitation data from the Swiss five radar, dual-polarisation C-band network (1 km, 5 min) to quantify summer precipitation extremes over Switzerland across durations from 30 min to 24 h and spatial aggregations from 1 to 500 km². Return levels are estimated using the Simplified Metastatistical Extreme Value (SMEV) framework, which is well-suited to short, error-prone records, and validated against corresponding estimates from 60 long-term quality-controlled rain gauges. The resulting Intensity–Duration–Area–Frequency (IDAF) relationships explicitly capture how extremes vary jointly across space and time.

Rainfall extremes exhibit a pronounced dependence on spatiotemporal scale, and vary spatially depending on the large-scale flows typical of the region and its topographic structure. For short durations and small areas, the largest return levels are concentrated in regions where strong orographic lifting is expected, including the Jura and north- and south-Alpine windward slopes, while lower values occur over the Plateau and inner Alpine valleys. As duration and area increase, small-scale peaks are progressively smoothed, and broader regions of high intensity return levels emerge, with the Southern Alps remaining a consistent hotspot across all scales. Analysis of three recent high-impact flood-producing storms illustrates how the spatiotemporal distribution of rainfall governs hydrological hazard, and how radar can capture localised extremes often missed by gauges.

Overall, the resulting multiscale return level maps provide an improved basis for hydrological design and risk assessment in complex terrain, demonstrating the value of radar-based IDAF analysis and the ability of the framework to derive scale-aware flood-relevant extremes from short radar records.

Received: 31 Mar 2026 – Discussion started: 18 May 2026

Competing interests: At least one of the (co-)authors is a member of the editorial board of Hydrology and Earth System Sciences.

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, 2118 KB)

Supplement (1604 KB)

Download & links

Talia Rosin, Francesco Marra, Marco Gabella, Urs Germann, Daniel Wolfensberger, and Efrat Morin

Status: open (until 30 Jun 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-1800', Anonymous Referee #1, 16 Jun 2026 reply
Review of Rosin et al. (2026) - Radar-Derived Intensity-Duration-Area-Frequency Relations for Assessing Hydrological Hazards in Complex Terrain
General assessment
The manuscript presents an application of the areal Simplified Metastatistical Extreme Value framework to derive radar-based Intensity–Duration–Area–Frequency relationships for summer precipitation extremes over Switzerland. The study uses 9 years of 5 min, 1 km² radar QPE data from the MeteoSwiss radar network, evaluates pixel-scale return levels against long-term rain-gauge estimates, and illustrates the hydrological relevance of the resulting IDAF fields through three recent high-impact flood-producing events. The topic is relevant for HESS because it addresses scale-dependent precipitation extremes, radar-based design rainfall estimation, and hydrological hazard assessment in complex Alpine terrain.
The manuscript has several strengths. It addresses an important problem: the estimation of spatially and temporally distributed rainfall extremes in mountainous terrain where gauge networks are sparse and not fully representative. The use of radar data for IDAF analysis is promising, and the event-based severity diagrams are potentially useful for linking rainfall scale dependence to hydrological impacts. The manuscript is also generally well connected to recent literature on SMEV, radar-based frequency analysis, areal reduction factors, and Swiss precipitation extremes.
However, in its present form, I think the manuscript requires major revision before it can be considered for publication. My main concerns are not with the overall topic, which is relevant and potentially valuable, but with the clarity of presentation, the organization of the results and discussion, and the level of methodological justification provided for several key choices. In particular, the manuscript should better justify the use of a 9-year radar record for estimating return periods up to 100 years, the choice of only 100 bootstrap repetitions, the spatial smoothing of the Weibull shape parameter, and the interpretation of radar–gauge discrepancies. Several figures are introduced without sufficient interpretation in the text, and some sections contain discussion-type material within the results section, while the discussion itself does not sufficiently refer back to the manuscript’s own figures and results.
I therefore recommend major revisions. The manuscript could become a strong contribution after improving the structure, tightening the interpretation, and adding clearer methodological support for the main assumptions.
Major comments
The manuscript applies an existing methodology, namely the areal SMEV framework, to Switzerland. This is a valid and potentially useful contribution, especially because the Swiss Alpine domain is topographically complex and hydrologically relevant. However, the manuscript should more explicitly distinguish between methodological novelty and regional/application novelty.At present, the introduction states that the study “further advances this framework” by applying it to Alpine topography and linking duration–area scaling to observed impacts. This statement would be stronger if the authors clarified exactly what is new relative to Rosin et al. (2024) and other radar-based IDAF or ARF studies. For example, is the main novelty the Swiss radar dataset, the Alpine complexity, the validation against long-term gauges, the event-severity application, or the interpretation of topographic controls? This distinction is important because the statistical framework itself appears to be largely adopted from previous work. I recommend adding a concise paragraph at the end of the introduction explicitly listing the paper’s main contributions.

A central assumption of the manuscript is that 9 years of summer-only radar data are sufficient to estimate return levels up to 100 years using SMEV. The authors cite previous SMEV work to support this, but the issue deserves a more explicit and critical treatment in this manuscript. The concern is especially important because the record includes only June–August precipitation, i.e. about 27 summer months in total. Even if SMEV uses ordinary events rather than annual maxima, estimation of 50- and 100-year return levels from such a record still depends strongly on the validity of the assumed Weibull tail, event independence, stationarity, radar-QPE stability, and the regionalization/smoothing strategy. The manuscript should therefore better quantify and discuss the extrapolation uncertainty associated with 50- and 100-year estimates. The current presentation risks giving the impression that 100-year estimates from a 9-year radar archive are as reliable as those obtained from much longer records. The manuscript should be more cautious and explicit about this limitation.
It is therefore suggested to consider the following points:
how many independent ordinary events are available per pixel or region;

how this number varies spatially and with event definition;

whether the estimated number of storms per year is stable across the 9-year period;

how sensitive 50- and 100-year return levels are to the 90 % censoring threshold, the 15 × 15 km² shape-smoothing window, and the hail-removal procedure;

whether the 100-year maps should be interpreted as robust design estimates or as exploratory radar-based estimates with substantial uncertainty.

Another ambiguity drives from if the manuscript compares radar-derived SMEV return levels with gauge-derived SMEV return levels, not with GEV-derived return levels. This should be stated consistently and clearly.the comparison is also limited to the 1 km² radar-pixel scale, while the main novelty of the manuscript concerns areal aggregation up to 500 km². This limitation is understandable because the gauge network cannot validate areal return levels directly, but it should be emphasized more strongly. The validation supports the radar product at the pixel scale only; it does not fully validate the areal IDAF estimates.
Third, the discussion of radar–gauge disagreement should separate different sources of discrepancy more explicitly: radar-QPE errors, gauge measurement errors, point-area representativeness differences, temporal aggregation effects, and sampling uncertainty. These processes have different implications. For example, a disagreement caused by sub-pixel variability does not necessarily imply radar bias, whereas disagreement caused by Z–R uncertainty or hail contamination does.
I suggest adding a short conceptual paragraph explaining what the validation can and cannot establish.
The authors smooth the SMEV Weibull shape parameter using a 15 × 15 km² moving window and then re-estimate the scale parameter. This is an important methodological choice because the shape parameter controls tail behaviour and therefore strongly affects high return periods.
The manuscript states that the window size was selected by sensitivity analysis, but the main text should provide more information on this sensitivity analysis. For example:
What window sizes were tested?

What metric was used to choose 15 × 15 km²?

Was the choice evaluated against gauge-derived return levels?

Does the optimal window size vary between the Plateau, Jura, Southern Alps, and inner-Alpine valleys?

Could smoothing across sharp topographic gradients mix physically distinct precipitation regimes?

How sensitive are 50- and 100-year return levels to this smoothing choice?

Because this smoothing step regularizes the results and narrows the uncertainty intervals, it should be described and justified more rigorously.
The manuscript uses 100 block-bootstrap repetitions to estimate uncertainty ranges. It is unclear whether 100 repetitions are sufficient, especially for high return periods and for spatially distributed estimates. With only 100 bootstrap samples, the empirical percentiles are based on the tails of a small bootstrap ensemble, which may be unstable.
The authors should justify the choice of 100 repetitions or increase the number substantially, for example to 500 or 1000 if computationally feasible. At minimum, they should include a sensitivity test showing that the uncertainty intervals are not materially affected by increasing the number of bootstrap repetitions.
This point is important because the uncertainty intervals in Fig. 6 appear narrow, and the authors attribute this partly to shape-parameter smoothing. The reader needs to know whether the narrow intervals reflect genuine stability or are partly an artefact of the bootstrap design and smoothing procedure.
Several parts of the results section contain interpretation and literature comparison that would fit better in the discussion. For example, the comparison of spatial patterns with previous Swiss radar climatologies and the physical explanation of orographic mechanisms could be moved from Section 4 to the discussion.
At the same time, the discussion section currently does not refer sufficiently to the manuscript’s own results and figures. It contains many useful ideas, but the connection to the figures is sometimes weak. The discussion should more consistently refer back to manuscript’s figures so that the reader can verify the claims.
I recommend reorganizing the manuscript as follows:
Present the main return-level maps and IDAF/ARF curves in the results section with direct description of what each figure shows.

Move broader physical explanations and comparisons with previous studies to the discussion.

In the discussion, explicitly refer back to the relevant figures when interpreting regional differences, ARF behaviour, and simple-scaling behaviour.

Figures 6, 7, and 8 are central to the manuscript, but the text does not provide enough detailed interpretation of them. Figure 8 in particular is introduced but not adequately discussed in the results section. This makes it unclear why the figure is included and what the reader should learn from it.
The event-based severity diagrams are interesting and potentially valuable. However, some statements are written too generally given that they are based on individual case studies. For example, the statement that the 21 June 2024 event implies highest hazard for small to medium-sized catchments with sub-hourly to multi-hour response times is physically plausible, but the authors should clarify whether this is a new finding, a confirmation of established hydrological understanding, or an illustration of how the IDAF framework diagnoses event severity. The event analyses should be framed as illustrative applications rather than as independent proof of general hydrological rules. The authors should also avoid wording that may appear contradictory, such as describing an event that caused fatalities and destructive impacts as not producing “major flooding.” If the intended meaning is that the event did not produce major flooding in larger river basins, this should be stated explicitly.
Specific and technical comments
The introduction is generally well motivated. However, the novelty statement should be sharpened. The authors should explicitly distinguish between the existing SMEV/areal-SMEV methodology and the new contribution of applying it to Swiss radar data and hydrological events.
Line 113: The term “inner-Alpine valleys” should be defined more clearly. Please refer explicitly to the relevant regions in Fig. 1, for example Western Alps and Eastern Alps, or mark representative valleys if the term refers to specific subregions.
Figure 1: The figure is important for the rest of the manuscript, but some graphical elements are unclear. What does the thin blue lines represent? Please define all lines and symbols in the caption. In addition, the six regional analysis boxes should be labelled directly on the map or in an inset. Since these boxes are used later for the IDAF curves, the reader should not have to infer which box corresponds to Jura, Plateau, Pre-Alps, Western Alps, Eastern Alps, or Southern Alps.
In section 3.3, please provide more detail on how the multiplicative summer bias adjustment factor was estimated. Was it calculated using all 60 gauges? Was this factor spatially uniform across all regions? If a single national correction factor is used, the authors should discuss whether this may introduce regional biases in complex terrain.
In section 3.4, the definition of independent storms as wet periods separated by at least 24 h of dry weather should be further justified for summer convective precipitation in Switzerland. Is 24 h appropriate for both short-lived convective events and longer synoptic systems? How sensitive are the results to this choice?
In section 3.5, the 90 % left-censoring threshold should be better justified in the main text. The manuscript states that this was selected from a sensitivity analysis using gauge data. Please summarize the main outcome of that sensitivity analysis and indicate whether the selected threshold is valid across all durations and regions.
The authors should also report the typical number of ordinary events used for fitting. This would help readers assess the effective sample size of the SMEV estimation.
In lines 221 – 224, The manuscript uses 100 bootstrap repetitions. Please justify this number or increase it. With only 100 repetitions, the 5th and 95th percentile estimates may be unstable, particularly for 100-year return levels. A sensitivity test using more repetitions would strengthen the uncertainty analysis.
In section 3.6, The ARF definition relative to a 1 km² radar pixel is reasonable for gridded data, but the terminology should be handled carefully. Since the denominator is not a true point measurement, the authors should consistently refer to “1 km²-reference ARFs” or “radar-pixel-reference ARFs” rather than classical point-to-area ARFs.
In Figure 2, the validation is useful, but the caption and text should specify clearly that both radar and gauge return levels are estimated using SMEV, unless a different method is actually used. If GEV estimates are also used anywhere, this needs to be stated explicitly and methodologically described.
In Figure 3, the authors should discuss whether the regional differences are related to radar beam geometry, altitude, gauge exposure, precipitation regime, or the uniform bias-adjustment factor. The current interpretation remains somewhat general.
In line 290 and Figure 4, when stating “across Switzerland,” please make clear in the caption that the boundary shown in Fig. 4 is the Swiss national boundary. This will help readers unfamiliar with the map.
In lines 292-295 and Figure 4, when referring to geographic regions such as the Jura Mountains, Alpine barrier, Southern Alps, Plateau, and inner-Alpine valleys, these should be visible or labelled in the figure, or the reader should be directed clearly to Fig. 1. Otherwise, the spatial interpretation is difficult to follow.
In lines 298 – 303, this paragraph compares the results with previous studies and explains physical mechanisms. It is relevant, but it fits better in the discussion section. The results section should first describe the manuscript’s own findings; comparison with previous work should then be used in the discussion to interpret them.
In line 316, the statement that there is “a narrowing of the Southern Alps maximum” is difficult to appreciate visually from Fig. 5. The spatial structure appears broadly similar across return periods. If the authors want to make this point, they should quantify it, for example by showing the area exceeding a given percentile or threshold as a function of return period. The same figure also raises a broader question: are 9 years of data sufficient to resolve the spatial differences among 50- and 100-year return levels? The authors should discuss this more explicitly.
Figures 6 and 7 convey partly overlapping information. Figure 6 shows IDAF curves by region, while Figure 7 reorganizes similar information by duration and adds the R² analysis. The authors should consider merging these figures or moving part of the information to the supplement. If both figures are retained, their distinct purposes should be explained more clearly. In addition, the panels in Fig. 6 could be reorganized geographically. For example, the top row could show Jura, Plateau, and Pre-Alps, and the bottom row could show Western Alps, Southern Alps, and Eastern Alps. This would make the regional comparison more intuitive.
The regional names should be made clearer. The reference to Fig. 1 is not sufficient because the boxes in Fig. 1 are not labelled directly. Please label the regional boxes in Fig. 1 or add a small map inset to Fig. 6.
The text should explicitly describe Fig. 7. In particular, the authors should explain what the high R² values imply.
Figure 8 is not sufficiently discussed in the results section. Since ARFs are a central component of IDAF analysis, the authors should describe the main patterns in detail. For example, which regions show the strongest areal reduction? How does ARF vary with duration? Are any ARFs greater than 1? What does this imply physically? How does this compare with previous studies?
In lines 334 – 338 , the opening paragraph of Section 5 explains the general applicability of IDAF relations and severity diagrams. This material may fit better in the methodology section, where the event-severity analysis is introduced. Section 5 could then begin directly with the selected events and the rationale for choosing them.
Figure 9 may be redundant with Fig. 1. The authors could consider merging the event-location information with Fig. 1 or providing Fig. 9 as an inset or supplementary figure. If Fig. 9 is retained, the caption should explain why a separate location figure is needed.
In lines 360 – 361, the statement that the patterns imply the highest hazard for small to medium-sized catchments with sub-hourly to multi-hour response times is hydrologically reasonable. However, the authors should clarify that this is an interpretation of this event’s scale-dependent rainfall structure, not a general conclusion derived from a single event. The statement could be rewritten as:
In page 16, and regarding Mesolcina event, the manuscript states that the event lacked the duration to drive major flooding in larger river basins. This should be clarified, because the event itself caused severe impacts and fatalities. The current wording could be misread as minimizing the severity of the event.
In line 365, the phrase “significant discrepancy” between gauge- and radar-derived return periods needs clarification. From Fig. 10, some gauges located near affected regions appear to show return periods broadly consistent with nearby radar estimates. If the discrepancy refers to gauges missing the storm core because of their spatial distribution, this should be stated more precisely. The issue may be sparse sampling rather than disagreement where radar and gauges are colocated.
In lines 383 – 384, the statement that the maximum radar-derived return period is only 33 years, compared with more than 3000 years for the downtown gauge, is important but not sufficiently supported by Fig. 11 as currently presented. Visually, the gauge dot and adjacent radar colours appear relatively similar. The authors should provide the numerical values explicitly, state which gauge is used, and explain how the gauge return period was calculated. If the difference is mainly due to sub-kilometre spatial variability and 10 min versus 30 min accumulation differences, this should be clearly stated. Otherwise, the reader may not understand how the manuscript supports the 3000-year claim.
In line 395, the term “northern Alpine slopes” is difficult to interpret from Fig. 12 alone. Please either mark this region in the figure or use terminology that directly corresponds to the regional divisions shown in Fig. 1.
In lines 426 – 431, this paragraph discusses Pre-Alps, Jura, Southern Alps, convective storms, stratiform precipitation, severe deep convection, and ARF behaviour. The ideas are relevant, but the paragraph is difficult to follow and should be rewritten. The authors should refer explicitly to their own results, especially Fig. 8, before invoking broader physical explanations.
The discussion section should be substantially revised. At present, it contains many relevant physical interpretations, but it does not sufficiently guide the reader through the manuscript’s own evidence. The authors should refer frequently to mansucript figures when discussing:
regional contrasts in return levels;

duration dependence;

area dependence;

ARF behaviour;

the role of orography;

differences between convective and longer-duration events;

the practical hydrological relevance of IDAF curves.

The discussion would be much clearer if each interpretive claim were linked to a specific figure or result.
Minor comments
“simplified-variant Metastatistical Exrteme Value” should be corrected to “simplified variant of the Metastatistical Extreme Value.”

“Intensity-duration-area-frequency” should be consistently written as “Intensity–Duration–Area–Frequency.”

Geographic terms should be used consistently: Jura Mountains, Swiss Plateau, Pre-Alps, Western Alps, Eastern Alps, Southern Alps, inner-Alpine valleys.

The manuscript should avoid introducing figures without substantive discussion in the text.

Figure captions should define all symbols, lines, regions, and boundaries.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1800-RC1
RC2:
'Comment on egusphere-2026-1800', Anonymous Referee #2, 21 Jun 2026 reply
This paper utilizes high spatiotemporal resolution radar quantitative precipitation estimation data from Switzerland during the summers of 2016–2024, combined with the areal SMEV framework, to establish the IDAF relationship for summer extreme precipitation with a return period of 2–100 years. Furthermore, it demonstrates its application value in hydrological hazard assessment in complex topographical areas through three recent high-impact flood events. The paper addresses a clear scientific question, provides valuable data resources, employs a highly targeted methodology, and the results have practical significance for multi-scale rainfall diagnosis of hazards such as mountain floods, urban flooding, and debris flows. I believe this paper has high publication potential overall, especially in applying areal SMEV to the complex Alpine terrain of Switzerland and linking multi-scale extreme precipitation statistics with actual hydrological hazard responses. However, several key issues in the current manuscript still require further attention, particularly the reliability of extrapolating 9 years of radar data to a 50–100 year return period, the propagation of radar QPE systematic errors to the return level and IDAF, the impact of SMEV shape parameter spatial smoothing on uncertainty and terrain gradient, and quantitative comparison with existing Swiss IDF/IDAF or CombiPrecip related results. These issues do not diminish the overall value of this paper, but they do affect the interpretability and reliability of the results. Therefore, I recommend a major revision.

The paper uses summer radar data from 2016–2024 to estimate the multi-scale return level up to a 100-year return period. The authors reason that SMEV can utilize the tail distribution of ordinary events rather than relying solely on annual maximum values, thus making it more suitable for short radar sequences; the paper also cites existing research indicating that approximately 10 years of records can be used for stable estimation within the SMEV framework. This argument is reasonable and forms the core basis for the methodological choice in this paper. However, from the perspective of engineering hydrology and risk management, extrapolating the 50-year and 100-year recurrence levels based on 9 years of data still exhibits significant model dependence. The uncertainties here include not only bootstrap sampling uncertainty, but also whether the tail of ordinary events follows a Weibull distribution, whether the assumption of event independence is sufficient, whether different weather patterns still exist within the summer, whether radar QPE errors affect the tail, and whether shape parameter spatial smoothing alters extreme tail behavior. It is recommended that the authors more clearly distinguish in the results and conclusions that: SMEV parameter estimation is statistically more stable than traditional GEV/POT; the 100-year recurrence level is still a significant extrapolation; and the reliability of this extrapolation depends on the SMEV model assumptions and the radar QPE error structure.

This paper uses five MeteoSwiss dual-polarization C-band radars with a data resolution of 1 km² and 5 min, and points out that complex terrain can bring problems such as ground clutter, mountain obstruction, bright bands, hail pollution, and uncertainties in the Z–R relationship. The authors also performed summer mean field bias corrections using a multiplicative factor of 0.87 and removed periods of high hail probability. These quality control steps were necessary and reasonable. However, the current uncertainty estimate primarily relies on resampling years to estimate sampling uncertainty. The paper explains that 100 bootstraps were used, with the 5–95th percentile as the uncertainty range. This method does not adequately reflect the systematic error of radar QPE, especially for short-duration, small-area extreme rainfall events, where QPE error may be more significant than sampling error. This was particularly evident in the Lausanne 2018 event. The paper points out that even at the smallest scale of 1 km² and 30 min, the radar-derived maximum return period was only 33 years, while the return period at the city center rain gauge exceeded 3000 years. The authors attribute this difference to the point-area scale mismatch and the significant underestimation of the peak rainfall intensity of the event by the standard Swiss-tuned Z–R relationship. It is recommended that the authors supplement at least one of the following analyses: conduct sensitivity tests on the 0.87 bias correction factor, such as ±10% or changing the correction factor by region/duration; conduct scenario tests for short-duration strong convective events with different Z–R relationships or raindrop spectrum assumptions; transform the radar-rain gauge bias structure into an uncertainty range for the recurrence level; and distinguish between sampling uncertainty and QPE-related uncertainty in the IDAF plot.

The paper uses a 15 × 15 km² moving window to smooth the Weibull shape parameters and then re-estimates the scale parameters under fixed shape conditions. The authors believe that this treatment can reduce sampling noise from the 9-year short record while preserving physically reasonable spatial gradients. Furthermore, the paper mentions that shape smoothing significantly reduces estimation uncertainty without significantly changing the magnitude of the recurrence level. I believe this is one of the key steps in the paper's method, but the current justification is still insufficient. Extreme precipitation in complex topographic regions often exhibits rapid spatial variations between windward slopes, leeward slopes, valleys, ridges, and rain shadow areas. A 15 km window may be reasonable in some regions, but it could also be overly smoothed in areas with strong topographic gradients, particularly affecting the tail of high return periods and the slope of small-scale IDAF. The narrow IDAF confidence intervals in Fig. 6–7 are explained by the authors as shape smoothing increasing the effective sample size and reducing sampling noise. However, statistically, spatial smoothing also introduces strong prior structures, potentially underestimating local uncertainties. The authors suggest supplementing the data with different smoothing windows, such as 5, 10, 15, and 25 km, to investigate their impact on 20-year and 100-year return levels.

The authors validated the radar-derived 1 km² pixel-scale return level using 60 long-term rain gauges. The validation results show overall consistency between radar and rain gauge data, but with a significant time-dependent bias: short-duration radar overestimates, while long-duration radar underestimates. Furthermore, the bias exhibits regional structure, with more pronounced overestimation in the Jura and Central Alps, and more likely underestimation in the Plateau, Pre-Alps, and Southern Alps. The main innovation of this paper is areal IDAF, namely, the areal return levels and ARF at 10, 50, 100, and 500 km² scales. Due to insufficient rain gauge network density, the authors could not directly validate larger area scales, which is understandable. However, some statements in the current paper may lead one to believe that 1 km² validation is sufficient to support the reliability of all area scales. In fact, area aggregation changes the error structure: random errors may be averaged, but systematic underestimation or overestimation, terrain-related errors, storm core localization errors, and occlusion errors do not necessarily decrease simply with area. It is recommended that the authors add a clearer limiting statement to the validation section: current independent validation mainly supports radar-derived pixel-scale return levels, while areal return level validation is indirect.

The paper points out that Western Alps show ARF > 1 over 24 hours, explaining that the reference pixel is located in an occluded valley, and as the aggregated area increases, high precipitation from adjacent exposed slopes is included, thus making the areal average rainfall intensity higher than that of the 1 km² reference pixel. This explanation is reasonable in complex terrain and consistent with the specific behavior of fixed-area ARF in areas with strong spatial gradients. However, ARF > 1 is counterintuitive for many readers, especially in engineering applications where ARF is often understood as a reduction factor that decreases or equals 1 with increasing area. It is recommended that the authors provide a clearer explanation to avoid misunderstanding by engineering readers.

It is recommended that the spatial variation of the number of common events, n, be more clearly explained in the methodology section. In SMEV, n is the average annual storm number; complex terrain and regional climate differences may cause significant variations in n across different regions of Switzerland. It is recommended that the spatial distribution or regional statistics of n be provided.

The left censoring threshold uses the lowest 90% of events as non-exceedances and is determined based on the sensitivity of rain gauge data. It is recommended that it be explained whether this threshold is equally applicable to radar areal samples, especially considering that the tail of the common event distribution may change after aggregation of different areas.

The text mentions that radar data was available from 2005, but due to system upgrades, it is limited to 2016–2024. It is recommended to more clearly state whether any remaining non-uniformities after system upgrades have been completely ruled out, such as algorithm version changes, radar maintenance, or missing data.

The 2018 Lausanne incident illustrates that 1 km² radar pixels may significantly smooth out sub-kilometer-scale peaks. It is recommended to explicitly state in the conclusion that the IDAF method described in this paper applies to spatial scales of 1 km² and above, and should not be used to interpret sub-kilometer urban storm peaks.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1800-RC2

Talia Rosin, Francesco Marra, Marco Gabella, Urs Germann, Daniel Wolfensberger, and Efrat Morin

Supplement

https://doi.org/10.5194/egusphere-2026-1800-supplement

Talia Rosin, Francesco Marra, Marco Gabella, Urs Germann, Daniel Wolfensberger, and Efrat Morin

Viewed

Total article views: 204 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
132	61	11	204	25	8	10

HTML: 132
PDF: 61
XML: 11
Total: 204
Supplement: 25
BibTeX: 8
EndNote: 10

Views and downloads (calculated since 18 May 2026)

Month	HTML	PDF	XML	Total
May 2026	119	49	8	176
Jun 2026	13	12	3	28

Cumulative views and downloads (calculated since 18 May 2026)

Month	HTML	PDF	XML	Total
May 2026	119	49	8	176
Jun 2026	13	12	3	28

Viewed (geographical distribution)

Total article views: 199 (including HTML, PDF, and XML) Thereof 199 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 25 Jun 2026

Short summary

Extreme rainfall in mountainous regions is difficult to assess as it varies strongly across space and time, and rain gauges are often too sparse to capture it. Using 9 years of radar data, we analyse summer rainfall extremes across Switzerland for different durations and areas. We show that rainfall intensity depends strongly on scale and topography, and that radar better captures local extremes, improving flood hazard assessment, as demonstrated for three recent major flood-producing storms.


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