https://orcid.org/0009-0008-3835-7861
Figures
Abstract
Sea-level rise has large impacts on coastal areas. Many approaches to quantify the net benefits of regional or global coastal adaptation rely on strong assumptions about economically efficient decision-making and may neglect critical uncertainties about future extreme water levels and socioeconomic development. In particular, the deep and dynamic uncertainties associated with future sea levels complicate efforts to adapt coastal areas and infrastructure to account for sea-level rise. Robust decision-making (RDM) provides a way to identify adaptation strategies that perform well across uncertain sea-level futures. In this work, we quantified the impacts of an RDM approach on regional and global coastal adaptation using an economic regret, defined as the difference in cost of a strategy compared to the “optimal” outcome. We modeled decision-making using a regret-based criterion and computed the economic regret of each adaptation decision candidate, which we then compared to classical decision-making approaches. We found that the majority of coastal segments that changed strategies under regret-based criteria opted for a higher level of adaptation, primarily by expanding their retreat elevation. Although the total adaptation costs remained comparable to those under the cost-minimizing criterion, the use of regret led to a 3–50% reduction in flood damage, highlighting the advantages of prioritizing robust outcomes over purely cost optimization. While our analysis assumed immediate implementation of adaptation and evaluated a limited set of decision criteria, the results demonstrate that incorporating regret-based decision-making into coastal impact models can provide policy-relevant insights for designing more robust adaptation strategies under deep uncertainty.
Citation: Estevez C, Feke K, Srikrishnan V, Wong TE (2026) Regret-based global coastal adaptation decision-making under sea level uncertainty. PLOS Clim 5(8): e0000947. https://doi.org/10.1371/journal.pclm.0000947
Editor: Jamie Males, PLOS Climate, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: January 5, 2026; Accepted: July 24, 2026; Published: August 21, 2026
Copyright: © 2026 Estevez et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The code and data associated with this study are publicly available via Zenodo and GitHub to support transparency and reproducibility. The materials are organized into three components. First, the repository MimiCIAMRegret.jl contains the implementation of the regret-based decision criteria and is available on GitHub (https://github.com/CarolinaEstevez/MimiCIAMRegret.jl). Second, the local sea-level rise (LSLR) projection datasets used in this study are publicly available through Zenodo (https://zenodo.org/records/18500003). Third, the processed results, summary datasets, and plotting scripts used to generate the figures are archived in a Zenodo repository (https://zenodo.org/records/18671987). These repositories provide permanent access to the code, data, and materials necessary to reproduce the analyses presented in this study, in accordance with the journal’s open science requirements.
Funding: This material is based upon work supported in part by the National Science Foundation (NSF) under Award No. DMS-2213432. Tony Wong, Carolina Estevez, and Kelly Feke received support from this grant. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
Climate change is a crucial issue that affects all of society, leading to higher temperatures, shifting precipitation patterns, and accelerating sea-level rise (SLR) [1–5]. If global warming exceeds 2.0°C, flood-related damages could reach 2.8% of global GDP by 2100 [6]. Even under lower-end projections, warming beyond 2.5°C may increase flood costs from $1.2 to $1.6 trillion by 2150 [7]. Regional impacts are also substantial; for example, the U.S. Gulf of Mexico Coast alone could face $134-$177 billion (2015 US dollars) in damages by 2030 [8]. However, there are possible strategies for managing the risks posed by SLR. These include new coastal management policies [9] and robust adaptation planning approaches [10]. The adaptation strategies are typically categorized as protection (e.g., seawalls), accommodation (e.g., building elevation), and relocation (proactive or reactive) [11].
It is critical to account for uncertainties in SLR, storm surges, or future development [12]. Previous studies have shown that sea-level rise is one of the most influential factors shaping coastal adaptation strategy selection [13]. These challenges often stem from conditions of deep uncertainty, a situation in which experts or decision-makers cannot agree on the probabilities associated with potential future outcomes or even on the set of potential outcomes itself [14,15]. In the context of coastal adaptation, deep uncertainty arises from complex, interacting climate and socioeconomic processes that make it difficult to identify a single “optimal” strategy. Even if a model points to a single optimal strategy, its performance (in terms of coastal damages, for example) can be fragile, producing potentially overconfident and biased outcomes [16]. Additionally, strategies derived solely from optimization may fail to achieve equitable outcomes if the distributional impacts on affected populations are not explicitly considered [17]. Although considerable progress has been made in understanding how deeply uncertain SLR can affect the performance of coastal adaptation strategies, much of this work has been conducted at the local level [10,16,18–20]. However, the goal of the present study is to examine how different decision-making frameworks enable the identification of more effective strategies for navigating the uncertainties surrounding coastal risks. Global-scale models offer a useful and internally consistent framework to interrogate uncertainties by considering multiple SLR projections. These models still face challenges in fully representing deep uncertainty, particularly in how decision-making responds to a wide range of plausible futures [4,21].
Large-scale models such as the Coastal Impact and Adaptation Model (CIAM) and its open-source versions, pyCIAM and MimiCIAM, offer the ability to analyze spatially heterogeneous adaptation decisions across thousands of coastline segments. This enables a direct and internally consistent comparison of coastal adaptation costs with other climate-related damages and the costs of mitigation [22–24]. However, the scalability and efficiency of these models present new challenges. For example, their simplified representation of decision-making may overestimate the reliability of the suggested “optimal” adaptation strategies, which typically focus on minimizing adaptation costs and anticipated future damages. This shortcoming of large-scale models such as CIAM raises the importance of examining how we can ensure that the strategy selected is not only cost-effective but also satisfactorily effective across a broad set of future environmental and socioeconomic scenarios.
One approach to improving coastal adaptation modeling under deep uncertainty is to implement robust strategies that perform well across a wide range of plausible futures [14,25,26]. Robust decision-making (RDM) stands in contrast to traditional optimization models that rely on precise probabilities and perfect foresight to determine a single “optimal” decision. Additionally, depending on the decision criteria used, the RDM can focus on minimizing vulnerability and avoiding poor outcomes in worst-case scenarios and resilient policy choices [25,27,28]. This shift is particularly important in long-term coastal governance, where SLR impacts are uncertain, path-dependent, and potentially irreversible, and where decisions made today shape future exposure, equity, and public expenditure [29]. Specifically, we implement an economic regret criterion, defined as the difference between a strategy’s realized performance and the ex-post optimal outcome, to inform adaptation decisions. In this context, regret-based decision criteria can be interpreted as a tool within broader risk governance frameworks, to help identify strategies that reduce the likelihood of severe damages across uncertain futures [28]. While this is still a simplified representation of reality, regret-based rules also reflect the behavioral tendency of decision-makers to place greater weight on avoiding outcomes that perform worse than expected [30]. By embedding regret-based decision-making within a global-scale coastal model, this study provides a policy-relevant framework for selecting adaptation strategies that are not only cost-efficient, but also can help navigate deep uncertainty.
By introducing a robust decision-making criterion to a coastal impacts and adaptation model, this work improves our understanding of how imperfect knowledge of future sea levels influences adaptation decisions and costs. In Sections 2.1 and 2.2, we review the MimiCIAM model and relevant data sets that are used to generate future states-of-the-world (SOW) under the environmental and socioeconomic scenarios. In Section 2.3, we describe the regret-based decision-making criteria that we incorporate into MimiCIAM. In Section 3, we report the results of experiments to compare how our new regret-based decision-making framework affects the coastal adaptation decisions made within MimiCIAM and the associated costs and damages. Sections 4 and 5 summarize our findings and offer a broader perspective on the implications of these results.
2. Methods
2.1. Mathematical model
We use a version of the Coastal Impact and Adaptation Model [CIAM; 23], implemented in the Mimi integrated modeling framework [MimiCIAM; 24] in the Julia programming language. For each segment of the global coastline, with a median length of 18 km and similar physical characteristics, MimiCIAM evaluates the strategy that minimizes the anticipated adaptation costs. MimiCIAM uses exogenous data sets, such as inundation area, wetland extent, land value, and SLR to account for local geophysical and socioeconomic characteristics [23].
For each segment (indexed by ) and local sea-level rise (LSLR) scenario (indexed by
), MimiCIAM considers the following adaptation strategies: protection (the construction of seawalls), retreat (a proactive retreat from the coastline), or no adaptation [which might include reactive retreat, see Supporting Information Fig A in S1 Appendix from 24]. For protection and retreat, MimiCIAM uses an exogenous input dataset [DINAS-Coast 31] for the N-year peak storm surge return level, where N is 1, 10, 100, 1,000, or 10,000 years. For instance, adapting to the 100-year flood hazard level means preparing for a storm that is expected to occur once every 100 years. It is important to note that, while extreme sea levels are a key uncertainty driving coastal hazards, previous work using MimiCIAM found that the costs associated with adaptation are not influenced by the specific storm surge dataset used in this analysis [Supporting Information Fig K in S1 Appendix from 24]. Additionally, MimiCIAM incorporates scenarios for future climatic and socioeconomic changes through combined Shared Socioeconomic Pathways (SSP) and Representative Concentration Pathways (RCP) [32]. The model then determined the most cost-effective adaptation strategy, either retreat or protection, over a 50-year decision horizon. We outline this process below, but direct the interested reader to Diaz [23] for further details.
For each level of adaptation strategy (indexed by ), as well as no adaptation, at each time step (indexed by
), MimiCIAM calculated the total costs of adaptation and damage across five categories: wetland loss (
, Fig A(b) in S1 Appendix), retreat and relocation expenses (
, Fig A(d) in S1 Appendix), inundation/dryland loss(
, Fig A(b) in S1 Appendix), flooding-related properties and human life losses (
, Fig A(a) in S1 Appendix), and protection expenses (
, Fig A(c) in S1 Appendix, costs related to constructing and maintenance of seawall protection). We follow previous work using MimiCIAM and its predecessor, CIAM, by using a constant discount rate (
) equal to 4% [23,24]. The discount rate reflects how future costs are valued relative to the present costs [33]. In the original MimiCIAM, the combination of an adaptation strategy and level that yields the lowest net present value (NPV, Eq. 1) of the total estimated cost is considered to be the “optimal” strategy for each segment at that adaptation period (
).
To reflect actual planning decision time horizons for coastal adaptation, the model only considers the NPV for the first adaptation period (2010–2050) when determining the adaptation strategy to pursue. To maintain the initial level of protection throughout the full model time horizon (2010–2200), the chosen adaptation measures were updated roughly every 50 years. Unless otherwise stated, the monetary units for our results are presented as 2010 U.S. dollars.
2.2. Data
MimiCIAM evaluates the costs of adaptation strategies against flooding damage and sea-level changes in 12,148 coastal segments globally using the Dynamic Interactive Vulnerability Assessment database [DIVA, 34] and SSP-RCP scenarios [32]. The DIVA segments are irregularly shaped and divide the coastline into sections with similar geological and socioeconomic characteristics. They have a median length of approximately 18 km. For this study, we analyze adaptation decisions and costs for the full global case, as well as for 178 segments that belong to the U.S. Gulf of Mexico coast (“Gulf Coast”, Fig B in S1 Appendix). This extends roughly from the southern tip of Texas to the southern tip of Florida. The results obtained for the Gulf Coast show a similar trend compared to the global case, therefore, they are included in the supplementary material.
SSP-RCP scenarios are used to explore potential future damages under various socioeconomic growth and climate change conditions. We use four scenarios to capture the breadth of possible socioeconomic and geophysical futures: SSP1-2.6 (lower emissions scenario indicating sustainability-focused growth), SSP2-4.5 (a middle-of-the-road scenario where socioeconomic trends do not shift substantially from historical patterns), SSP4-6.0 (scenario with increased inequality and emphasis on regional solutions), and SSP5-8.5 (high-emissions scenario with rapid and fossil-fueled development) [35]. Within each SSP-RCP scenario, uncertainty in LSLR is represented through a large ensemble of projections, and the robustness analysis is conducted across these SLR futures rather than across alternative socioeconomic pathways. In addition, the model evaluates adaptation decisions independently for each coastal segment under the socioeconomic trajectories defined by the SSP scenarios. However, it does not explicitly represent differences in governance quality, institutional capacity, or social vulnerability across countries, which may influence adaptive capacity and the feasibility of adaptation strategies in practice.
Additionally, to account for uncertainty in future LSLR, we use a dataset of 10,000 sea-level projections from the MimiBRICK (Building Blocks for Relevant Ice and Climate Knowledge) sea-level model [36]. MimiBRICK’s model structure accounts for possible rapid ice sheet responses consistent with sea-level projections reported in the Sixth Assessment Report (AR6) [37]. MimiBRICK, coupled with SNEASY, incorporates uncertainties related to the carbon cycle, climate system, and sea-level changes, which are influenced by the model assumptions regarding future greenhouse gas emissions [24,36]. The major factors contributing to global mean sea-level changes, such as meltwater from the Greenland and Antarctic ice sheets and glaciers and ice caps, thermal expansion, and alterations in land water storage, are considered on MimiBRICK [38]. More details about MimiBRICK are provided in the supplementary material accompanying this work (Text A in S1 Appendix). We use a set of local sea-level fingerprints to downscale projections of these sea-level components to their impacts on the local SLR on a 1-degree grid [39].
In order to endogenously account for uncertainties in LSLR within MimiCIAM, we used a multivariate log-normal probability distribution to represent uncertainty in sea-level projections. We fit these distributions for each of the SSP1-2.6, SSP2-4.5, SSP4-6.0, and SSP5-8.5 scenarios using 10,000 posterior samples from the locally downscaled MimiBRICK projections of the SLR contributors. At each coastal segment, the dimensions of the multivariate fit were sea-level height and time. A multivariate log-normal distribution is employed and directly compared with the conventional multivariate normal assumption to represent LSLR uncertainty. The log-normal specification provides a better fit to the distribution of LSLR (Fig C-E in S1 Appendix), particularly by capturing its skewness, which are not well represented under the normal approximation. However, despite these differences in statistical fit, the resulting regret-based performance metrics and adaptation decisions remain effectively unchanged across both distributions (Fig F-I in S1 Appendix). This indicates that the overall conclusions are robust to the choice of distribution.
The MimiCIAM decision regarding adaptation choices, timing, and levels is made by minimizing a target metric across a subsample of these posterior samples for a given emission scenario (e.g., minimizing expected damage in the original version of CIAM). These LSLR projections are the basis for the RDM process, which allows us to implement an adaptation strategy under a single SOW and evaluate how well that decision performs across other possible SOWs, capturing the potential regret if the original assumptions are incorrect. To determine what size LSLR subsample from the multivariate normal distribution is needed to have stable adaptation decisions, we conducted experiments using 10–1,000 LSLR random samples for each segment and SSP-RCP scenario. The results show that future damage does not change substantially beyond 500 LSLR samples (Fig J in S1 Appendix). The results shown in this work use 500 LSLR samples.
2.3. Regret-based decision-making
Economic regret (subsequently referred as regret for brevity) quantifies the difference in performance between a decision made under uncertainty and the optimal possible decision if the decision-maker could have anticipated the outcome. Here, the strategy performance is measured by the NPV of future adaptation costs and damages using strategy , in segment
, facing the LSLR scenario
(NPVn,m,s, Eq. 1). For segment
and LSLR scenario
, there is some strategy that minimizes NPVn,m,s. Thus, the regret of using strategy
, in segment
, in LSLR scenario
is:
We use three decision criteria based on regret: (i) selecting the strategy that minimizes the expected regret over the ensemble of LSLR scenarios (
) (MinExpRegret, Eq. 3); (ii) picking the strategy
that minimizes the expected value of regret among the LSLR simulations that yield the highest 5% of regret (”conditional value at risk”; MiniCVaRRegret, Eq. 4); or (iii) selecting the strategy
that minimizes the maximum regret over LSLR scenarios (MiniMaxRegret, Eq. 5). The principal goal of MimiCIAM’s original decision-making criterion is to minimize expected net costs (adaptation costs - avoided damages) through an adaptation decision (where the expectation is over the distribution of extreme sea-levels). In the original model decision process, only a single LSLR simulation is considered each time that the model is run. Instead, we use the following regret-based criteria: MinExpRegret, MiniMaxRegret, and MiniCVaRRegret using ensembles of LSLR simulations. Defining
as the 95th percentile of the regret distribution (distributed over the LSLR ensemble), our three decision criteria are given by the following equations.
Evaluating multiple regret criteria also serves as a sensitivity analysis to the choice of decision rule, allowing us to assess how adaptation strategies change under different levels of risk aversion. In this study, MinExpRegret corresponds to the 50th percentile, MinCVaRRegret is evaluated at the 95th percentile, and MiniMaxRegret represents the limiting case at the 100th percentile. These selections can be interpreted as points along a continuum of increasing risk aversion. As the percentile level decreases from 95% toward 50%, the results obtained with MinCVaRRegret progressively converge toward those of MinExpRegret. Conversely, as the percentile increases from 95% toward 100%, the results become increasingly similar to those of MiniMaxRegret. We present the results obtained using MiniCVaRRegret and MiniMaxRegret in the main text, and we show results obtained using MinExpRegret as the decision criterion in the supplementary material. The results are broadly similar to those using MiniCVaRRegret and MiniMaxRegret.
3. Results
3.1. Changes in selected adaptation strategies
In the global case, we find that the majority of the segments pursue a retreat strategy, specifically at the 1-year and 10-year return levels (Fig 1, Fig M in S1 Appendix). Between 68–76% of coastal segments find retreat as the optimal strategy, while protection is optimal for 8–9% of segments, and no adaptation for 16–22% (Fig 1, and Fig M, P, Q and Table B in S1 Appendix). When applying regret-based decision criteria (MinExpRegret, MiniCVaRRegret, and MiniMaxRegret), the proportion of segments that opt for a different strategy compared with the original MimiCIAM model (MinExpDamage) ranges from 1–22%, with the highest level of change (22%, Table D in S1 Appendix) occurring under MiniMaxRegret in SSP5-8.5. Among the segments that followed a different adaptation strategy compared to the original MimiCIAM, 66–94% followed a higher level of adaptation (Fig 1, Table F in S1 Appendix). For example, changes involve shifting from no adaptation to retreat (Fig 1d), or maintaining retreat but increasing the return level (Fig 1c). Additionally, a few segments switch to no adaptation (Fig 1, and Fig M, P, Q and Table F in S1 Appendix), a phenomenon that we explore in more detail in Section 3.4.
Sankey diagram for all global coastal segments comparing strategies for coastal adaptation using MinExpDamage (left column of panels(a), (c), (e), and (g)) vs MiniMaxRegret (right column of panels (a), (c), (e), and (g)) and MinExpDamage (left column of panels (b), (d), (f), and (h)) vs MiniCVaRRegret (right column of panels (b), (d), (f), and (h)) under different SSP-RCP scenarios: SSP1-2.6 (a, b), SSP2-4.5 (c, d), SSP4-6.0 (e, f), and SSP5-8.5 (g, h). The adaptation strategy is represented in orange for protection, blue for retreat, and green for no adaptation. The darkness of the color increased as the level of adaptation increased.
3.2. Increases in adaptation levels
The implementation of regret-based decision-making criteria leads to higher levels of adaptation compared with the MinExpDamage criterion. This trend is evident in global simulations, particularly when adaptation decisions are made using the MiniMaxRegret and MiniCVaRRegret criteria (Fig 2). These more risk-averse criteria increase the frequency of higher adaptation levels (e.g., 1000- and 10,000-year return levels), especially in the high-end SSP5-8.5 scenario (Fig 2d). Overall, the increase in the adaptation level distribution reflects the tendency of regret-based approaches to prioritize robustness over optimality, favoring strategies that hedge against worst-case outcomes.
Frequency bar charts of the adaptation level of the strategies implemented in each of the decision criteria in the global coastal segments (MinExpDamage, blue bars; MiniCVaRRegret, green bars; MiniMaxRegret, purple bars) under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d).
The increase in adaptation is further supported by the observed growth in retreat elevations among segments that change strategies to retreat. The retreat elevation is the vertical elevation (in meters) above sea level, indicating the relocation of infrastructure, assets, or populations away from vulnerable coastal areas. This increase is notable globally, with 66–94% of the segments (Table F in S1 Appendix) modifying their strategy, either by increasing the level of retreat or switching to a retreat strategy. While MiniCVaRRegret increases the number of retreat implementations, it reduces median retreat elevation by 7% under SSP4-6.0 and 9–11% under SSP1-2.5, indicating a shift toward more spatially distributed but less intensive retreat. Nevertheless, compared to MinExpDamage, it yields higher median retreat elevations over time, by 1–239% in 2050, 8–74% in 2100, and 6–26% in 2150 under SSP2-4.5 and SSP5-8.5, suggesting precautionary retreat in high-risk segments. MiniMaxRegret, being more risk averse than MiniCVaRRegret, results in even larger increases in retreat elevation of 12–397% in 2050, 12–132% in 2100, and 13–42% in 2150 (Fig 3, Table J in S1 Appendix). These findings indicate that regret-informed decisions do not just shift strategies; they also lead to larger-scale implementations, particularly in highly exposed segments. Additionally, the smaller percentage difference in median retreat elevation by 2150 compared to 2050 reflects the fact that the MiniCVaRRegret and MiniMaxRegret criteria hedge more strongly against worst-case outcomes, leading to earlier and more extensive retreat. In contrast, under the MinExpDamage criterion, retreat is implemented gradually, thereby minimizing immediate costs.
Boxplot of the retreat elevation (m) for the global coastal segments that switch to a retreat strategy or continue under different SSP-RCP scenarios: (a) SSP1-2.6, (b) SSP2-4.5, (c) SSP4-6.0, and (d) SSP5-8.5. For each scenario, the retreat elevation is shown at three adaptation decision periods: 2050, 2100, and 2150. Results are grouped by decision-making criterion: MinExpDamage (blue), MiniCVaRRegret (green), and MiniMaxRegret (purple).
3.3. Adaptation costs using regret-based decision-making
Despite the differences in strategy selection, the total cost of adaptation remained relatively consistent across each SSP-RCP scenario over the four decision criteria. While regret-based criteria generally lead to higher costs incurred to pay for adaptation up front, these costs are offset by reduced flood damages in the longer term, resulting in comparable total costs (NPV) across scenarios (Fig X, Y and Tables M, N in S1 Appendix). This suggests that the trade-off between proactive investment and avoided damage tends to balance out. Specifically, MiniMaxRegret and MiniCVaRRegret generally use more proactive strategies that reduce exposure through relocation and long-term flood damages, at greater upfront cost. This tendency is amplified under higher LSLR scenarios, where larger potential regret values favor strategies that implement higher protection levels earlier in the planning horizon. Although these strategies require larger upfront investments, they can reduce long-term maintenance and adjustment costs across multiple scenarios. This highlights how robust decision-making enables a wider range of adaptation pathways at comparable costs, helps to balance short and long-term damages under deep uncertainty.
In terms of median flood damage costs, globally, MiniCVaRRegret achieves 4–18% reductions compared to flood damage costs by implementing MinExpDamage. However, there is an exception where flood damage under SSP2-4.5 where the median cost was similar (Fig 4, Fig Z and Table P in S1 Appendix). In contrast, MiniMaxRegret consistently lowered flood damages in all global cases, achieving reductions of 5–26% (Fig 4, Fig Z and Table P in S1 Appendix), with the largest improvements seen in the most extreme scenarios. Furthermore, MinExpDamage results showed a greater variability (Fig 4), suggesting less predictable performance across future SLR conditions. In contrast, regret-based strategies, particularly MiniMaxRegret, not only achieve lower median flood damages but also exhibit narrower uncertainty ranges, indicating more reliable performance across a wide range of potential futures. These results underscore the effectiveness of MiniMaxRegret in identifying robust strategies that perform consistently across diverse future scenarios, rather than relying on a single “optimal” adaptation decision.
Boxplots for the flood damage cost for all the segments in all global segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MiniCVaRRegret (green), and MiniMaxRegret (purple).
3.4. Shift to no adaptation
While we have found that regret-based decision-making typically results in higher levels of adaptation and lower flood damages, there are limited but important exceptions. As a representative case of these exceptions, a segment near Charlotte County, Florida, shifted from a 10-year retreat strategy to no adaptation under MiniMaxRegret and MiniCVaRRegret (Fig 5). In this case, the reactive retreat strategy associated with no adaptation yields a lower NPV for adaptation costs between 2010 and 2050, largely due to the smaller retreat elevation required at that time (Fig 5c). However, this comes at the expense of higher flood damages and increased retreat costs in later periods. The decision reflects the influence of discounting, which reduces the weight of future damages and costs in the NPV calculation. Similar dynamics were observed globally, where several segments, especially those with low population densities, shifted from proactive strategies to no adaptation (Fig AH in S1 Appendix). These outcomes highlight how discounting and population exposure can potentially drive counterintuitive adaptation choices with regret frameworks.
(a,b) Comparison of the bar plots of the net present value of the adaptation cost for Charlotte County in Florida for 2050 (a) and 2100 (b) using MinExpDamage, MinExpRegret, MiniMaxRegret, and MiniCVaRRegret. The NPV is divided into retreat cost (red), wetland cost (blue), flood cost (purple), and inundation cost (green). (c) Intervals of values of retreat elevation in 2050 for Charlotte County, considering a 10-year retreat or no-adapt strategy. (d) Gulf Coast map with the location of Charlotte County (green circle). The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md).
4. Discussion
We implemented new regret-based decision criteria to determine coastal adaptation strategies within a coastal impacts and adaptation model. We have also estimated future adaptation costs for the full global case, as well as the U.S. Gulf of Mexico Coast (see supplemental material). Across both the Gulf Coast and global cases, we find that segments that pursue different adaptation strategies under a regret-based decision criterion predominantly choose higher levels of adaptation compared to the approach that minimizes expected damages (Fig 1-3, Fig K-S in S1 Appendix). This behavior is consistent with real-world coastal planning frameworks that prioritize strategies performing well across multiple uncertain futures, such as the Louisiana Coastal Master Plan and the Dutch Delta Programme [40,41]. Notably, this increase in adaptation level did not result in higher overall costs (Fig X, Y and Table M, N in S1 Appendix), although in high-variability scenarios such as SSP5-8.5, the use of a single robust strategy per segment can lead to higher damages in some cases, reflecting a trade-off between robustness and optimality. In fact, the median total costs remained similar across all the criteria in its corresponding SSP-RCP scenario, showing that the higher adaptation investment under regret-based approaches was generally offset by corresponding reductions in flood damages (Fig 4, and Fig Z-AA in S1 Appendix). This finding has implications beyond the economic considerations. While our model focuses on quantifiable costs, including loss of life, it is crucial to recognize that lower future damages also translate to reduced human suffering, which is an aspect not fully captured by economic metrics alone. The fact that increased investment in climate adaptation reduces economic and human costs is critical for the development of policies guiding how we adapt to and manage environmental hazards affecting human systems.
There were differences among the three regret-based criteria. Adopting the MiniMaxRegret decision criterion leads to more changes in adaptation strategy compared to MinExpRegret and MiniCVaRRegret. This reflects that MiniMaxRegret uses a higher level of risk aversion than the other approaches (Fig 1-2, Fig K-S in S1 Appendix). By implementing MinExpRegret, 1–4% of the total segments change their strategy, where the ranges are dependent on the SSP-RCP scenario (Table C-D in S1 Appendix). On the other hand, by implementing MiniCVaRRegret, 2–14% of the total segments change their strategy. Additionally, using the compromise MiniMaxRegret decision criterion, 4–22% of the total segments change their strategy. In this way, the conditional value-at-risk approach balances risk aversion and overprotection, generally leading to similar costs across various scenarios while providing better protection against extreme outcomes (Fig X, Y in S1 Appendix). In practice, the choice among these criteria may reflect the risk tolerance of decision-makers, with MinExpRegret emphasizing average performance across scenarios and MiniCVaRRegret and MiniMaxRegret prioritizing robustness against severe but uncertain outcomes.
Planned relocation dominates as the optimal adaptation strategy across all SSP-RCP scenarios, in all decision criteria, and for both the global and Gulf Coast cases, particularly using 10-year and 100-year retreat levels (Fig 1, and Fig K-M in S1 Appendix). This is consistent with previous work that found the use of relocation can reduce the global adaptation cost of the 21st century compared to focusing on protection [23,42,43]. However, these results must be interpreted with caution, as relocation policies can disproportionately affect vulnerable communities and raise important environmental justice concerns [17]. Our model a simplification of the reality that relocation requires multi-level governance and broader societal engagement involving both coastal and inland communities [29].
A key caveat that accompanies the results presented here is that MimiCIAM and similar coastal adaptation models assume that not only can the optimal adaptation strategy be known to the model decision-maker, but this strategy is pursued without delay. In practice, it is likely that there may be delays associated with the implementation of any adaptation measures, for example, political, financial, and institutional constraints [17,44]. Moreover, factors such as the importance of social attachment and community identity will reduce the use of relocation as an adaptation measure, or at the very least, will increase the inertia that must be overcome for relocation. Therefore, the results presented here should not be interpreted as prescriptive policy recommendations. Instead, they should be viewed as a tool to explore possible adaptation pathways and support informed discussions with communities about future risks and adaptation options. Incorporating community perspectives into these discussions can improve the equity and feasibility of adaptation strategies [17,40,44]. In addition, because the model aggregates costs and damages at broad spatial scales and relies primarily on monetary metrics, it may obscure important distributional effects and does not fully capture non-economic losses such as cultural heritage, ecosystem services, or place attachment. Finally, MimiCIAM currently lacks explicit equity indicators that would allow the model to transparently assess how adaptation decisions distribute costs and benefits across populations and reflect underlying ethical considerations.
Despite the general tendency towards increasing levels of adaptation, we also find that a small number of segments show a decrease in the level of adaptation when using regret-based decision rules as compared to minimizing expected damage (Fig 1, and Fig K-S in S1 Appendix). These areas, which tend to have low population density (Fig AH in S1 Appendix), trade off the higher future damages and adaptation costs (generally reactive relocation) against lower near-term adaptation costs. Given that reactive adaptation is eventually pursued for these segments (Fig 5a-b), this implies that adopting flexible adaptation strategies, such as proactive and reactive during different adaptation phases, would lead to a lower overall cost of adaptation. This behavior is influenced by the assumed 4% discount rate, which reduces the present value of future costs (e.g., $100 in 50 years is valued at approximately $14). While the choice of discount rate affects absolute cost estimates, we find that the overall trends in adaptation strategies remain consistent across alternative rates (2% and 6%, Fig AJ, AK in S1 Appendix), particularly the tendency of regret-based approaches to encourage earlier and higher adaptation. This sensitivity points to the importance of adaptive decision-making to manage dynamic risk [45] and carefully considering the trade-off between valuing present costs against future benefits [42]. Additionally, extending the initial adaptation period from 2010 to 2100 may offer a more comprehensive understanding of the long-term implications of SLR, while also presenting consistent trends in adaptation strategies and total costs (Fig AL in S1 Appendix). However, such a 90-year horizon exceeds the typical planning timeframes used in many local-scale projects, such as Louisiana’s Coastal Master Plan, which looks ahead 50 years [46]. Moreover, the definition of regret relies on a perfect foresight benchmark, which may introduce a conceptual limitation, as sea level projections tend to diverge based on SSP-RCP scenario by mid-century [2]. To assess sensitivity, we tested an alternative formulation based on a single expected-value reference, finding that the overall results remain consistent (Fig AM, AN in S1 Appendix). This suggests that our main conclusions, particularly the preference for more proactive adaptation without substantial cost increases, are robust to the choice of regret definition.
5. Conclusion
Adopting regret-based criteria in coastal adaptation planning can improve the efficacy of our strategies to manage flood risks by decreasing the potential damages associated with SLR while not appreciably increasing the overall adaptation costs. In broad agreement with the conclusions of Lincke and Hinkel [42], we found that planned retreat is a robust adaptation strategy for the largest share of coastal segments in both the U.S. Gulf Coast and globally, highlighting the importance of managed retreat in many coastal areas. Implementing regret in the decision-making process provides a more realistic consideration of uncertainties in SLR projections and extreme sea level events, which is crucial for effective coastal management, principally for high-risk areas like the Gulf Coast. However, translating these modeling insights into practice will require engagement with stakeholders, governance institutions, and policymakers to ensure that adaptation strategies address issues of equity and environmental justice. Our results show that higher levels of protection can be attained at similar overall costs, underscoring the benefits of improved accounting for both uncertainties and decision-maker aversion to the risks associated with these uncertainties. These results highlight that addressing global coastal risk through proactive adaptation is a robust and economically efficient strategy in the face of global climate change.
Supporting information
S1 Appendix. Fig A. Components of MimiCIAM cost.
Representation of flooding damage of a hurricane (a), inundation, dry land loss, and wetland loss (b). Two adaptation scenarios are protection using a sea wall (c) and retreat from the coastal line. Fig B. Gulf Coast segments. Scatter map of the centroids of the 178 U.S. Gulf Coast segments. The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig C. Q-Q plot for SSP2-4.5. Quantile–quantile (Q–Q) comparison of local sea-level rise (LSLR) projections between MimiBRICK and MimiCIAM under the SSP2–4.5 scenario using multivariate normal distribution (red dots) and multivariate log-normal distribution (blue dots). Each panel compares the empirical distributions of LSLR across selected coastal segments and time horizons: 2010 (a–c), 2050 (d–f), and 2100 (g–i). Using three random selected segments: Cape Coral (FL, UnitedStates4484), Pascagoula (MS, UnitedStates4607), and Corpus Christi, (TX, UnitedStates4697). The horizontal axis represents MimiBRICK projections, while the vertical axis shows the corresponding MimiCIAM values. The green line denotes the 1:1 reference. Fig D. Q-Q plot for SSP4-6.0. Quantile–quantile (Q–Q) comparison of local sea-level rise (LSLR) projections between MimiBRICK and MimiCIAM under the SSP4–6.0 scenario using multivariate normal distribution (red dots) and multivariate log-normal distribution (blue dots). Each panel compares the empirical distributions of LSLR across selected coastal segments and time horizons: 2010 (a–c), 2050 (d–f), and 2100 (g–i). Using three random selected segments: Cape Coral (FL, UnitedStates4484), Pascagoula (MS, UnitedStates4607), and Corpus Christi, (TX, UnitedStates4697). The horizontal axis represents MimiBRICK projections, while the vertical axis shows the corresponding MimiCIAM values. The green line denotes the 1:1 reference. Fig E. Q-Q plot for SSP5-8.5. Quantile–quantile (Q–Q) comparison of local sea-level rise (LSLR) projections between MimiBRICK and MimiCIAM under the SSP5–8.5 scenario using multivariate normal distribution (red dots) and multivariate log-normal distribution (blue dots). Each panel compares the empirical distributions of LSLR across selected coastal segments and time horizons: 2010 (a–c), 2050 (d–f), and 2100 (g–i). Using three random selected segments: Cape Coral (FL, UnitedStates4484), Pascagoula (MS, UnitedStates4607), and Corpus Christi, (TX, UnitedStates4697). The horizontal axis represents MimiBRICK projections, while the vertical axis shows the corresponding MimiCIAM values. The green line denotes the 1:1 reference. Fig F. Sankey diagram for all global coastal segments comparing strategies using multivariate normal distribution. Sankey diagram for all global coastal segments comparing strategies for coastal adaptation using MinExpDamage (left column of (a), (c), (e), and (g) panels) vs MiniMaxRegret (right column of (a), (c), (e), and (g) panels) and MinExpDamage (left column of (b), (d), (f), and (h) panels) vs MiniCVaRRegret (right column of (b), (d), (f), and (h) panels) under different SSP-RCP scenarios: SSP1-2.6 (a, b), SSP2-4.5 (c, d), SSP4-6.0 (e, f), and SSP5-8.5 (g, h). The adaptation strategy is represented by orange for protection, blue for retreat, and green for no adaptation. The darkness of the color increases as the level of adaptation increases. Fig G. Total adaptation cost for all global coastal segments using multivariate normal distribution. Boxplots for the total NPV of future adaptation costs and damages for all global segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig H. Boxplots of sensitivity test using original MimiBRICK data for the total NPV of future adaptation costs. For all the segments in the Gulf Coast measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig I. Sankey diagram of sensitivity test using original MimiBRICK data for all Gulf Coast segments comparing strategies. Sankey diagram of sensitivity test using original MimiBRICK data for all Gulf Coast segments comparing strategies for coastal adaptation using MinExpDamage (left column of all panels) and MiniMaxRegret (right column of all panels) under SSP5-8.5. The adaptation strategy is represented by orange for Protect, blue for Retreat, and green for No Adapt. The darkness of the color increases as the level of adaptation increases. Fig J. Sensitivity test for LSLR sample size. Comparison of the mean optimal adaptation costs, represented as the net present value (NPV Optimal total), conducted across 10 randomly selected segments. This analysis employs 12 distinct sample sizes of local sea level rise (LSLR) cases between 10–1000, using SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). These plots illustrate the variations in adaptation costs under different future climate scenarios and sample sizes, and how they converge around 500 LSLR cases. Fig K. Sankey diagram for all Gulf Coast segments comparing strategies. Sankey diagram for all Gulf Coast segments comparing strategies for coastal adaptation using MinExpDamage (left column of all panels) and MinExpRegret (right column of all panels) under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). The adaptation strategy is represented by orange for Protect, blue for Retreat, and green for No Adapt. The darkness of the color increases as the level of adaptation increases. Fig L. Sankey diagram for all Gulf Coast segments comparing strategies. Sankey diagram for all Gulf Coast segments comparing strategies for coastal adaptation using MinExpDamage (left column of (a), (c), (e), and (g) panels) vs MiniMaxRegret (right column of (a), (c), (e), and (g) panels) and MinExpDamage (left column of (b), (d), (f), and (h) panels) vs MiniCVaRRegret (right column of (b), (d), (f), and (h) panels) under different SSP-RCP scenarios: SSP1-2.6 (a, b), SSP2-4.5 (c, d), SSP4-6.0 (e, f), and SSP5-8.5 (g, h). The adaptation strategy is represented by orange for protection, blue for retreat, and green for no adaptation. The darkness of the color increases as the level of adaptation increases. Fig M. Sankey diagram for all global segments comparing strategies. Sankey diagram for all global coastal segments comparing strategies for coastal adaptation using MinExpDamage (left column of all diagrams) and MinExpRegret (right column of all diagrams) under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). The adaptation strategy is represented by orange for protect, blue for retreat, and green for no adaptation. The darkness of the color increases as the level of adaptation increases. Fig N. Scatter map of Gulf Coast segments adaptation strategy. Scatter map of Gulf Coast region segments that change their adaptation strategy using MinExpRegret (right column) relative to MinExpDamage (Original MimiCIAM; left column) under different SSP-RCP scenarios: SSP1-2.6 (1st row), SSP2-4.5 (2nd row), SSP4-6.0 (3rd row), and SSP5-8.5 (4th row). The adaptation strategy is represented by an orange circle for protection, a blue diamond for retreat, and a green triangle for no adaptation. The darkness of the color increases as the level of adaptation increases. The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig O. Scatter map of Gulf Coast segments adaptation strategy. Scatter map of Gulf Coast region segments that change their adaptation strategy when using MiniMaxRegret (middle column) or MiniCVaRRegret (right column), relative to MinExpDamage (Original MimiCIAM; left column), under different SSP-RCP scenarios: SSP1-2.6 (1st row), SSP2-4.5 (2nd row), SSP4-6.0 (3rd row), and SSP5-8.5 (4th row). The adaptation strategy is represented by an orange circle for protection, a blue diamond for retreat, and a green triangle for no adaptation. The darkness of the color increases as the level of adaptation increases. The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig P. Scatter map of global segments adaptation strategy. Scatter map of global coastal segments that change their adaptation strategy using MinExpRegret (right column) relative to MinExpDamage (Original MimiCIAM; left column) under different SSP-RCP scenarios: SSP1-2.6 (1st row), SSP2-4.5 (2nd row), SSP4-6.0 (3rd row), and SSP5-8.5 (4th row). The adaptation strategy is represented by an orange circle for protection, a blue diamond for retreat, and a green triangle for no adaptation. The darkness of the color increases as the level of adaptation increases. The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig Q. Scatter map of global segments adaptation strategy. Scatter map of global coastal segments that change their adaptation strategy when using MiniMaxRegret (middle column) or MiniCVaRRegret (right column), relative to MinExpDamage (Original MimiCIAM; left column), under different SSP-RCP scenarios: SSP1-2.6 (1st row), SSP2-4.5 (2nd row), SSP4-6.0 (3rd row), and SSP5-8.5 (4th row). The adaptation strategy is represented by an orange circle for protection, a blue diamond for retreat, and a green triangle for no adaptation. The darkness of the color increases as the level of adaptation increases. The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig R. Adaptation level strategies comparison in the Gulf Coast. Frequency bar charts of the adaptation level of the strategies implemented in each of the decision criteria in the Gulf Coast region segments (MinExpDamage, blue bars; MinExpRegret, red bars; MiniCVaRRegret, green; MiniMaxRegret, purple) under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Fig S. Adaptation level strategies comparison of global segments. Frequency bar charts of the adaptation level of the strategies implemented in each of the decision criteria in the global coastal segments (MinExpDamage, blue bars; MinExpRegret, red bars) under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Fig T. Boxplot of the retreat elevation implemented in the Gulf Coast segments. Boxplot of the retreat elevation (m) for the Gulf Coast segments that switch to a retreat strategy or continue under different SSP-RCP scenarios: (a) SSP1-2.6, (b) SSP2-4.5, (c) SSP4-6.0, and (d) SSP5-8.5. For each scenario, the retreat elevation is shown at three adaptation decision periods: 2050, 2100, and 2150. Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig U. Boxplot of the retreat elevation implemented in the global segments. Boxplot of the retreat elevation (m) for the global coastal segments that switch to a retreat strategy or continue under different SSP-RCP scenarios: (a) SSP1-2.6, (b) SSP2-4.5, (c) SSP4-6.0, and (d) SSP5-8.5. For each scenario, the retreat elevation is shown at three adaptation decision periods: 2050, 2100, and 2150. Results are grouped by decision-making criterion: MinExpDamage (blue) and MinExpRegret (red). Fig V. Boxplot of the sea wall height implemented in the Gulf Coast segments. Boxplot of the height implemented for sea walls (m) for the Gulf Coast segments that switch to a protecting strategy or continue under different SSP-RCP scenarios: (a) SSP1-2.6, (b) SSP2-4.5, (c) SSP4-6.0, and (d) SSP5-8.5. For each scenario, the height is shown at three adaptation decision periods: 2050, 2100, and 2150. Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig W. Boxplot of the sea wall height implemented in the global segments. Boxplots of the height implemented for sea walls (m) for the global coastal segments that switch to a protecting strategy or continue under different SSP-RCP scenarios: (a) SSP1-2.6, (b) SSP2-4.5, (c) SSP4-6.0, and (d) SSP5-8.5. For each scenario, the height is shown at three adaptation decision periods: 2050, 2100, and 2150. Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig X. Total adaptation cost for the Gulf Coast segments. Boxplots for the total NPV of future adaptation costs and damages for all the segments in the Gulf Coast measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig Y. Total adaptation cost for the global segments. Boxplots for the total NPV of future adaptation costs and damages for all the segments in the global case measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig Z. Optimal flood cost for the Gulf Coast segments. Boxplots for the flood damage cost for all the segments in the Gulf Coast measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AA. Optimal flood cost for all global coastal segments. Boxplots for the flood damage cost for all the segments in all global segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue) and MinExpRegret (red). Fig AB. Optimal inundation cost for the Gulf Coast segments. Boxplots for the optimal inundation costs for the Gulf Coast segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AC. Optimal wetland cost for the Gulf Coast segments. Boxplots for the optimal wetland costs for the Gulf Coast segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AD. Optimal adaptation cost for the Gulf Coast segments. Boxplots for the optimal adaptation costs (Retreat cost + Construction cost) for the Gulf Coast segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AE. Optimal inundation cost for all coastal segments. Boxplots for the optimal inundation costs for global coastal segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AF. Optimal wetland cost for all coastal segments. Boxplots for the optimal wetland costs for global coastal segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AG. Optimal adaptation cost for all coastal segments. Boxplots for the optimal adaptation costs (Retreat cost + Construction cost) for global coastal segments measured in billion 2010 US dollars under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AH. Comparison of GDP per capita and population density per segment. Scatter plots of the GDP per capita vs population density of the 12,148 segments in 2050 (a,c) and 2100 (b,d) comparing the values of all the segments (blue dots) and segments that change from any strategy to No-Adapt using MiniMaxRegret (a,b, red dots) and MiniCVaRRegret(a,d, red dots). Fig AI. Results comparison for Port Jefferson. (a,b) Comparison of the bar plots of the net present value of the adaptation cost for Port Jefferson in New York for 2050 (a) and 2100 (b) using MinExpDamage, MinExpRegret, MiniMaxRegret, and MiniCVaRRegret. The NPV is divided into retreat cost (red), wetland cost (blue), flood cost (purple), and inundation cost (green). (c) Intervals of values of retreat elevation in 2050 for Port Jefferson 2050 considering a 1-year retreat or a 1,000-year retreat strategy. (d) Map with the location of Port Jefferson (green circle). The map base layer was generated by PlotlyJS in Julia, sourced from public domain Natural Earth datasets and available at https://www.naturalearthdata.com/downloads/; https://github.com/JuliaPlots/PlotlyJS.jl/blob/master/LICENSE.md). Fig AJ. Total adaptation cost for the Gulf Coast segments using discount rate equal to 2%. Boxplots for the total NPV of future adaptation costs and damages for all the segments in the Gulf Coast measured in billion 2010 US dollars using a discount rate equal to 2% under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AK. Total adaptation cost for the Gulf Coast segments using discount rate equal to 6%. Boxplots for the total NPV of future adaptation costs and damages for all the segments in the Gulf Coast measured in billion 2010 US dollars using a discount rate equal to 6% under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AL. Sankey diagram for all Gulf Coast segments comparing strategies using first decision horizon between 2010 and 2100. Sankey diagram for all Gulf Coast segments comparing strategies for coastal adaptation using MinExpDamage (left column of all panels) and MiniMaxRegret (right column of all panels) under different SSP5-8.5 using the first decision horizon between 2010 and 2100. The adaptation strategy is represented by orange for Protect, blue for Retreat, and green for No Adapt. The darkness of the color increases as the level of adaptation increases. Fig AM. Total adaptation cost for the Gulf Coast segments using a different regret definition. Boxplots for the total NPV of future adaptation costs and damages for all the segments in the Gulf Coast measured in billion 2010 US dollars using under different SSP-RCP scenarios: SSP1-2.6 (a), SSP2-4.5 (b), SSP4-6.0 (c), and SSP5-8.5 (d). Results are grouped by decision-making criterion: MinExpDamage (blue), MinExpRegret (red), MiniCVaRRegret (green), and MiniMaxRegret (purple). Fig AN. Sankey diagram for all Gulf Coast segments comparing strategies using a different regret definition. Sankey diagram for all Gulf Coast segments comparing strategies for coastal adaptation using MinExpDamage (left column of all panels) and MiniMaxRegret (right column of all panels) under different SSP5-8.5 using
. The adaptation strategy is represented by orange for Protect, blue for Retreat, and green for No Adapt. The darkness of the color increases as the level of adaptation increases. Table A. Percentage of the segments in the Gulf Coast that opt for Protect, Retreat, or No Adaptation. Percentage of the segments in the Gulf Coast that opt for Protect, Retreat, or No Adaptation as their adaptation strategy under each SSP-RCP scenario, based on four different decision-making criteria: MinExpDamage (3rd column), MinExpRegret (4th column), MiniCVaRRegret (5th column), and MiniMaxRegret (6th column). Values are reported for each SSP-RCP scenario (rows). Table B. Percentage of all global coastal segments that opt for Protect, Retreat, or No Adaptation. Percentage of the global coastal segments that opt for Protect, Retreat, or No Adaptation as their adaptation strategy under each SSP-RCP scenario, based on four different decision-making criteria: MinExpDamage (3rd column), MinExpRegret (4th column), MiniCVaRRegret (5th column), and MiniMaxRegret (6th column). Values are reported for each SSP-RCP scenario (rows). Table C. Percentage of Gulf Coast segments that pursue a different adaptation strategy. Percentage of Gulf Coast segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (2nd column), MiniCVaRRegret(3rd column), and MiniMaxRegret (4th column). Values are reported for each SSP-RCP scenario (rows). Table D. Percentage of global coastal segments that pursue a different adaptation strategy. Percentage of global coastal segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (2nd column), MiniCVaRRegret(3rd column), and MiniMaxRegret (4th column). Values are reported for each SSP-RCP scenario (rows). Table E. Percentage of Gulf Coast segments that follow a higher level of adaptation. Percentage of Gulf Coast segments that follow a higher level of adaptation over the ones that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (2nd column), MiniCVaRRegret(3rd column), and MiniMaxRegret (4th column). Values are reported for each SSP-RCP scenario (rows). Table F. Percentage of global coastal segments that follow a higher level of adaptation. Percentage of global coastal segments that follow a higher level of adaptation over the ones that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (2nd column), MiniCVaRRegret(3rd column), and MiniMaxRegret (4th column). Values are reported for each SSP-RCP scenario (rows). Table G. Percentage of Gulf Coast segments that opt for Protect, Retreat, or No Adapt over the ones that pursue a different adaptation strategy. Percentage of Gulf Coast segments that opt for Protect, Retreat, or No Adapt over the ones that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Table H. Percentage of global coastal segments that opt for Protect, Retreat, or No Adapt over the ones that pursue a different adaptation strategy. Percentage of global coastal segments that opt for Protect, Retreat, or No Adapt over the ones that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Table I. Median percentage difference in retreat elevation for the Gulf Coast segments. Median percentage difference in retreat elevation for the Gulf Coast segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario and for three time horizons: 2050, 2100, and 2150. Differences are calculated as the median retreat elevation under each regret-based criterion minus the median retreat elevation under MinExpDamage. Table J. Median percentage difference in retreat elevation for the global coastal segments. Median percentage difference in retreat elevation for the global coastal segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario and for three time horizons: 2050, 2100, and 2150. Differences are calculated as the median retreat elevation under each regret-based criterion minus the median retreat elevation under MinExpDamage. Table K. Median percentage difference in sea wall height for the Gulf Coast segments. Median percentage difference in sea wall height for the Gulf Coast segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario and for three time horizons: 2050, 2100, and 2150. Differences are calculated as the median sea wall height under each regret-based criterion minus the median sea wall height under MinExpDamage. Table L. Median percentage difference in sea wall height for the global coastal segments. Median percentage difference in sea wall height for the global coastal segments that pursue a different adaptation strategy relative to the MinExpDamage criterion, under three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario and for three time horizons: 2050, 2100, and 2150. Differences are calculated as the median sea wall height under each regret-based criterion minus the median sea wall height under MinExpDamage. Table M. Median percentage difference in total adaptation cost for Gulf Coast segments. Median percentage difference in total net present value of adaptation cost (NPV) over the period 2010–2200 of Gulf Coast segments, relative to the baseline MinExpDamage criterion. Results are represented for three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Differences are calculated as the median total NPV of adaptation cost under each regret-based criterion minus the median total NPV under MinExpDamage. Table N. Median percentage difference in total adaptation cost for all global coastal segments. Median percentage difference in total net present value of adaptation cost (NPV) over the period 2010–2200 of global coastal segments, relative to the baseline MinExpDamage criterion. Results are represented for three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Differences are calculated as the median total NPV of adaptation cost under each regret-based criterion minus the median total NPV under MinExpDamage. Table O. Median percentage difference in flood damage cost for Gulf Coast segments. Median percentage difference in flood damage cost over the period 2010–2200 of Gulf Coast segments, relative to the baseline MinExpDamage criterion. Results are represented for three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Differences are calculated as the median flood damage cost under each regret-based criterion minus the median flood damage cost under MinExpDamage. Table P. Median percentage difference in total adaptation cost for all global coastal segments. Median percentage difference in flood damage cost over the period 2010–2200 of global coastal segments, relative to the baseline MinExpDamage criterion. Results are represented for three regret-based decision-making criteria: MinExpRegret (3rd column), MiniCVaRRegret (4th column), and MiniMaxRegret (5th column). Values are reported for each SSP-RCP scenario. Differences are calculated as the median flood damage cost under each regret-based criterion minus the median flood damage cost under MinExpDamage.
https://doi.org/10.1371/journal.pclm.0000947.s001
(PDF)
Acknowledgments
We thank Frank Errickson, Matthew Hoffman, Basca Jadamba, Radley Powers, Brian Prest, Lisa Rennels, and Jordan Wingenroth for useful discussions and feedback. This material is based upon work supported by the National Science Foundation under Award No. DMS-2213432. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
References
- 1.
Church JA, Clark PU, Cazenave A, Gregory JM, Jevrejeva S, Levermann A, et al. Sea Level Change. Stocker TF, Qin D, Plattner G-K, Tignor M, Allen SK, Boschung J, et al. Climate Change 2013: The Physical Science Basis. Cambridge University Press. 2013.
- 2.
Fox-Kemper B, Hewitt HT, Xiao C, Aðalgeirsdóttir G, Drijfhout SS, Edwards TL, et al. Ocean, Cryosphere and Sea Level Change. Climate Change 2021 – The Physical Science Basis: Working Group I Contribution to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. 1.a ed. Cambridge University Press. 2021. https://doi.org/10.1017/9781009157896
- 3. Hartin C, McDuffie EE, Noiva K, Sarofim M, Parthum B, Martinich J, et al. Advancing the estimation of future climate impacts within the United States. EGUsphere. 2023;2023:1–32.
- 4. Hinkel J, Lincke D, Vafeidis AT, Perrette M, Nicholls RJ, Tol RSJ, et al. Coastal flood damage and adaptation costs under 21st century sea-level rise. Proc Natl Acad Sci U S A. 2014;111(9):3292–7. pmid:24596428
- 5. Riera-Spiegelhalder M, Campos-Rodrigues L, Enseñado EM, Dekker-Arlain J den, Papadopoulou O, Arampatzis S, et al. Socio-economic assessment of ecosystem-based and other adaptation strategies in coastal areas: A systematic review. JMSE. 2023;11(2):319.
- 6. Jevrejeva S, Jackson LP, Grinsted A, Lincke D, Marzeion B. Flood damage costs under the sea level rise with warming of 1.5 °C and 2 °C. Environ Res Lett. 2018;13(7):074014.
- 7. Wong TE, Dake S, Feke K, Darnell C, Srikrishnan V. Coastal adaptation and damage costs at different global warming thresholds. npj Nat Hazards. 2025;2(1).
- 8. Reguero BG, Beck MW, Bresch DN, Calil J, Meliane I. Comparing the cost effectiveness of nature-based and coastal adaptation: A case study from the Gulf Coast of the United States. PLoS One. 2018;13(4):e0192132. pmid:29641611
- 9. Nicholls R, Brown S, Goodwin P, Wahl T, Lowe J, Solan M, et al. Stabilization of global temperature at 1.5°C and 2.0°C: Implications for coastal areas. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2018;376(2119):Article 2119.
- 10. Haasnoot M, Kwadijk J, van Alphen J, Le Bars D, van den Hurk B, Diermanse F, et al. Adaptation to uncertain sea-level rise; how uncertainty in Antarctic mass-loss impacts the coastal adaptation strategy of the Netherlands. Environ Res Lett. 2020;15(3):034007.
- 11. Duijndam SJ, Botzen WJW, Hagedoorn LC, Aerts JCJH. Anticipating sea-level rise and human migration: A review of empirical evidence and avenues for future research. Wiley Interdiscip Rev Clim Change. 2022;13(1):e747. pmid:35865647
- 12.
Oddo PC, Lee BS, Garner GG, Srikrishnan V, Reed PM, Forest CE, et al. Deep uncertainties in sea-level rise and storm surge projections: Implications for coastal flood risk management. Risk Analysis. 2017. https://doi.org/10.1111/risa.12888
- 13. Fischbach J, Johnson D, Kuhn K. Bias and efficiency tradeoffs in the selection of storm suites used to estimate flood risk. JMSE. 2016;4(1):10.
- 14.
Lempert RJ. Shaping the next one hundred years: New methods for quantitative, long-term policy analysis. 2003.
- 15.
Walker WE, Lempert RJ, Kwakkel JH. Deep uncertainty. Encyclopedia of Operations Research and Management Science. 3.a ed. Springer US. 2013. 1641.
- 16. Ruckert KL, Srikrishnan V, Keller K. Characterizing the deep uncertainties surrounding coastal flood hazard projections: A case study for Norfolk, VA. Sci Rep. 2019;9(1):11373. pmid:31388022
- 17. Siders AR. Social justice implications of US managed retreat buyout programs. Climatic Change. 2018;152(2):239–57.
- 18. Abadie LM, Galarraga I, Markandya A, Sainz de Murieta E. Risk measures and the distribution of damage curves for 600 European coastal cities. Environ Res Lett. 2019;14(6):064021.
- 19. Hummel MA, Griffin R, Arkema K, Guerry AD. Economic evaluation of sea-level rise adaptation strongly influenced by hydrodynamic feedbacks. Proc Natl Acad Sci U S A. 2021;118(29):e2025961118. pmid:34253614
- 20. Lorie M, Neumann JE, Sarofim MC, Jones R, Horton RM, Kopp RE, et al. Modeling coastal flood risk and adaptation response under future climate conditions. Clim Risk Manag. 2020;29:100233. pmid:32832376
- 21. Tiggeloven T, de Moel H, Winsemius HC, Eilander D, Erkens G, Gebremedhin E, et al. Global-scale benefit–cost analysis of coastal flood adaptation to different flood risk drivers using structural measures. Nat Hazards Earth Syst Sci. 2020;20(4):1025–44.
- 22.
Depsky N, Bolliger I, Allen D, Choi JH, Delgado M, Greenstone M, et al. DSCIM-Coastal v1.0: An open-source modeling platform for global impacts of sea level rise. EGUsphere. 2022;:1–47. https://doi.org/10.5194/egusphere-2022-198
- 23. Diaz DB. Estimating global damages from sea level rise with the Coastal Impact and Adaptation Model (CIAM). Climatic Change. 2016;137(1–2):143–56.
- 24. Wong TE, Ledna C, Rennels L, Sheets H, Errickson FC, Diaz D, et al. Sea Level and socioeconomic uncertainty drives high-end coastal adaptation costs. Earths Future. 2022;10(12):e2022EF003061. pmid:37035442
- 25. McInerney D, Lempert R, Keller K. What are robust strategies in the face of uncertain climate threshold responses?. Climatic Change. 2012;112(3–4):547–68.
- 26.
Walker WE, Lempert RJ, Kwakkel JH. Deep uncertainty. Encyclopedia of Operations Research and Management Science. 3 ed. Springer US. 2016. 1641.
- 27. Kasprzyk JR, Nataraj S, Reed PM, Lempert RJ. Many objective robust decision making for complex environmental systems undergoing change. Environmental Modelling & Software. 2013;42:55–71.
- 28. Shwom R, Kopp RE. Long-term risk governance: when do societies act before crisis?. Journal of Risk Research. 2019;22(11):1374–90.
- 29. Bongarts Lebbe T, Rey-Valette H, Chaumillon É, Camus G, Almar R, Cazenave A, et al. Designing coastal adaptation strategies to tackle sea level rise. Front Mar Sci. 2021;8.
- 30. Rockafellar RT, Uryasev S. The fundamental risk quadrangle in risk management, optimization and statistical estimation. Surveys in Operations Research and Management Science. 2013;18(1–2):33–53.
- 31.
DINAS-Coast Consortium. DIVA coastal database version 1.0. Potsdam Institute for Climate Impacts Research (PIK). 2006.
- 32. O’Neill BC. The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6. Geoscientific Model Development. 2016;9(9):3461–82.
- 33.
Prest B. Discounting 101: How does discounting help decisionmakers understand the costs and benefits of choices and policies—and how does it apply to climate change?. https://media.rff.org/documents/Discounting_Explainer_-_Final.pdf. 2020.
- 34. Vafeidis AT, Nicholls RJ, McFadden L, Tol RSJ, Hinkel J, Spencer T, et al. A new global coastal database for impact and vulnerability analysis to sea-level rise. Journal of Coastal Research. 2008;244:917–24.
- 35.
Hausfather Z. Socioeconomic pathways’ explore future climate change. Carbon Brief: Clear on Climate. https://www.carbonbrief.org/explainer-how-shared-socioeconomic-pathways-explore-future-climate-change/. 2018.
- 36. Wong TE, Rennels L, Errickson F, Srikrishnan V, Bakker A, Keller K, et al. MimiBRICK.jl: A Julia package for the BRICK model for sea-level change in the Mimi integrated modeling framework. JOSS. 2022;7(76):4556.
- 37. Wong TE, Bakker AMR, Ruckert K, Applegate P, Slangen ABA, Keller K. BRICK v0.2, a simple, accessible, and transparent model framework for climate and regional sea-level projections. Geoscientific Model Development. 2017;10(7):2741–60.
- 38. Urban NM, Keller K. Probabilistic hindcasts and projections of the coupled climate, carbon cycle and Atlantic meridional overturning circulation system: A Bayesian fusion of century-scale observations with a simple model. Tellus A: Dynamic Meteorology and Oceanography. 2010;62(5):737.
- 39. Slangen ABA, Carson M, Katsman CA, van de Wal RSW, Köhl A, Vermeersen LLA, et al. Projecting twenty-first century regional sea-level changes. Climatic Change. 2014;124(1–2):317–32.
- 40.
Coastal Protection and Restoration Authority of Louisiana. Louisiana’s Comprehensive Master Plan for a Sustainable Coast. Coastal Protection and Restoration Authority of Louisiana. 2023. https://coastal.la.gov/wp-content/uploads/2023/06/230531_CPRA_MP_Final-for-web_spreads.pdf
- 41.
Mahmood S. Flood Risk Reduction: Cutting-Edge Strategies and Innovative Technologies. Springer Nature Switzerland. 2026. https://doi.org/10.1007/978-3-032-12225-4
- 42. Lincke D, Hinkel J. Coastal Migration due to 21st Century Sea‐Level Rise. Earth’s Future. 2021;9(5).
- 43. Xu L, Yang X, Chen D, Sornette D, Prishchepov AV, Ding S, et al. Global coastal human settlement retreat driven by vulnerability to coastal climate hazards. Nat Clim Chang. 2025;15(10):1060–70.
- 44. Rasmussen DJ, Kopp RE, Shwom R, Oppenheimer M. The political complexity of coastal flood risk reduction: Lessons for climate adaptation public works in the U.S. Earth’s Future. 2021;9(2):e2020EF001575.
- 45. Völz V, Hinkel J, Kupfer S, MacPherson LR, Wulff Norrby CJ. Learning about sea level rise uncertainty improves coastal adaptation decisions. Earth’s Future. 2024;12(10).
- 46.
CPRA CP. 2023 Coastal Master Plan. State of Louisiana. 2023. https://coastal.la.gov/our-plan/2023-coastal-master-plan/