Sea walls make economic sense on only 2.8% of the world’s coastline. Sea-level, subsidence and insurance data on a retreat already under way, unnamed.
The Retreat Already Under Way
The seas are rising faster, and most of the world’s coastline will never be walled
Global mean sea level is now rising by about 4.1 mm a year, roughly double the rate of the early 1990s ✓ Established Fact [1]. A 2025 global study finds that building coastal protection is economically justified along only 2.8% of the world’s shoreline [2]. Along the rest, the coast will be abandoned, either by plan or by default.
Coastal retreat is not a choice that governments will make some day in the future. It is already happening, in pieces, under other names. Between 1989 and 2017 the US Federal Emergency Management Agency funded 43,633 buyouts of flood-prone homes across 49 states, which were then demolished and returned to open land ✓ Established Fact [3]. On North Carolina’s Outer Banks, 27 privately owned houses in Rodanthe and Buxton collapsed into the Atlantic between May 2020 and October 2025 [4]. In June 2024, 300 Guna families, more than 1,000 people, left the island of Gardi Sugdub for new houses on the Panamanian mainland [5]. Fiji has already moved six communities and says that up to 676 may eventually need to follow [6]. The withdrawal is real. What is missing is the word for it. Governments announce seawalls in public and buy out houses quietly, one parcel at a time. Planning documents refer to phased adaptation or resilient relocation, and in 2018 one California town voted to forbid its future councils from planning for retreat at all [7] [8]. This report measures the gap between what the data require and what politics will say out loud.
The physical driver is accelerating. In 2024 global sea level rose by 0.59 cm, well above the 0.43 cm that NASA scientists had expected, and the annual rate has more than doubled since satellite measurements began in 1993 ✓ Established Fact [9]. Since that year, global sea level has risen by about 10 cm. The World Meteorological Organization’s report on 2025 puts the current rate at about 4.1 mm a year, against about 2.1 mm in the early satellite era [1]. In 2024 most of the increase came from ocean warming rather than melting ice, an unusual reversal driven by a record-hot year [9]. As NASA sea-level researcher Josh Willis put it, the rise seen in 2024 was higher than expected [9]. The acceleration matters more than any single year. A doubling of the rate in three decades means that infrastructure designed for the sea level of the 1990s is already working against a moving target, and that the gap between design standards and conditions on the ground widens every year that standards are not revised.
For the United States, the reference projection is the 2022 interagency technical report led by NOAA with NASA, the EPA, USGS, FEMA and the Army Corps of Engineers. It projects 0.25 to 0.30 m of rise along the US coastline by 2050, as much as the entire previous century, and moderate flooding more than ten times as often as today ✓ Established Fact [10]. Without deep emissions cuts, a further 0.5 to 1.5 m is projected by 2100 [10]. At global scale, the IPCC’s Sixth Assessment Report gives a likely range of 0.63 to 1.01 m by 2100 under its very high emissions pathway, and states that rise approaching 2 m by 2100 and 5 m by 2150 cannot be ruled out ◈ Strong Evidence [11]. These ranges are not exotic. They are the planning inputs that coastal engineers, insurers and mortgage lenders are now asked to work with, and they are the reason that a growing number of public agencies publish high-end scenarios alongside central estimates.
The number of people exposed depends heavily on which elevation data are used, and the best available data keep raising the estimate. Kulp and Strauss, using an improved elevation model, found in 2019 that 110 million people already live below the current high-tide line, protected by defences or natural barriers, and that 190 million live on land projected to fall below the high-tide line by 2100 even under low emissions ◈ Strong Evidence [12]. Under high emissions, up to 630 million people live below projected annual flood levels for 2100 [12]. A separate assessment of the low-elevation coastal zone, land less than 10 m above sea level, put its 2015 population between 750 million and nearly 1.1 billion people, with a core estimate of 815 million, roughly 10% of humanity [13]. Research published in 2025 counts about 230 million people within 1 m of current sea level [14]. The range itself is a finding. When the best estimates differ by hundreds of millions of people, the uncertainty is not about whether exposure is large, but about how large.
Flooding is arriving before the sea reaches any dramatic threshold. NOAA’s annual outlook predicts a national median of 7 to 12 high-tide flood days in 2026 and 2027, a new record and more than 250% above the level of 2000, rising to 55 to 85 days a year by 2050 ✓ Established Fact [15]. High-tide flooding, often called nuisance flooding, closes roads, corrodes cars and overwhelms storm drains on sunny days, and it is the form in which most coastal residents first meet sea-level rise. In Guam, Apra Harbor recorded 77 flood days in the 2025 and 2026 season [15]. For Pacific atolls the trajectory is now fixed. A NASA analysis found that Tuvalu, Kiribati, Fiji, Nauru and Niue will see at least 15 cm of rise within 30 years regardless of future emissions, and that parts of Tuvalu with fewer than five flood days a year will see about 25 by the 2050s [16]. By 2050 daily tides are projected to submerge half of Funafuti, home to 60% of Tuvalu’s residents [17].
A NASA analysis published in September 2024 found that the Pacific island states will see at least 15 cm of additional rise within 30 years irrespective of emissions [16]. The International Thwaites Glacier Collaboration reported in 2024 that the glacier’s retreat has been under way for more than 80 years and that its full collapse would add 65 cm, with the West Antarctic Ice Sheet holding 3.3 m [18]. Emissions cuts change the pace and the eventual height, not the direction.
Set against this physical trajectory, the economic arithmetic of defence is stark. A 2025 global analysis by Ballesteros, Lincke, Nicholls and colleagues, using the IPCC’s latest scenarios, finds that protection is economically optimal along only 2.8% of the world’s coastline, yet that this thin slice holds 78% of the coastal population and 91% of coastal assets ◈ Strong Evidence [2]. An earlier analysis by Lincke and Hinkel reached a near-identical result, about 3% of coastline covering about 75% of coastal residents [19]. The implication is rarely stated. For about 97% of the world’s shoreline, the economically rational response to rising seas is not a wall but accommodation or withdrawal. Where seawalls make sense, they make a great deal of sense, around dense cities, ports and valuable districts. Everywhere else, the question is not whether people and property will move back from the water, but whether that movement will be planned, financed and fair, or chaotic, delayed and paid for by whoever happens to own the last house on the shore.
The Land Is Sinking Too
Subsidence is the half of the problem that governments can actually control
Coastal residents experience relative sea-level rise of 7.8 to 9.9 mm a year on average, up to four times the global-mean rate, because so much of the inhabited coast is sinking ◈ Strong Evidence [20]. In many delta cities the ground falls faster than the ocean rises.
Global averages hide the variable that matters most to people living on the coast: the movement of the land beneath them. A 2021 analysis led by Robert Nicholls of the Tyndall Centre found that coastal residents experienced average relative sea-level rise of 7.8 to 9.9 mm a year over two decades, compared with a global-mean relative rate of about 2.5 mm, because about 58% of the world’s coastal population lives on subsiding deltas ◈ Strong Evidence [20]. Less than 1% lives where the land is rising. Relative sea-level rise, the combination of ocean rise and land movement, is the quantity that floods streets, salinates wells and overtops defences. In January 2026, a study in Nature by Leonard Ohenhen and colleagues examined 40 major river deltas and found that land subsidence exceeds local sea-level rise in 18 of them, putting more than 236 million people at heightened near-term flood risk ✓ Established Fact [21]. As Nicholls observed of that work, from a risk point of view it does not really matter whether the sea level rises or the land sinks [21].
The causes of subsidence are mostly human. In the 2026 delta study, groundwater extraction was the main driver in 35% of deltas, with sediment starvation from upstream dams and the weight of urban development also contributing [21]. Parts of the Mekong, Ganges-Brahmaputra, Chao Phraya and Yellow River deltas are sinking by more than 1 cm a year [21]. The Mekong illustrates how badly the hazard was underestimated. In 2019, Minderhoud and colleagues showed that the delta’s mean elevation is about 0.8 m above sea level, not the 2.6 m assumed in earlier impact assessments, an error larger than a century of projected ocean rise ✓ Established Fact [22]. A delta home to about 20 million people turned out to be far closer to the water than its planners had believed. Bangladesh, where 41% of the population lives below 10 m, faces a modelled 1.3 million displaced people by 2050 from sea-level rise alone [23].
China provides the largest data set. A 2024 study in Science covering 82 cities found that 45% of urban land is subsiding and 16% is sinking by more than 10 mm a year, affecting about 270 million urban residents ✓ Established Fact [24]. Beijing recorded subsidence of up to 45 mm a year near major infrastructure, and Tianjin was named among the national hotspots [24]. The authors projected that the share of China’s urban land lying below sea level could triple by 2120, affecting 55 to 128 million people [24]. In the United States, a 2025 study in Nature Cities found that all 28 of the country’s most populous cities are sinking, with about 34 million people living on subsiding ground and about 80% of the motion attributable to groundwater extraction [25]. Houston is the extreme case, with 40% of its area sinking faster than 5 mm a year and local zones dropping 5 cm a year [25]. On the East Coast, more than half of the infrastructure in New York, Baltimore and Norfolk sits on land sinking by 1 to 2 mm a year [26].
- Jakarta: average subsidence of 3.5 cm a year; parts of north Jakarta have dropped up to 4 m in four decades and about 40% of north Jakarta is now below sea level [27]
- Beijing and Tianjin: local rates up to 45 mm a year near infrastructure, with 16% of the urban land of 82 Chinese cities sinking faster than 10 mm a year [24]
- Houston: 40% of the city sinking faster than 5 mm a year and local zones dropping 5 cm a year [25]
- Mekong delta: mean elevation about 0.8 m above sea level, with parts of the delta sinking by more than 1 cm a year [22] [21]
- Venice: the MOSE barrier was raised 25 times in 2023 and 28 times in 2024 as relative sea level in the lagoon keeps climbing [28]
Jakarta is the case where subsidence became a national strategy. The city has sunk by an average of 3.5 cm a year, north Jakarta has dropped by up to 4 m in four decades, and about 40% of the north now lies below sea level ✓ Established Fact [27]. Flood losses are estimated at US$186 million a year, rising to US$421 million by 2030 without intervention [27]. Indonesia’s response has been two of the largest coastal decisions anywhere. In 2019 it announced that the national capital would move to Nusantara in East Kalimantan, citing Jakarta’s sinking among its reasons [29]. In June 2025 President Prabowo Subianto recommitted to a Giant Sea Wall of about 500 km along the north coast of Java, with a total cost of about US$80 billion and a first Jakarta Bay phase of US$8 to 10 billion over 8 to 10 years [30]. Jakarta’s 10 million residents are not moving. The capital’s institutions are, slowly, while the city behind them prepares to be defended.
Japan shows that subsidence can be stopped. Tokyo first identified land sinking in Koto Ward in the 1910s, driven by groundwater pumping and natural gas extraction [31]. The national government responded with the Industrial Water Law of 1956 and the Building Water Law of 1962, which restricted pumping and supplied alternative surface water, and by the 1970s the sinking had largely halted ✓ Established Fact [31] [32]. Nicholls credits the outcome directly: the authorities provided a good alternative water supply and passed laws to stop groundwater use [21]. Japanese engineers are now applying that experience in Jakarta through JICA-supported projects [27]. The lesson has a second half, however. Stopping subsidence does not raise the land that has already sunk. Tokyo’s zero-metre zones cover about 124 km², about 20% of the 23 special wards, and are home to about 1.5 million people who depend permanently on about 60 km of continuous seawall and 15 floodgates [32]. Halting subsidence turned Tokyo into a city that must be defended for as long as it exists.
Ocean rise is global and fixed for centuries. Subsidence is local, mostly caused by groundwater pumping, and can be stopped by law within a generation, as Tokyo showed. Attributing every flood to climate change lets local water politics escape scrutiny, while blaming only local pumping understates the global trend. Both halves need separate policies, and only one of them is fully under the control of a city government.
The two halves of the problem therefore call for different decisions. Ocean rise is a global trend that individual governments cannot change and must plan around. Subsidence is a local choice that can be reversed in trend through water law, as Tokyo did half a century ago [31] [32]. Where subsidence is ignored, relative rise of several centimetres a year can make retreat unavoidable within decades, as in north Jakarta. Where it is controlled, the reward is not safety but a permanent dependence on defences. Either way, the subsidence data sharpen the central conclusion of this report. The coastlines most exposed to the sea are often those where the land itself is moving, and the populations living there are among the least able to pay for walls. Recognising subsidence as a policy failure rather than a natural hazard is the first step towards an honest discussion of which coasts can be held.
The Arithmetic of Defence
Where walls pay for themselves, and where they never will
Global models find that dike protection would cost US$12 to 71 billion a year by 2100, far less than the flood damage it would avoid ✓ Established Fact [33]. But those savings are concentrated on a thin, dense slice of coast, and the megaprojects designed to deliver them keep growing in cost.
The economic case for defence is real and should be stated plainly. In 2014, Hinkel and colleagues estimated in PNAS that without adaptation, 0.2 to 4.6% of the global population would be flooded every year by 2100 under 25 to 123 cm of sea-level rise, with expected annual losses of 0.3 to 9.3% of global GDP ✓ Established Fact [33]. Investment in and maintenance of dikes would cost US$12 to 71 billion a year by 2100, much less than the damages avoided [33]. On these numbers, building defences around cities is one of the best investments available to coastal governments. The Netherlands is the standing proof. After the North Sea flood of 1953 killed 1,836 people in the Netherlands and 307 in England, the Dutch built the Delta Works and turned the country, about a quarter of which lies below sea level, into the most defended coastline on Earth [34] [35]. Rotterdam, Tokyo and London are viable at their current sea levels because they were walled.
The same global models, read in full, show where the logic ends. Lincke and Hinkel found that protection is economically robust along about 3% of the world’s coastline, covering about 75% of the global coastal population [19]. The 2025 update by Ballesteros and colleagues, using IPCC AR6 scenarios, puts the figure at 2.8% of coastline, safeguarding 78% of coastal people and 91% of coastal assets ◈ Strong Evidence [2]. Under the same analysis, between 4 and 72 million people could migrate away from the coast over the century, and net land loss could range from 2,800 to 490,000 km², with India and Vietnam facing the largest absolute migration [2]. The models do not say that the rest of the coast is worthless. They say that along most of it, the value protected would not repay the cost of protecting it, so that money spent on walls there would be better spent moving people and assets, raising buildings, or accepting periodic flooding.
Two independent analyses using different sea-level scenarios reach the same order of magnitude: about 3% of coastline protecting about 75% of coastal residents [19], and 2.8% of coastline protecting 78% of residents and 91% of assets [2]. The corollary is that along roughly 97% of global shoreline, withdrawal or accommodation is the cheaper long-run strategy.
Where defence is attempted, the price tags have become some of the largest civil-engineering commitments in history. The US Army Corps of Engineers’ 2022 draft plan for the New York and New Jersey harbour proposed 12 storm-surge barriers and more than 41 miles of shoreline walls at a cost of US$52.6 billion, with 14 years of construction starting in 2030 [36]. The Coastal Texas Project, often called the Ike Dike, was authorised by Congress in the Water Resources Development Act of 2022, but its cost estimate has risen from US$34 billion to as much as US$57 billion once inflation over a 20-year build is included, and no construction money has been appropriated ✓ Established Fact [37]. Indonesia’s Java sea wall is priced at about US$80 billion over 15 to 20 years [30]. Each of these figures is a promise to protect particular places, and by implication a decision about which places will not be protected.
The projects that have been built show how defence costs grow after the ribbon is cut. Venice’s MOSE barrier, with 78 mobile gates, became operational in 2020 and reached its 100th raising in March 2025; Italy’s transport ministry says it has prevented more than €2.6 billion in damage [38]. But it cost about €7 billion against an original estimate of about €2 billion [39], each raising costs about €200,000, and it was raised 25 times in 2023 and 28 times in 2024 [28]. As sea level rises, the gates must close more often, isolating the lagoon from the sea for longer. England’s Thames Estuary 2100 plan now costs £16.2 billion for 2023 to 2100, including a replacement for the Thames Barrier by 2070, an increase of about 50% since 2012 ✓ Established Fact [40]. In the Netherlands, the Delta Programme’s budget of about €29 billion for 2015 to 2050 falls about €9 billion short of the expected cost of about €38 billion, driven by the need to strengthen dikes [41]. Even the richest defenders are running behind the water.
Defence projects also fail politically, especially when they make visible which neighbourhoods are inside and which are outside the line. In June 2020 the Army Corps released a US$4.6 billion draft plan for Miami-Dade County that included about 6 miles of walls, three surge barriers and the elevation of thousands of homes [42]. Residents, environmental groups, business leaders and the county itself rejected it, with critics calling it a US$5 billion sticking plaster and the county arguing that the Corps had ignored the region’s hydrology, and the study was reworked towards elevation and floodproofing [42]. Charleston’s proposal for a US$1.1 billion wall of about 8 miles around the historic peninsula, with the city liable for about US$385 million, raised the same question in a different form: whether a wall around the most valuable district is a plan for the city or only for its core [43]. Benefit-cost analysis is the hidden engine of these choices, and it rewards property value.
Federal benefit-cost rules measure the value of what a defence protects, mostly as property value. Expensive districts therefore qualify for walls, while cheaper neighbourhoods on the same coast qualify for buyouts or nothing. Mach and colleagues found that within the richer counties that run buyouts, the homes bought out sit in poorer and more diverse neighbourhoods. Defence and retreat are being allocated by wealth before anyone uses either word.
The arithmetic of defence therefore produces a map, not a single answer. Dense, wealthy, economically central coasts will be held, at rising cost and with increasing dependence on barriers that close more often each decade. Mach and colleagues found that within the counties that run buyout programmes, which tend to be richer and denser, the homes bought out were concentrated in poorer and more racially diverse neighbourhoods [3]. Rural coasts, small towns, Indigenous communities and lower-income suburbs are unlikely to meet the benefit-cost threshold for major defences, and will face a different set of options: elevation, floodproofing, managed relocation, or decline. The thin slice of coast that will be defended is not a natural fact. It is the result of valuation rules that weight property over people, and of budgets that are already falling short where defences exist. Recognising that the economically rational response for most of the coastline is withdrawal does not settle how withdrawal should happen. It does make clear that pretending every coast can be held is itself a policy choice, with predictable winners and losers.
The Dominoes
How coastal risk moves through prices, insurance, mortgages and local tax bases
US homes exposed to projected sea-level rise sell for about 7% less than comparable unexposed homes ✓ Established Fact [44], yet researchers estimate that the US housing market is still overvalued by US$121 to 237 billion because flood risk is not fully priced ⚖ Contested [45].
Before the sea takes a house, the market starts to discount it, unevenly and with a delay. Bernstein, Gustafson and Lewis found in the Journal of Financial Economics that US homes exposed to projected sea-level rise sell for about 7% less than equivalent unexposed homes the same distance from the beach, and that a 4% discount applies even to homes not projected to flood for about a century ✓ Established Fact [44]. The discount has grown over time and is driven by sophisticated buyers and by communities most concerned about climate change [44]. In Florida, Keys and Mulder found that the market adjusted first through volume rather than price: sales in the 187 most exposed coastal census tracts fell by about 16,500 transactions between 2013 and 2018, and from 2018 prices in exposed tracts fell 5 to 10% relative to less exposed areas [46]. In Miami-Dade, Keenan and colleagues found that since 2000 homes at higher elevations have appreciated slightly faster than low-lying ones, a pattern consistent with what they call climate gentrification [47].
Other researchers argue that the adjustment has barely begun. A 2023 study in Nature Climate Change by Gourevitch and colleagues estimated that the US housing market is overvalued by US$121 to 237 billion because flood risk is not reflected in prices, across 14.6 million properties with at least a 1% annual chance of flooding ⚖ Contested [45]. Florida alone accounted for about US$50 billion of the overvaluation, 69% of the overvalued dollars lay outside FEMA’s official flood zones, and low-income households could lose about 10% of their home equity if prices corrected [45]. Penny Liao of Resources for the Future warned that the result is a problem for cities and towns that could struggle financially if property values fall [45]. The two findings are compatible. Exposed homes already trade at a discount, but the discount is smaller than the expected losses, especially where buyers are uninformed and where flood maps understate risk. A partial price signal is the most dangerous kind, because it reassures owners while leaving the largest correction ahead of them.
- Insurers reprice or withdraw: in Louisiana more than 20 insurers withdrew or went insolvent within two years after four hurricanes in 2020 and 2021 caused about US$23 billion in insured losses [48]
- Public insurers of last resort absorb the risk: Florida’s Citizens Property Insurance peaked at about 1.4 million policies in 2023 before private carriers took most of them back by late 2025 [49]
- Federal flood insurance runs a deficit: the National Flood Insurance Program owes US$22.525 billion to the Treasury [50]
- Lenders shift risk to taxpayers: after billion-dollar disasters, lenders sell more conforming mortgages to Fannie Mae and Freddie Mac [51]
- Prices and local tax bases adjust: municipalities that depend on property tax face shortfalls if coastal values correct [45]
Insurance is the channel through which future risk becomes a current cost. In the United States, flood damage to homes is excluded from standard homeowners policies and covered mainly by the federal National Flood Insurance Program. The program owed the Treasury US$22.525 billion in March 2026, with US$7.9 billion of remaining borrowing authority ✓ Established Fact [50]. Since 2005 it has repaid only US$2.82 billion of principal while paying US$6.17 billion in interest, currently about US$619 million a year, and its authorisation was due to expire on 30 September 2026 [50]. The cost is concentrated in a small set of properties that flood again and again. About 45,000 properties met FEMA’s criteria for severe repetitive loss between 1968 and 2022 [52]. Each of these represents a public decision, repeated after every flood, to pay for rebuilding in the same place. As OsakaWire’s earlier investigation of the insurance retreat documented, private insurers declined to renew 1.9 million US homeowners policies between 2018 and 2023, and coastal and flood risk is a large part of that withdrawal [53].
According to the Congressional Research Service, NFIP debt stood at US$22.525 billion in March 2026. Since 2005 the program has repaid US$2.82 billion of principal and paid US$6.17 billion in interest, about US$1.7 million a day [50]. The debt is a measure of how long federal policy has subsidised the occupation of flood-prone land rather than withdrawal from it.
FEMA’s attempt to price flood risk honestly shows what happens when the signal reaches households. Risk Rating 2.0, introduced from October 2021, set premiums according to the flood risk of each property, with increases capped at 18% a year for primary residences. A study published in December 2025 linked it to an 11 to 39% decline in new NFIP policies and a 5 to 13% decline in existing ones, depending on the size of the premium increase, with the lowest-income zip codes most likely to drop coverage ✓ Established Fact [54]. The NFIP has fallen from 5.7 million policies at its 2009 peak to fewer than 4.7 million, and only about 4% of US homeowners carry flood insurance [54]. Jesse Gourevitch of the Environmental Defense Fund, a co-author, said that rising premiums are driving many households, especially lower-income ones, to forgo coverage [54]. Accurate pricing is a precondition of planned retreat, because it tells owners what staying will cost. But on its own it does not move anyone.
When insurance becomes unaffordable, households do not leave. They drop coverage and stay, carrying uninsured risk on what is often their largest asset. The first phase of coastal retreat is therefore financial, not physical: the withdrawal of insurers, lenders and buyers from places where people still live. Without buyout funding to match, risk-based pricing converts a slow physical hazard into a sudden wealth loss for those least able to absorb it.
The mortgage system adds a further transfer. Ouazad and Kahn found that after billion-dollar disasters, lenders become more likely to originate loans that meet the conforming limits of Fannie Mae and Freddie Mac and to sell them on, moving climate risk from private balance sheets to government-sponsored enterprises and ultimately to taxpayers [51]. In England, the government-backed reinsurer Flood Re has kept household flood cover affordable since 2016, but it is designed to end in 2039, on the assumption that the market will by then be able to price risk directly; about 8 million properties in England, one in four, could face some flood risk by 2050 [55]. The local tax base is the last domino. Climate Central’s analysis found that more than 648,000 US land parcels will lie at least partly below the tidal boundary by 2050, with more than 130,000 in Florida and 106,000 in Louisiana [56]. New Jersey, where about 40,000 properties face permanent coastal flooding by 2050, began requiring sellers and landlords to disclose flood history and risk in March 2024 [57]. Disclosure accelerates price discovery, which is its purpose, and it brings the fiscal reckoning for coastal municipalities forward.
Taken together, the dominoes describe a retreat that has started in balance sheets long before it shows up in moving vans. Insurers withdraw first, public insurers absorb the risk until political pressure forces them to shed it, flood insurance is repriced until the households least able to pay drop out, lenders pass risk to public agencies, and prices adjust in fits and starts as information spreads. Each step is rational for the institution taking it. None of them is designed to move people out of harm’s way, and several of them make it harder to do so, because a household whose home has lost value has less money with which to buy a safer one. A coherent retreat policy would sequence these steps the other way round: first a credible offer to buy or relocate, then honest pricing and disclosure, and finally the withdrawal of subsidies for rebuilding in place. At present the sequence runs in reverse, and its costs fall on the households with the least capacity to bear them.
The Places Already Leaving
From Louisiana to Funafuti, relocation is happening without a playbook
After the 2011 tsunami, Japan completed 8,373 housing plots in 324 relocation districts, 99.8% of the plan ✓ Established Fact [58]. It remains the clearest proof that collective relocation can be done at scale. Most climate relocations elsewhere involve dozens of families, years of delay and budgets that keep growing.
Isle de Jean Charles, in Louisiana’s Terrebonne Parish, became the symbol of US climate relocation because the land under its residents disappeared in a single lifetime. The island shrank from more than 22,000 acres to about 320 acres, and in 2016 the Department of Housing and Urban Development awarded Louisiana US$92 million under its National Disaster Resilience competition, US$48 million of it for resettlement ✓ Established Fact [59]. The state bought a 515-acre site near Schriever, about 40 miles north, for US$11.7 million, residents began moving in during August 2022, and 37 homes had been completed by the autumn of 2024, with nine more planned [59]. An investigation published in March 2025 counted about 36 households at the new site, called New Isle, mostly members of the Jean Charles Choctaw Nation [60]. Reporting on the project has documented disputes between tribal leaders and the state over eligibility and control of the design [60]. It moved a community, but it also showed that a relocation designed by agencies rather than residents can leave lasting disputes about who it was for.
In Alaska, the village of Newtok spent two decades leaving a site that climate change was making uninhabitable. Congress approved a land exchange in 2003 so that the village could move to Mertarvik; the move began in 2019 and the last residents left Newtok at the end of 2024 [61]. Nearly 300 residents now live in 46 new homes, at a cost expected to exceed US$150 million drawn from grants by at least seven federal agencies [61]. The money did not buy a functioning town. More than US$6 million of road funding was spent without installing water pipes, and no house in Mertarvik has a flush toilet or a shower; one building inspector called the housing some of the worst new construction the inspector had ever seen [61]. A new school is due in 2026, with construction of about US$56 million and a total project cost of more than US$81 million [62]. Mertarvik demonstrates that relocation without a lead agency, a single budget and a construction standard becomes slower and more expensive than anyone planned.
Panama’s relocation of Gardi Sugdub is a counter-example of speed. Over three days in June 2024, 300 Guna families, more than 1,000 people, moved from the crowded island in the Guna Yala archipelago to a new settlement on the mainland with housing for 1,350 people ✓ Established Fact [5]. The housing ministry had committed to the project in 2017, and the new town includes more than 300 two-bedroom homes, a school for 650 students and a ceremonial house [63]. About 200 people chose to stay on the island [5]. Overcrowding as well as sea-level rise drove the move, a reminder that relocation decisions are rarely about a single hazard [63]. President Laurentino Cortizo said that the island’s population had experienced great concern because of rising seas [5]. The Guna case shows what makes relocation work: a community that chose to move, a government that built first, and a destination close enough to preserve ties to the old place.
In the Pacific, relocation has become an explicit part of national planning. Fiji has already relocated six communities, has 17 on a priority list, and says that up to 676 communities may eventually need to move, in a country where 76% of the population lives within 5 km of the coast ✓ Established Fact [6]. It created a Climate Relocation of Communities Trust Fund in 2019 and a Climate Change Act in 2021 [6]. David Lund, Fiji’s special adviser on climate action, described the challenge directly in August 2025 [6]. Tuvalu, whose land averages about 2 m above sea level, has taken the next step, from internal relocation to international mobility [17]. Under the Falepili Union treaty signed with Australia in 2023, 280 Tuvaluans a year can win permanent residence by ballot. The first ballot, in mid-2025, drew 8,750 entries from a resident population of 10,643, although family duplicates may inflate the count [64]. The two governments describe the arrangement as mobility with dignity [65]. Kiribati bought about 6,000 acres on Vanua Levu in Fiji in 2014 as a hedge [66].
It’s not adaptation in the traditional sense, it’s about leaving your traditional land and livelihoods behind.
— David Lund, Special Adviser on Climate Action, Government of Fiji, August 2025Europe’s cases are smaller but legally revealing. In the French overseas territory of Saint-Pierre-et-Miquelon, the village of Miquelon, about 600 people living about 2 m above sea level, is being rebuilt about 1 km away at an elevation of 10 m or more, a decision made firm after Storm Fiona in 2022 [67]. In Soulac-sur-Mer, the Signal apartment building, built in 1967 about 200 m from the sea, stood less than 20 m from the water by 2023; its 75 co-owners were evacuated in 2014, compensated at 70% of value after years of litigation, and the building was demolished in February 2023 ✓ Established Fact [68]. In Wales, the 2014 shoreline management plan concluded that the defences of Fairbourne would not be maintained indefinitely and that the village could become uninhabitable from 2054, but in 2022 Gwynedd Council said that it had no formal plan to decommission it [69]. Fairbourne is the clearest example of a retreat announced in a technical document and then denied in public.
Japan’s experience after the Great East Japan Earthquake remains the largest planned coastal relocation in a wealthy democracy. Under the collective relocation programme for disaster prevention, 324 districts in 27 municipalities planned 8,389 household plots on higher ground; by 2020, 8,373 plots, 99.8% of the plan, had been completed, including 5,638 in Miyagi Prefecture alone ✓ Established Fact [58]. The programme was triggered by catastrophe rather than projection, funded centrally, and implemented through municipalities that kept communities together. Elsewhere, relocation remains piecemeal. In Washington State, the Quinault Indian Nation has spent more than a decade trying to move its Lower Village at Taholah uphill, building its first new structure on an US$8 million loan after two federal grants were refused in 2023 [70]. In Saint-Louis, Senegal, a World Bank-funded project moved families from the Langue de Barbarie to Diougop [71], while Human Rights Watch found other displaced families left out of the relocation for nine years [72]. Indonesia’s capital move, partly a response to Jakarta’s sinking, has seen its 2026 budget cut to about Rp6.3 trillion from Rp43.4 trillion in 2024 [29].
Set side by side, these cases reveal a consistent pattern. Relocation succeeds where a community chooses to move, where the destination is built before people arrive, where one public body holds the budget and the responsibility, and where the move happens before the final disaster. Japan and Gardi Sugdub met most of these conditions. Isle de Jean Charles and Mertarvik met some of them and paid for the gaps in delay, cost and conflict. Fairbourne met none, and has been left in limbo by a forecast that no authority is willing to act on. The scale of what is coming dwarfs every case so far. Japan’s 8,373 plots and Fiji’s six relocated communities are small compared with the 110 million people who already live below the high-tide line [12] [58] [6]. The precedents show that retreat can be done with dignity. They do not show that it can be done at the pace the data imply without a step change in funding and institutional commitment.
The Policy Machinery
Buyouts, budgets and laws that avoid saying the word
FEMA funded 43,633 buyouts between 1989 and 2017, taking an average of 5.7 years from disaster declaration to closeout ✓ Established Fact [3]. In April 2025 the agency cancelled its main pre-disaster resilience programme, a decision a federal court later ruled unlawful [75].
The main instrument of managed retreat in the United States is the voluntary buyout, in which a public body buys a flood-prone home at pre-flood market value, demolishes it and keeps the land as open space. Mach and colleagues counted 43,633 FEMA-funded buyouts between 1989 and 2017 in 49 states, three territories and 1,148 counties, 94% of them administered by local governments ✓ Established Fact [3]. The programme is large in aggregate and small in each place. The average project size fell from 19 to 7 properties over the period, and it took an average of 5.7 years from a disaster declaration to the closeout of a buyout [3]. An analysis by NRDC found a median of more than five years between a flood and a completed buyout, with fewer than half of projects closed within five years [77]. For a household that has just been flooded, five years is long enough to rebuild, take on new debt or give up and walk away. New Jersey’s Blue Acres programme, one of the most active state efforts, had bought its 700th home about seven years after Hurricane Sandy, none of them on the ocean [78].
The funding behind these programmes proved fragile in 2025. On 4 April 2025, FEMA ended its Building Resilient Infrastructure and Communities programme, known as BRIC, the main federal source of money for projects that reduce risk before disasters, and moved to return about US$882 million of infrastructure-law funding to the Treasury [75]. Twenty states sued in July 2025. A federal judge in Massachusetts, Richard Stearns, issued a preliminary injunction in August, ruled the termination unlawful on 11 December 2025, and on 6 March 2026 ordered FEMA to issue a new funding notice within 21 days; FEMA then published a combined notice for 2024 and 2025 worth US$1 billion ✓ Established Fact [75]. North Carolina alone had lost US$200 million in grants [79]. Its attorney general, Jeff Jackson, said that the state took FEMA to court, won, and then saw the agency defy the court order [79]. For the localities that plan buyouts years in advance, a programme that can be cancelled by memo and restored by injunction is not a reliable foundation.
FEMA ended BRIC on 4 April 2025. After a 20-state lawsuit, Judge Richard Stearns of the District of Massachusetts ruled the termination unlawful on 11 December 2025 and, on 6 March 2026, ordered the agency to reissue a funding notice, which it did with a combined US$1 billion offer [75]. The episode shows that the money for planned retreat depends on political decisions that can reverse within a single year.
England has gone further than most countries in stating its expected losses, while still avoiding the language of retreat. The Environment Agency’s 2024 national assessment found 3,500 properties at risk of coastal erosion by 2055, rising to about 10,100 by 2105, on the assumption that shoreline management plans are funded and delivered ✓ Established Fact [80]. Across all sources of flooding, 6.3 million properties in England are at risk today, and about 8 million, one in four, could be by mid-century [81]. France’s public expertise agency, Cerema, has modelled what erosion alone will cost. With defences maintained, about 5,200 homes and 1,400 business premises worth about €1.2 billion would be lost by 2050; in a scenario of inaction to 2100, about 450,000 homes, 55,000 business premises and €86 billion of property would be exposed [82]. These numbers are precise enough to guide a national relocation policy. Neither country yet has one, and both rely on local authorities that lack the money to buy out more than a handful of properties.
New Zealand illustrates the distance between recognising retreat and legislating for it. Its Climate Change Commission noted in its May 2026 national risk assessment that the country has no legislative framework for managed retreat [83]. The Climate Change Response Amendment Bill introduced on 15 July 2026 requires councils to prepare adaptation plans for priority locations, but provides no mechanism to require, facilitate or fund relocation [83]. The United States has moved through disclosure rather than relocation law. New Jersey requires sellers and landlords to disclose flood history and risk since March 2024, and other states, including Texas, Louisiana and South Carolina, have stronger disclosure rules than most [57]. Disclosure is cheaper than buyouts and politically easier to pass, which is why it spreads. It moves information, not people, and its main effect is to bring forward the price adjustment described in section four.
International law has begun to address the extreme case of sea-level rise: the loss of a state’s territory. In November 2023 the Pacific Islands Forum declared that the statehood and sovereignty of its members will continue notwithstanding the impact of sea-level rise [74]. In July 2025, the International Court of Justice issued a unanimous advisory opinion on states’ climate obligations, finding that once a state is established, the disappearance of one of its constituent elements, such as territory, would not necessarily entail the loss of its statehood ◈ Strong Evidence [76]. These statements matter to atoll states such as Tuvalu and Kiribati, whose maritime zones and seats at the United Nations depend on their legal continuity. In 2021, Tuvalu’s foreign minister Simon Kofe delivered his COP26 address standing knee-deep in seawater to make the point visible [84]. The legal presumption of continuity is a safeguard against the worst outcome. It does not resettle a single household.
| Risk | Severity | Assessment |
|---|---|---|
| Funding reversals | BRIC was cancelled in April 2025 and restored only by court order, Indonesia cut the 2026 Nusantara budget by about 85% and New Zealand’s bill funds no relocation. Retreat requires money over decades; current budgets change within a year. | |
| Insurance-led abandonment | Risk-based pricing has cut new NFIP policies by up to 39% in the hardest-hit areas. Households stay without cover, turning physical risk into uninsured wealth loss rather than movement. | |
| Inequitable allocation | Benefit-cost rules reward property value, so wealthy districts get walls while buyouts concentrate in poorer and more diverse neighbourhoods. | |
| Defence cost escalation | MOSE more than tripled its budget, the Coastal Texas Project rose by about 68% before construction and the Dutch Delta Programme faces a €9 billion gap. | |
| Community fragmentation | About 200 people stayed on Gardi Sugdub, Isle de Jean Charles remains disputed and Mertarvik houses lack plumbing. Poorly designed moves can break the communities they aim to save. |
The common feature of these instruments is that each is designed to avoid an explicit collective decision. Buyouts are voluntary and parcel by parcel. Disclosure laws leave the decision to buyers. Adaptation plans identify priority locations without powers or money to move anyone. Shoreline management plans, as in Fairbourne, predict abandonment but leave councils to deny it. Insurance reforms send price signals but stop short of offering an exit. The result is a policy machinery that can manage a slow trickle of departures but not the scale implied by the data. Japan’s post-tsunami relocation shows the alternative: a national programme with a single funding stream, planned at the level of whole districts and delivered by municipalities. Its limitation is that it followed a catastrophe. The open question for every coastal democracy is whether such a programme can be built before the disaster rather than after it, while the people affected still have homes with value to trade.
The Contested Ground
Four disputes that will decide how retreat happens
The disagreement is no longer about whether seas will rise. It is about who decides which coasts are held, who pays, whether markets already price the risk, and whether the word retreat itself makes the policy impossible ⚖ Contested.
The first dispute is whether defence or retreat is the better investment. Advocates of defence point to the global models, in which dike protection costs far less than the damage it avoids, and to the record of Venice’s MOSE barrier, which officials say has prevented more than €2.6 billion in damage for operating costs of about €20 million ⚖ Contested [33] [38] [28]. They note that the same models find that defending under 3% of the coast protects most people and assets [19] [2]. Critics reply that those same models imply retreat is economically rational along about 97% of the coastline, and that megaprojects systematically exceed their budgets: MOSE cost about €7 billion against an original €2 billion, the Coastal Texas Project rose from US$34 to 57 billion, and the Thames estuary plan rose about 50% in a decade [39] [37] [40]. Both sides are right about different places. The dispute is really about where the line between them is drawn and who draws it.
The second dispute is whether property markets already price coastal risk. The evidence for efficient pricing is substantial: a 7% discount on exposed homes nationally, falling sales volumes and then 5 to 10% relative price declines in exposed Florida tracts, and an elevation premium in Miami-Dade ⚖ Contested [44] [46] [47]. The evidence for a bubble is also substantial: an estimated US$121 to 237 billion of overvaluation, most of it outside FEMA flood zones, with risk transferred to government-sponsored mortgage enterprises after disasters [45] [51]. The most consistent reading is that markets price coastal risk partially, mainly where buyers are sophisticated and information is good, and that subsidised insurance and flood maps that understate risk suppress the rest of the signal. The policy consequence is important. If markets already price most of the risk, buyouts at market value are fair. If they do not, buyouts at current prices transfer public money to owners before the correction.
The third dispute concerns consent: whether planned retreat is voluntary or coercive in practice. Supporters point to the Falepili visas, allocated by ballot without any requirement for a job offer, to Miquelon’s first phase of plots for households that choose to move, and to Gardi Sugdub, where families chose to leave and about 200 people chose to stay ⚖ Contested [64] [67] [5]. Critics point to the Signal building, whose owners were evacuated by order in 2014 and compensated at 70% of value after years of litigation, and to Fairbourne, whose residents learned of a possible decommissioning date from a technical plan [68] [69]. Human Rights Watch found Senegalese families displaced by floods who were left out of the planned relocation altogether [72]. Voluntary often means that a buyout is the only offer on the table, made after the value of staying has already collapsed. The distinction between managed retreat and abandonment depends on whether the offer arrives before or after that collapse.
The fourth dispute is about language itself. In 2018, Del Mar in California removed managed retreat from its sea-level plan and passed a resolution barring future councils from planning for it, after a senior planner reported that even discussing the idea devalued property ⚖ Contested [7]. Planners have tried alternatives such as resilient relocation, graceful retreat and phased adaptation [8]. Amanda Stoltz, a researcher on retreat communication, observed that the word retreat can be seen as synonymous with failure, and that no one really wants to be managed [8]. Others argue that the label matters less than funding and facts on the ground [8]. The California Coastal Commission has been blunter than most governments, stating that there is no realistic scenario for California to adapt to rising seas without some form of removal or relocation [8]. The scientific literature has moved in the same direction: in 2019 Siders, Hino and Mach argued in Science that retreat should be treated as a strategic set of options rather than a last resort [73].
The Case for Holding the Line
Global models find that defending about 3% of coastline protects about 75 to 78% of coastal residents and 91% of coastal assets [19] [2].
Dike protection would cost US$12 to 71 billion a year by 2100, far less than losses of up to 9.3% of global GDP without adaptation [33].
MOSE has been raised more than 100 times and is credited with preventing more than €2.6 billion in damage [38].
The Case for Planned Retreat
Along about 97% of global shoreline, protection does not repay its cost [2].
MOSE rose from about €2 billion to €7 billion, and the Coastal Texas Project from US$34 to 57 billion before construction [39] [37].
Every wall commits future budgets to maintenance and replacement as seas rise; the Thames Barrier must be replaced by 2070 [40].
Japan completed 99.8% of its post-tsunami relocation plots, and Gardi Sugdub moved 300 families in three days [58] [5].
Without buyouts, rising premiums push lower-income households to drop flood insurance and stay in place [54].
A fifth, more technical dispute sits beneath the others: how fast and how far the seas will rise. Glaciologists including Chris Stokes of Durham University argued in 2025 that even 1.5°C of warming is too high for the polar ice sheets, and that sea-level rise is likely to accelerate to rates that are very difficult to adapt to, possibly about 1 cm a year by 2100 ⚖ Contested [14]. The IPCC states that rise approaching 2 m by 2100 cannot be ruled out under very high emissions [11]. On the other side, the International Thwaites Glacier Collaboration concluded in 2024 that the most extreme ice-cliff collapse scenarios are less likely this century than feared, pushing the risk of a West Antarctic collapse towards the 23rd century [18]. Estimates of exposed population also vary widely with the data used, from 750 million to 1.1 billion in the low-elevation coastal zone [13] [12]. For planning, the disagreement about the tail of the distribution matters less than its implication: any coastal decision with a lifetime beyond 2100 must be robust to outcomes that cannot be excluded.
What the Evidence Tells Us
Retreat is a decision being made by default
The evidence points in one direction ◈ Strong Evidence. For most of the world’s coastline, the choice is not between defence and retreat. It is between a retreat that is planned, financed and fair, and one that happens anyway, through insurance withdrawals, collapsing prices and the next storm.
Five findings stand out from the data assembled in this report. First, sea-level rise is accelerating and locked in for centuries: the rate has doubled since the early 1990s, and Pacific islands will see at least 15 cm more within 30 years whatever happens to emissions ✓ Established Fact [1] [16]. Second, in much of the inhabited coast the land is sinking faster than the sea is rising, mostly because of groundwater extraction, which makes subsidence the part of the problem that local governments can control [20] [21]. Third, defence is economically justified along less than 3% of the world’s coastline, although that slice holds most coastal people and assets [2]. Fourth, the financial system is already withdrawing from the rest, through insurance non-renewals, risk-based premiums and partial price discounts [54] [44] [53]. Fifth, communities that relocate collectively, with a destination built in advance and a single public funder, can do so with dignity, as Japan and Gardi Sugdub showed [58] [5].
Taken together, these findings mean that retreat is already the default policy for most of the world’s coastline. It is simply not being administered. The physical trend ensures that low-lying and sinking land will flood more often. The economic models ensure that most of it will not be walled. The financial system ensures that insurers and lenders will price the risk before governments do. What remains undecided is the sequence and the distribution of losses. If governments wait, the sequence is set by markets and storms: coverage first, then values, then the house. If they act, the sequence can be reversed, with buyout offers before prices collapse, relocation sites before disasters and honest disclosure before sales. The California Coastal Commission’s statement that adaptation is impossible without some removal or relocation [8] is one of the few official admissions of what the data imply. Most governments still prefer to fund walls in public and buyouts in private.
There is no realistic scenario for California to adapt to future sea-level rise without some form of removal and/or relocation.
— Noaki Schwartz, spokesperson, California Coastal Commission, November 2021A retreat policy that matches the evidence would share a small number of features, drawn from the cases that have worked and those that have failed. None of them requires new science. All of them require political willingness to name the problem and fund it over decades rather than budget cycles.
- Name the line: publish, for each coastline, where defence is planned and where it is not, as England’s shoreline management plans do, and then fund the consequences rather than deny them [80] [69]
- Pay before the price falls: offer buyouts while homes still have value to trade, rather than five years after a flood [3] [77]
- Build first, then move: relocation succeeds when the destination exists before people leave, as in Japan and Gardi Sugdub [58] [63]
- Stop the controllable half: regulate groundwater pumping in sinking cities, as Japan did from 1956 [31] [21]
- Protect funding from political cycles: programmes that can be cancelled by memo, as BRIC was in 2025, cannot support decisions that take a decade [75]
The question of who pays will dominate the next decade. The NFIP’s US$22.5 billion debt, the €9 billion gap in the Dutch Delta Programme and the US$80 billion cost of the Java sea wall show that defence is expensive and getting more so [50] [41] [30]. The US$121 to 237 billion of estimated overvaluation in US housing shows that the loss on the retreat side is also large, and that it will fall on someone [45]. Without planning, it falls on the households that own the most exposed homes and on the municipalities that depend on their taxes. With planning, a share of it is socialised, through buyouts, relocation grants and transitional support. The choice is not between paying and not paying. It is between paying in an orderly way, in advance and with public accountability, or paying later in the form of uninsured losses, distressed sales and emergency relief. The record of buyouts, with average completion times of 5.7 years, suggests that even orderly payment needs to start much earlier than it does today [3].
Del Mar banned planning for retreat, Fairbourne’s council denied the plan that predicted its decline, and the policy literature is full of euphemisms. The evidence suggests that avoiding the word does not avoid the outcome. It only removes the planning, so that retreat arrives as a market event rather than a public decision. Naming retreat is the precondition for financing it fairly.
For coastal residents, the evidence carries a practical message. Exposure to sea-level rise is already partially reflected in prices, more so where buyers are informed, and disclosure laws are spreading [44] [57]. Flood insurance premiums are moving towards risk-based levels, and public backstops such as Flood Re have set end dates [54] [55]. Buyout programmes exist but are slow and politically fragile [3] [75]. Households in low-lying areas therefore face a narrowing window in which their homes still have value that can be used to move. For governments, the message is that the cheapest moment to plan retreat is before it is forced, while relocation can be voluntary, communities can move together and public money buys land at prices that have not yet collapsed. Japan built 8,373 plots after a catastrophe [58]. The test for every coastal country is whether it can do the same before one.
OsakaWire’s assessment is that coastal retreat is the most important adaptation policy that most governments refuse to name. The physical evidence is strong, the economic evidence is consistent across independent models, and the financial withdrawal is already measurable. The contested questions, about pricing, consent and the upper tail of sea-level rise, change the timing and the distribution of costs, not the direction ◈ Strong Evidence [11] [18] [45]. As Robert Nicholls put it, from a risk perspective it does not matter whether the sea rises or the land sinks [21]. It matters even less whether the response is called retreat, relocation or phased adaptation. What matters is whether it is planned. On current evidence, for most of the world’s coastline, it is not.