by Judith Curry
Best practices in adapting to sea level rise use a framework suitable for decision making under deep uncertainty.
This post is the third (and final) part in the series on New Jersey sea level rise:
- Climate adaptation follies. Part I: The New Jersey challenge
- Climate adaptation follies. Part II: scenarios of future sea level rise
The posts are drawn from my report Assessment of projected sea level rise scenarios for the New Jersey coast.
As described in Parts I and II, sea level rise projections for the 21st century are characterized by deep uncertainty:
- Deep uncertainty (recognized ignorance) – fundamental uncertainty in the mechanisms being studied and a weak scientific basis for developing scenarios; future outcomes may lie outside of the realm of regular or quantifiable expectations; no agreement on how to define the possible outcomes.
Apart from uncertainties in emissions scenarios, there are substantial uncertainties in: climate sensitivity to increasing CO2, future volcanic eruptions, solar variability, multi-decadal ocean oscillations, and possible instabilities in ice sheets. Kopp et al. (2017) state:
“The breadth of published projections, as well as of remaining structural uncertainties, highlight the fact that future sea-level rise remains an arena of deep uncertainty.”
The following text is drawn from my report “Assessment of projected sea level rise scenarios for the New Jersey Coast”, which includes references.
Decision making under deep uncertainty
Making good decisions under conditions of deep uncertainty is far more complex than merely selecting the ‘best’ scenario for a specific application, which is the recommendation provided in the Rutgers Report.
Because of the clear expectation for continued sea-level rise, proactive coastal management approaches can be developed and deployed. A range of adaptation measures can be used, depending on the local vulnerabilities, land use and nature of the assets at risk: protection, accommodation, reclamation, retreat.
However, the large range of potential future sea levels poses the question: “When and how much to adapt?” Deep uncertainties in the rate and magnitude of sea level rise, particularly in the second half of the century, complicate decision making on coastal adaptation.
The deep uncertainty associated with future sea level rise poses substantial challenges for long-lived decisions with high stakes and high sunk (irreversible) costs: major infrastructure, building developments and land use planning. If long-term sea level change is not accounted for appropriately, it could mean greater risks, locking into greater costs, or wasted investments.
Uncertainty in climate projections and potential instability in the West Antarctic ice sheet is not expected to narrow in the near term. The challenge facing policy makers is how to make good decisions in the near term, while ensuring that long-term options for addressing uncertain future conditions are not pre-empted or made unnecessarily costly by earlier decisions.
Deep uncertainty due to climate change requires moving away from the ‘predict then act’ paradigm to one of ‘robust decision making,’ characterized by continuous learning and dynamic adaptation.
The ‘dynamic robustness’ approach incorporates flexibility into adaptation plans that can be changed over time as more is learnt or as conditions change. The ‘Adaptation Pathways’ approach (Ranger et al. 2013) identifies the timing and sequencing of possible ‘pathways’ of adaptation measures over time under different scenarios. These concepts have been integrated into an overall ‘Dynamic Adaptation Policy Pathways’ (DAPP) approach (Haasnoot et al. 2013). DAPP is a framework for identifying present and future uncertainties, evaluating vulnerabilities and alternative solutions, taking necessary actions in the short term, and monitoring changes and gathering insights that might indicate that new decisions or reassessments are required. Flexibility and iterative planning are core elements of the approach.
In the DAPP approach, a plan includes an initial action, emphasis on monitoring data, and a series of actions over time (pathways) depending on future scenarios that may emerge. DAPP is predicated on a strong understanding of the decision problem itself, rather than focusing on climate projections. The decision-centered approach of DAPP focuses on understanding the characteristics of the decision problem (the objectives and values of stakeholders, trade-offs, constraints and decision criteria), the vulnerability of the system and the adaptation options themselves.
Prominent applications of an adaptive approach to uncertain sea level rise include:
- The Thames Estuary study to protect the city of London (Ranger et al. 2013)
- New Zealand with a national guidance to coastal adaptation (Bell et al. 2018)
- In the Netherlands an adaptive approach has been put into practice for adaptation to SLR within the Delta Program (Van Alphen, 2016)
An example of the DAPP approach applied locally in the U.S. is provided by Obeysekera et al. (2020), for adaptation to sea level rise in the Little River Basin, Miami, Florida.
An important conclusion from these studies is that the DAPP approach implies different needs from climate science:
- a shift in emphasis away from probabilistic modeling;
- greater investment in observations and monitoring;
- improved understanding of historical climate variability; and
- improved understanding of relevant processes and their representation in models to enhance ‘best guess’ models and to better bound future projections using narrative scenarios.
Dynamic Adaptive Policy Pathways
In the decision-centered DAPP approach, scenarios of climate change are not the main driver for the process. Nevertheless, scenarios of future change play an obvious and important role in the decision making process.
The Rutgers Report provides a full probability distribution of future sea-level change that incorporates both the likely range and worst-case scenarios, conditional upon an emissions scenario. A problem with this approach is that the probability density functions (PDFs) are highly conditional on the methods that produced them and provide only a limited sampling of the uncertainties. Kopp et al. (2019) recognizes this by stating: “For processes subject to deep uncertainty, alternative justifiable approaches to constructing a probability distribution can yield quite divergent answers.”
While the motivation for probabilistic approaches driven by climate model simulations is to support decision making in the context of cost-benefit analysis, this approach can be counter productive for DAPP (apart from the issue of non-uniqueness of the probability distribution). Climate models are at best partial scenario-generators, which is at odds with the requirement of robust decision making to map the range of plausible outcomes. Further, DAPP approaches are scenario neutral, in that decisions do not require information about the probability or likelihood of different future scenarios.
Smith and Stern (2011) argue that there is value in scientific speculation on policy-relevant aspects of plausible, high-impact scenarios, even though we can neither model them realistically nor provide a precise estimate of their probability. A set of narrative scenarios can be formulated that use empirical models and expert judgment to complement outputs from climate models. This approach produces a much wider range of scenarios than would be generated by climate models. The potential problem of generating a plethora of potentially useless future scenarios is avoided if they are focused on scenarios that are expected to be significant in a specific decision making context.
The objective of DAPP is to develop an iterative, learning decision process that cost-effectively reduces risk today while avoiding foreclosing future options. Considering the full range of plausible scenarios, the DAPP approach provides clear information on the effectiveness and timing of options, enabling analysts to assess under what conditions and on what timescale a plan could fail. The approach explicitly recognizes that adaptation over time will be determined not only by what can be anticipated today, but also what is observed and learned in the future. The approach ensures that the short- to medium-term plan is set in a framework that will not be maladaptive if climate change progresses at a rate that is different from current expectations.
DAPP is designed to perform adequately under a wide range of possible future states. ‘Low-regret’ measures are implemented in the near-term. Low-regret measures are those that reduce risk immediately and cost-effectively under a wide range of climate/sea level rise scenarios. Low-regret measures can buy time to monitor and learn before making a major investment. DAPP plans are designed to be adjusted over time as more is learnt about the future. In this way, flexibility is built into the long-term strategy—the timing of new interventions and the interventions themselves can be changed over time.
DAPP planning provides a framework for incorporating flexibility, so that infrastructure can be adjusted or enhanced in the future at minimal additional cost. This includes include safety margins, where infrastructure is over-engineered to cope with greater than expected change; this approach is effective where the marginal cost is low.
A route-map lays out the options and provides information on when and how decisions should be made. The route-map is used to identify a set of a decision points, triggering specific options or pathways, conditional on observations of sea level rise and other indicators.
The DAPP approach is robust not only to climate change, but also to all other sources of risk and uncertainty, including socioeconomic uncertainties and uncertainties resulting from a lack of data. As long as the pathways account for such potential surprises and learning, allowances for adjustment can be incorporated into the plan.
Scenarios for DAPP
Scenarios used in DAPP include a ‘likely’ range and estimates of the plausible worst case.
Best practices in developing scenario outcomes for climate change adaptation start with the scenarios provided by the IPCC assessment reports. Experts or other practitioners generating scenario outcomes for a specific application may choose to select specific IPCC scenarios or generate scenarios beyond what the IPCC provides, but these choices should be justified relative to what the IPCC has provided.
Reasons for thinking that climate models are predicting too much warming include:
- The RCP8.5 emissions scenario is implausible.
- Observed warming for the past two decades is less than the average rate of warming predicted by climate models.
- The ensemble of climate model simulations does not sample the full range of likely values of equilibrium climate sensitivity, neglecting the lowest 20% of the likely range from the IPCC AR5.
- Climate models do not include solar variability and volcanic eruptions, with plausible scenarios for a cooling effect in the 21st century. Ignoring volcanic eruptions ignores their cooling effects (Bethge et al, 2017. Most projections of solar variability for the 21st century expect cooling relative to the 20th century (Matthes et al. 2017).
Given the implausibility of the RCP8.5 emissions scenario, use of RCP4.5 (moderate emissions) is justified by the IEA Report, at least out 2050. Specifically considering the amount of warming associated with the RCP4.5 scenario, my assessment is that temperature change is very unlikely to exceed the upper bound of the IPCC AR5 likely range, for the reasons cited in the above bullets.
Worst case scenarios, Dragon Kings and gray swans
The plausible worst-case scenario can play an important role in certain decision making frameworks. However, considerable care is needed in formulating the plausible worst-case outcome so as to be relevant and useful for decision makers.
Outcomes of future climate change are associated with deep uncertainty, and plausible outcomes (especially on the high end) are weakly constrained. Experts inevitably disagree on what constitutes a plausible worst-case scenario when the knowledge base is uncertain (Bamber et al. 2019 is a case in point). Curry (2019) has developed a classification of worst-case scenarios based on the extent to which borderline implausible parameters or inputs are employed in developing the scenario via physical or mental models. This classification is inspired by the Queen in “Alice in Wonderland:” “Why, sometimes I’ve believed as many as six impossible things before breakfast.” This classification articulates three categories of worst-case scenarios:
- Conceivable worst case: formulated by incorporating all worst-case parameters/inputs into a model; the outcome does not survive refutation efforts.
- Possible worst case (borderline impossible): Includes multiple worst-case parameters/inputs in model-derived scenarios; the outcome survives refutation efforts (at least temporarily).
- Plausible worst case: Includes at most one borderline implausible assumption in model-derived scenarios.
The plausible worst-case scenario is most relevant for decision making. Candidates for the plausible worst-case scenario can be evaluated by assessing the input assumptions and parameters that are used in developing the scenario. Inevitably, there will be disagreement as to what constitutes an implausible input, and hence there is a range of candidate worst-case scenarios to consider.
A ‘black swan event’ (Taleb, 2007) is a metaphor that describes an event that comes as a surprise, has a major effect, and is often inappropriately rationalized after the fact with the benefit of hindsight. In assessing the climate change impacts on sea level rise and coastal storms, attempts are made to foresee worst-case scenarios. There are two different types of plausible worst-case scenarios of relevance to the assessment of coastal threats from climate change:
- Gray swan: a high-impact event that may be foreseeable using historical data combined with physical knowledge. Gray swan scenarios are of relevance for worst-case impacts from a single landfalling hurricane (Lin and Emanuel, 2015)
- Dragon King: an event that is extremely large in size or impact, occurring in nonlinear and complex systems that is generated from positive feedbacks, tipping points, bifurcations, regime shifts. By understanding the underlying dynamics, there may be some potential predictability. Major instabilities in West Antarctic ice sheet fall into the Dragon King category (Sornette, 2009).
Gray swans are somewhat different from Dragon Kings in that our understanding is sufficient to formulate plausible gray swan scenarios of individual extreme events, whereas Dragon Kings imply a large-scale event arising from instability or a regime shift.
A number of different scenarios should be formulated for plausible gray swan and Dragon King events. Probability distributions can be formulated for gray swan and Dragon King events, based on a distribution of inputs. However, it is important to keep in mind that such probability distributions do not relate directly to outcomes, but rather to the plausibility of the individual scenarios as the worst case. For Dragon Kings, any estimated probabilities will evolve with increasing knowledge. When there is sufficient reason to believe that a Dragon King event could occur, it is best for decision making purposes if the distribution for the Dragon King regime are presented separately from the probabilities for the range of outcomes that are better understood (Ranger et al 2013).
While speculative scenarios can be useful in support of the decision making process, formulation of the plausible worst case scenario(s) for decision making applications requires justification for the assumptions that went into the model (physical or mental), including the plausibility of the assumptions. A major concern about the Bamber et al. (2019) expert elicitation that was used in the Rutgers Report is that the individual respondents were not required to provide justification for their predicted outcomes.
Due to ignorance, misaligned incentives, and cognitive biases, there is often a failure to adequately anticipate Dragon King and gray swan events. However, when explicit efforts are undertaken to anticipate such events, their importance and likelihood can be over-emphasized and there is a great deal of uncertainty and speculation that needs to be acknowledged.
Plausibility of major instability in the West Antarctic ice sheet
The primary concern over future sea level rise in the 21st century is related to potential dynamical instabilities in the West Antarctic Ice Sheet. The West Antarctic Ice Sheet rests on bedrock below sea level, making the ice sheet vulnerable to melting from the ocean. If these marine ice shelves – the floating extensions of glacial ice flowing into the ocean – lose mass, their buttressing capacity is reduced, accelerating seaward ice flow. This self-sustaining process is known as Marine Ice Sheet Instability (MISI).
The IPCC AR5 (2013) has medium confidence that this additional contribution from the West Antarctic ice sheet would not exceed several tenths of a meter of sea level rise during the 21st century [IPCC AR5 WG1 Chapter 13]. Subsequent to the IPCC AR5, there has been considerable focus on the worst-case scenario for global sea level rise, and our ‘background knowledge’ is rapidly changing. DeConto and Pollard (2016) articulated a mechanism whereby disappearance of ice shelves allows formation of ice cliffs, which may be inherently unstable if they are tall enough to generate stresses that exceed the strength of the ice. This ice cliff failure can lead to ice sheet retreat via a process called marine ice cliff instability (MICI), that is hypothesized to cause partial collapse of the West Antarctic Ice Sheet with increased warming.
The IPCC SROCC (2019) provides an updated summary on the potential contribution of dynamical instabilities in the West Antarctic Ice Sheet to global sea level rise. The IPCC SROCC assessed the amount of sea level rise increase from dynamical instability of the West Antarctic ice sheet to be 16 centimeters (range: 2–37 cm). The SROCC notes that the expert elicitation approach (Bamber et al., 2019; used in the Rutgers Report) suggests considerably higher values for sea level rise from the West Antarctic ice sheet than provided in Table 4.3 of the IPCC SROCC.
If RCP8.5 is assumed to be implausible and the focus is on the moderate emissions scenarios (RCP4.5), what constitutes the plausible worst-case scenario for sea level rise? Specifically with regards to the DeConto and Pollard (2016) mechanism which heavily influenced the Rutgers Report, the SROCC makes the following statement:
“The results by DeConto and Pollard (2016) indicate significantly higher mass loss even for RCP4.5, potentially related to their high surface melt rates on the ice shelves as contested by Trusel et al. (2015). This early onset of high surface melt rates in DeConto and Pollard (2016) leads to extensive hydrofracturing of ice shelves before the end of the 21st century and therefore to rapid ice mass loss. For this reason, their results and probabilistic (e.g., Kopp et al., 2017; Le Bars et al., 2017) and statistical emulation estimates that build on them (Edwards et al., 2019), are not used in SROCC sea level projections.”
A recent publication by Donat-Magnin et al. (2021) uses improved estimates of surface melt rates, and finds that for RCP4.5 only the Abbot glacier in the Amundsen sector is expected to become unstable to hydrofracturing (the DeConto-Pollard mechanism) during the 21st century. Edwards et al. (2019) further supports at most a small contribution in the 21st century from RCP4.5 for the DeConto-Pollard mechanism.
Specifically with regards to the Marine Ice Cliff Instability (MICI) of DeConto and Pollard (2016), the IPCC SROCC makes the following statement:
“Overall, there is low agreement on the exact MICI mechanism and limited evidence of its occurrence in the present or the past. Thus the potential of MICI to impact the future sea level remains very uncertain” [Cross-Chapter Box 8]
At this point, there isn’t an obviously plausible Dragon King scenario for sea level rise in the 21st century under RCP4.5.
Grey swan scenarios for hurricanes
Superstorm Sandy and the 1893 hurricane striking New Jersey are reminders that a storm nominally having Category 1 force winds can produce a greater storm surge and overall more damage than a more intense Category 3 hurricane. The strong surge from Sandy was associated with extratropical transition and its subsequent very large horizontal extent, a westward track that directly struck the coast, and landfall at high tide.
Sandy was not a worst-case scenario for New Jersey; a substantially higher storm surge (estimated at 13 feet) occurred for the 1821 Cape May hurricane. If the 1821 hurricane had occurred at high tide, slower forward motion and with a larger horizontal extent, the storm tide would have been substantially higher.
Lin and Emanuel (2015) define ‘gray swan’ hurricanes as high-impact storms that would not be predicted based on history but may be foreseeable using physical knowledge together with historical data.
There are several strategies for generating scenarios of gray swan hurricanes that should be considered in assessing risks impacting the New Jersey coast:
- Consider the occurrences of previous storms in the historical, archaeological and geologic records that impacted the mid-Atlantic states. If it has happened before, it can happen again.
- Synthetic scenarios can be created by combining plausible worst-case storm elements into individual scenarios.
- The intensity (maximum wind speed) can be increased by 5% and 10% to account for possible global warming impacts on hurricane intensity.
In developing grey swan scenarios of relevance to storm surge in a particular location, the following storm parameters can be varied within a physically plausible range:
- Intensity (maximum winds): up to Category 4 (a category 5 landfall as far north as New Jersey is judged to implausible).
- Horizontal size: up to the size of Hurricane Sandy, although such a large horizontal size is inconsistent with the strongest hurricane intensities.
- Speed of forward motion of the storm: slower forward motionproduces the largest surge.
- Angle of approach to the coastline: the straight east-west track is the worst case for NJ.
- Time of landfall relative to the astronomical tide: high tide is worst case.
Timescales of adaptation
DAPP frameworks have mostly been applied to more gradual shifts of climate change, rather than extreme and abrupt changes. Hasnoot et al. (2020) addresses the concern of adaptation for extreme scenarios of sea level change from instability of the West Antarctic ice sheet (a Dragon King scenario) that could involve rapid onset and high rates of change. Such an event would be associated with a short time to adapt, which can have large consequences for decision making.
Hasnoot et al. (2020) identify the decision making challenges arising from potentially accelerated sea level rise. Decisions may need to be taken when there is still large uncertainty about the sea level rise at the end of the envisioned lifetime and the lead time of follow-up interventions. The time required for planning and implementation can be decades for large coastal defense projects and other major infrastructure (e.g. bridges), which are designed for a lifetime exceeding a century. Most coastal defense and major infrastructure decisions have a long lifetime and cannot easily be solved with incremental or flexible measures, and these decisions will thus have to account for high amounts of sea level rise at once.
Worst-case sea level rise scenarios can be used to assess under what conditions alternative adaptation pathways are needed, which can help to prepare and enable timely adaptation. This can be accomplished through flexible measures and preparatory actions to keep options open (e.g. spatial reservations for future options), and in the design of structures to enable long-term adaptation (e.g. a large foundation of a structure to build higher later).
The time horizon of a pathways study should be chosen by considering the envisioned functional lifetime. For decisions with a long lifetime (>100 years), the focus should not be on projections of sea level rise for a specific time horizon. When looking at longer time horizons, it is more useful to consider the perspective that for some decisions it is not a matter of whether SLR will rise to certain levels, but when this will occur. This perspective may help to overcome decision paralysis due to uncertainty.
JC’s reflections on best practices for adaptation
I have been working in the climate adaptation space since 1999, mostly through my company Climate Forecast Applications Network (CFAN), but also on several university-based projects prior to 2006 (when CFAN was formed). During this period I have worked with development banks, corporations, government agencies (local/state/national) and NGOs. I have worked on range of different projects in different sectors and for different countries that address different types of vulnerabilities.
The single biggest problem that I see in climate adaptation is getting the climate community (broadly defined as scientists and adaptation decision makers) to move away from the ‘predict then act’ paradigm. Apart from potentially misleading the decision making process, the ‘predict then act’ paradigm places undue emphasis on the ‘correctness’ of the prediction. This gives rise to acrimonious disagreements and motivates the ‘consensus’ approach and the stifling of disagreement by the climate establishment, all in the name of promoting good decision making. By contrast, robust decision making approaches explicitly welcome (and actively seek) all plausible scenarios, rendering most disagreements about projected outcomes to be moot.
The second biggest problem is the perceived urgency of action, which exacerbates the problems associated with ‘predict then act.’ The incremental approach of robust decision making builds in flexibility to the adaptation planning.
The decision-centric mode that characterizes robust decision making focuses scenarios around specific vulnerabilities or concerns of the decision maker. Too often, climate adaptation interventions focus excessively on climate change and less on examinations of what drives local vulnerability. Reducing vulnerability is the central criterion of adaptation success. Top down interventions by development banks or U.N. agencies have many well-documented failures, owing to failures to consult adequately with local stakeholders and to truly understand the causes of local vulnerabilities (both environmental and societal).
And finally, it needs to be re-emphasized that climate models aren’t particularly useful at generating future outcomes of regional climate change and extreme weather events, which are targets of adaptation. Hence narrative scenarios developed from historical/paleo data and guided by simple process models and climate model simulations provide a much richer set of scenarios. Which experts are used in developing scenarios also matters quite a bit to the suite of scenarios that are provided. Best practice is to use 2 to 3 different teams with different perspectives/expertise in generating scenarios and evaluating scenarios of the other teams. I have been involved in three projects that used the multiple teams approach, and all parties learned much and the decision makers ended up with a much better understanding of the uncertainties and different factors in play.