the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Sep 2022
Multi-scenario multi-objective analysis of downscaled shared socio-economic pathways (SSPs) for robust policy development in coupled human-water systems
Abstract. Shared socio-economic pathways (SSP) scenario analysis is concerned with developing climate change adaptation strategies that perform well across a wide range of plausible future socio-economic and climate change conditions. However, downscaled/localized SSP scenarios, most relevant for regional climate adaptation, are poorly understood in terms of their deep uncertainties and how these scenarios can contribute to the development of robust regional policies in coupled human-water systems. In the present study, we propose a new framework that integrates a multi-scenario multi-objective (meta-criteria) optimization analysis of a set of downscaled/localized SSP storylines with the robust decision-making concept to find optimal robust solutions under deep uncertainty concerning regional climate adaptation. By developing an integrated dynamic simulation-optimization model, potential policy alternatives are investigated, and their robustness evaluated based on four key objectives: farm income, groundwater depletion, soil salinity, and reliability. Scenario-based multi-objective optimization for multiple SSP scenarios is merged into a robust optimization problem and evaluated in parallel. The proposed framework is applied to study potential robust solutions for vulnerabilities of a real-world human-water system in Pakistan's Rechna Doab region that has multiple stakeholders and conflicting objectives. The results revealed Pareto optimal solutions that are both optimally feasible and robustly efficient. The socio-environmental conditions of SSPs have a significant influence on the estimated robustness. The candidate solutions under scenario SSP1 are remarkably comparable to those offered by scenario SSP5, which was deemed to be the best among the SSPs evaluated. SSP3 was the least desirable of the SSP scenarios examined and solutions resulted in undesirable soil salinity, groundwater depletion, and reliability values. By incorporating SSP narratives and quantitative scenario analysis, the proposed framework revealed advantages for integrated dynamic modelling of human-water systems with a high level of uncertainty and complex interconnections to discover robust climate change adaptation solutions.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(2016 KB)
Interactive discussion
Status: closed
-
RC1: 'Comment on hess-2022-297', Anonymous Referee #1, 11 Dec 2022
Alizadeh et al. illustrate a scenario-based multi-objective optimization of water management strategies to be robust to different localized SSP scenarios. The framework builds upon multi-objective robust optimization methods in the literature by grounding the optimization in localized SSP scenarios co-designed by stakeholders.
I like the idea of using localized SSP scenarios for multi-objective robust optimization, as it has both computational benefits from using only a handful of scenarios as opposed to hundreds of possible scenarios generated across deeply uncertain parameter values and interpretability benefits for stakeholders. However, as currently presented, these benefits are not illustrated and many of the methods are unclear. With respect to the Methods, the decision variables were never mentioned (groundwater pumping?) and it was unclear to me whether a robust optimization across the 5 SSPs was performed or 5 separate optimizations, one to each SSP. The notation in the equations was also confusing, inconsistent, and not always explained.
With respect to the contribution, the benefits of the new framework over existing frameworks should be illustrated if the framework is the novelty. I recommend the authors compare their approach to multi-objective robust optimization and multi-scenario robust decision making (see Bartholomew and Kwakkel, 2020) to show its computational benefits. Including a survey of stakeholders with respect to the interpretability of these different approaches would also be informative if they do in fact gravitate to this approach because of the localized SSPs. Without such a comparison, this paper is simply another case study on multi-objective robust optimization of water management strategies for climatic and socioeconomic change and more appropriate to Journal of Hydrology: Regional Studies or as a Case Study in Journal of Water Resources Planning and Management, with a focus on the localized insights from the study (which are thin as currently written).
As such, I recommend this paper be rejected but with encouragement to resubmit if the authors refocus the paper on comparing their framework to existing multi-objective robust optimization approaches to illustrate its benefits. Please see the attached annotated manuscript for additional comments.
Reference:
Bartholomew, E., & Kwakkel, J. H. (2020). On considering robustness in the search phase of robust decision making: a comparison of many-objective robust decision making, multi-scenario many-objective robust decision making, and many objective robust optimization. Environmental Modelling & Software, 127, 104699.
-
RC2: 'Comment on hess-2022-297', Anonymous Referee #2, 17 May 2023
Â
Â
The paper, if I understand it correctly, tries to use DMDU approaches to assess the robustness of policy interventions given downscaled SSP scenarios.
Â
Â
Â
The first problem is that the SSP framework, in its original inception, was designed to enable the comparison of global IAMs. True, a wide literature on using SSPs and downscaling them has emerged, but I fundamentally question the policy relevance of this approach. Scenario planning aims at revealing vulnerabilities of proposed policies. Those vulnerabilities often depend significantly on local conditions and case-specific details. Whether or not the SSP framework has any relevance for this is questionable at best. So, why even bother with the SSP framework? Merely using it because others do so and because it is available makes no sense.
Â
Second, the authors use a case study. They claim but don’t substantiate that this case is suitable. So, can you give me scientific, rather than pragmatic, arguments substantiating the claim that this case is suitable?
Â
Third, in the methods section, I struggle to clearly understand exactly what is done, how it is done, and where the novelty is. There are several reasons for this. First, there are various statements I don’t understand. What do you mean when you say that local (I guess you mean localized or downscaled) SSP scenarios are deeply uncertain, or write about uncertain scenarios? Scenarios are not deeply uncertain. We are deeply uncertain about the future and use scenarios to bound and characterize this uncertainty. Any given scenario itself should be unambiguous and clear. Likewise, the authors speak of probabilities under deep uncertainty. This is a contradiction. Deep uncertainty implies that the various parties to a decision do not know or cannot agree on prior probabilities. Second, the equations are wrong. For example, in equation 1, both x and k are an element of P, but then the indexing makes no sense. Likewise, in the text below this equation, i refers first scenario instance and then a specific objective. Third, the authors copy an open loop intertemporal formulation for the decision problem from the highly simplistic lake problem test function but don’t reflect or acknowledge that this formulation implicitly means that you have perfect foresight. Likewise, the authors copy a statical-based robustness metric but fail to argue why for this problem, this metric is appropriate. I question this, given the statement in line 391 that only 5 scenarios are used.  Fourth, no implementation details are provided, so I have no way of checking the implementation of any of this. In short, the methodological presentation falls far short of what I consider to be minimally adequate.
Â
Fourth, the presentation of the results falls short of standard practice in the DMDU literature. Figure 4a and b show the same data with just another projection. So either a or b is redundant. I suggest removing A because 3D plots are utterly incomprehensible in print. Second, it is standard practice to align the direction of desirability in parallel axis visualizations. So, the ideal solution should be a straight line at either the top or bottom of the figure. Here that is, I think, not done because, e.g., farm income should be high while, e.g., soil salinity should be low. Moreover, in figure 5, the color bar is redundant because reliability is already shown in the parallel axis plot. Moreover, what remains unclear to me is whether there is a single set of Pareto optimal solutions across the SSP scenarios or whether there are separate sets for each scenario (goes back to the previous paragraph, and the lack of clarity on the methods). My previous comments likewise apply to figure 6: why have 2 panels here? Why is there a colorbar? Is the direction of desirability correct (and have the robustness metrics been calculated correctly, given that statistical robustness metrics can be tricky when mixing minimization and maximization objectives)? I fail to understand the point of figure 7. There are standard visualizations for scenario discovery results in the literature. Still, even a simple table with attention also for the statistical significance (Bryant and Lempert 2010) of the identified limits would suffice here. In the caption of figure 8, I am utterly at a loss. Suddenly we have SSP scenarios and final scenarios discovered for the five SSPs. I don’t even know what this might mean.
Â
Last, there are various strange citations. Just 2 examples:
           e.g., line 48, Miettinen, 2012 for deep uncertainty
           e.g., line 63, 64 on clustering of exploratory modeling results.
Â
Bryant, B. P. and R. J. Lempert (2010). "Thinking Inside the Box: a participatory computer-assisted approach to scenario discovery." Technological Forecasting and Social Change 77(1): 34-49.
Â
Citation: https://doi.org/10.5194/hess-2022-297-RC2
Interactive discussion
Status: closed
-
RC1: 'Comment on hess-2022-297', Anonymous Referee #1, 11 Dec 2022
Alizadeh et al. illustrate a scenario-based multi-objective optimization of water management strategies to be robust to different localized SSP scenarios. The framework builds upon multi-objective robust optimization methods in the literature by grounding the optimization in localized SSP scenarios co-designed by stakeholders.
I like the idea of using localized SSP scenarios for multi-objective robust optimization, as it has both computational benefits from using only a handful of scenarios as opposed to hundreds of possible scenarios generated across deeply uncertain parameter values and interpretability benefits for stakeholders. However, as currently presented, these benefits are not illustrated and many of the methods are unclear. With respect to the Methods, the decision variables were never mentioned (groundwater pumping?) and it was unclear to me whether a robust optimization across the 5 SSPs was performed or 5 separate optimizations, one to each SSP. The notation in the equations was also confusing, inconsistent, and not always explained.
With respect to the contribution, the benefits of the new framework over existing frameworks should be illustrated if the framework is the novelty. I recommend the authors compare their approach to multi-objective robust optimization and multi-scenario robust decision making (see Bartholomew and Kwakkel, 2020) to show its computational benefits. Including a survey of stakeholders with respect to the interpretability of these different approaches would also be informative if they do in fact gravitate to this approach because of the localized SSPs. Without such a comparison, this paper is simply another case study on multi-objective robust optimization of water management strategies for climatic and socioeconomic change and more appropriate to Journal of Hydrology: Regional Studies or as a Case Study in Journal of Water Resources Planning and Management, with a focus on the localized insights from the study (which are thin as currently written).
As such, I recommend this paper be rejected but with encouragement to resubmit if the authors refocus the paper on comparing their framework to existing multi-objective robust optimization approaches to illustrate its benefits. Please see the attached annotated manuscript for additional comments.
Reference:
Bartholomew, E., & Kwakkel, J. H. (2020). On considering robustness in the search phase of robust decision making: a comparison of many-objective robust decision making, multi-scenario many-objective robust decision making, and many objective robust optimization. Environmental Modelling & Software, 127, 104699.
-
RC2: 'Comment on hess-2022-297', Anonymous Referee #2, 17 May 2023
Â
Â
The paper, if I understand it correctly, tries to use DMDU approaches to assess the robustness of policy interventions given downscaled SSP scenarios.
Â
Â
Â
The first problem is that the SSP framework, in its original inception, was designed to enable the comparison of global IAMs. True, a wide literature on using SSPs and downscaling them has emerged, but I fundamentally question the policy relevance of this approach. Scenario planning aims at revealing vulnerabilities of proposed policies. Those vulnerabilities often depend significantly on local conditions and case-specific details. Whether or not the SSP framework has any relevance for this is questionable at best. So, why even bother with the SSP framework? Merely using it because others do so and because it is available makes no sense.
Â
Second, the authors use a case study. They claim but don’t substantiate that this case is suitable. So, can you give me scientific, rather than pragmatic, arguments substantiating the claim that this case is suitable?
Â
Third, in the methods section, I struggle to clearly understand exactly what is done, how it is done, and where the novelty is. There are several reasons for this. First, there are various statements I don’t understand. What do you mean when you say that local (I guess you mean localized or downscaled) SSP scenarios are deeply uncertain, or write about uncertain scenarios? Scenarios are not deeply uncertain. We are deeply uncertain about the future and use scenarios to bound and characterize this uncertainty. Any given scenario itself should be unambiguous and clear. Likewise, the authors speak of probabilities under deep uncertainty. This is a contradiction. Deep uncertainty implies that the various parties to a decision do not know or cannot agree on prior probabilities. Second, the equations are wrong. For example, in equation 1, both x and k are an element of P, but then the indexing makes no sense. Likewise, in the text below this equation, i refers first scenario instance and then a specific objective. Third, the authors copy an open loop intertemporal formulation for the decision problem from the highly simplistic lake problem test function but don’t reflect or acknowledge that this formulation implicitly means that you have perfect foresight. Likewise, the authors copy a statical-based robustness metric but fail to argue why for this problem, this metric is appropriate. I question this, given the statement in line 391 that only 5 scenarios are used.  Fourth, no implementation details are provided, so I have no way of checking the implementation of any of this. In short, the methodological presentation falls far short of what I consider to be minimally adequate.
Â
Fourth, the presentation of the results falls short of standard practice in the DMDU literature. Figure 4a and b show the same data with just another projection. So either a or b is redundant. I suggest removing A because 3D plots are utterly incomprehensible in print. Second, it is standard practice to align the direction of desirability in parallel axis visualizations. So, the ideal solution should be a straight line at either the top or bottom of the figure. Here that is, I think, not done because, e.g., farm income should be high while, e.g., soil salinity should be low. Moreover, in figure 5, the color bar is redundant because reliability is already shown in the parallel axis plot. Moreover, what remains unclear to me is whether there is a single set of Pareto optimal solutions across the SSP scenarios or whether there are separate sets for each scenario (goes back to the previous paragraph, and the lack of clarity on the methods). My previous comments likewise apply to figure 6: why have 2 panels here? Why is there a colorbar? Is the direction of desirability correct (and have the robustness metrics been calculated correctly, given that statistical robustness metrics can be tricky when mixing minimization and maximization objectives)? I fail to understand the point of figure 7. There are standard visualizations for scenario discovery results in the literature. Still, even a simple table with attention also for the statistical significance (Bryant and Lempert 2010) of the identified limits would suffice here. In the caption of figure 8, I am utterly at a loss. Suddenly we have SSP scenarios and final scenarios discovered for the five SSPs. I don’t even know what this might mean.
Â
Last, there are various strange citations. Just 2 examples:
           e.g., line 48, Miettinen, 2012 for deep uncertainty
           e.g., line 63, 64 on clustering of exploratory modeling results.
Â
Bryant, B. P. and R. J. Lempert (2010). "Thinking Inside the Box: a participatory computer-assisted approach to scenario discovery." Technological Forecasting and Social Change 77(1): 34-49.
Â
Citation: https://doi.org/10.5194/hess-2022-297-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 533 | 222 | 39 | 794 | 38 | 45 |
- HTML: 533
- PDF: 222
- XML: 39
- Total: 794
- BibTeX: 38
- EndNote: 45
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1