Adapting to sea level rise and storms Authors: Robert Mendelsohn, Jinge Li, Jonghyun Yoo Published: 2026 ABSTRACT It is well known that sea level rise (SLR) will force retreat along rural coastlines but that cities will need to protect their coasts. But how exactly should seawalls, flood insurance, and retreat be used over time to minimize the overall cost of coastal flooding to society? This paper examines dynamic coastal adaptation in six cities along the eastern seaboard of the United States through 2180. We argue that seawalls should be designed to minimize the sum of seawall costs and expected residual flood damage. Flood insurance for residual damage can address risk aversion. The results reveal that only areas with high expected damage per kilometer of coastline require walls. Only a few segments of most cities need a wall now. The optimal wall height is about 1–2 m. SLR will force more urban seawalls to be built over the next 100 years. The faster SLR rises, the sooner they will need to be built.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-50397-2. Introduction Tidal stations in the United States reveal that sea level rise (SLR) has been following an upward path from as early as the late 19th century1. Tidal stations from around the world confirm SLR is a global phenomenon2. Under climate change, SLR is projected to increase3. In the 1990’s, coastal studies began to investigate how to adapt to the permanent inundation from rising seas4–6. These early SLR studies suggested that rural areas should retreat as the seas advance but that the more valuable land in cities should be protected by seawalls. As SLR proceeded over time, the studies recommended building ever higher walls. This early research on SLR, however, did not address storms. Later studies showed that incorporating a 100-year storm to SLR dramatically increases flood damage7–9. This study contributes to the adaptation literature by examining how coastal flooding should be managed when all storm surges gradually become higher over time. Even before SLR was considered a problem, economic studies evaluated how to protect against storms. The first formal economic analysis of coastal storms evaluated a model where the marginal cost of protection was equated to the expected marginal damage from all storms10. The model suggested that seawalls were appropriate protection but should allow some storms to overtop the walls. The author disagreed with his own model because it implied that the Netherlands should tolerate frequent catastrophic flooding. He instead recommended a more precautionary approach to flood protection which has been widely adopted. The Netherlands consequently build seawalls to stop a storm that happens once every 1000 years. The precautionary approach may be appropriate if a country (like The Netherlands) is below sea level because any flood that tops the seawall leads to catastrophe. However, most coastal cities are not below mean sea level. For most cities in the world, this paper argues that the expected-value model is the appropriate framework for designing seawalls. We demonstrate in this paper that the expected value approach is generally less costly than building either precautionary tall walls or no walls at all. The low walls implied by the expected-value approach may occasionally be overtopped, but fair insurance for residual damage can be less costly than building precautionary walls. The expected-damage storm model, however, must be updated to account for Sea Level Rise (SLR). SLR adds a dynamic dimension to the problem of flooding: as seas rise, the expected damage increases annually. We build a new simulation model called Coastal REsilience to Storm Surge (CRESS). The simulation model manages storm surges that gradually intensify as SLR proceeds. The model also includes economic growth along the shoreline. Every 50 years, the model re-evaluates what to do over the next 50 years. It calculates the present value of the expected cost and the residual damage of a wide range of alternative seawall choices. Armed with this information, a decision maker can compare “optimal walls” that minimize wall cost plus residual damage, no walls at all, and precautionary walls that are effectively safe. Another issue in the literature concerns how spatially detailed coastal models should be. National and global coastal models have relied on large coastal segments7,9,11–14. This study examines a spatially detailed model to assess the importance of scale. The model does not explicitly measure the value of wetlands and beaches, but it does take care not to damage those coastal resources. We assume that walls will be built at the high tide line so that the walls do not cause beach erosion or prevent wetlands from access to the sea. When combined with SLR, the high tide rule implies walls will retreat inland as they are replaced over time. Even in cities, there should be a slow retreat from the coastline. The model is calibrated to six cities spread along the Atlantic coastline of the United States (Bridgeport, Connecticut; Middletown Township, New Jersey; Norfolk, Virginia; Wilmington, North Carolina; Charleston, South Carolina; and Jacksonville, Florida). There is 814 km of coastline in these six cities. Only Charleston has an existing seawall. The cities were chosen because they are near long-standing National Oceanic and Atmospheric Administration (NOAA) tidal stations that allow estimation of the probability distribution of storm surges. The analysis compares the cost and damage of no walls at all, optimal seawalls, and precautionary seawalls of 3.7 m in height above ground. We start by calculating the expected damage if no urban seawalls are built. We then calculate whether to build a seawall in 2030, 2080, and 2130. We then look for the seawall that minimizes the sum of cost and expected damage. Note that this calculation identifies not only what seawalls to build but also where not to build them at all. Methods: theory We begin with a simple one-period example to understand how to capture the probability distribution of storms in a flood model10. For each coastal site i, there is a probability density function, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\pi\:}_{i}\left(h\right)$$\end{document}, of storm surges of heights, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:h$$\end{document}, up to some maximum height, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{max}$$\end{document}. The height of the surge is effectively measuring the severity of the surge. When a surge of height \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:h$$\end{document} occurs, it causes damage \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:D\left(h|X\right)$$\end{document} per event that depends on the characteristics of the site, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}_{it}$$\end{document}. These characteristics measure the vulnerability of each site. The expected value of the total annual damage in an unprotected site is:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E\left[{TD}_{i}\right]={\int\:}_{0}^{{h}_{max}}{\pi\:}_{i}\left(h\right){D}_{i}\left(h|{X}_{i}\right)dh$$\end{document} The expected value of the annual residual damage, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E\left[{RD}_{i}\left({H}_{i}\right)\right]$$\end{document}, of a seawall of height, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{H}_{i}$$\end{document}, that stops surges below \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{H}_{i}$$\end{document} is:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E\left[{RD}_{i}\left({H}_{i}\right)\right]={\int\:}_{{H}_{i}}^{{h}_{max}}{\pi\:}_{i}\left(h,t\right){D}_{i}\left(h|X\right)dh$$\end{document} This model suggests that the optimal choice of seawall height, Hi, should minimize the sum of the annual seawall cost, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{i}\left({H}_{i}\right)$$\end{document}, and the expected residual flood damage:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\underset{H}{\mathrm{min}}\left\{{C}_{i}\left({H}_{i}\right)+E\left[{RD}_{i}\left({H}_{i}\right)\right]\right\}$$\end{document} This implies that wall height should be chosen so that the expected marginal reduction in damage from increasing wall height just equals the marginal cost. Note that the author of this approach did not want this model applied because it appeared to suggest low walls for the Netherlands. However, once one accounts for the fact that The Netherlands is below sea level, any major coastal flood would lead to catastrophic damage. That is, the expected damage model when applied to the Netherlands suggests a very tall and highly protective wall is best. However, much of the world’s coastline, unlike the Netherlands, is above sea level. Flooding of individual segments of the coastline is not catastrophic. It is not necessary or even desirable to build very tall seawalls. Fair insurance can be less costly than very tall walls for managing rare but powerful storm surges. Risk averse potential victims have a less expensive option in flood insurance. The damage to them would be the certain premium they must pay in advance for insurance. With fair insurance, the value of the flood risk is the expected value of flood damage. For people who choose not to buy flood insurance, it is reasonable for society to use the expected value of flood damage as a measure of the damage they are expected to incur. Society can therefore value residual flood damage using the expected value of the remaining flood damage. Assuming that a seawall is a capital investment with a finite lifetime of T years, the residual damage with a wall of height H is the present value of the annual residual damage over the wall’s lifetime:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PV\left(E\left[{RD}_{i}\left({H}_{i}\right)\right]\right)=\:\sum\:_{t=1}^{T}\left\{\underset{H}{\overset{{h}_{max}}{\int\:}}\pi\:\left(h,{SLR}_{i}\left(t\right)\right){D}_{it}\left(h\right)dh\right\}/{(1+r)}^{t}$$\end{document} where r is a discount rate. Expected annual damage each year changes as SLR progresses. We model this non-stationary process by increasing the surge height of each storm by SLR given the probability distribution of storm surge15–17. So \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:\left(h\right)$$\end{document} is now a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:\left(h,SLR\left(t\right)\right)$$\end{document}. SLR will increase the expected value of flood damage, the expected value of avoided flood damage, and the expected value of residual flood damage over the lifetime of the seawall. The higher the rate of SLR, the more the expected flood damage increases. The present value of the cost of seawalls, PV(C(H)), is made of two components. An initial upfront capital investment to build the seawall, K(H), and a stream of annual maintenance costs M(H):5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PV\left({C}_{i}\left({H}_{i}\right)\right)=K\left({H}_{i}\right)+{\int\:}_{t=1}^{T}M\left({H}_{i}\right){e}^{-rt}dt$$\end{document} The objective of the optimal wall is to minimize the sum of the present value of the expected residual damage and the present value of the cost of the seawall:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\underset{{H}_{i}}{\mathrm{min}}\left\{PV\left(E\left[{RD}_{i}\left({H}_{i}\right)\right]\right)+PV\left({C}_{i}\left({H}_{i}\right)\right)\right\}$$\end{document} Note that this decision involves a large expenditure upfront for the seawall in return for a stream of reduced flood damage. The decision is consequently sensitive to the discount rate (interest rate for a long-term municipal bond). Higher discount rates reduce the present value of the stream of maintenance costs and especially flood damage. We assume that damage and cost beyond the lifetime of the seawall do not affect the initial wall decision. Near the end of the lifetime of each seawall, we assume society makes a new decision to consider retreat or building a new seawall. Coastal protection is effectively a sequence of decisions made over time. We follow earlier SLR literature and examine this sequence of choices for different SLR scenarios as though they are known. The scenarios bracket plausible outcomes so that decision-makers can understand the role SLR might play. In practice, we believe actual future decisions would be based on conditions observed in the future. The overall choice therefore involves acting and then learning. We leave modeling this act and learn strategy under uncertainty for future research. Throughout, we assume ‘fair insurance’ is available to all households. We assume risk-averse households and firms purchase this fair flood insurance and therefore use the premium as the value of flooding. We assume society should also use the expected value of damage when evaluating a seawall for households or firms who do not buy insurance. We discuss the limitations of this insurance assumption in the Discussion. Methods: model The Coastal REsilience to Storm Surge (CRESS) model is an integrated assessment tool that combines information from oceanography, geography, and economics to predict the cost and expected flood benefits of seawall construction. CRESS measures the expected benefit, residual damage, and cost of alternative seawall designs, allowing for comparisons between strategies such as maximizing net benefits, opting for no protection, or building seawalls to stop a 100-year storm. This paper applies the CRESS model to six representative cities along the eastern seaboard of the United States (see Fig. 1). These cities were selected due to their proximity to longstanding NOAA tidal stations, providing robust data for analysis. The precise definitions and boundaries of each city are provided in Sect. 1 (“Site Description”) of the Supplementary Information. Fig. 1Location of six cities. Note: Tidal stations are located nearby each of these cities along the eastern coast of the United States. Location of six cities. Note: Tidal stations are located nearby each of these cities along the eastern coast of the United States. A common assumption in the literature is that coastal segments of 70–100 km are homogeneous7,14, implying that a uniform wall could be built for each large segment. CRESS examines this homogeneity assumption by measuring the desired height of walls in numerous smaller segments within each city (ranging from 1 to 20 km). The model examines a 4 m flood in each city and determines which coastal land masses could be independently protected by a wall along their coastlines. For each of these segments, the model calculates the expected damage with no wall, the optimal length and height of a wall, and the cost of a 3.7 m high precautionary wall. In some segments, the optimum may well be no wall at all if the cost exceeds the benefit. For each coastal segment, we measure the area at each elevation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:\left(\epsilon\:\right)$$\end{document}, (up to 5 m) using Amazon Web Services (AWS) Open Data “Terrain Tiles”. This tileset is a mosaic generated by the United States Geological Survey using LiDAR and alternative data. Using the AWS data, we measure the area at each elevation (slice) in each segment. Figure 2 reveals the elevations throughout Norfolk, Virginia. The high-elevation part of the city, shown in white, is not vulnerable to coastal flooding over the next 150 years. The purple areas are low elevation lands that are highly vulnerable to flooding. Along the coast and up the rivers, there are also slivers of land in orange and red that are flood vulnerable. Note that streams and rivers along the coast allow surges to travel into the interior of the city. Elevation maps of the remaining cities are in Sect. 2: Elevation Maps (Figures S1: S5) in the Supplementary Information. We rely on a bathtub model of flooding. This implies that a surge of height, h, will flood all the land up to that height. In practice, if a surge is short-lived, the flood levels may not reach deep into the land all the way to the surge height. The bathtub model may overestimate some of the flooding deep into the land. Google Earth was used to determine the fraction of land, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\eta\:$$\end{document}, that is developed in each segment. From a sample of buildings in each segment, we estimate the average lot size, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document}, and property value. The aggregate number of buildings, ni, in each slice was calculated by dividing the area by the average lot size. The aggregate value of buildings in each slice was determined by multiplying the average value of a building in the segment by the number of buildings in the slice. We assume the lot size, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document}, in developed areas and the value per building, v, are homogeneous within each segment. Fig. 2Flood vulnerable land in Norfolk, Virginia. Note: The black line marks the boundary of the study area which is all of the city of Norfolk. The coastline to the north is the Atlantic Ocean. The interior coastline is adjacent to the Elizabeth River. The purple areas are particularly low but the orange and red areas are also flood vulnerable. The white area is largely safe from flooding. Figure generated by authors from Amazon Web Services (AWS) Open Data “Terrain Tiles”. Flood vulnerable land in Norfolk, Virginia. Note: The black line marks the boundary of the study area which is all of the city of Norfolk. The coastline to the north is the Atlantic Ocean. The interior coastline is adjacent to the Elizabeth River. The purple areas are particularly low but the orange and red areas are also flood vulnerable. The white area is largely safe from flooding. Figure generated by authors from Amazon Web Services (AWS) Open Data “Terrain Tiles”. By flooding land up to 4 m, we can detect which segments are effectively independent of their neighbors. These are the smallest geographic units that are independent. We then number each segment and study them separately. The boundaries are determined by geography, often by a river or ridge. Each segment has a stretch of coastline or riverfront that makes it vulnerable to storm surges. If a wall were built along all the low-lying parts of this shoreline, it would prevent surges below the height of the wall. We identify 54 segments across the six cities as shown in Sect. 3: Segment Maps (Figures S6: S10) in Supplementary Information. Figure 3 displays the ten segments in Norfolk, Virginia. The colored lines track the coastline of each segment. The area landward of each colored coastline is the section of the city protected by a wall along that coastline. Only the low-lying land identified in the elevation maps is vulnerable to flooding. For each segment, we calculate the height of the land near the coast. It is only necessary to build a seawall across the low parts of the coastal edge, so the length of the seawall depends on the height of the seawall. Higher seawalls must be longer because the elevation of more of the segment’s coast is below the targeted height. We assume that each seawall is built at the elevation of Mean Higher High Water (MHHW), which is the mean high tide caused solely by the moon and sun. We choose MHHW for the wall location for several reasons. First, placing the seawalls at MHHW protects beaches and saltwater wetlands from damage by the wall. Historical seawalls below MHHW often damaged saltwater wetlands by cutting them off from tidal flows. The seawall still could potentially limit the landward expansion of wetlands. But we are anticipating that seawalls would retreat when being replaced at the end of their lifetime. The new seawall would be built at the new MHHW. Similarly, walls built on beaches can cause erosion. By building the wall behind the beach, one can limit this problem. Second, building a seawall at MHHW will keep it away from daily tides. The seawall will be subject to much less wave action and therefore have lower maintenance costs. Third, many states control the land below MHHW. Building a seawall at the boundary of state land would avoid taking of private land. Fig. 3Ten segments in Norfolk VA. Note: The colored lines in Fig. 3 mark the coastline of each segment. The gray lines help identify the boundary of each segment. Each segment is numbered. The colors of the coastline refer to when each segment gets a wall with Low SLR. How much of the land in each segment that is flooded depends on the height of the surge. Figure generated by authors from Google Earth. Ten segments in Norfolk VA. Note: The colored lines in Fig. 3 mark the coastline of each segment. The gray lines help identify the boundary of each segment. Each segment is numbered. The colors of the coastline refer to when each segment gets a wall with Low SLR. How much of the land in each segment that is flooded depends on the height of the surge. Figure generated by authors from Google Earth. An important topic for future research is whether to build seawalls even farther inland. The height of the surge that is blocked depends on the elevation of the top of the wall which is the sum of the elevation of the ground and the height of the wall above ground. Because seawalls built on higher ground can achieve the same level of protection with less height above ground, they may cost less for any given surge level. This encourages seawalls to be built inland. However, as seawalls get closer to existing buildings, there may be complications that require modification to either the seawall or the buildings. When existing buildings end up on the seward side of a seawall, the seawall cuts them off from the rest of the city. There may be serious legal and political issues as a result. We leave the controversial choice of moving seawalls further inland than MHHW for future research. Damage The damage \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{d}_{i}$$\end{document}, that a storm surge of height, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:h$$\end{document}, would cause to a specific building i depends on the value of the affected building, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{i}$$\end{document}. We follow the literature and assume that the building value is 70% of the property value. We rely on previously estimated damage functions from Hazus18. We focus on the damage function that applies to 1–2-story buildings because it describes most of the buildings in the dataset. An important limitation of this approach is that it does not capture how flooding affects tall buildings. This introduces uncertainty into our analysis of downtown segments. The damage function, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:d\left(h-{\epsilon\:}_{i},{n}_{i}*v\right)$$\end{document}, depends on the depth of the flooding at the building (the surge height, h, minus the elevation of the building site, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{i}$$\end{document}), the number of buildings at each elevation, ni, and the average value of a building in the segment, v. Note that elevation in this study is defined relative to ground level whereas the depth damage literature generally measures elevation relative to the first floor of each building. Buildings without basements are assumed to be flooded once the flood reaches the building:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:D\left(h,\epsilon\:,{n}_{i},v\right)=\left\{\begin{array}{c}0,\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:h<\epsilon\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\\\:\frac{\left(h-\epsilon\:\right)}{8}{n}_{i}v,\:\:\:\:\:\:\:\:if\:\epsilon\:\le\:h\le\:\epsilon\:+8\:\:\:\:\:\:\:\\\:\:\:{n}_{i}v,\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:h>\epsilon\:+8\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\end{array}\right.$$\end{document} Buildings with basements are assumed to be flooded at 1 m below ground elevation:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:D\left(h,\epsilon\:,{n}_{i}v\right)=\left\{\begin{array}{c}0,\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:h<\epsilon\:-1\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\\\:\frac{\left(h-\epsilon\:\right)}{8}{n}_{i}v,\:\:\:\:\:\:\:\:if\:\epsilon\:-1\le\:h\le\:\epsilon\:+7\:\:\:\:\:\:\:\\\:\:\:{n}_{i}v,\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:if\:h>\epsilon\:+7\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\end{array}\right.$$\end{document} The expected annual damage at time t for buildings at elevation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\epsilon\:$$\end{document}i with aggregate value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{n}_{i}v$$\end{document} is therefore:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E\left[ {D\left( {\epsilon _{i} ,n_{i} v} \right),t} \right] = \int_{{\underline{\epsilon } }}^{8} {\pi _{t} \left( h \right)D\left( {h,\varepsilon _{i} ,n_{i} v} \right)e^{{\alpha t}} dh}$$\end{document} where we integrate from the lowest relevant elevation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\underline{\varepsilon}$$\end{document}, to 8 m, reflecting all possible surge heights that will cause additional harm. The total annual damage, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:D\left(h\right)$$\end{document}, in each segment from each specific surge of height \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:h$$\end{document} is the integral of the damage across the elevations:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:D\left(h\right)={\int\:}_{0}^{h}d\left(h-{\epsilon\:}_{i},{n}_{i\mathrm{*}}v\right)d\epsilon\:$$\end{document} The annual expected total damage from all surges up to a maximum (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{max}$$\end{document}) in an unprotected segment is the integral of the damage times the probability of each surge as shown in Eq. (1). The annual expected residual damage is given by Eq. (2). Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E\left[TD\right]$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E\left[RD\left(H\right)\right]$$\end{document} vary for every segment depending on the aggregate building value at each elevation. The analysis focuses on measuring flood damage to buildings. In practice, flood events cause other damages as well. Building contents are lost. Infrastructure (roads, bridges, and utilities) is often damaged. There is a high cost to cleaning up parks and debris. There are overtime costs for police and fire units. There are also deaths and injuries. Previous assessments of specific floods, such as Hurricane Sandy, suggest that these other damages are about the same size as the flood damage to buildings19. They are likely to be proportional to the building damage in each segment. For example, as more buildings are flooded, more electricity and water utility lines, as well as more kilometers of streets, are likely to be affected. We double the building damage to estimate the total damage from flooding. Note that these additional damages tend to accrue to the local or state governments and utilities rather than to vulnerable private property owners. We estimate the cumulative probability function of surges, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:\left(h\right)$$\end{document}, by fitting extreme tides to a generalized extreme value distribution15 using the maximum tide observed in each year at the nearby tidal station. The cumulative density function is:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:F\left(h|\:\mu\:,\:\sigma\:,k\right)=\mathrm{exp}\left\{-{\left[1+k\frac{\left(h-\mu\:\right)}{\sigma\:}\right]}^{-\frac{1}{k}}\:\right\}$$\end{document} The parameters of this GEV regression are shown in Table S.1 in the Supplementary Information. The probability density function, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:\left(h\right)$$\end{document}, shown in Fig. 4, is the derivative of this cumulative density function. This distribution captures the probability of common low storm surges but also estimates the probability of rare but high storm surges. Although lower, more common tides cause no harm today, with SLR, they can eventually become damaging. We therefore extrapolate from the peak of the GEV tides to more common tides that might occur up to 9 times a year. Over the next 150 years, the model suggests that the highest surges will remain below 5 m, and below 3.5 m at most sites. Properties above these peak surges are not flooded. The GEV results for each city presented in Fig. 4 reveal that storm-surge distributions differ across cities. Bridgeport has much higher surges because of the funnel shape of Long Island Sound, whereas Wilmington has much lower surges because it lies farther inland than the other cities. Fig. 4The probability distribution of storm surge height. Note: The probability of storm surges is calculated from tidal station extreme values collected by NOAA (2023a) at each site. The area underneath each curve sums to 1. Figure generated by authors. The probability distribution of storm surge height. Note: The probability of storm surges is calculated from tidal station extreme values collected by NOAA (2023a) at each site. The area underneath each curve sums to 1. Figure generated by authors. We examine two SLR scenarios in this paper. The Low SLR scenario follows the predicted path associated with the relatively low RCP2.6 IPCC global emission scenario3. The High SLR scenario follows the predicted path with the relatively high RCP8.5 IPCC emission scenario3. Relative to the MHHW in 2025, the Low SLR scenario leads to an increase of 0.8 m by 2180 and the High SLR scenario leads to a 2 m increase by 2180. The precise path is shown in Supplementary Information Figure S.11. These scenarios should be interpreted as upper and lower bounds, with the most likely outcomes lying between these extremes. We use them to illustrate how adaptation changes across plausible SLR pathways, not because either extreme is itself the most likely outcome. The relevant measure for coastal protection is relative SLR, which measures changes in ocean height relative to a fixed location on land. We rely on NAVD88 to measure elevation of both surges and buildings. This is the traditional measure of elevation on land. Although a great deal of attention is focused on changes in the elevation of the ocean, in many places, the land is also changing in elevation. The land could be subsiding because of human activities such as withdrawing groundwater or fossil fuels from beneath the city. But the land can also rise in response to melting since the last glacial period. These changes in land elevation are location-specific and they are independent of the effects of climate change on oceanic elevation. To measure local changes in land elevation, we examine the difference between the observed relative rate of SLR21 and the average observed SLR of oceanic height. We assume that the historic changes in land elevation are a linear function of time. Each city has its own rate of land elevation change, LEi, as listed in Table S.1 in the Supplementary Information. The resulting SLR pathways are slightly different for each city as shown in Fig. 5. Each SLR scenario is incorporated into the model by increasing the flood height of each storm by the relative SLR measured in Fig. 5. This effectively shifts the probability distribution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:\left(h\right)$$\end{document} of storm surges to the right over time. It corresponds to moving the GEV location parameter, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mu\:}_{i,0}$$\end{document}, from its initial observed level in 2025, to ever higher levels over time.12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mu\:}_{it}={\mu\:}_{i,0}+\:SL{R}_{it}$$\end{document} Figure 6 illustrates this principle by adding the relative sea level rise one would expect after several decades to the underlying probability distribution of storm surge. Note that the probability of seeing common surges increases dramatically whereas the probability of seeing large surges increases just slightly. This change causes the expected damage from low surges to increase dramatically, whereas the expected damage from high surges increases only slightly. In this analysis, we cap the present value of the maximum flood damage each period at each surge height to be no more than the cumulative building value up to that elevation. Fig. 5MHHW trajectories by city under low/high SLR. Note: The curves above identify the Low and High SLR scenarios for the height of MHHW relative to land in each city over the next 150 years starting in 2030. It is very likely that the actual outcome will lie between these scenarios. Bridgeport, CT is the highest. Figure generated by authors. MHHW trajectories by city under low/high SLR. Note: The curves above identify the Low and High SLR scenarios for the height of MHHW relative to land in each city over the next 150 years starting in 2030. It is very likely that the actual outcome will lie between these scenarios. Bridgeport, CT is the highest. Figure generated by authors. Fig. 6SLR shifts probability distribution of surges. Note: Relative sea level rise over several decades shifts the probability distribution of storm surges at each site. The effect leads to a large increase in the frequency of low surges but only a small change in the frequency of large surges. Figure generated by authors. SLR shifts probability distribution of surges. Note: Relative sea level rise over several decades shifts the probability distribution of storm surges at each site. The effect leads to a large increase in the frequency of low surges but only a small change in the frequency of large surges. Figure generated by authors. The present value of the residual damage, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PV\left(E\left[RD\left(H\right)\right]\right)$$\end{document}, in each segment is equal to:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PV\left(E\left[{RD}_{i}\left({H}_{i}\right)\right]\right)=\:\sum\:_{t=1}^{50}\left\{\underset{{H}_{i}}{\overset{{h}_{max}}{\int\:}}\pi\:\left(h,{SLR}_{i}\left(t\right)\right){D}_{i}\left(h\right)dh\right\}/{(1+r)}^{t}$$\end{document} The term inside the brackets is the expected annual residual damage each year. The expected annual residual damage rises over time as the relative sea level rises. Assuming that seawalls would be funded by long-term bonds, we use a long-term real interest rate of 4% for a discount rate. Seawall cost The computation of the cost of the seawall follows the literature4,5,22. The cost of constructing the wall rises with the volume of the wall. Coastal walls tend to have a triangular shape because the base must get wider as the wall gets taller to prevent it from falling over. The cost of walls rises with the square of the wall height times the wall length. However, we rely on updated estimates of construction cost22 which suggests that a wall 1 m high costs $5.2 million per kilometer. We modify both height and length in the construction-cost equations. Each wall in each segment has a uniform elevation at the top. But because the ground level undulates along the coast, the height of the wall above ground will vary. The elevation of the top of the seawall, Hi, is the sum of the seawall height, WHi, and the elevation of the land, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\epsilon\:}_{i}$$\end{document}, upon which the seawall is built (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{H}_{i}={WH}_{i}+{\epsilon\:}_{i}$$\end{document}). The length of the wall also varies slightly with the elevation of the top of the wall. A greater fraction of the segment requires protection as wall height increases. For each segment, we measure the length of the coastline, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{L}_{i}\left({\epsilon\:}_{i}\right)$$\end{document}, at each height. The construction cost is:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{i}\left({H}_{i}\right)=5200{\int\:}_{\epsilon\:\:}^{{H}_{i}}{\left({H}_{i}-{\epsilon\:}_{i}\right)}^{2}L\left({\epsilon\:}_{i}\right)d{\epsilon\:}_{i},\:\:$$\end{document} A coastal wall is expected to last 50 years, after which it will have to be replaced22. The maintenance cost over the lifetime of the wall is the present value of annual maintenance for 50 years. Annual maintenance (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\psi\:$$\end{document}) is expected to be 0.5% of construction cost per year22. Given these assumptions, the present value of maintenance costs is:15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\omega\:=\psi\:\frac{1-{\mathrm{e}}^{-50r}}{r}C\left(H\right)$$\end{document} The present value of maintenance adds about 10% to the total cost of a wall. The present value of the total cost of the seawall with height \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:H$$\end{document} in each segment is therefore:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PV\left(C\left(H\right)\right)=C\left(H\right)\mathrm{*}\left(1.10\right)$$\end{document} The present value of the cost of a 1 m wall is $5.7 million/km, and that of a 2 m wall is $22.8 million/km. Results We begin by examining the expected damage in the two SLR scenarios over time. Annual expected damage rises steadily in both scenarios, partly because of growth in aggregate property value in the flood zone, but mostly because of sea level rise. The analysis then considers whether to build a wall in each segment. The model considers all plausible wall heights for each segment in each period. If the expected cost exceeds the expected benefit for all wall heights, no wall is built in this period. Table 1 shows the number of segments for which the model recommends seawalls in each period and city. In these six cities with their 54 segments, 15 segments should have a seawall by 2030 in the low SLR scenario, and 17 segments should have a seawall in the High SLR scenario. The speed of SLR has only a small impact at first. Only Charleston already has a seawall. The gap between what is needed by 2030 and what currently exists may partly reflect the federal emphasis on precautionary seawalls. The general factor that determines which segments need protection is the expected damage per length of coastline. When this is low, it is not worth the cost of building a wall. As shown in Table S.2 in the Supplementary Information, segments with high expected damage per length generally warrant a wall. In 2030, the segments that need walls have an average present value of expected damage per length of $20.3 million/km. The segments that do not require a wall have an average expected damage per length of only $0.8 million/km. All the scenarios call for more segments to be protected over time as SLR proceeds. Whether SLR remains close to historic rates or rapidly accelerates, more seawalls are needed over time. With the High SLR scenario, however, more seawalls are needed earlier. The main difference between the SLR scenarios is timing. Table 1Predicted number of segments requiring seawalls by city, year, and SLR scenario.SLR scenarioCityTotal segmentsSegments with seawalls in203020802130LowBridgeport5014Middletown7455Norfolk10149Wilmington8678Charleston1111Jacksonville233711Total54152538HighBridgeport5055Middletown7466Norfolk10489Wilmington8588Charleston1111Jacksonville2331323Total54174152Note “Total Segments” reports the total number of coastal segments identified in each city. The columns for 2030, 2080, and 2130 report the number of segments in that city for which the model recommends a seawall in that year. The effect of faster SLR is limited in the near term but becomes more important over time. Results from CRESS2.2. Table 1Predicted number of segments requiring seawalls by city, year, and SLR scenario.SLR scenarioCityTotal segmentsSegments with seawalls in203020802130LowBridgeport5014Middletown7455Norfolk10149Wilmington8678Charleston1111Jacksonville233711Total54152538HighBridgeport5055Middletown7466Norfolk10489Wilmington8588Charleston1111Jacksonville2331323Total54174152Note “Total Segments” reports the total number of coastal segments identified in each city. The columns for 2030, 2080, and 2130 report the number of segments in that city for which the model recommends a seawall in that year. The effect of faster SLR is limited in the near term but becomes more important over time. Results from CRESS2.2. Predicted number of segments requiring seawalls by city, year, and SLR scenario. Note “Total Segments” reports the total number of coastal segments identified in each city. The columns for 2030, 2080, and 2130 report the number of segments in that city for which the model recommends a seawall in that year. The effect of faster SLR is limited in the near term but becomes more important over time. Results from CRESS2.2. Table 2 describes the characteristics of the predicted seawalls. The elevation of the land where the wall is built is equal to MHHW. This elevation rises over time with SLR. How far inland the wall retreats depends on the slope of the land. But with the height of the ocean rising by only small amounts every 50 years, the retreat will generally be about 5–50 m every 50 years depending on the SLR scenario and land slope. The heights of walls within each city vary across segments. Cities should not necessarily build all their seawalls at the same height. Segments with higher expected damage per km will often need taller seawalls. Cities should carefully examine every segment separately to determine both when they need a wall and how high it should be. Table 2Average wall characteristics each period by SLR scenario.SLR scenarioVariable203020802130LowElevation wall (m)0.790.971.24Height of walls (m)0.971.110.99Wall Length (km)126.7279.6385.7HighElevation wall (m)0.741.081.77Height of walls (m)1.051.411.52Wall Length (km)146.0472.6653.0Note: Average characteristics of optimal walls across segments in each time period. Elevation of the wall is the elevation of the land upon which the wall is built which we assume is the high tide line in each city. The height of the wall is the difference between the top of the wall and the ground. The elevation of the top of the wall is the sum of ground elevation and wall height. Wall length is the aggregate length across the six cities. Results from CRESS2.2. Table 2Average wall characteristics each period by SLR scenario.SLR scenarioVariable203020802130LowElevation wall (m)0.790.971.24Height of walls (m)0.971.110.99Wall Length (km)126.7279.6385.7HighElevation wall (m)0.741.081.77Height of walls (m)1.051.411.52Wall Length (km)146.0472.6653.0Note: Average characteristics of optimal walls across segments in each time period. Elevation of the wall is the elevation of the land upon which the wall is built which we assume is the high tide line in each city. The height of the wall is the difference between the top of the wall and the ground. The elevation of the top of the wall is the sum of ground elevation and wall height. Wall length is the aggregate length across the six cities. Results from CRESS2.2. Average wall characteristics each period by SLR scenario. Note: Average characteristics of optimal walls across segments in each time period. Elevation of the wall is the elevation of the land upon which the wall is built which we assume is the high tide line in each city. The height of the wall is the difference between the top of the wall and the ground. The elevation of the top of the wall is the sum of ground elevation and wall height. Wall length is the aggregate length across the six cities. Results from CRESS2.2. The major change in walls observed over time in Table 2 is their increasing length. This is largely due to more segments requiring walls over time. The key role of seawalls is providing extensive protection, not zero risk protection. More of the urban coastline must be protected. But by keeping the walls low, the cost per km of walls can be kept low enough to afford more coastline being protected. The lowest existing buildings largely remain intact in the Low SLR scenario through 2180. However, in the High SLR scenario, after 2090, hundreds of buildings start to fall below the MHHW in both Wilmington and Jacksonville. In the High SLR scenario, these will likely be destroyed by storms before 2090, but in general, they should not be replaced. It is important that the market is aware of the rising risk of building at low elevation in order to discourage inefficient rebuilding. With fair insurance, premiums will rise sufficiently to produce that signal. But subsidies to flood insurance will distort that signal and can eventually lead to buildings below MHHW. After 2090, if the new wall is built at MHHW, these buildings will be on the wrong side of the seawall. Because owners have a long time to depreciate these buildings, the expected damage should be low. But it is important that landowners be warned of future risks so that vulnerable buildings are depreciated. Table 3Aggregate costs of walls over time by SLR scenario.SLR ScenarioVariable203020802130LowBenefit of Walls2,4018,04517,083Cost of Walls5531,7311,787Residual Damage7851,403456HighBenefit of Walls2,71034,165113,146Cost of Walls8254,5659,436Residual Damage8214,1152,429Note: All costs are the present value in each 50 year period of the aggregate outcome across the six cities. Values are in millions of USD. Results from CRESS2.2. Table 3Aggregate costs of walls over time by SLR scenario.SLR ScenarioVariable203020802130LowBenefit of Walls2,4018,04517,083Cost of Walls5531,7311,787Residual Damage7851,403456HighBenefit of Walls2,71034,165113,146Cost of Walls8254,5659,436Residual Damage8214,1152,429Note: All costs are the present value in each 50 year period of the aggregate outcome across the six cities. Values are in millions of USD. Results from CRESS2.2. Aggregate costs of walls over time by SLR scenario. Note: All costs are the present value in each 50 year period of the aggregate outcome across the six cities. Values are in millions of USD. Results from CRESS2.2. Table 3 presents the aggregate expected benefit, cost, and residual damage across the six cities for each period of time in each scenario. The cost of building optimal seawalls rises over time as there are more walled segments. The benefit of the walls is the damage prevented. The benefit is equal to expected damage minus residual damage. With the optimal walls, the benefit far exceeds the cost of the seawalls. In the Low SLR scenario, the aggregate benefit-cost ratio of the seawalls rises from 3 in 2030 to over 6 in 2130. In the High SLR scenario, the aggregate benefit-cost ratio rises from 3 in 2030 to over 10 in 2130. The net benefit of building optimal seawalls rises more rapidly in the High SLR scenario. Table 4 calculates the total cost of the four prominent adaptation strategies: just flood insurance (no walls), 3.7 m safe seawalls everywhere, 3.7 m safe walls only in high damage/km segments, and optimal seawalls. Residual risk is assumed to be covered by fair flood insurance. We assume that all seawalls are built at MHHW at the beginning of each period. The total cost is the sum of the adaptation cost and the residual expected damage. In the Low SLR scenario, building no walls is cheaper than building safe walls through 2180. In the High SLR scenario, building no walls is cheaper than building safe walls until 2130 when the safe walls become the less costly of these expensive choices. However, in both the Low SLR scenario and the High SLR scenario, the least expensive choice in these simulations is always the optimal seawall with residual insurance. Initially, the total cost of the optimal seawalls (with residual insurance) is 44% and 59% of the cost of building no walls in the Low and High SLR scenarios respectively. By 2130, the cost of the optimal seawalls is just 22% and 12% of the cost of insurance alone in the Low and High SLR scenarios respectively. A similar result applies when comparing the optimal seawalls (with residual insurance) versus building safe walls. In 2030, the total cost of the optimal walls (with residual insurance) is 1.5% and 3% of the total cost of the “safe” walls in the Low and High SLR scenarios respectively. In 2130, the total cost of the optimal walls (with residual insurance) is 8% and 20% of the total cost of the “safe” walls in the Low and High SLR scenarios respectively. Even when safe walls are built only in places with high damage per kilometer, they remain an expensive alternative, except in the High SLR scenario for 2130. Table 4Total cost to society of alternative adaptation strategies.SLR ScenarioPolicy Option203020802130LowNo walls3,1478,97315,3003.7 m walls54,19962,55973,351Limited 3.7 m walls15,95335,98656,695Optimal Walls1,4573,1523,047HighNo walls3,53168,236115,5763.7 m walls62,53664,60668,398Limited 3.7 m walls20,19754,00066,165Optimal Walls1,6478,68011,865Note: The values are the present value of 50 years of wall cost plus residual damage in millions of USD at the beginning of each period. Results from CRESS2.2. Table 4Total cost to society of alternative adaptation strategies.SLR ScenarioPolicy Option203020802130LowNo walls3,1478,97315,3003.7 m walls54,19962,55973,351Limited 3.7 m walls15,95335,98656,695Optimal Walls1,4573,1523,047HighNo walls3,53168,236115,5763.7 m walls62,53664,60668,398Limited 3.7 m walls20,19754,00066,165Optimal Walls1,6478,68011,865Note: The values are the present value of 50 years of wall cost plus residual damage in millions of USD at the beginning of each period. Results from CRESS2.2. Total cost to society of alternative adaptation strategies. Note: The values are the present value of 50 years of wall cost plus residual damage in millions of USD at the beginning of each period. Results from CRESS2.2. Sensitivity analysis The parameters in CRESS are based on the available evidence. However, the parameters are nonetheless uncertain. The analysis already evaluates alternative sea level rise scenarios. In this section, we explore how sensitive the model results are to alternative assumptions about cost estimates, damage estimates, the discount rate, coastal waves, and tall buildings. We do a sensitivity analysis by varying one parameter at a time to understand how that parameter alters the results. The results of the Low SLR scenario for cost, damage, and interest rates are shown in Table 5 and the results of the High SLR scenario are shown in Table S3 in the Supplementary Information. Increasing (decreasing) cost, reducing (increasing) damage, and raising (lowering) the discount rate all reduce (increase) the number of walls and their size and length. Increasing (decreasing) the cost per unit of walls, increasing (decreasing) the damage per flood, or lowering (raising) the interest rate, all increase (decrease) the sum of wall cost and residual damage. The size of these effects varies depending on the parameter, time, and the SLR scenario. Most sensitivity scenarios lead to one or two walls changing. However, decreasing damage in 2130 in the Low SLR scenario leads to 6 fewer walls being built. Decreasing damage reduces wall length by about 20% in 2030 and 10% in 2130 and to 2% in 2130 with High SLR. Increasing damage increases wall length by 17% in 2030 and 2% in 2130 under Low SLR, and by 10% in 2030 and almost zero in 2130 under High SLR. The effect of cost on length is slightly smaller than the effect of damage on length with Low SLR and cost has a larger effect than damage on length with High SLR. Decreasing damage reduces height by about 3 cm in 2030 and by 12 cm in 2130 with Low SLR. Decreasing damage reduces height by 9 cm in 2030 and by 5 cm in 2130 with High SLR. Increasing damage increases wall height by 3 cm in 2030 and 5 cm in 2130 under Low SLR, and by 10 cm in 2030 and 2 cm in 2130 under High SLR. Decreasing cost increases the height of the walls by 12 cm in 2030 and by zero in 2130. With High SLR, decreasing cost increases wall height by 20 cm in 2030 and by 15 cm in 2130. Increasing cost causes wall height to fall by 2 cm in 2030 and by 6 cm in 2130 with Low SLR but with High SLR, wall height falls by 10 cm in 2030 and by 17 cm in 2130. With Low SLR, decreasing damage by 50% reduces the overall total cost (wall cost plus residual damage) by 28% in 2030 and by 20% in 2130. Increasing damage by 50% increases total cost by 17% in 2030 and by 8% in 2130. With High SLR, decreasing damage by 50% reduces total cost by 28% in 2030 falling to 20% in 2130. Increasing damage by 50% increases total cost by 17% in 2030 falling to 8% in 2130. In 2030, lowering cost reduces total costs by 26% with Low SLR but by 32% with High SLR. By 2130, lowering cost reduces total costs by about 45% in both SLR scenarios. In 2030, increasing cost by 50% leads to higher total costs of 19% with Low SLR and 22% with High SLR. By 2130, increasing cost by 50% leads to higher total costs of 32% with Low SLR and 36% with High SLR. Lower interest rates increase total costs because total costs reflect the discounted value of a stream of costs over 50 years. A 3% interest rate (relative to the assumed 4% real rate) increases total costs by 12% in 2030 for both SLR scenarios, and by 14% under Low SLR and 8% under High SLR in 2130. A 5% interest rate decreases total costs by 10% in 2030 for both SLR scenarios, and by 12% under Low SLR and 5% under High SLR in 2130. Table 5Sensitivity analysis of low SLR scenario.TypeCost 1.5XCost 0.5XDamage 1.5XDamage 0.5Xr = 3%r = 5%2030# Walls 15131717121714Δ % Length-9.8%14.8%16.8%-19.5%15.3%-6.0%Δ Height cm-2.012.03.0-3.00.00.0Δ% Benefits20.8%-29.2%27.8%-31.1%17.8%-11.6%Δ% Wall Cost17.2%-23.3%16.2%-25.1%8.1%-9.0%Δ % Residual Damage18.7%-25.8%21.1%-27.6%12.2%-10.1%2130# Walls 38364139323936Δ % Length-4.1%3.7%2.2%-10.1%8.0%-14.9%Δ Height cm-6.0-1.05.0-12.04.0-1.0Δ% Benefits24.4%-44.8%-0.8%-28.8%15.0%-12.6%Δ% Wall Cost48.9%-44.8%8.0%-0.3%11.0%-9.1%Δ % Residual Damage32.0%-44.8%1.9%-20.0%13.7%-11.5%Note: The Sensitivity Analysis with the High SLR Scenario is in Table S3 in the Supplementary Information. Each factor was tested one by one. Results from CRESS 2.2. Table 5Sensitivity analysis of low SLR scenario.TypeCost 1.5XCost 0.5XDamage 1.5XDamage 0.5Xr = 3%r = 5%2030# Walls 15131717121714Δ % Length-9.8%14.8%16.8%-19.5%15.3%-6.0%Δ Height cm-2.012.03.0-3.00.00.0Δ% Benefits20.8%-29.2%27.8%-31.1%17.8%-11.6%Δ% Wall Cost17.2%-23.3%16.2%-25.1%8.1%-9.0%Δ % Residual Damage18.7%-25.8%21.1%-27.6%12.2%-10.1%2130# Walls 38364139323936Δ % Length-4.1%3.7%2.2%-10.1%8.0%-14.9%Δ Height cm-6.0-1.05.0-12.04.0-1.0Δ% Benefits24.4%-44.8%-0.8%-28.8%15.0%-12.6%Δ% Wall Cost48.9%-44.8%8.0%-0.3%11.0%-9.1%Δ % Residual Damage32.0%-44.8%1.9%-20.0%13.7%-11.5%Note: The Sensitivity Analysis with the High SLR Scenario is in Table S3 in the Supplementary Information. Each factor was tested one by one. Results from CRESS 2.2. Sensitivity analysis of low SLR scenario. Note: The Sensitivity Analysis with the High SLR Scenario is in Table S3 in the Supplementary Information. Each factor was tested one by one. Results from CRESS 2.2. We also test the possible influence of waves along coastal segments facing the open water. We first measure the fraction of the coast in each segment that faces open water. We then explore what would happen along open water if each surge was 0.67 m higher because of waves. The detailed results by segment are shown in Tables S4-S6 in the Supplementary Information. Table S4 describes when waves encourage new walls to be built. Waves cause 4 new walls to be built along the open water in 2030 and 2 new walls in 2080 in the Low SLR scenario. Waves cause three new walls to be built along open water in the High SLR scenario in 2030. Waves also cause walls along the open ocean to be higher. The Low SLR results in Table S5 suggest waves would cause walls to be 18 cm higher on average. The High SLR results in Table S6 suggest waves would cause walls to be 8 cm higher on average. Waves have a larger effect in the Low SLR scenario because the baseline walls are lower so that adding height is less expensive compared to the High SLR scenario. A final issue concerns the flooding in downtown areas with tall buildings. The flood damage function in CRESS was based on 1–2 story buildings. It is highly likely this methodology is inadequate to measure how sensitive the downtown results might be to flooding. The analysis of the downtowns of Bridgeport (Segment 5), Norfolk (Segment 1), Wilmington (Segment 2), and Jacksonville (Segment 11) should therefore be interpreted with caution. Future research must be able to assess the effects of flooding on tall buildings to determine whether downtown areas need more flood protection than the rest of the city. Discussion This paper re-examines how coastal protection against storms can be modeled most effectively. The paper suggests that relying solely on flood insurance or solely on “safe” walls is unlikely to be cost-minimizing. The optimal coastal flooding strategy is to minimize the sum of the present value of the cost of seawalls and the present value of the expected residual flood damage. Seawalls are more effective than fair insurance at addressing common low-surge storms. Fair insurance, in turn, may be more cost-effective than tall seawalls for managing the damage associated with rare but powerful storm surges. In many urban contexts, modest walls together with residual insurance provide an effective combination. The modest seawalls are often less costly than fair insurance at addressing low surges. In turn, fair insurance is typically less costly than tall “safe” walls for rare but violent surges. In our framework, modest seawalls with residual insurance are less costly than either insurance alone or tall seawalls alone. An important exception arises when a nation is below sea level, in which case every major storm can be catastrophic and the tall walls may be necessary. Models of coastal protection must not only consider storms, but they must also consider sea level rise (SLR). The paper develops an integrated assessment model, CRESS, to measure both the impact of storms and the impact of SLR. The model measures the present value of expected flood damage, expected residual damage, and the cost of seawalls of varying heights. The model compares the total cost (construction, maintenance, and residual damage) of different combinations of flood insurance and various seawalls. CRESS is applied to six cities along the Atlantic Coast of the United States. The model accounts for the probability distribution of storm surges at each site as well as SLR. We examine seawalls built at MHHW (the high tide line) to avoid harming saltwater wetlands and beaches. Seawalls are assumed to retreat as MHHW moves inland. The model finds that without seawalls, the present value of flood damage over the next 50 years in these six cities would sum to $3-$3.5 billion depending on the SLR scenario. However, by 2130, the present value of flooding without seawalls could increase to $15.3 billion with Low SLR and $115 billion under High SLR. Tall seawalls would prevent much of this coastal damage, but they would cost between $16 billion and $60 billion every 50 years, depending on how many segments built them. The optimal choice is a set of low walls. These walls would gradually cover each city as SLR proceeds. Flood insurance would cover the small expected residual damage from storms that overtop the walls. The total cost of optimal seawalls and residual flood insurance for these six cities is about $1.5 billion in 2030. By 2130, these costs would rise to $3 to $12 billion for the Low and High SLR scenarios. The optimal seawall building strategy is considerably less expensive than building no walls at all. An important insight of the results is that some segments in cities need walls right away, others later when the sea is high enough, and a few segments are best without a wall at all. Getting the timing right for when to build a wall saves a lot of resources. The timing depends on the rate at which the sea rises. If SLR proceeds slowly, many walls can be delayed far into the future. If SLR is rapid, most coastal cities will need a lot of walls this century. Looking forward 150 years, it is uncertain how SLR will evolve. However, even in the rapid SLR scenario, there is ample warning that SLR is increasing quickly in the annual measurements of SLR. Every city will know decades in advance if rapid SLR is occurring. Another useful result is that the optimal height of seawalls should vary from segment to segment between cities and even within cities. Cities that face a higher probability of high surges need taller walls. Segments with high expected damage per length generally need taller walls. It is useful to design the walls that are best for each segment, rather than to assume one single height works in all cases. One flaw in this analysis is that it could not properly assess downtown segments with tall buildings. The flood damage function in the paper was based on data from 1 to 2 story buildings and could not detect whether tall buildings were more flood vulnerable. Future studies need to assess the flooding of tall buildings to obtain reliable estimates of flood risk in downtown segments. The model suggests that urban segments with high expected damage per length may warrant seawalls. In this framework, the preferred seawalls should be built no lower than the high tide line. The model suggests that these seawalls should generally be 1–2 m above ground, depending on the site. This is far lower than the 3.7 m wall height needed for walls to be considered “safe” from a hundred-year storm. However, using walls to protect against rare storms is not generally cost-effective. Further, the optimal seawall may also be more compatible with existing urban uses and therefore more politically feasible. The only city in our sample with a seawall close to the model’s estimated optimum is Charleston, as shown in Fig. 7. Charleston funded this seawall with tourism taxes. Fig. 7Seawall in Charleston, SC. Note: This is a picture of the rebuilding of the Charleston seawall between 2019 and 2026. Charleston is the only city in the study with a seawall close to the model’s estimated optimum. Photo by Charleston-SC.gov. Seawall in Charleston, SC. Note: This is a picture of the rebuilding of the Charleston seawall between 2019 and 2026. Charleston is the only city in the study with a seawall close to the model’s estimated optimum. Photo by Charleston-SC.gov. By minimizing the sum of construction cost, maintenance cost, and expected residual damage, in a model with both storms and SLR, CRESS offers a practical city-scale tool for coastal flood defense. However, the analysis in this paper can be improved further. First, the model does not fully capture all the properties that are at risk of flooding. The model still relies on an area-wide definition of properties at risk. Future studies might be able to do much better with new data sets that include all buildings. Second, CRESS does not yet capture details of every property such as the type of building or the number of floors of the building. Third, there may be different flood vulnerabilities for each type of building. Capturing this additional detail may lead to much more accurate flood damage estimates. Fourth, most studies do not measure the value of buildings and their contents. Real estate data often just reports the total value of a property but not the building. The value of the building as a fraction of the property value can vary widely and can be lower in proximity to water. Contents can vary by building type and value as well. Fifth, flood damage may depend on more than just flood depth. Other factors such as water velocity and waves can play a role. Addressing these issues would help increase the accuracy of the damage estimates. The analysis suggests that, under the insurance assumptions used here, insurance may be a less costly way to address risk aversion than the construction of tall walls. However, fair insurance is not a perfect tool. Many people may choose not to purchase actuarially fair insurance. They may not be risk averse or they may be misinformed. The analysis of the seawall treats these people as though they value the risk at its expected value. This may well be appropriate for a social decision such as building a seawall. But there remains a concern about what happens when a flood overtops a seawall. Whether someone has purchased insurance or not matters a great deal to the consequences. Rather than replying on individuals to purchase flood insurance, local governments could do so instead. This would cover the damage from catastrophic storms that overtop optimal walls. Another question that this analysis does not answer is who will pay for seawalls? One can ask the private landowners who benefit from each seawall to pay a share, however the costs of floods extend to the local community, so local governments should probably pay a share as well. Finally, state and federal government might also want to contribute to flood protection given that it enhances the quality of life to people in their constituencies. The analysis does not examine all possible strategies to reduce coastal flooding. The study does not examine storm gates, although they are promising tools in select places23. Buying and destroying vulnerable buildings that are not worth insuring, raising buildings on stilts, beach nourishment, and ecosystem restoration are alternative strategies that should be explored24–26. It is possible that some of these alternative strategies might be more effective than seawalls27. But many of these alternative strategies would be particularly useful in places that cannot be protected by seawalls28. Furthermore, these alternative strategies may be useful to address other coastal issues besides flooding, such as erosion and saltwater intrusion, that seawalls are not intended to address. The results suggest that the United States may wish to reconsider some existing policies for coastal flood protection. Federal agencies, such as the U.S. Corps of Engineers and the Federal Emergency Management Agency, currently recommend only “safe” seawalls that are at least 3.7 m high. These walls can stop the storm surge of a 100-year storm (major hurricane) but they are expensive. In our calibration, the total cost of these “safe” walls is substantially higher than that of optimal-sized walls even when including residual damage. The United States does not generally require such high walls in these settings. Instead, more widespread protection from carefully targeted modest walls may be more cost-effective. Another important issue concerns flood insurance. There have recently been significant improvements in the design of flood insurance in the United States. Congress has moved federal flood insurance from a subsidy program towards property-specific risk-based “fair insurance”. By charging premiums equal to expected damage, fair insurance premiums provide an important signal to the real estate market concerning how “safe” a property actually is. Of course, if a neighborhood builds a modest seawall, fair insurance should lower premiums to reflect the new residual damage. Aligning flood insurance premiums with residual risk would encourage local communities to build modest seawalls. This study examines how coastal resilience might adapt as sea level rise continues. Depending on how SLR is observed to evolve, the optimal strategy should change over time. Possible responses to climate change include migration29, changes in land use30,31, and economic development32. The relationship between flood risk and development adds complexity and uncertainty to the analysis9,33,34, which deserves careful study. Finally, SLR will become increasingly important over time. Coastal planning for the distant future must take all these factors into account20,35. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1