Climate Change Adaptation Planning for a Rail Transit Line through Multicriteria Decision Analysis

Sea-level rise (SLR) and the attendant increases in the frequency and intensity of coastal flood events are expected to significantly increase the risk to coastal communities and critical infrastructure (IPCC 2022; Martello and Whittle 2023). Hence, investments in climate change adaptation will become increasingly important, particularly if property owners and infrastructure managers are to avoid the unacceptably high damage costs associated with future flood events. Prudent investments in these climate change adaptation projects will require cost–benefit analysis (CBA) to ensure that the benefits (e.g., flood risk reduction benefits) outweigh the costs (i.e., net present value; NPV > 0). For public agencies, which are typically not profit-oriented, conventional CBA is often an inadequate screening tool, because other competing nonmonetary priorities (e.g., equity, safety, mobility, etc.) also influence and inform capital investment decisions.

Multicriteria decision analysis (MCDA) is a scalable, generalizable, transparent, and well-established decision-support technique that can enable the consideration and balancing of such competing and potentially conflicting monetary and nonmonetary priorities when evaluating the relative merits of potential decision alternatives (Kiker et al. 2005; Porthin et al. 2013; Linkov 2021). MCDA can enable a holistic evaluation of alternatives through a consideration of technical, economic, environmental, and social criteria (Wang et al. 2009) while simultaneously incorporating the priorities and preferences of decision makers and engaged stakeholders (Linkov 2021; Skdimore and Cohon 2023). There is substantial prior precedent of applying MCDA in the context of environmental planning (Kiker et al. 2005; Hajkowicz and Collins 2007; Wang et al. 2009; Huang et al. 2011), flood protection (Meyer et al. 2009; da Silva et al. 2020; Bagheri et al. 2021), infrastructure management (Kabir et al. 2014), and climate adaptation planning (Porthin et al. 2013; Haque 2016; Lawrence et al. 2019; Zucaro et al. 2021; Skidmore and Cohon 2023; da Silva et al. 2022), although there has thus far been a lack of application in the context of infrastructure adaptation planning.

MCDA analyses vary in their degree of sophistication according to the structuring of the decision space and evaluation of objective functions. Examples include the weighted sum method (WSM; inclusive of multiattribute utility theory, MAUT), outranking methods (e.g., ELECTRE I, TOPSIS, PROMETHEE), analytical hierarchy process (AHP), and several others (Zopounidis and Doumpos 2002; Bouyssou 2009; Meyer et al. 2009; Wang et al. 2009; Kabir et al. 2014; Samstad et al. 2019). MAUT methods have been previously implemented in the context of climate adaptation planning (Porthin et al. 2013; Lawrence et al. 2019) and infrastructure management (Kabir et al. 2014), although existing applications to date have not applied this framework for climate adaptation planning within the institutional context of public transportation agencies. Consequently, questions relating to transportation-specific adaptation decision criteria, capital investment priorities, and risk preferences have remained largely unexamined to date.

This gap persists, despite a separate body of research and practice which suggests that, in addition to other capital investment priorities (Pollack et al. 2015; MBTA 2022b), public transportation infrastructure managers are placing increasing importance on system resilience (Wilbanks and Fernandez 2014; ASCE and Ayyub 2018), particularly with respect to climate change (Miao et al. 2018; NASEM 2017, 2021a, b). Climate change resilience is supported by investments in climate change adaptation (Martello et al. 2021; Martello and Whittle 2023), yet the current literature offers little guidance as to how resilience can be systematically incorporated into capital investment planning and decision-making. Addressing these gaps in the literature, we develop and apply a stochastic MCDA framework for transit infrastructure climate change adaptation planning, considering decision criteria relating to system performance, economic, and environmental impacts. We implement an MAUT approach, considering variability in stakeholder priorities, while also applying a novel characterization of uncertainty in stakeholder risk preferences.

We apply this framework to a case study, in which we investigate a set of potential climate change adaptation alternatives for three geographically and hydraulically distinct segments of a regional rail rapid transit line, the Massachusetts Bay Transportation Authority (MBTA) Blue Line (one of four lines that form the core transit network in Greater Boston, owned and operated by the MBTA). The 9.7 km (6.0 mi) long Blue Line extends from downtown Boston beneath Boston Harbor (through a 120-year-old tunnel) northward to Revere, traversing coastally adjacent low-lying areas that will be increasingly exposed to coastal flood risks with SLR. Through a sample survey of an academic panel of transportation experts, we calibrate a stochastic MCDA model, investigating how uncertainty in stakeholder priorities and risk preferences can influence the rank preference of adaptation alternatives across locations. We further compare the resulting variance in rank preference against an agnostic weighting and risk preference baseline, in which weights are assigned based on random criteria weights and random assignment of risk preferences.

In alignment with prior literature (Porthin et al. 2013; Wang et al. 2009; Linkov 2021), we conduct an MCDA in three basic steps: (1) problem identification (inclusive of identifying potential alternatives and specific decision criteria); (2) model construction (where criteria weights and scoring methods, e.g., utility functions, are established); and (3) model implementation (i.e., evaluation of decision criteria for each alternative, as well as overall performance of alternatives). Fig. 1 provides a hierarchical overview of these three steps.

We illustrate the proposed MCDA approach through a real-world case study by considering several potential adaptation alternatives for the Blue Line (Fig. 3). For the purposes of this analysis, we discretize the Blue Line into three separate segments based on their hydraulic characteristics, as shown in Fig. 3: (1) Central Tunnel (CT) segment (from Bowdoin to Airport Portal); (2) East Boston (EB) segment (from Airport Portal to the Beachmont Elevated); and (3) Revere (R) segment (from the Beachmont Elevated to Wonderland).

Before developing adaptation alternatives, we establish a DFE for the purposes of preliminary design. We apply this DFE across all alternatives under consideration to enable direct comparison of the relative merits of the design alternatives. Selection of this DFE requires the exercise of significant engineering judgment, because the design flood (taken to be the 1–100-year coastal flood event, consistent with ASCE 2015) will increase over time with future SLR, which is deeply uncertain. As shown in Fig. 4(a), sea levels in Boston Harbor have already risen by approximately 0.3 m since 1920 (NOAA 2023), and based on current trends (consistent with the Intermediate SLR projection provided by Sweet et al. 2022), 1.2 m of SLR is expected in Boston Harbor by the year 2100. Based on the 95th percentile of the intermediate pathway (Sweet et al. 2022), we estimate that a DFE of +4.6 m (+15 ft) NAVD88 would provide sufficient protection through 2100. As shown in Fig. 4(b), this DFE is adequate till the end of this century under the 95th percentile SLR of most Sweet et al. (2022) and IPCC (2022) SLR scenarios.

For each of the Blue Line segments, we consider four adaptation alternatives, as shown in Fig. 5. These four alternatives are also compared with a no-action baseline (Alternative 1), as shown in Fig. 3, in which no adaptation investments are undertaken. Alternative 2 involves shore-based measures (many of which have already been proposed through the Climate Ready Boston program; City of Boston 2017, 2020, 2022; WHG 2022), by which municipalities and regional public agencies develop flood risk reduction measures, providing exogenous coastal flood exposure benefits to the Blue Line. Under this alternative, we assume that the MBTA will be a partner to a cost share agreement that is responsible for an uncertain portion of capital and maintenance costs [cost share uncertainty assumed to follow a beta distribution, β(1.5, 3.0), implying an expected 33% cost share; Martello et al. 2023]. Alternative 3 presents a robustness-based improvement that involves flood-proofing stations and the construction of a floodwall around the perimeter of the at-grade right-of-way (ROW). Alternative 4 focuses on improving rapidity of recovery through the construction of new tunnels to cover additional sections of ROW and elevation of the existing ROW (e.g., construction of a viaduct), in combination with station floodproofing, to avoid flood exposure. Alternative 5 corresponds to a managed retreat in which rail service in East Boston and Revere is replaced by bus rapid transit (BRT) service and the existing Blue Line maintenance facility (Orient Heights Yard) is relocated, enhancing adaptive capacity by enabling flexible rerouting of the proposed BRT service. Further details of each alternative, including preliminary design requirements and detailed preliminary cost estimates, can be found in Appendix B.

We consider a flexible implementation of all alternatives through an exposure-based decision rule (i.e., alternatives are constructed if the 1–20-year flood event is expected to exceed a critical threshold elevation, based on short-term SLR trends). Further details of the preliminary design of each alternative, as well as the ROA valuation and flexible implementation, are provided in Appendix B.

Respondent ranks for each criterion are shown in Fig. 6, along with the average weight corresponding to the survey responses, as calculated by using Eq. (2). The expert academic panel (n = 17 respondents) collectively ranked safety impacts as the highest priority decision criteria, followed by mobility impacts and system preservation. Indirect benefits were ranked slightly higher on average than direct benefits, which, in turn, were ranked slightly higher than equity impacts. Environmental impacts and budget impacts were ranked least important on average by respondents and were therefore assigned the smallest average criteria weights, as shown on the left column in Fig. 6.

Presenting decision criteria in this same rank order, Fig. 7 summarizes the risk preferences specified by respondents. Respondents were most risk averse with respect to the safety impacts and system preservation criteria. Respondents were most risk prone with respect to mobility impacts. Respondents were approximately evenly distributed between risk averse, neutral, and prone for the remaining decision criteria, which is reflective of an overall risk neutral stance in aggregate across most decision criteria.

Table 3 summarizes the performance of each adaptation alternative for each subcriterion along each Blue Line segment. These results readily enable performance comparisons along specific subcriteria, although a meaningful comparison between alternatives is difficult at this stage (i.e., prior to the evaluation of alternative performance). However, we note that across each of the three segments, the no action (Alternative 1) clearly underperforms all other alternatives across each of the subcriteria. Notably, the cumulative cost variability of Alternative 1 is at least an order of magnitude greater than all other alternatives across each of the three segments, reflecting the significantly higher exposure to future coastal flood events under this no-action alternative.

Implementing the stochastic MAUT framework under Scenario 1 (agnostic baseline, randomized criteria ranking, and risk preferences), we find that shored based measures (SBMs) (Alternative 2) provide a significantly higher expected utility than all other alternatives across all three segments, as shown in Fig. 8(a). Although we observe an appreciable overlap in the range of utility scores between Alternatives 2, 4, and 5 across all three segments [Figs. 8(b–d)], according to the RAI matrices in Fig. 8(e), Alternative 2 is the highest-ranked (i.e., most favorable) alternative more than 95% of the time. Aside from the no-action alternative (1), which is clearly the least favorable ( $\mathrm{RA}{\mathrm{I}}_{5,1}=100\text{\%}$ across all segments), there is considerable overlap in performance between the remaining alternatives.

The range of utility scores for Alternatives, 3, 4, and 5 overlap significantly in the Central Tunnel [Fig. 8(b)] and East Boston [Fig. 8(c)] segments. In Revere, Alternatives 3 and 5 perform nearly identically, as shown by their directly overlapping PDFs in Fig. 8(d), with Alternative 4 sharing a similar mean value and a slightly wider range of utility scores. As a consequence of this nearly identical performance, RAI values for Alternatives 3, 4, and 5 in the Revere segment are uniformly distributed, as shown in Fig. 8(e) (i.e., there is an approximately equal probability of any of the three alternatives ranking 2nd, 3rd, or 4th). In the East Boston segment, the appreciable overlap between Alternatives 3, 4, and 5 yields a similar distribution of RAI values, with Alternative 3 most likely ranking 4th $(\mathrm{RA}{\mathrm{I}}_{4,3}=77\text{\%})$. The RAI values for Alternatives 3, 4, and 5 in the Central Tunnel segment show a clearer expected ranking, with the cover/elevate alternative (4) ranking 2nd in 62% of the simulations $(\mathrm{RA}{\mathrm{I}}_{2,4}=62\text{\%})$, floodwalls (Alternative 3) ranking 3rd $(\mathrm{RA}{\mathrm{I}}_{5,1}=52\text{\%})$, and managed retreat (Alternative 5) ranking 4th $(\mathrm{RA}{\mathrm{I}}_{5,1}=74\text{\%})$.

Next, we consider the full range of priorities and risk preferences of the sample set of stakeholders (Scenario 2, survey sample). In this scenario, Alternative 4 (cover/elevate) provides a noticeably higher-expected utility across all segments, as shown in Fig. 9(a). We further observe a greater overlap in the range of utility scores, particularly between Alternatives 2 and 4 in East Boston [Fig. 9(c)] and Revere [Fig. 9(d)]. This appreciably lowers the RAI dominance of Alternative 2 across all segments, reducing RAI_1,2 to 81% in the Central Tunnel, 73% in East Boston, and 75% in Revere [Fig. 9(e)]. As a consequence of the higher utility scores of Alternative 4, it ranks as the preferred alternative for an appreciable subset of the simulations across each segment (Central Tunnel, $\mathrm{RA}{\mathrm{I}}_{1,4}=14\text{\%}$; East Boston, $\mathrm{RA}{\mathrm{I}}_{1,4}=25\text{\%}$; Revere, $\mathrm{RA}{\mathrm{I}}_{1,4}=21\text{\%}$). Under Scenario 2, we further observe a decrease in the relative favorability of the managed retreat alternative in East Boston and Revere [i.e., the PDF shifts to the left in Figs. 9(c and d)]. RAI values for the managed retreat alternative indicate appreciably higher probabilities of lower rankings, as compared to those of the agnostic baseline (Scenario 1). As shown in Fig. 9(e), Alternative 5 is most likely to rank 4th in the Central Tunnel $(\mathrm{RA}{\mathrm{I}}_{4,5}=74\text{\%})$ and Revere $(\mathrm{RA}{\mathrm{I}}_{4,5}=77\text{\%})$ and is more likely to Rank 3rd or 4th in the East Boston segment when compared with the results presented in Scenario 1.

Lastly, we consider the relative performance of the five alternatives using the average criteria weights and risk preferences of the sample set of stakeholders (Scenario 3) in which we observe nearly identical expected utility values, as shown in Fig. 10(a). However, as a consequence of eliminating uncertainty in criteria priorities and reducing uncertainty in risk preferences, there is considerably less spread in utility scores, wherein we observe only an appreciable overlap between Alternatives 3 and 4 in the Central Tunnel segment [Fig. 10(b)] and between Alternatives 3 and 5 in the East Boston segment [Fig. 10(c)]. This considerably reduces the spread in RAI values across all three locations, with SBMs (Alternative 2) ranking as the preferred alternative in more than 98% of the MCS trials across all three locations, as shown in Fig. 10(e) (i.e., $\mathrm{RA}{\mathrm{I}}_{1,2}\ge 98\text{\%}$). We can also observe the consequence of overlapping utility scores between Alternatives 3 and 4 in the Central Tunnel segment in Fig. 10(e), wherein the RAI values show a less certain relative ranking, with the Cover/Elevate alternative (4) ranking 2nd in 88% of trials (i.e., $\mathrm{RA}{\mathrm{I}}_{2,4}\ge 88\text{\%}$). Similarly, the greater overlap between Alternatives 3 and 5 in the East Boston segment yields a less certain relative ranking, with the floodwall alternative (3) ranking 3rd only 57% of the time (i.e., $\mathrm{RA}{\mathrm{I}}_{3,3}\ge 57\text{\%}$).

These results suggest that, even in the absence of clearly defined investment priorities or risk preferences (Scenario 1), SBMs (Alternative 2) are the preferred alternative across all three segments. Further specification of decision criteria priority and risk preferences can appreciably change the relative ranking of alternatives, potentially leading to lower-rank acceptability of a preferred alternative, if a consensus is not reached (Scenario 2). Yet, if stakeholders are able to establish a consensus on the relative priority of decision criteria and risk preferences (e.g., an agreement to equally weight stakeholder preferences), then a clearer rank prioritization can be established, with a clear preferred alternative (Scenario 3). The contrast between these three scenarios underscores the importance of understanding and considering stakeholder preferences in decision-making, as well as the importance of reaching a consensus between stakeholders when weighing the relative merit of potential adaptation investments. Although some degree of uncertainty in stakeholder risk preferences remains, we demonstrate that a clear rank preferability of adaptation alternatives can be established, even in the absence of precise utility functions.

The results of the case study presented suggest that adaptation investments that reduce exposure to transit infrastructure through adaptation measures exogenous to the system (Alternative 2) are generally preferable to other types of resilience-enhancing adaptation measures. These results further suggest that adaptation investments aimed at improving system rapidity (i.e., Alternative 4, cover/elevate) are preferable to robustness (Alternative 3, floodwalls) or adaptive capacity-enhancing investments (Alternative 5, managed retreat), despite the highest expected capital costs and lower-expected NPV across all three Blue Line segments. Further, although the managed retreat, Alternative 5, had among the highest-expected NPV in the Revere segment and second highest in the East Boston segment, it was nonetheless the worst-performing alternative when considering performance across the full set of subcriteria, particularly as stakeholder priorities and preferences were increasingly brought into focus (i.e., in Scenarios 2 and 3). Given that this managed retreat alternative considers only a moderate service reduction, it is clear that a more drastic managed retreat measure (e.g., one in which transit service is East Boston and Revere is removed entirely) would be perceived as untenable, based on the perspective of the expert panel. Although the relative performance of these different types of resilience-enhancing adaptation investments is probably not generalizable beyond the case study, these results, particularly those under Scenario 1 in East Boston (where we see nearly identical performance between the robustness, rapidity, and adaptive capacity enhancing Alternatives, 3, 4, and 5 respectively), suggest that a consideration of all types of resilience-enhancing adaptation alternatives is worthwhile, insofar as they have the potential to provide similar levels of utility, despite their varied typologies and investment valuations (i.e., NPV).

The preferability of the SBMs, Alternative 2, underscores the latent value of interagency collaboration in the public sector, particularly when planning flood risk reduction investments that provide direct benefits to public infrastructure. At the same time, we note that this collaboration necessarily requires some degree of dependence and reliance on external agencies and municipalities, which may be perceived as unfavorable by some decision makers, particularly if a history of poor cooperation exists between some or all of the agencies involved. Further, the practical difficulty of executing such an alternative is probably greater than the other endogenous (i.e., self-contained and internally planned) adaptation schemes. This can introduce a considerably greater risk of inaction, particularly if there is significant difficulty in finding financing, negotiating cost-share agreements, or sharing project management responsibility between public sector agencies. Although this dependence on other agencies may present additional coordination challenges, such coordination is likely to bring additional benefits beyond cost sharing. For example, the establishment of cost-sharing expectations, as well as the required interagency dialog and cooperation, can facilitate future joint investment in adaptation projects, insofar as the institutional experience gained can reduce prevailing institutional barriers to cooperation, such as the lack of clearly defined roles and financing expectations (Mesdaghi et al. 2022).

If institutional barriers prevent interagency cooperation and the SBM alternative is considered infeasible, the cover/elevate alternative (4) would be the preferred alternative, based on the survey average weights and risk preferences (Scenario 3). However, in the absence of stakeholder priorities and preferences (Scenario 1) or a lack of consensus among stakeholders (Scenario 2), it is markedly less clear which among the remaining is the preferred alternative. In such situations, further refinement of alternatives and subsequent reanalysis of relative performance is probably warranted. Such an iterative planning approach is in alignment with existing resilience-based design and flood risk management planning approaches and philosophies (Yoe and Orth 1996; ASCE and Ayyub 2018); the proposed MCDA framework can better facilitate stakeholder engagement in such iterative planning processes.

While the current case study is limited to a set of five alternative adaptation schemes, other alternatives could be considered. For example, alternatives with varying levels of protection (i.e., different design flood elevations) could readily be accommodated, enabling further exploration of trade-offs between different types of resilience-enhancing adaptation measures. The proposed framework can also address budgetary constraints (by way of a lower maximum criteria scale value), thereby allowing decision makers to explore how such constraints could alter the relative ranking of alternatives. Similarly, the lower and upper bounds of any particular quantitative subcriteria could be further refined and calibrated to align with additional real-world project constraints. Further extensions could include a consideration of uncertainties in the subcriteria performance, consistent with the evaluation of utility function uncertainty. Alternatively, the qualitative criteria performance for each alternative could be evaluated directly by decision makers and stakeholders, thereby enabling a consideration of subcriteria performance uncertainty and average performance values in a similar manner to stakeholder preferences in Scenarios 2 and 3 (i.e., consider the full-range subcriteria performance levels by sampling survey results and assessment considering the average of survey responses).

Such extensions would allow for variance in subcriteria performance, although there are additional limitations associated with the limited range and stepwise nature of the qualitative scoring metric (which causes sensitivity to changes in utility under minor perturbations of qualitative subcriteria performance). We note that in the current methodology, relatively minor changes in qualitative subcriteria performance can be highly consequential, whereas minor perturbations in quantitative subcriteria performance cause minimal changes in the overall utility of a given alternative. If a wider range of qualitative performance is considered (e.g., evaluation over the range of −10, 10 instead of −3, 3), then the impact of marginal changes in subcriteria performance could be reduced. However, such a finergrained qualitative scoring metric is probably more difficult to apply and consequently less transparent to stakeholders and decision makers. Where possible, further quantitative evaluation of additional subcriteria would also further reduce this sensitivity, although at the expense of significantly increasing the complexity of evaluating alternatives.

Lastly, although we survey a sample set of decision makers in the case study, the analysis framework presented is readily extensible to participatory planning approaches. In addition to considering the diversity of viewpoints, priorities, and risk preferences within an organization or agency, a consideration of the viewpoints of additional external stakeholders, including members of the community and general public, provides a mechanism for enhancing the procedural equity of the planning process (Malloy and Ashcraft 2020). In addition to the equity benefits brought by such engagement, such participatory planning approaches are more likely to yield results with sustained public support, which prior experiences and case studies suggest are critical for the completion of flood risk reduction projects (Rasmussen et al. 2023).

Enhancing the climate resilience of transportation infrastructure requires investments in climate change adaptation projects. In this study, we develop a transit-specific MCDA framework for evaluating the relative merit of resilience-enhancing climate adaptation investments while considering the priorities and risk preferences of decision makers and uncertainty in the utility of subcriteria. Considering a set of transit-specific set of adaptation decision criteria and a novel approach to the development of uncertain utility functions, we apply this framework to a case study considering a set of potential adaptation alternatives for a rail rapid transit line that is vulnerable to coastal flooding. Our results suggest that investments in exposure reduction through shore-based measures are the preferred alternative, although the relative performance of the remaining alternatives is relatively uncertain and dependent on the priorities and preferences of stakeholders. Through a consideration of three separate scenarios, we demonstrate the importance of soliciting stakeholder priorities and preferences, as well as the uncertainty reduction of establishing consensus among stakeholders.

The case study for the MBTA Blue Line in Boston results suggest that resilience-enhancing adaptation investments that aim to reduce exposure (i.e., shore-based measures exogenous to the system) are preferable to endogenous (i.e., internal) adaptation investments. The proposed (adaptive capacity enhancing) managed retreat was least preferable, given that this alternative yielded no improvement in safety and less beneficial impacts for mobility compared with other alternatives. The preferability of the shore-based measure alternative underscores the value of interagency collaboration, although implementation beyond the planning phase is liable to carry more incompletion risk as compared to an internal adaptation investment. The MCDA framework presented readily enables transit agencies to evaluate the relative merit of potential climate resilience–enhancing adaptation investments through a multifaceted approach that is particularly useful for differentiating between several adaptation investments with positive net present values (i.e., a positive return on investment).

The MCDA framework presented enables transit agencies to develop strategic adaptation investment plans more holistically in alignment with internal agency priorities, while simultaneously allowing for the inclusion and consideration of additional community stakeholders. Such a holistic and collaborative evaluation and comparison of alternatives can enable decision makers to select the most palatable alternatives, thereby minimizing the likelihood of negative consequences associated with inaction or maladaptive investment in the face of continuing climate change in the coming decades.

All data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request.

This study was funded by the Massachusetts Bay Transportation Authority (MBTA) and the United States Department of Defense (via the DoD SMART Scholarship-for-Service Program). The opinions expressed in this paper are those of the authors and do not represent those of the MBTA, US Army Corps of Engineers, or DoD.