the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
An appraisal of the value of simulated weather data for quantifying coastal flood hazard in the Netherlands
Abstract. With recently updated safety norms, assessment of flood safety in the Netherlands requires return values for coastal sea level and surface stress for a wide range of return periods up to 107 years. Estimates from measurements are highly uncertain. To reduce the uncertainty, a possible solution could be to replace measurements by simulated weather datasets much larger than a typical measurement record. However, systematic errors in simulations can easily outweigh any gains in precision. Combining insights from physics and extreme value theory with evidence from data, we argue that even as stress from present-day weather prediction models may be too high or too low, these data are suitable for estimating the shape of the upper tail of the distribution function of stress, and that this extends to simulated data of water level along the Dutch coast. As scale and location parameters can be estimated with sufficient precision from relatively short measurement records, we estimate return values from a combination of measurements (for scale/location of water level) and simulated data (for shape), assess their uncertainty, and discuss strengths and limitations of the approach and prospects for further exploiting simulated weather data to assess flood risk.
- Preprint
(7226 KB) - Metadata XML
-
Supplement
(1541 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on egusphere-2024-912', Agustín Sánchez-Arcilla, 17 Jun 2024
The paper touches an interesting topic but it should improve the format (the reading is a bit cumbersome with lots of geographical names not located) and internal cross references (e.g. line 395 to the same section when looking for further information).
In more detail, the paper should address:
- Whether the shape of the probability distribution and the characteristics of the tale will be the same under future climates, particularly in the case of disruptions.
- The analysis of tail shape differences using the Gamma parameter or the Beta parameter should be better explained, not only making reference to the data source or probability distribution but also to the underlying physics.
- The assumptions that the development of extra tropical cyclones will constitute a continuum under future climates and that tail anomalies will not be affected by different extreme populations, should also be discussed making reference to future climate projections, particularly depending on scenario and horizon.
- The discussion about the benefits of increasing resolution should also make reference to the level of numerical diffusivity that is associated to resolution. This is an important point for all the following discussion on the shear stress which is directly related to the energetic transfer between air and sea.
- The difference in stress tail shape depending on resolution (e.g. figure 5 and associated paragraph) should also be discussed in more detail, explaining the large difference in scale and relating that to the underlying physics.
- The assumption that the extremal index for stress should be the same as for wind speed (line 320 and following) should also be discussed considering the non linear relationship and the reasons to question a single extremal index for both variables.
- The conclusion that large spatial variations in return values, from measurement analysis, is due to sampling variability is rather questionable (line 345 and following). This should be discussed considering sources of spatial variability and the dynamic evolution of the storm event over different spatial positions.
- The same discussion should be carried out for the limited universality of U10 tail shapes when compared to stress tail shapes. A physical based discussion on how the variability from the wind speed is transferred to the stress field, considering their non linear relation would greatly benefit that paper. The scale ratios from a number of positions along the coast (line 375 and following) should also be discussed considering the specific geographical constraints of each of those locations, particularly when in some of the locations larger ratios are found (line 380). I think that finding there is no large systematic impact of resolution is probably too strong a conclusion since very little is said about the conditions applied for that assessment.
- The systematic over estimation of stress from land sectors appears to be attributed to limited resolution of SEAS5 (line 390 and following). I wonder if that is the main reason since land roughness could also play a role in here.
In summary, the paper presents a lot of solid and well executed work and I think with some extra effort it could become an interesting publication in this field.
Citation: https://doi.org/10.5194/egusphere-2024-912-CC1 -
AC3: 'Reply on CC1', Cees de Valk, 10 Nov 2024
Reply: We are grateful to Agustín Sánchez-Arcilla for contributing valuable comments on our manuscript. As a first step, we will check and improve the structure and wording, and references. Below, we will answer the detailed comments point by point, citing the comment.1. Whether the shape of the probability distribution and the characteristics of the tail will be the same under future climates, particularly in the case of disruptions.A: This is an important issue, but would require an entire study in itself. One could analyse very large datasets of climate model projections for the present and future epochs, and compare shape estimates from both. This would also need to involve several models to check if results from different models are consistent. We would recommend this as a follow-up to the present study: the current manuscript is already quite a long read, and to us it makes sense to first study the use of simulated data to estimate return values for present climate (as we do in our manuscript) before moving to the topic of possible effects of climate change, which could involve greater prominence of cyclones of (sub)tropical origin (If remnants of tropical cyclones are not considered, we hypothesize that the shape will not change (much) under climate change, as the physical principles will be unchanged for the storms of extratropical origin).2. The analysis of tail shape differences using the Gamma parameter or the Beta parameter should be better explained, not only making reference to the data source or probability distribution but also to the underlying physics.A: Explaining the tail shape of stress based on physical principles is something we would very much like to do. However, we have not been able to do this until now. The cited literature indicates that the shape of the tail of stress is primarily controlled by the statistics of the jet stream velocity and position, even as other processes may strengthen or weaken the cyclone over its lifetime.3. The assumptions that the development of extra tropical cyclones will constitute a continuum under future climates and that tail anomalies will not be affected by different extreme populations, should also be discussed making reference to future climate projections, particularly depending on scenario and horizon.A: Indeed, the literature indicates that storms of (sub)tropical origin may become more prominent in the future, which may require explicit consideration of different populations of cyclones. However, we do not intend to discuss future climates at this moment, as it will widen the scope of the manuscript too much and will require much additional research. It will certainly be a focus of future work, as the problem is highly relevant.4. The discussion about the benefits of increasing resolution should also make reference to the level of numerical diffusivity that is associated to resolution. This is an important point for all the following discussion on the shear stress which is directly related to the energetic transfer between air and sea.A: Numerical diffusivity is indeed an important aspect of resolution, depending also on the discretisation scheme applied: due to numerical diffusion, the effective resolution is often coarser than the mesh size. We consider this background knowledge that does not warrant a discussion in our manuscript. For the most important layers close to the sea surface, vertical momentum transfer has been parameterised in numerical weather prediction models, so numerical diffusion will not play a role here. This is the reason that we focus on differences between drag parameterisations employed in the models discussed. Additionally, the wind fields that generate the extreme surges are much larger than the model resolution, which reduces the influence of horizontal numerical diffusion on the outcomes.5. The difference in stress tail shape depending on resolution (e.g. figure 5 and associated paragraph) should also be discussed in more detail, explaining the large difference in scale and relating that to the underlying physics.A: We agree with this comment. Indeed, we cannot relate the difference in scale to resolution: based on resolution only, we would expect the difference in scale to have the opposite sign of what we observe. Therefore, the difference in scale is much more likely related to the different drag parameterizations in these models, which are indeed very different. We intend to discuss this point in somewhat more detail in the revised manuscript.Henk: Ik stel voor om de figuur opnieuw te maken met SEAS5*1.1 en RACMO*1.25. De verschillen die je dan nog ziet zullen grotendeels verband houden met de 6-hr versus 3-hr tijdstap. Eventueel nemen we ook voor RACMO een 6hr tijdstap voor de Gumbelplot van extreme stress. Zo niet, zou ik in het het antwoord zeggen: The difference in absolute values is mainty due the the difference in timestep (6hr versus 3hr).6. The assumption that the extremal index for stress should be the same as for wind speed (line 320 and following) should also be discussed considering the non linear relationship and the reasons to question a single extremal index for both variables.A: Nonlinearity of the relationship between u10 and stress does not affect the extremal index, and it does not affect estimators of the extremal index: they are invariant to a continuously increasing transformation. Therefore, we do not intend to elaborate on this point.7. The conclusion that large spatial variations in return values, from measurement analysis, is due to sampling variability is rather questionable (line 345 and following). This should be discussed considering sources of spatial variability and the dynamic evolution of the storm event over different spatial positions.A: We consider sampling variability a likely explanation of the large spatial variability of return value estimates from measurements in Caires (2009) because (a) the samples are relatively small (typically covering about 40 years), and in addition (b) the thresholds for these estimates were selected by an adaptive scheme, which selects a threshold based on the observed variation of estimates as functions of threshold. Furthermore, large "apparent outliers" in Figure 9 appear to be limited to a single site; at neighbouring sites, the simultaneous values tend to be much less extreme, so we suspect that they may be random (sampling variability) and possibly in some cases erroneous and/or related to specific local conditions at the site (e.g. funneling; most sites are at or near the coast). Furthermore, a number of the "outliers" in these plots are related to the same storms. In addition, we find that spatial variability of wind speed tails is not really needed for a close fit to the empirical tails; see Figure 9. In response to the above comment, we intend to address this issue in some more detail in the revision of the manuscript.8. The same discussion should be carried out for the limited universality of U10 tail shapes when compared to stress tail shapes. A physical based discussion on how the variability from the wind speed is transferred to the stress field, considering their non linear relation would greatly benefit that paper.A: The limited universality of u10 tail shapes is a topic we discussed in an earlier report (de Valk and van den Brink, 2023b) in some more detail. It can be explained by the simple model of Blackadar and Tennekes (1968) of the relationship of the wind profile and stress to the geostrophic wind in a neutrally stable boundary layer. We mention this model in Section S4 in the context of drag saturation, but it also explains the limited universality of u10 tail shapes. Therefore, we will mention this in the revised manuscript.9. The scale ratios from a number of positions along the coast (line 375 and following) should also be discussed considering the specific geographical constraints of each of those locations, particularly when in some of the locations larger ratios are found (line 380). I think that finding there is no large systematic impact of resolution is probably too strong a conclusion since very little is said about the conditions applied for that assessment.A: For this reason, we do not explicitly exclude a resolution effect, and only present the alternative explanation of inadequate representation of the hydrodynamics during severe storm conditions in the shallow Waddenzee as a working hypothesis. Furthermore, we now have more detailed figures showing that the effect of hydrodynamic model resolution on bias in surge and water levels is linear, and therefore resolution of the hydrodynamic model does not offer an explanation. We will add this as Supplementary Materials and refer to it in the msin text.10. The systematic over estimation of stress from land sectors appears to be attributed to limited resolution of SEAS5 (line 390 and following). I wonder if that is the main reason since land roughness could also play a role in here.A: Errors in roughness over land may indeed contribute, but the difference in roughness over land and sea is very large: the roughness length over sea is about 100 times smaller than over land; whereas variation over land is of O(10). Therefore, we expect that the poor approximation of the surface roughness due to limited resolution of SEAS5 is the main cause (the model uses a weighted average of the roughnesses over land and sea within a grid cell). We will expand a little more on this in the revision.Citation: https://doi.org/
10.5194/egusphere-2024-912-AC3
-
RC1: 'Comment on egusphere-2024-912', Anonymous Referee #1, 22 Aug 2024
The problem addressed by the authors is highly relevant, and the results presented are quite promising regarding the effectiveness of the proposed methodology in reducing uncertainty when estimating very high return period quantiles.
However, the manuscript is challenging to follow, especially concerning the methodology. I believe it would benefit from a dedicated Methods section that clearly presents the following aspects, considering that not all readers of the journal are specialists in the statistical methods used:
- The probability distributions used.
- The method(s) employed to estimate the parameters of these distributions, including the shape, scale, and position parameters, along with how these parameters are combined.
- The method used to estimate confidence intervals, with clarification of terms like “sampling error” and “total uncertainty.”
On the other hand, the range of variables and models analyzed is quite broad (e.g., wind stress, wind velocity, HW, HW conditional on wind direction). In my view, this diversity requires discussing many specific aspects of each variable and model, which distracts from what I consider the main contribution of the work: demonstrating how the use of data from weather simulations significantly reduces uncertainty in the analysis of extreme values.
In this regard, I strongly recommend that the authors focus on detailing the proposed methodology and then demonstrate its application to a single variable (or at most two), setting aside complementary analyses that do not seem fundamental (e.g., directional analysis, relationships between wind stress and wind velocity).
In summary, I believe this work has the potential for significant impact, but the wording and structure of the manuscript need thorough revision.
Citation: https://doi.org/10.5194/egusphere-2024-912-RC1 -
AC1: 'Reply on RC1', Cees de Valk, 10 Nov 2024
We are grateful for the helpful comments on the structure and wording of the manuscript, and will implement the suggested changes. In particular,
(a) We will restructure and shorten the introduction to focus directly on the general question of whether simulated weather data can be used to improve the accuracy of return value estimates, and then discuss the specific case to be considered (coastal water level, wind stress, Dutch coast, etc.). We will add a basic introduction of the concepts (e.g. the meaning of a probability distribution in relation to frequencies of exceedance of levels), to make clear what we mean with "tail" etc.
(b) We will drop de discussion of directional dependence and of dependence between stress and high water (HW) from the manuscript.
(c) A section will be added about the methods used to address the problem (e.g. relation between exceedance frequencies and distribution, classes of distributions considered and relationship between their shapes, estimators for these, sampling uncertainly estimation by bootstrapping and model-related error estimation)
We still feel that discussing both water level and stress would be helpful. Without considering HW, the practical relevance of the manuscript to coastal flood risk (and hence, to the journal) would be limited. Also, HW is used to check the estimates of the scale correction applied to the tail of stress, which is important because there is no alternative for it.
On the other hand, without considering stress, the connection between extremes of HW and the generating storms will be lost as well, which is an essential part of our argument that the shape of the tail of HW can be determined from simulated data. But we understand that the presentation of these different aspects may need to be simplified or better structured, for example by including one or more diagrams summarising the proposed method (containing the datasets and models involved).Citation: https://doi.org/10.5194/egusphere-2024-912-AC1
-
RC2: 'Comment on egusphere-2024-912', Anonymous Referee #2, 29 Sep 2024
This work focuses on the usefulness of available datasets of simulated weather variables for extreme values analyses in the framework of coastal flooding and protection design. For this, different sources of data and assessment of descriptors of statistical distributions are used along the Dutch coast. The potential of the method and the benefits when compared to other approaches are assessed. The topic is relevant for different applications and the work goal is a good attempt to improve the representation of extreme data probabilities, which fits very well within the scope of this journal.
However, the manuscript needs further in-depth refinement before being considered for publication. There is huge and worthy work in it, undoubtedly, but the work flow must be presented more orderly and with less circular presentation of issues and analyses.
For example, the introduction could be more focused from the beginning. It is not until line 80 that we learn the goal of the work; this could be anticipated earlier to facilitate a comprehensive reading. Additionally, there are different references to later sections when presenting the relevant points in the justification of the work objectives and approach; in my opinion, this is not necessary and does not add clarity but rather makes it less smooth to follow. The work structure itself is not easy to follow, especially regarding methods, and within each section the flow of explanation or discussion goes to and fro presented many different issues without a clear linkage or order.
The work will largely benefit from a thorough revision of English writing and phrasing. In some cases, reading is difficult and some paragraphs do need clear corrections. This is generalized, so no specific cases are listed.
Other minor issues regarding edition: “et al.” is missing the final point in many citations along the text; please, check and add; citations along the text are not orderly listed with a uniform criterion, please, revise following the journal’s indications.
Citation: https://doi.org/10.5194/egusphere-2024-912-RC2 -
AC2: 'Reply on RC2', Cees de Valk, 10 Nov 2024
We are grateful for the helpful comments and suggestions for improvement of the structure and writing. In response, we will make the following changes:
(a) We will restructure and shorten the introduction to focus directly on the general question of whether simulated weather data can be used to improve the accuracy of return value estimates, and then discuss the specific case to be considered (coastal water level, wind stress, Dutch coast, etc.). We will add a basic introduction of the concepts (e.g. the meaning of a probability distribution in relation to frequencies of exceedance of levels), to make clear what we mean with "tail" etc.
(b) References to later sections will be checked and removed.
(c) A section will be added about the methods used to address the problem (e.g. relation between exceedance frequencies and distribution, classes of distributions considered and relationship between their shapes, estimators for these, sampling uncertainly estimation by bootstrapping and model-related error estimation)
(d) We will add a diagram summarising the proposed method and the data and models involved.
(e) Following the suggestion of another reviewer, we will drop de discussion of directional dependence and of dependence between stress and high water (HW) from the manuscript in order to improve focus.
(f) We will check and improve the writing and phrasing and correct the citations.
Citation: https://doi.org/10.5194/egusphere-2024-912-AC2
-
AC2: 'Reply on RC2', Cees de Valk, 10 Nov 2024
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
267 | 97 | 39 | 403 | 29 | 16 | 17 |
- HTML: 267
- PDF: 97
- XML: 39
- Total: 403
- Supplement: 29
- BibTeX: 16
- EndNote: 17
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1