Modeling Coastal Flood Risk and Adaptation Response under Future Climate Conditions

Abstract

The National Coastal Property Model (NCPM) simulates flood damages resulting from sea level rise and storm surge along the contiguous U.S. coastline. The model also projects local-level investments in a set of adaptation measures under the assumption that these measures will be adopted when benefits exceed the costs over a 30-year period. However, it has been observed that individuals and communities often underinvest in adaptive measures relative to standard cost-benefit assumptions due to financial, psychological, sociopolitical, and technological factors. This study applies an updated version of the NCPM to incorporate improved cost-benefit tests and to approximate observed sub-optimal flood risk reduction behavior. The updated NCPM is tested for two multi-county sites: Virginia Beach, VA and Tampa, FL. Sub-optimal adaptation approaches slow the implementation of adaptation measures throughout the 100-year simulation and they increase the amount of flood damages, especially early in the simulation. The net effect is an increase in total present value cost of $1.1 to $1.3 billion (2015 USD), representing about a 10% increase compared to optimal adaptation approaches. Future calibrations against historical data and incorporation of non-economic factors driving adaptation decisions could prove useful in better understanding the impacts of continued sub-optimal behavior.

1. Introduction

Coastal floods are among the most damaging natural disasters in the United States (Gall et al., 2011; Smith and Katz, 2013). The five costliest natural disasters in U.S. history were tropical cyclones that caused substantial coastal flood damage; all occurred within the last 20 years (NOAA, 2019). Coastal flood damages will likely increase over time due to climate-driven sea level rise and possibly due to stronger and more frequent coastal storms (Jongman et al., 2012; Hallegatte et al., 2013; Dinan, 2017). In addition to these climate-driven increases in damages, continued coastal development may put even more property and infrastructure at risk in the coming decades (Kousky and Michel-Kerjan, 2015; Dinan, 2017; Zagorsky, 2017).

While coastal flooding is likely to increase, some of the economic and social impacts may be avoided through future investments in adaptation measures (Houser et al., 2015). The United States has already invested many billions of dollars to reduce flood risk and this is likely to continue. Many studies estimate the extent and impact of past adaptation (Godschalk et al., 2009; Bakkensen and Mendelsohn, 2016) and others predict avoided losses due to potential future adaptation (Hinkel et al., 2014; Neumann et al., 2015; Dinan, 2017).

Neumann et al. (2010; 2015) developed the National Coastal Property Model (NCPM) to estimate coastal flood damages in the United States for different sea level rise and storm surge scenarios, at an aggregate level. The NCPM estimates flood damage, the cost of future investment in flood adaptation, and the avoided cost of flood damage associated with future adaptation. Studies conducted with the NCPM provided input for the Third and Fourth U.S. National Climate Assessments (U.S.GCRP, 2014; 2018).

This paper builds on the work of Neumann et al. (2015) by refining the NCPM’s methods for simulating adaptation decisions so that they better reflect traditional cost-benefit tests. By testing for adaptation before testing for abandonment, the new approach leads to substantially more adaptation and less abandonment. In addition, the refinement improves the cost-benefit test by comparing the cost of adaptation against the potential reduction in flood damages, rather than damages alone. These two changes improve the ability of the model to optimize net present benefits.

Despite the body of evidence showing substantial economic benefits from adaptation to coastal flood risk, recent studies suggest that there is significant underinvestment in cost-effective adaptation (Bakkensen and Mendelsohn, 2016; Mendelsohn et al., 2020). The reasons for this underinvestment are unclear, but may include financial, psychological, sociopolitical, and technological barriers to adopting adaptive measures (Chambwera et al. 2014; Bubeck et al., 2012; Poussin et al., 2014). Therefore, a key challenge is to model the future impact of coastal adaptation when a traditional benefit-cost approach will not accurately reflect on-the-ground decision making. Assuming all adaptation measures will be implemented if benefits of reduced damages exceed costs of installation may lead to unrealistically high projections of adaptation investment and, therefore, understate the net costs of coastal flooding. To address this, the new NCPM formulation is modified to estimate the aggregate effects of economically sub-optimal adaptation response.

To begin, this paper provides background on sea level rise and storm surge trends under climate change, and the potential increases in flooding that may result. Next, we summarize research on future damages of coastal flooding and future adaptation. Then we describe updates and changes to the NCPM, including the new approaches for simulating adaptation choices. The core of the paper discusses how predicted coastal flood damages and adaptation investment vary under different adaptation approaches, with a specific application to two multi-county study areas, Virginia Beach, VA and Tampa, FL. Finally, we discuss implications of the findings and offer ideas for future research.

1.1. Background

1.1.1. Sea Level Rise

Global mean sea level (GMSL) change is driven primarily by land-ice changes and thermal expansion, with a minor contribution from changes in global land-water storage (Church et al., 2013). Local, relative sea level changes are additionally affected by (1) gravitational, rotational, and deformational effects of redistribution of mass between land-ice and the ocean, as well as between land-water reservoirs and the ocean; (2) changes in the distribution of heat and salinity in the ocean (i.e., steric effects); (3) changes in winds and currents (i.e., dynamic sea levels); and (4) non-climatic processes such as glacial-isostatic adjustment, tectonics, and sediment compaction (see Kopp et al., 2015).

For the Fourth U.S. National Climate Assessment, Sweet et al. (2017) developed a set of six global sea level rise scenarios. These scenarios range from a low of 0.3 m (1 ft), corresponding to a continuation of current trends, to a high of 2.5 m (8.2 ft) of GMSL rise between the years 2000 and 2100. The 2.5 m scenario, while higher than the Third National Climate Assessment highest scenario of 2.0 m, was based on newer observational and modeling evidence about the contributions of the Antarctic ice sheet to sea level rise (e.g., Kopp et al., 2017) and arguments that 2.3–2.7 m (7.5–8.9 ft) of GMSL rise by 2100 was within the realm of physical plausibility (e.g., Miller et al., 2013).

Kopp et al. (2014) developed probabilistic, location-specific relative sea level projections. By conditioning the Kopp et al. (2014) projections on different GMSL values, Sweet et al. (2017) mapped their six GMSL scenarios onto projections of local sea level rise. This paper incorporates the local sea level rise scenarios from Sweet et al. (2017) into the NCPM.

1.1.2. Extreme High Water Levels

Coastal flooding is driven by stochastic high water events, such as storm surge and waves caused by tropical cyclones, other coastal storms, and astronomical high tides. A complete understanding of changes in the frequency and magnitude of high water requires consideration of the interaction of long-term mean sea level rise with shorter-duration high water events.

A small shift in local mean sea level can substantially impact the frequency of coastal high water levels at any given location. For example, Sweet and Park (2014) demonstrated that, due to observed sea level rise, some parts of the U.S. coast are already experiencing nuisance flooding nearly 10 times more frequently than they did in the middle of the 20th century. Under an intermediate sea level rise scenario [0.34 m (1.1 ft) of GMSL rise], Sweet et al. (2017) reported that by 2050 the majority of the contiguous U.S. coastline will see high water levels that historically occurred once every five years occurring at least five times per year. Sea level rise will also increase the frequency of more rare and damaging extreme high water levels. Based on the sea level rise projections of Kopp et al. (2014), Buchanan et al. (2017) found that the number of current “100-year floods” is expected to occur a median of 40 times more often by 2050 (ranging from 1 to 1,314 times more often) at different sites along the U.S. coast. A study of New York City’s historical 1-in-100 year coastal flood event found that it could occur as often as once every 20 years with just 0.76 m (2.5 ft) of local sea level rise (Horton et al., 2015).

Sea level rise will increase the severity of flooding by raising the baseline water level over which storms and other high water level events create a surge. For example, one study (Miller et al., 2013) estimated that if sea levels were a foot lower when Hurricane Sandy struck (i.e., at their levels of a century ago), approximately 80,000 fewer people in the New York metropolitan region would have been living below the storm tide level. Thus, elevated water levels due to sea level rise will put many more people and assets at risk of flooding along coastal flood plains.

Independent of sea level changes, the frequency and severity of coastal storms and their associated storm surges may change in the future. Among warm season coastal storms, the balance of evidence suggests that major hurricanes (Categories 3–5) are likely to become more frequent and intense. However, there are substantial uncertainties about future changes in the total number of tropical cyclones in the North Atlantic, as well as how these storms may change in intensity (Kossin et al., 2017). Tropical cyclones are driven in part by the warmth of the upper ocean and there is high confidence that the upper ocean will continue to warm with climate change (Church et al., 2013). However, many highly uncertain factors influence tropical cyclone frequency and intensity, including (1) easterly waves in Africa; (2) vertical wind shear; (3) vertical atmospheric temperature gradients; (4) atmospheric moisture; (5) atmospheric particulates such as aerosols, black carbon, and dust; and (6) localized anomalies of sea surface and upper ocean temperatures (Horton and Liu, 2014). Moreover, an increase in the intensity of tropical cyclones may not translate into an increase in the number of intense cyclones making landfall in a particular location, or even overall (e.g., Garner et al., 2017).

Cold season storms are a major driver of coastal flood risk in some regions such as the U.S. West and Northeast coasts (e.g., Catalano and Broccoli, 2018). Winter storms have increased in frequency and intensity since 1950, but models do not agree in projecting future changes of these storms (Kossin et al., 2017). Therefore, implications for the coastlines also remain uncertain, especially regarding individual strong storms. Winter storm patterns depend on the frequency and magnitude of anomalous jet stream configurations (Liu et al., 2012). Whether these will increase or decrease under climate change remains an unsettled topic in climate science (Francis and Vavrus, 2012; Wallace et al., 2014). Changes in storm tracks may also affect how changes in extratropical cyclones lead to surge events (Emanuel et al., 2008; Colle et al., 2013; Emanuel, 2013).

The interaction between these two changing drivers of high water levels – sea level rise and coastal storms – is an important area of study. For example, the two drivers can interact in non-linear ways, leading to larger changes in the frequency of high water levels events than a separate analysis of the two components treated in an additive way would suggest. Furthermore, there is the possibility that the two components are correlated with each other, which could further modify the frequencies and intensities of high water levels (Little et al., 2015).

1.1.3. Damages and Adaptation

Sea level rise projections have been used to assess national-scale impacts to U.S. coastal resources since at least the late 1980s (Park et al., 1989; Smith and Tirpak, 1989; Yohe, 1990; Titus et al., 1991), and more recent analyses have included the joint effects of sea level rise and storm surge (Frumhoff et al., 2007; Kirshen et al., 2012; Lin et al., 2012; Houser et al., 2015; Neumann et al., 2015; Hsiang et al., 2017). These studies evaluated monetary damages to assets that would be exposed to inundation (e.g., property and infrastructure) and, in some cases, losses due to business interruption (e.g., Houser et al., 2015).

Many recent estimates of the impacts of sea level rise and storm surge on coastal resources include the costs and benefits (i.e., avoided damages) of adaptation actions. Several studies demonstrate that adaptation actions – including protective structures, beach nourishment, elevating and flood-proofing structures, and abandoning properties – are often cost-effective (Godschalk et al., 2009; Eisenack et al., 2014; Multihazard Mitigation Council, 2017).

However, it is not clear how extensively property owners and communities in the United States will adapt to increasing flood risk, posing challenges for computational projections of future adaptation. Individuals and communities make adaptation decisions based on a wide range of psychological, economic, and sociopolitical factors, and how those factors influence decisions is poorly understood (Bubeck et al., 2012; Poussin et al., 2014). Furthermore, predicting these factors decades into the future is extremely difficult. For example, individuals’ and communities’ adaptation decisions are highly dependent on government policies and other incentive structures (Kousky and Shabman, 2015; Moss et al. 2019), which are constantly changing. Generally, observations have shown under-investment in flood protection. Bakkensen and Mendelsohn (2016) found little evidence of adaptation to coastal flood risk in the United States. Similarly, a survey-based study showed very little investment to mitigate hurricane risk and little intention to invest among coastal homeowners in North Carolina (Javeline and Kijewski-Correa, 2019). Kousky and Shabman (2017) discuss how the bulk of investment in flood protection generally occurs in the aftermath of a disaster rather than as part of carefully planned programs.

Despite these challenges, studies that do not account for the avoided damages of adaptation almost certainly over-estimate the aggregate impacts of climate change (Diaz, 2016; Ward et al., 2017), and likely miss meaningful spatial patterns as some locations will be better able to adapt than others (Moser et al., 2014). Accordingly, several approaches have been tested to estimate the combined effect of future flood risk and adaptation. Most of these studies apply a top-down approach in which a model predicts rates of adaptation investment across large geographic areas. For example, Hinkel et al. (2014) specified a demand function for adaptation that predicts adaptation in terms of the level of protection (i.e., the return interval or magnitude of a flood for which a levee is designed) as a function of avoided damages, gross domestic product, per capita income, sea level rise, and population density. They also specified a cost function for levees (the only adaptation option included) and equated marginal costs and marginal benefits of adaptation to derive optimal adaptation conditions, with estimates of elasticities taken from available empirical results. Combined with flood damage modeling, this approach was applied to coastal segments worldwide that are roughly 10–20 km (6–12 miles) in length. Diaz (2016) built upon a similar coastal-segment database but along each segment identified a benefit-cost-optimal mixture of levee construction and retreat (i.e., abandoning property).

Instead of explicitly modeling future adaptation response, Dinan (2017) addressed it using empirical elasticities of flood damage with respect to income and population density. This approach captures how damages would systematically vary as a result of multiple underlying drivers, including development in flood-prone areas and adaptation. While Hinkel et al. (2014) specifically estimated investment in levees, Dinan (2017) estimated changes in realized damages at the county-level, but did not predict rates or types of adaptation investment. The Dinan (2017) study incorporated sea level projections from Kopp et al. (2014), projections of storm surge from Emmanuel (2013), and elasticities derived from Bakkensen and Mendelsohn (2016). Diaz (2016), Dinan (2017), and Hinkel et al. (2014) all showed the potential for substantial increases in coastal flood damage from sea level rise and storm surge, but also demonstrated the potential benefits of adaptation.

Ward et al. (2017) developed and tested an approach for estimating how levees built to different protection standards would affect expected annual damages (EAD) associated with riverine floods at a subnational level. That study tested three different policy scenarios, including one to maximize the net present value of adaptation and two that maintain risk at constant levels. Their results suggest that levees are likely cost-effective risk management options regardless of their design standard and future climate scenarios.

In contrast, bottom-up approaches tailor a methodology to a specific location (e.g., Aerts et al., 2014) or propose general frameworks that can be applied anywhere (e.g., Kirshen et al., 2012). The NCPM employs a bottom-up approach designed to simulate how property owners might respond to local flood risk. Instead of predicting adaptation over large regions using statistical models, the NCPM simulates local flood damages and selects local adaptation measures based on potential costs and benefits. Some details about the NCPM are provided in Section 2.4; a full description of the model can be found in Neumann et al. (2010; 2015).

1.2. Study Approach

The NCPM is a well-established model that was designed for national-scale analysis. However, the examination of detailed results in conjunction with other recent research revealed several opportunities for improving the model. This paper first makes adjustment to the NCPM adaptation simulation to better reflect traditional risk-based, cost-benefit analysis. Then, in order to account for the observed underinvestment in flood protection (as summarized above), we formulated alternative approaches to estimate the impact of decisions that do not focus solely on maximizing cost-effective risk reduction. These alternatives approaches require higher benefit-cost ratios in order to trigger adaptation and enable us to evaluate how sub-optimal decisions might impact aggregate economic cost to society from sea level rise and storm surge flooding. With these model changes, we are able to calculate the increase in benefits due to improvements in the NCPM, and estimate the decrease in net present value resulting from the sub-optimal approach (i.e., where increases in flood damages outweigh reductions in adaptation investments).

2. Methods

2.1. Study Locations

The NCPM is designed to estimate the aggregate impacts of sea level rise and storm surge along the contiguous U.S. coastline. We chose two NCPM sites as the focus of this study: the Tampa and Virginia Beach regions. The Tampa site, which includes Hillsborough and Pinellas counties, represents the second-highest estimated sea level rise and storm surge damage among all sites modeled in Neumann et al. (2015). The greater Tampa area also has an interesting combination of open ocean and back bay shoreline; it includes urban, rural, and beachfront settings; and it includes complex canal-style development that can be particularly challenging for the type of adaptation modeling described here. Finally, Tampa has enough low-lying, valuable property (Figure 1) that decisions to adapt are made relatively early in the simulation period. These factors make Tampa a good modeling case.

Figure 1:

Property value share within elevation ranges relative to mean higher high water for the two sites included in this study.

We chose the Virginia Beach site because it ranks high on most sea level rise and storm surge vulnerability assessments (e.g., Kulp and Strauss, 2017), and includes a mix of urban, suburban, and beachfront development. Further, the Virginia Beach area has experienced relatively high relative sea level rise for Atlantic Ocean sites due to high rates of subsidence (Eggleston and Pope, 2013). The Virginia Beach site also has low-lying valuable property, although less than the Tampa site, so decisions to armor and elevate are also taken early, and throughout, the century-long simulation. The Virginia Beach site includes Hampton, Portsmouth, Norfolk, and Virginia Beach counties.

2.2. Sea Level Rise Scenarios

We used the local sea level rise scenarios developed by Sweet et al. (2017) for the Fourth U.S. National Climate Assessment. Specifically, we used local sea level rise scenarios based on GMSLs of 0.5, 1.0, 1.5, and 2.0 m (1.6, 3.3, 4.9, and 6.6 ft) by 2100, representing low, intermediate, intermediate-high, and high scenarios. For each GMSL scenario, Sweet et al. (2017) identified a large ensemble of consistent local sea level rise realizations. From those realizations, they developed low, median, and high scenarios of local sea level rise for each global sea level rise scenario. These projections were produced on a 1° grid along U.S. coastlines, at decadal intervals from 2000 to 2100. For the purposes of testing alternative adaptation assumptions in the NCPM, we used median local conditions for each global mean listed above. The NCPM is designed to use a single sea level rise value per county, so gridded local sea level rise is spatially averaged for each county.

2.3. Storm Surge Probabilities

Previous studies using the NCPM relied on projected storm surge probabilities under climate change conditions derived from Emanuel (2013). The advantage of this approach is that it considers forecasts of future climatic conditions, especially sea surface temperatures, to estimate storminess and surge heights at any particular location. A disadvantage is that the results are not available currently for all sites in the United States. As a result, the NCPM post-processing includes an extrapolation of storm surge impact results across sites based on key impact drivers such as property value intensity in vulnerable zones. Further, Emanuel’s (2013) work is only one of several forecasts of future storminess associated with climatic forecasts, each of which reveals slightly different results (see Dinan, 2017) for a comparison of the impacts from alternative sources of future storminess).

An alternative approach, applied here, is to rely on current storm surge probabilities based on recent historical tide gauge data. These were developed using records from six tide gauges at the Virginia Beach site, and two tide gauges at the Tampa site. The records from these gauging sites ranged from 9 to 40 years in duration, with the majority of the records extending to the present day. Comparing storm surge exceedance curves from the alternative historical tide gauge version to Emanuel’s (2013) results reveals differing outcomes for our two locations: the Emanuel approach yields higher storm surge heights across the frequency domain for the Tampa site, but lower storm surge heights for the Virginia Beach site.

Using these tide gauge data, we extracted the maximum daily water level from each record, and detrended the resulting set of maximum gauge heights from each time series. From the detrended data, we then calculated a distribution of storm surge heights by fitting a generalized extreme value distribution to the annual maximum time series from each gauge. This provided an estimate of the surge heights associated with return intervals ranging from 2 to 500 years.

2.4. National Coastal Property Model

2.4.1. Description

The NCPM divides the potentially vulnerable U.S. coastal area [counties that include any land that is both 20 m (66 ft) or less in elevation (relative to North American Vertical Datum of 1988) and hydrologically connected to coastal waters] into a uniform 150-m grid, with each grid cell effectively representing a small group of property owners who are assumed to act as coordinated agents in adaptation decisions. The NCPM estimates flood hazard (storm surge probability plus sea level rise), property value exposure (land and structure), and EAD for each grid cell. The property value is estimated from a database of market-adjusted assessed values, and is scaled into the future based on projected changes in economic growth. The model simulates damages and local adaptation decisions over the 21st century, with annual decisions for adaptation to permanent inundation due to sea level rise, and decadal decisions for adaptation to storm surge combined with sea level rise. The NCPM does not consider indirect or induced damages and expenditures (e.g., costs of temporary relocation, lost wages, tax revenue effects, jobs gained from adaptation investments or rebuilding) beyond direct property damage.

The NCPM includes four adaptation options: (1) beach nourishment, which is applicable to open-ocean-facing cells only; (2) armoring (i.e., floodwalls); (3) structure elevation; and (4) abandonment. The model evaluates these adaptation options in each grid cell independently to determine whether an adaptation response is warranted in that cell and which option is most effective (see the description under Simulating Adaptation Decisions). Beach nourishment is triggered only for cells that include both a beach land cover designation and a non-zero property value, and requires replenishment once every 10 years thereafter, with amounts of sand consistent with maintaining a constant beach profile in response to rising seas (i.e., increasing amounts of sand to replenish the beach are needed as sea level rise progresses). Beach nourishment is assumed to be effective for up to 0.3 m (1 ft) of sea level rise, after which hard structure armoring is required along with continued beach replenishment with sand. Armoring and structure elevation are assumed to be adequate to protect against the 100-year flood magnitude for the duration of the simulation. Therefore, when the model selects armoring or structure elevation, it eliminates damages for all floods up to and including the 100-year flood for all subsequent time steps. The implicit assumption is that armoring will be augmented in response to rising seas and that structure elevation is sufficient for future sea level rise-driven increases in the 100-year flood. The model does not consider whether armoring to protect against smaller or larger floods would be more cost-effective (e.g., Mendelsohn et al. 2020), nor does it have the capability to model protective measures such as sea-gates that could address larger geographic regions than a single cell. The model incorporates a maintenance cost for augmenting armoring over time in response to sea level rise. The cost of armoring is different for open-ocean-facing locations than for bay-front locations because of the higher wave energy and intensity anticipated for the former. Abandoning a cell eliminates all damages for the remainder of the simulation. The costs of each option vary by grid cell and are derived from a variety of data sources (see Neumann et al., 2015); the cost of abandoning a cell is assumed to equal the property value of the cell.

The NCPM tracks adaptation investments and residual flood damage at each grid cell over a 100-year simulation period, covering 2001 to 2100. Residual damages occur in cells where no adaptation has been selected by the model, as well as in cells with adaptation due to extreme water levels exceeding the assumed 100-year-flood capacity of adaptation measures. Results are summed across each study site so that the NCPM reports total investment in (a) nourishment, armoring, and elevation, (b) total cost of property abandonment (land and structure), and (c) total cost of residual flooding by study site and by decade over the simulation period.

2.4.2. Simulating Adaptation Decisions

The original formulation of the NCPM had two key limitations that we seek to improve on:

1. Adaptation by abandoning property was considered first in the model sequence, regardless of whether other adaption measures would be cost-effective, and

2. Nourishment, armoring, or elevation were evaluated by comparing their cost to damages rather than expected avoided damages.

This original formulation was not always able to identify the most cost-effective adaptation approach for each location. As a result, the NCPM was modified in two ways. First, the order was modified so that armoring and structure elevation are considered before, and separate from, the decision to abandon. If it is cost-effective to armor a cell or elevate structures in the cell, the cell will not be abandoned. Second, a new approach compares the cost of different adaptation options within each cell to the expected reduction in damages that would result from those adaption options. The new cost-benefit test compares an estimate of discounted avoided damages over the next 30 years with the cost of each adaptation option (reference Figure 2 for the simulated decision sequence).

Figure 2:

Simulated adaptation response in the NCPM. First, the model determines whether each cell will be affected by storm surge. If so, a cost-benefit test determines if armoring or elevation should be implemented. If neither armoring nor elevation pass the cost-benefit test, the model determines whether expected damages exceed the property value. If so, the cell is abandoned.

The cost-benefit test relies on an estimate of EAD and expected annual benefits (EAB) of adaptation. To estimate EAD from episodic flooding, the NCPM calculates expected damages for eight points on the surge exceedance curve (Figure 3). The area under the curve provides an estimate of EAD for the current decade in the simulation.

Figure 3:

Illustration of how the NCPM estimates expected annual damage. The area under the curve represents EAD; area A is the EAD for floods that have at least 1% annual chance exceedance and area B is the EAD for floods that have less than 1% annual chance exceedance. The NCPM estimates damages for flood events shown by the larger points along the curve.

Adaptation measures are triggered by EAB of adaptation. We assume that armoring and elevation are implemented to protect against the 100-year flood and EAB of adaptation is equivalent to the EAD for all events up to and including the 100-year flood.

The simulated decision to implement adaptation measures relies on the following variables:

• P = Property value in a cell; varies across cells and is inflated over the model run.

• C = Cost of adaptation; varies across cells and adaptation type. For abandonment, C = P. For armoring and elevation, the cost includes capital and the present value of maintenance.

• EAD = Expected annual damages = Area A + Area B from Figure 3.

• EAD_P = Expected annual damages with adaptation in place = Area B (i.e., damages for events larger than the 100-year event).

• EAB = Expected annual benefits of adaptation = EAD – EAD_P = Area A.

The model selects the lowest-cost adaptation option if EAB for that option exceed C. This represents a traditional cost-benefit test for optimal risk-reduction investment at an individual property level (represented here by the 150-m grid cells).

Ideally, such a decision would consider all future costs and benefits discounted to present value (at the time of the decision). While surge exceedance does not vary over time in this study, damages will grow due to local sea level rise. In any given model timestep, incorporating the future effects of sea level rise into the cost-benefit estimate would complicate the NCPM, as it would require estimating EAD and EAB using at least three different surge exceedance curves (one for each decade with growing increments of sea level rise); in reality, future sea level rise is always uncertain and incorporating that uncertainty into the adaptation response during each model timestep would further complicate the model.

Therefore, the model makes a simplifying assumption that EAD in the current timestep (decade) is a reasonable approximation of EAD over the ensuing three decades. This means that simulated adaptation decisions account for realized sea level rise up through the timestep (i.e., decade) of each decision, but they do not incorporate an estimate of how sea level rise in subsequent timesteps will affect flood damages. In each timestep, the NCPM looks ahead 30 years and uses current EAD as an approximation of future damages. The NCPM discounts future EAD at 3% to obtain a present value of expected damages over 30 years. Expected damages are estimated with and without adaptation for the 100-year event to estimate a present value of the expected benefits of adaptation.

As noted above, available evidence (Bakkensen and Mendelsohn, 2016; Javeline and Kijewski-Correa, 2019) suggests that there has been less adaptation than would result from an economic cost-benefit analysis. Rather, several other factors influence flood risk management choices, including financial constraints, risk attitudes, experience with floods, perceived efficacy of available adaptation measures, and trust in the source of risk information (Bubeck et al., 2012; Poussin et al, 2014; Wachinger et al., 2013). Even though a quantifiable understanding of these factors is not currently available, it is still valuable to consider how realistic adaptation decisions might affect the economic impacts of coastal flooding under climate change.

We investigated the potential impacts of sub-optimal adaptation response (i.e., decisions that do not maximize net benefits) by changing the required benefit-cost ratio to implement adaptation as one way to approximate the observed under-investment in flood protection. The approach described above requires that the benefit-cost ratio exceed 1.0 (henceforth, S1); in addition, we tested benefit-cost ratios of 2:1 (S2) and 4:1 (S4). An alternate approach to approximating sub-optimality would be to consider the possibility that individuals and communities apply higher discount rates than the social rate of time preference of 3% used by the model, In the case of the analysis in this paper, S2 corresponds to a discount rate of 10% and S4 to a discount rate of 21%.

To be clear, there is no evidence that people perform a modified benefit-cost test to guide adaptation decisions. Observed sub-optimal adaptation decisions may be driven by risk attitudes, financial constraints, or other factors. In addition, this approach glosses over potentially important spatial or temporal patterns of decisions because social and psychological drivers of decisions vary geographically. However, our approach does provide a simplified way to approximate the aggregate effects of decisions that do not adhere to formal benefit-cost criteria. Additional options for simulating realistic decisions are addressed in the Discussion and Conclusion.

3. Results

3.1. Adaptation Investment and Residual Flood Damages

Several patterns emerge from the results (Table 1 and Figures 4, 5, 6, and 7). First, the model rarely selects structure elevation and so this adaptation option makes up only a small portion of total cost across all six counties. This is due to the fact that in most locations the unit cost of elevation is higher than the cost of armoring. Second, in all six counties across the two selected sites, abandonment is rare and therefore also only makes up a small portion of overall cost. This result is expected with the new adaptation decision sequence described above (i.e., cells are abandoned only if there is no cost-beneficial option to reduce damages and only if damages for the decade exceed property value). Third, in the two oceanfront counties (Pinellas and Virginia Beach), investment in nourishment, investments in armoring, and residual flood damages each comprise substantial portions of total cost. In the four counties without oceanfront shoreline (where nourishment is not applicable), investment in armoring and residual flood damage make up the bulk of total costs. Depending on the magnitude of global sea level rise, and assuming an optimal (S1) investment strategy, the NCPM projects total costs between $7.6 billion and $10.5 billion (discounted at 3% to 2015) for the Tampa site, and $2.7 billion to $4.6 billion for the Virginia Beach site (Table 1). Undiscounted costs roughly double between the 0.5 m (1.6 ft) and the 2.0 m (6.6-ft) scenarios. The increase of storm surge damage with sea level rise varies according to site-specific characteristics – mainly the distribution of property values with elevation (Figure 1) and the timing of armoring, which protects inland properties from storms up to the 100-year return period. This is in part due to the static storm surge exceedance curves used here. In reality, as Emanuel (2013) concludes, these are likely to evolve over time as climate changes.

Table 1:

Total costs (adaptation plus residual flood damage) from 2001 to 2100 for the two study locations. The total net costs and damages are presented both undiscounted and discounted at 3%.

		Undiscounted, millions of nominal dollars		Discounted at 3%, millions of 2015 dollars
Sea level rise	Benefit-cost ratio	Total - Tampa site	Total - Virginia Beach site	Total - Tampa site	Total - Virginia Beach site
0.5 m	S=1	$17,941	$7,056	$7,574	$2,713
	S=2	$18,727	$7,424	$7,894	$2,756
	S=4	$20,603	$8,260	$8,440	$2,909
1.0 m	S=1	$21,604	$9,079	$8,564	$3,356
	S=2	$22,245	$9,451	$8,888	$3,413
	S=4	$23,879	$10,563	$9,463	$3,675
1.5 m	S=1	$25,190	$11,047	$9,566	$3,934
	S=2	$25,602	$11,342	$9,837	$3,996
	S=4	$26,953	$12,272	$10,453	$4,284
2.0 m	S=1	$29,012	$13,669	$10,498	$4,558
	S=2	$29,397	$13,883	$10,777	$4,614
	S=4	$30,632	$14,659	$11,432	$4,917

Figure 4:

Sample results comparing original and updated NCPM adaptation approach; total adaptation investment, abandonment, and residual storm surge damage through 2100 discounted at 3% to 2015 (Note: SLR = sea level rise)

Figure 5:

A sample of NCPM results for one area of Pinellas County. The zoomed-in sections of the map detail NCPM model results for the same area for 2030, 2050, and 2080. The red cells have been abandoned, the blue cells have implemented structure elevation, the yellow cells have implemented beach nourishment, and the black cells have implemented armoring.

Figure 6:

Undiscounted adaptation and residual storm surge damage by S and sea level rise scenario, millions of dollars (note: SLR = sea level rise)

Figure 7:

Discounted adaptation and residual storm surge damage by S and sea level rise scenario, millions of 2015 dollars discounted at 3% (SLR: sea level rise)

3.2. Updated Adaptation Decision Rules

The new cost-benefit adaptation approach reduces net costs (the sum of adaptation investments and flood damages) compared to the NCPM’s original formulation (Figure 4). Overall, compared to the original model formulation, the new approach substantially increases investment in armoring and substantially reduces property abandonment.

3.3. Non-Optimal Adaptation Approach

Compared to S1, S2 and S4 result in lower total adaptation investment (armoring and elevation) and higher residual flood damages across all six counties, with the total net cost increasing from S1 to S4. In the model, nourishment is a response to inundation, not to storm surge, and does not change with higher S values. From S1 to S2, the discounted total cost increases by $271 million to $324 million (depending on the sea level rise scenario) for the Tampa site, and by $43 million to $62 million for the Virginia Beach site. From S1 to S4, discounted total costs increase by $866 million to $899 million for the Tampa Site (depending on sea level rise scenario) and by $196 million to $350 million for the Virginia Beach site. The total increase of $1.1 to $1.3 billion in net present total costs across both sites represents an increase in costs of about 10% over the optimal adaptation scenario.

In addition, though it represents a very small proportion of total cost across all scenarios and locations, abandoning does increase slightly with S2 and S4 and with higher sea level rise scenarios. This increase in abandoning is driven by the fact that for some cells, the higher benefit-cost ratio is never achieved and adaptation is not implemented. However, as storm surge damage increases with each decade, damages eventually exceed the property value, triggering the model to abandon those cells. More abandonment is seen in Virginia Beach than in Tampa, and mainly for higher rates of sea level rise.

While the sub-optimal adaptation approach does increase total costs, the effect is smaller than the increase caused by 0.5 m (1.6-ft) increments of sea level rise. For example, for the Tampa site, the increase in discounted total costs from S1 to S2 is about 30% as large as the increase with an additional 0.5 m sea level rise (e.g., roughly $300 million versus $900 million in net present dollars), whereas for Virginia Beach the increase from S1 to S2 is less than 10% as large as it is with an additional 0.5 m sea level rise ($50 million). For S4, the non-optimality penalty in Tampa is almost as large as the damages from an additional 0.5 m sea level rise (about $900 million), whereas for Virginia Beach it is slightly more than half as large (about $300 million). Note that even when the model requires expected damages to be four times as large as the benefits, the model still predicts a significant amount of adaptation occurring, but generally later in the simulation.

Figure 8 shows the cumulative investment in armoring and the cumulative damage from storm surge over time for one county (Hampton, part of the Virginia Beach site), under one sea level rise scenario (1.5 m, 4.9 ft) for S1, S2, and S4. Figure 8 depicts S1 results in earlier investment in armoring, while S2 and S4 delay investment and result in less total investment over the entire simulation. The earlier investment in armoring leads to lower storm surge damage over the whole period, with higher total costs (armoring plus storm surge damage) for S1 relative to the other cases for the first two decades but reduced costs after the 2020s.

Figure 8:

Examples of decadal armoring investment, storm surge damage, and their sum by decade. Hampton County, sea level rise = 1.5 m. Results reported in millions of 2015 dollars.

The effects of S1, S2, and S4 differ across the two study sites and across the six counties included in those sites. Virginia Beach responds to higher S levels by cutting armoring much more sharply than Tampa. With sea level rise of 1.0 m, investment in armoring declines by 38% from S1 to S2 in Virginia Beach, but only 17% for Tampa; and for S4, the decline is 55% in Virginia Beach compared to 35% in Tampa. At the same time, residual storm surge damage increases much more in absolute terms in Tampa as armoring is reduced. For example, with 1.0 m of sea level rise, S2 causes a decrease in armoring of $245 million, but an increase in storm surge damage of $567 million, a ratio of 2.3:1; in Virginia Beach the decrease in armoring is $410 million, within an increase in damages of $458 million, a ratio of 1.1. The increase in total present value cost (adaption plus damages) for Virginia Beach is 1 to 2% for S=2, and 7 to 10% for S=4. For Tampa the total cost increase is 3 to 4% for S=2, and 9 to 11% for S=4.

At the county level, Hillsborough in Tampa sees the largest relative impacts. The ratio of increase in storm surge damage to decrease in spending on armoring ranges from 2.6 to 3.9. This leads to an increase in net relative impacts of 11 to 14% for S=2, and 34 to 36% for S=4. No other county has an increase of more than 4% for S=2, or 19% for S=4.

4. Discussion and Conclusion

The updated cost-benefit-based approach in the NCPM reduces net present costs of storm surge damage, adaptation investments, and abandonment in two ways. First, by implementing adaptive measures in all cases where benefits exceed costs and, second, by applying an abandonment test only after the potential for adaptation has been considered. Using the updated NCPM, damages were estimated for two multi-county sites, Tampa and Virginia Beach, with global sea level rise of 0.5 m (1.6 ft) to 2.0 m (6.6 ft) by 2100. The results show how armoring and nourishment adaptation decisions and residual storm surge damages vary by location. The model finds that elevation is rarely cost-effective for any cells in these locations, and that property abandonment is rare. Total costs are highest for the ocean-facing counties in each location (Pinellas and Virginia Beach) and increase roughly linearly with increasing sea level rise in 2100 in both locations (i.e., about $1 billion/0.5 m for the Tampa site and $600 million/0.5 m for the Virginia Beach site).

As described earlier, there is substantial evidence that people make risk-reduction decisions based on a wide range of cost, cognitive, social and emotional factors. S1 represents a traditional cost-benefit test, whereas S2 and S4 were designed to approximate the aggregate impact of decisions that, from a cost-benefit standpoint, could be viewed as sub-optimal. This was done in order to account for the fact that people do not make decisions solely to maximize expected net present value of benefits. Using the non-optimal adaptation approach, the model finds a decrease in armoring with a resulting increase in storm surge damages (along with some increase in abandonment).

Results suggest that a lack of proactive, optimal investment in adaptation may increase total discounted costs through the year 2100 by as much as $1.24 billion across just six coastal counties in the Tampa and Virginia Beach regions, or about a 10% increase above the optimal cost. For Tampa, when requiring expected benefits to be double the costs of adaptation (S2), the non-optimal penalty is equivalent to about 0.15 m (0.5 ft) of additional sea level rise by 2100. If adaptive measures are not implemented until expected benefits are four times as large as the damages, the additional damage would be equivalent to about 0.47 m (1.5 ft) of additional sea level rise by 2100. For Virginia Beach, the equivalent penalties are 0.05 m (0.2 ft) and 0.23 m (0.75 ft) of additional sea level rise, respectively. Considering the high-risk regions not included in this study (e.g., Miami, FL; New York City, NY; Houston, TX; New Orleans, LA), the national costs associated with sub-optimal flood risk adaptation investment could run into many tens of billions of dollars by the end of the century (not accounting for indirect or induced damages).

The sub-optimal benefit-cost ratios tested here are compatible with available evidence that property owners in the United States tend to under-invest in adaptation (Bakkensen and Mendelsohn, 2016; Javeline and Kijewski-Correa, 2019). Previous work (Multihazard Mitigation Council, 2017) showed that the benefit-cost ratio of actual flood risk adaptation investments in the U.S. is between 4:1 and 7:1. Mendelsohn et al. (2020) tested that general finding for New Haven, CT and found that new seawalls would have a cost-benefit ratio over two, with potential net benefits over 30 years totaling more than $60 million. It’s important to note that property owners and communities may not be using a modified cost-benefit test to guide their investment decision; in fact, standard cost-benefit criteria remain a reference point in deciding NOT to invest in adaptation (Kirshen et al. 2018) and passing a cost-benefit test is often required for investment in adaptation (see Moss et al. 2019).

The inclusion of non-optimal adaptation rules is meant to approximate the aggregate effects of sub-optimal adaptation behavior and represents an important first step in creating a model that can more accurately represent human decision-making. However, there is a lack of research to date that quantitatively describes the non-optimality of present-day adaptation. Javeline and Kijewski-Correa (2019) suggested that future analyses of a survey dataset may help to identify factors that predict this lack of investment in adaptation, which could explain behavioral and societal constraints. This type of research could be used to refine the sub-optimal decision rules used in the NCPM. For example, sub-optimal decision rules could vary spatially with property value or socioeconomic characteristics such as median household income or local economic output. The simple approach used here will make it easier to extrapolate results beyond the initial two regions studied, which is an important step for making damage estimates at the national level. However, adding additional variables to the model could improve our understanding of how non-economic factors influence adaptation response and how those responses might vary across locations and regions. Adding such factors to the model would enhance the model but it would also increase the computational complexity of the model.

Furthermore, the model makes two assumptions about coastal property in the future that may impact model results. First, the density and spatial distribution of property in flood-prone areas is assumed to remain fixed throughout the 21st century. In recent decades, development in coastal areas has outpaced other areas, but it is highly uncertain whether this will continue or how it might change in the decades to come. Such development may even be a function of adaptation decisions, with protected regions seeing greater real estate investment. The NCPM also assumes that property values will grow with future per-capita gross domestic product [with an elasticity of 0.45; details are provided in Neumann et al. (2010) and the projection of gross domestic product -per-capita is outlined in Neumann et al. (2015)]. While coastal property values continue to increase in many areas, some evidence now suggests that increasing coastal risk is putting downward pressure on property values (Bernstein et al., 2018; Keenan et al., 2018; McAlpine and Porter, 2018) and there may be greater pressure on property values in the future (Hino and Burke, 2020).

As discussed throughout this paper, existing research (e.g., Bubeck et al., 2012) suggests that many factors other than objective risk affect people’s choice. Furthermore, there is extensive evidence that people often have a difficult time understanding risk and acting on that information (Slovic, 1987; Kousky and Shabman, 2015). It may be reasonable to assume that this will continue and that property owners will have limited risk information and/or limited understanding of the available risk information. This study provides one approach for approximating this kind of non-optimal adaptation, examining the implications in two regions. Future work with the NCPM will extrapolate this approach nationally. However, better understanding of existing adaptation choices will continue to be important in order to calibrate and test this new methodology.

Highlights.

• Long-term flood risk projections are improved by incorporating reasonable estimates of the mitigating effects of adaptation.

• Drivers of risk-reduction decisions are complex and poorly understood.

• While cost-benefit is often used to project coastal flood adaptation response it often overestimates adaptation compared to observed risk reduction behavior.

• This study approximates sub-optimal adaptation responses using the National Coastal Property Model.

• Inefficient adaptation in Tampa, FL and Virginia Beach, VA can cost over a billion dollars.