the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Jan 2025
Robust Adaptive Pathways for Long-Term Flood Control in Delta Cities: Addressing Pluvial Flood Risks under Future Deep Uncertainty
Abstract. Delta cities are increasingly vulnerable to flood risks due to the uncertainties surrounding climate change and socioeconomic development. Decision-makers face significant challenges in determining whether to invest in high-level flood defenses for long-term planning. Adaptation solutions should be given considerable attention not only to robustness but also to adaptiveness if the future unfolds not as expectation. To support decision-making and meet long-term multi-objective targets, we propose a synthesized framework that integrates robustness analysis, adaptiveness analysis, and pathway generation. This framework was applied to evaluate alternative solutions for managing pluvial flood risk in central Shanghai. The results show that using a single-objective decision-making approach (focused only on robustness) tends to yield biased options. By examining the valid period and flexibility of candidate solutions, we assessed whether alternative solutions could meet long-term flood control targets. The analysis reveals that a combined option—incorporating increased green areas, an improved drainage system, and a deep tunnel with a 30 % runoff absorption capacity (D+G+Tun30)—is the most robust and adaptive pathway, based on multi-objective trade-off analysis. This study highlights the importance of considering valid period within predefined control targets and retaining flexibility to avoid path-dependency and minimize long-term regrets. The proposed framework can be applied to other delta cities to guide adaptive responses to future flood risks.
- Preprint
(1340 KB) - Metadata XML
-
Supplement
(852 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on hess-2024-391', Anonymous Referee #1, 23 Feb 2025
The study contained in this paper is well-conceived and well-conducted. It is a conscientious application of the robust decision making (RDM) decision making under deep uncertainty (DMDU) method to an important problem lying at the conjunction of earth science and public policy. Beyond this, the paper might be considered significant and important because it is, to this reviewer's knowledge, the first such application of DMDU concepts and methods to an East Asian policy decision problem carried out by a domestic resarch team. DMDU methods have disseminated rapidly since their first application in the US in the 1990s. However, their advance has been notably slow in China, Japan, Republic of Korea, and other nations in their proximity. This gives the paper an importance beyond its individual merits.
There are two principal reasons that the paper should undergo revision before publication. One is substantive and the other stylistic. Happily, both issues can be addressed without any revisions to the actual scientific design or analysis. Therefore, while the list of needed changes is long, they have been characterized as minor (if extensive). The bones of the study require little or no change and the scientific merits are readily apparent.
The substantive issue is one that I as a reader had a difficult time understanding at first. The authors characterize their study as not solely relying upon robustness criteria for evaluating alternative courses of action but also engaging adaptivity considerations as well. This is most curious and a bit solipsistic. Adaptation and flexibility have long been understood as being some of the principal tools for ensuring the robustness of a planned course of action. Indeed, under conditions of deep uncertainty they are almost invariably major components of a course of action deemed to be robust. Yet, in this paper they are principally presented as alternative approaches with which the authors innovative by taking both into consideration.
There may be several misunderstandings occurring. One is that robustness is a much abused term taking on different meanings in different fields. There are robust statistics, for example, while in engineering design for robustness may be to determine maximum stresses and then construct the item to withstand 3x of that force. This may be why the authors speak of the considerable expense of robust solutions in their problem space. But this is not so much true when robustness is used in the application of policy.
Another possibility is that the authors are confusing the concept of robustness with that of regret in terms of decision analysis. They discuss at one point different approaches to measuring "robustness" when in fact they are citing different methods for calculating regret. Regret is a very useful concept for determining the relative merits of alternative courses of action under deep uncertainty. When optimization is not feasible, as the authors correctly state, and the project of forecasting and prediction becomes fraught, then regret is the key for understanding choices. It is also a uni-dimensional concept. One chooses individual metrics and then use them to successively assess relative regret for alternative courses of action according to that metric. I think what they authors are saying, again correctly, is that such regret measures may be myopic when applied solely to cost-benefit analysis of those things that are easily measured, such as NPV. They introduce alternative metrics such as adaptability and do so convincingly. The problem is that they do so in a false opposition to a straw man that they term 'robustness'.
The second problem is one of syntax, grammar, and English usage. The authors have chosen to publish their work in an English-language journal for the benefit of a wider, global audience. They are to be applauded and we non-Mandarin speakers will be made richer for that decision. However, there are syntactical problems with the use of articles, verb agreement, incomplete and awkward sentences and other issues that it simply was not within the resources of this reviewer to correct. At least one of the authors is associated with an institution of higher education in the English-speaking world. Once the manuscript has been revised in light of my comments, the additional and more detailed marginalia that I have placed in the .pdf document attached to this review, and the insights of other reviewers, it would be wise for the authors to engage the services of a native English speaker colleague or copy editor to provide the light edit that would immensely improve the readability of this draft.
Please find comments of a more detailed character in the attached .pdf.
-
RC2: 'Comment on hess-2024-391', Anonymous Referee #2, 07 May 2025
The manuscript presents a methodological framework for flood control decision-making under uncertain future conditions, considering both the robustness and adaptiveness of candidate strategies. The methodology is particularly well-suited for delta megacities, as it explicitly accounts for the detrimental effects of rising sea levels and land subsidence—both of which typically impact these urban environments. Additionally, it addresses other climate change-induced threats, such as increasing extreme precipitation.
At this stage, however, the manuscript presents several issues and requires major revisions before it can be considered for final publication. My primary concern lies in the way adaptiveness is analyzed, specifically the use of a questionable metric for flexibility. This limitation negatively affects the results of the proposed decision-making framework, leading to biased conclusions regarding the optimal set of flood control strategies (see comments #2, #3, and #4). Other concerns include a general lack of clarity in the description of the methodology and an inadequate outline of the research background. These issues make it difficult to frame the novelty of this work in relation to existing literature, particularly when compared to studies from which portions of the methodology have been borrowed.
More detailed comments are provided below.
- Lines 56—89; 114—118: In the introduction, a description of the mentioned DMDU methods (e.g., RDM, DAPP, ROA, MORDM, Adaptive Policymaking, Adaptation pathway, and the concept of tipping point) would be beneficial. In this way, the reader will be able to compare these different methods, see their limitations, and better understand how the proposed methodology tries to address them. Probably, a scheme outlining the most popular approaches (at least RDM and DAPP; L78) would facilitate the comparison. At this stage, the authors talk about these existing methods assuming that the reader is already familiar with the topic.
- The way flexibility is defined in this work—i.e., as the number of combined flood-control measures gradually implemented in a given flood-control plan (L453-456; Fig. 4; L276-286) —provides a misleading assessment of the real adaptability of a given flood control implementation. This is because implementations based on a single flood control measure—hence, with the lowest flexibility rank, according to this adopted metric—do not necessarily preclude future, not-originally-planned updates by adding other control measures at later stages (e.g., if the original intervention was drainage improvement, by later adding green areas and/or the tunnel, if we consider candidate options from the specific case study; L455), as worse conditions are predicted for the years to come. Therefore, the actual difference between 1) solutions considering only one flood control strategy in the short term and 2) adaptive solutions where different strategies are all planned ahead and gradually implemented is just in the time of planning, not in their inherent flexibility. It might be interesting to study the effects of making plans for upgrades “on the fly” (i.e., when only one flood control approach is planned in the short run and additions are planned at later stages) vs. planning ahead all the different, gradual implementations. However, measuring flexibility using the proposed approach does not even help studying this. There are two major consequences related to this way of measuring flexibility, as detailed in the following two comments #3 and #4. To address these issues, the authors should either reframe the purpose of the manuscript, avoiding considering flexibility and centering the adaptability analysis around flexible pathways only (Fig. 5 and 6), or else re-run their analysis considering a more appropriate flexibility metric.
- In the light of my previous comment #2, comparing solutions with different numbers of combined flood control measures is misleading (e.g., Fig. 4), because strategies with more combined options (e.g., D+G+Tun30) will be systematically ranked as more flexible, although single-option solutions (such as GA and Dr) can also ensure high flexibility, in principle.
- Another implication of my previous comment #2 is that the proposed methodology does not really address one of its declared main purposes, i.e., finding trade-offs between robustness and adaptiveness (L195-196; L129-130; L146-147; L645-646; L653). In general, by combining a larger number of flood control measures over time, it should be expected that the resulting (gradually increasing) mitigation effects will be greater, during the infrastructure lifespan under the considered dynamic operating conditions (precipitation, sea level, etc.), compared with mitigation strategies using fewer gradual interventions; those solutions with many gradual interventions will be also ranked as more “flexible”, because of the ill-defined metric for flexibility. Because of this, solutions that implement many gradual updates (and hence that tend to achieve greater performance over time and more convenient cost-benefit ratios, through deferring expenses over time) will always tend to be preferred by this methodology, as they will be systematically regarded as more flexible too, with the risk of introducing bias. This is evident from Fig. 4, where the authors take D+G+Tun30 (brown curve) as the best flood mitigation approach (L473-475), due to its “well-balanced overall risk control performance and high value of flexibility”. However, if it were not for its high value of “flexibility”, that solution would be regarded as sub-optimal. Based on all the other objective values shown in Fig. 4, the actual best solution is instead Tun70 (blue curve), since it displays better values than D+G+Tun30 for average risk reduction rate (ARRR), Cost-Benefit ratio, and “valid period” (i.e., the period during which that flood control approach is effective). The low flexibility rank of Tun70 is only due to how flexibility is measured and does not reflect the fact that Tun70 would not preclude, in principle, combining other flood control strategies in the long period, if someday in the distant future upgrades were deemed necessary.
- The entire methodology section is vague and unclear, in the sense that it does not give a sense of logical flow starting from raw data processing to the final product. Different subsections describe different parts of the methodology, but they look somehow disconnected from each other. Given that the purpose of the paper is to provide a novel methodological framework for adaptive decision making, I strongly recommend revising this section in a way to guide the reader step by step through the proposed analysis framework. It may be beneficial to move Fig. 2 to the methodology section and organize the section around that Figure.
- Much information contained in the supplementary material is critical for understanding the proposed methodology and the presented case-study implementation; it is therefore suggested to suitably include parts of it into the revised methodology section, when addressing my previous comment #5.
- Lines 176-179: candidate hydrologic-hydraulic models suggested for use in the proposed decision-making procedure are quite different from the simpler SCS-CN model that is ultimately adopted in the case-study implementation (L395). I would state upfront that the methodology can leverage either sophisticated 1D-2D or simpler models, with likely major effects on the computational times. I would also state in the methodology section what model was ultimately selected for the case-study application, instead of mentioning it in the results (L395).
- (related to the previous comment) It important to note that model choice is expected to dramatically affect the overall computational times of the proposed procedure, but this is not clearly and exhaustively discussed at lines 618-620. In that paragraph, the authors emphasize the moderate computational and data requirements of the proposed methodology; however, this really depends on what hydrologic model is used within the methodology, and those considerations may not apply to cases where more sophisticated 1D-2D models (like those mentioned in the Methodology section; L176-179) are deployed.
- The term “valid period”, repeatedly used in the paper (e.g., L188; L200; L303; L308; L438; L489; L507; L520; L551; Table 1, etc.) to refer to the finite time period during which a given (single or combined) flood control measure is effective (L438-439), is confusing. I would change it to something easier to interpret, such as “effectiveness period” of a given flood control measure.
- L661-665: I tend to disagree with the authors’ claim that their work provides a theoretical foundation for decision-making methods in flood mitigation for coastal megacities. A “theoretical foundation” should imply the development of an entirely novel mathematical framework for addressing decision-making problems. While the resulting decision-making framework is novel, the authors primarily build upon existing metrics, models, and algorithms. Because of this, I would characterize the contribution of this work as a synthesis and application of established methodologies rather than the creation of a theoretical foundation.
- Eq. 1 (L221) and related text (L222-224) are unclear, because apparently the word “option” is used to refer to both individual flood control measures (a_i) and combinations of them (a or a_p) (L222-224). If the authors intend to determine the combination a_p of one or more options a_i that maximizes flood control performance across the range of scenarios w_j, chosen from the set of combinations a, then I suggest using different notations to indicate individual options and combinations of them (for example, keep a_i for individual options, and indicate combinations as c and c_p, respectively; also clearly distinguish between options and combinations of options in the text L222-224). Using the current notation, a_p could be mistakenly interpreted as a specific option, not as a combination of options. On the other hand, if the authors indeed intend to determine what individual option a_p among all candidates a_i shows the best performance across all operating condition scenarios w_j, then the current notation is formally incorrect, since the max function is applied to a single value, not a set of values. The correct notation should exclude the summation for i=1 to i=m from the current version of the equation and the range of variability of i should be specified below the max operator (in place of a).
- What performance metric is ultimately adopted for calculating f(a_i, w_j) in Eq. 1 and 2?
- Eq. 4 (L297) is unclear as many of the variables that appear in it are not defined anywhere. Eq. 4 is also formally incorrect, as distinct conflicting objectives considered in multi-objective problem formulations (for which trade-off solutions should be determined) should be expressed as separate objective functions to maximize or minimize.
- Fig. 5 and 6: what is the maximum time horizon considered for normalizing gamma in the x-axis (see Supplementary materials)? In other words, how many years correspond to gamma=1? For both figures, I suggest showing the time scale both in terms of gamma and the number of years.
- Fig. 5, L501. Why, from the time of GA and Tun30 installation (black dots, if I am not mistaken), is there an initial gradual performance increase (i.e., risk reduction) over time? Given that gradually worsening conditions (in terms of extreme precipitation trends, land subsidence, and sea level rise) are considered, should not performance (i.e., risk reduction) steadily decrease over time, with only “instantaneous” jumps associated with the installation? Even if we consider a finite period for the installation time, shouldn’t the performance increase only after the installation is complete? Please clarify.
- Fig. 5, L501. In the decreasing limbs, why does performance decrease non-linearly for Dr alone but then later it decreases linearly for Dr+GA as well as for Dr+GA+Tun30? I would have expected more non-linearity in performance decrease when multiple flood control measures operate together, since the curve would be reflecting the cumulative effects of supposedly different performance reduction dynamics over time. Please clarify.
- Notation used in equations 1, 2 and 3 (L221-247) should be consistent with the notation adopted later in the Results section (L406-459).
- Eq. 4 in the Supplementary material, Fig 3 in the manuscript: In the case of combined flood control measures implemented gradually (e.g., D+G, D+G+Tun30), do the authors consider the effects of the discount rate on the investment costs for additional interventions deferred over time? Are these effects reflected in the benefit-cost ratios plotted in Fig. 6?
- L432-450, Table 3; L615-616: the authors should provide the definition of density and coverage in the context of PRIM algorithm. Also, it is not clear how Table 3 is obtained by applying the condition given in Eq. 3.
- L352: the authors state that they performed a trade-off optimization, balancing robustness and adaptiveness. However, later in the paper the authors clearly state that they actually did not run any optimization, which is deferred to future work (L637-641). I would clarify from the beginning that the contribution of the manuscript only focuses on the formulation and not on the solution of a specific complex optimization problem defined using the proposed framework. It is worth highlighting here that the case study considered in this work had a very small number of candidate solutions, which made it possible to "manually" enumerate and compare all of them, to identify the optimal solution; however, more realistic applications, with possibly hundreds of candidate flood control scenarios, will require using an optimization algorithm to reduce the number of hydrologic simulation that would be otherwise required if all scenarios were simulated. Incidentally, L639-640 include an incorrect statement, as genetic algorithms are not machine learning methods.
- L617-627: these limitations should be discussed earlier, possibly in a dedicated subsection of the revised Methodology section.
- L676-677: The code used to perform the analysis would be an important asset for the reviewers to evaluate the proposed decision-making framework and the case-study implementation, especially in light of the unclear methodology section. In line with HESS journal policies, I encourage the authors to share polished open-source code and the complete dataset to enable the reproducibility of their analysis.
Minor comments:
- L98-101 The flow in this sentence is broken, like if there were some missing connectors or else two individual sentences got messed up into one.
- L182 “the calculation of indicators such as the average risk reduction rate (ARRR) and benefit-cost ratio (BCR) for each alternative option was performed for each scenario”. In this paragraph, some metrics are mentioned but their mathematical expression is not provided in the Methodology; additionally, the wording “indicators such as” is not appropriate in the Methodology section, as it is ambiguous. Does it mean that other metrics are calculated, besides ARRR and BCR? Or else do the authors intend that ARRR and BCR represent just examples of possible robustness metrics that the modeler may decide to use?
- L185 “Subsequently, the performance of each option and its combination was evaluated by quantitative comparison and ranking stability”. In this sentence, the notion of stability of the options is suddenly introduced, but it is never defined in the manuscript.
- L188 “Valid periods of the alternative options were determined based on the conditions of the successful scenarios under each (individual or combined) option, in conjunction with a specific flood control objective.” From this sentence, the concept of “valid period”, quite redundant throughout the paper, is not clear (also see my main comment #9). A clearer description of such concept is given in the Results section (lines 438-439); however, it would be beneficial to the reader if the concept of “valid period” were introduced earlier, from the first time that it is mentioned.
- L200-205: concepts such as “transient scenarios” and “signposts” are unclear in this section of the methodology. They become clear to the reader after reaching the Results section (Fig. 6), but the article should be organized in a way that all concepts unroll smoothly following sections order.
- L208: “The choice of robustness option is the meta-problem of how to decide (Herman et al., 2005)”. This sentence is too vague for and not related to the methodology section. I suggest either removing it or else moving it to the Introduction and elaborate more.
- L362-375: in this paragraph, parameters alpha, beta, and gamma are mentioned for the first time without prior introduction. These parameters are defined in the Supplementary material but should be also properly introduced in the manuscript.
- L408-415: “Benefit-cost is the evaluation dimension for the robustness metrics” needs rewording; “benefit-cost, was defined as the average risk reduction rate (ARRR) per unit cost based on the robustness metrics of Laplace’s Insufficient Reason” does not explain with sufficient detail how benefits and costs are calculated; also, the concept of Laplace’s Insufficient Reason is mentioned a few times in the paper but never introduced.
Citation: https://doi.org/10.5194/hess-2024-391-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 178 | 37 | 12 | 227 | 21 | 6 | 8 |
- HTML: 178
- PDF: 37
- XML: 12
- Total: 227
- Supplement: 21
- BibTeX: 6
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1