Introduction

Movement of people, planned or unplanned, may occur in some low-lying coastal areas in response to accelerating sea level rise (SLR)1,2,3. In this context, planned relocation, also called managed retreat or resettlement, is the permanent movement of entire communities from sites of presumably higher to lower exposure to coastal hazards – namely, SLR and related flooding. It is often considered a drastic “measure of last resort”4 as relocating can be traumatic, and livelihood outcomes are often negative5. Yet planned relocation is increasingly discussed as a viable adaptation option6 by governments and communities working to reduce people’s exposure to coastal hazards7,8,9 and avoid costs of forced displacement10. Unlike other more incremental strategies to protect (e.g., seawall) or accommodate (e.g., elevation of housing) coastal hazards that may lead to residual risk, planned relocation has the potential to be transformational11 and eliminate flood exposure12. But planned relocations may have maladaptive outcomes – when relocation unintentionally creates new harms13 – including exposure to similar or new hazards at the destination site14, which may even worsen with time. A post-tsunami planned relocation in the Maldives involved a move to an island, Dhuvaafaru, highly exposed to SLR15; post-disaster planned relocations in Fiji, Sri Lanka, India, and the Philippines8,16,17,18,19 involved moves to destination sites where flooding and landslides later occurred. Are these maladaptive examples the exception or the rule? Understanding if relocation reduces future risk by comparing origin to destination exposure is essential because these moves are costly and disruptive. Scarce resources must prioritize situations where, at a minimum, people will be safer7,20 and their sustainable livelihood trajectories improved5,21.

While a growing body of qualitative research has examined whether planned relocation achieves its socioeconomic goals9,21,22,23, and we recently evaluated diverse metrics for the success of relocations5, little quantitative research has assessed whether it achieves its primary goal of reducing exposure today, and no research has examined future exposure. In one of the few studies aimed at the former, Pinter and Rees estimated that planned relocation of eight non-coastal relocation cases in the Midwest of the United States (US) between 1937 and 2008 reduced riverine flood damage costs by over 95%24. Bodine et al. found that twelve of thirteen relocation sites were less exposed to coastal storm surge in Tacloban Philippines8. For a series of hypothetical moves, Duvat et al. found that future SLR exposure could be reduced by relocations in Rangiroa Atoll of French Polynesia25. Despite the clear need8, no study has systemically assessed future coastal hazard exposure reduction across completed or underway planned relocation cases at a global level throughout the 21st century. This dearth of evidence on hazard exposure reduction exists for adaptation efforts generally: a recent global stocktake reviewing climate adaptation responses in academic literature found little detailed evaluation of how enacted or planned strategies lower exposure26, with some exceptions (e.g., dikes27). This kind of evaluation is complicated by the need for counterfactuals – what would the outcome be if no adaptation occurred – and multiple future time horizons and climate change scenarios26.

Here, we quantitatively evaluated the effectiveness of coastal planned relocations to date at reducing communities’ exposure to projected coastal hazards – SLR and flooding – throughout the 21st century. We assessed all of the planned relocation cases from a global database28 meeting the following criteria: (i) initiated in the context of one or more coastal hazard(s), (ii) origin site was located adjacent to a coastline (within 1 kilometer), (iii) involved a whole community movement from a single origin to a single destination site, (iv) had available geolocated site data, and (v) had comparable elevation models. Of the 408 cases in the database, 17 met our inclusion criteria (Fig. 1). Combining maps of each communities’ origin and destination sites with a new high-resolution coastal digital elevation model29, SLR projections from the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report30, we considered the effects of future SLR at each site in 2030, 2050, and 2100 and across emissions scenarios (‘low emissions’ SSP1-2.6, ‘middle emissions’ SSP2-4.5, and ‘high emissions’ SSP5-8.5). We considered SLR effects alone (see Supplementary Tables 2–4) and in combination with a once-a-year flooding event29. Such events, which are expected to become more intense and frequent in low-lying coastal areas31, do not capture the entire range of flood risks, but they provide a robust standard for comparing among sites. Our measure of inundation is the fraction of the area of a site with projected exposure to SLR and flood. To estimate reduction in exposure, we calculated the difference between the fraction of site inundated at the origin (counterfactual for if the planned relocation adaptation response had not occurred) and destination site (in 2030, 2050, and 2100).

Fig. 1: Global map of 17 planned relocation cases.
figure 1

Further details about each case, including location and estimated population, are included in Supplementary Table 1. Map data: Wikimedia Commons.

Full size image

Results

The 17 coastal planned relocation cases analyzed span the Pacific, South and Southeast Asia, and North, South, and Central America (Fig. 1). Most cases are in island nations (e.g., Fiji, Maldives) and in the USA. Fourteen cases are completed while three are ongoing. While we assessed exposure to SLR and flooding, other coastal hazards such as storm hazards (e.g., wind and rain), erosion, saltwater intrusion, wetland loss, and tsunamis also informed actual relocation decisions. Supplementary Table 1 summarizes further details about each relocation case, including the primary coastal hazards, specific locations, estimated population size, timelines, status, and source for geolocating origin and destination sites.

Four example planned relocation cases illustrate the range of future risk reduction outcomes at the end of the century (2100) under a high emissions scenario (SSP5-8.5) (Fig. 2). Exposure reduction was complete in Gardi Sugdub where a 3.5-km relocation from a small island origin to a mainland destination decreased inundation from 100% to 0%. Vunidogoloa and Taholah, two sites with mainland origins and destinations, experienced substantial exposure reduction, with origin sites subject to 60% and 59% inundation and destination sites 0% inundation. Exposure reduction was modest in the small island-to-small island relocation of Kandholhudhoo, from 100% inundation in the origin to 66% inundation in the destination.

Fig. 2: Projected exposure in four illustrative case studies.
figure 2

The four planned relocation cases include: A Gardi Sugdub, Panama; B Taholah, United States; C Vunidogoloa, Fiji; and D Kandholhudhoo, Maldives. Each map shows projected SLR and floods at the end of the century (2100) under a high emissions scenario (SSP5-8.5), and the intersections with origin (yellow polygons) and destination sites (black polygons).

Full size image

The effectiveness of planned relocation at reducing SLR and flooding over time (Fig. 3) requires asking three questions. First, if a community remains at origin site, will it be exposed over time? In all cases, origin sites are projected to be exposed to SLR and flooding, with inundation increasing towards the end of the century (Fig. 3, Supplementary Table 2). Second, if a community relocates to the new site, will it be exposed over time? Destination sites in nine cases are exposed to SLR and flooding, increasing late in the century (Fig. 3, Supplementary Table 3). Notably, some cases (Kandholhudhoo, Gemendhoo, Madifushi, Isle de Jean Charles, El Choncho, Narikoso, and Denimanu) involve destination sites that are projected to be considerably inundated whereas others (Enseada, Vunisavisavi) have more marginal inundation, ≤4% of destination site area. Finally, how much less exposed is a community after the planned relocation? By the end of the century in a mid-emissions scenario, exposure decreases by an average of 47%, with a range from 100% in Gardi Sugdub to 2% in Jakarta (Supplementary Table 4).

Fig. 3: Percent area exposed to SLR and floods at origin and destination sites over time under medium emissions scenario, by case.
figure 3

Exposure to SLR and a once-a-year flooding event at origin and destination varies over time (2030, 2050 and 2100) under a mid-emissions scenario (SSP2-4.5). Each line chart represents the percentage of the origin site area (blue) or destination site area (orange) that is inundated. The x axis on all charts represents time (2030, 2050, 2100), while the y axis represents percentage of site inundated. The first three cases (A) involve moves from a small island-to-small island, the next four cases (B) involve moves from small island-to-mainland/large island, while the final ten cases (C) involve moves from mainland-to-mainland. Map data: Google Earth.

Full size image

The cases mapped in Fig. 2 are examples of three classes of planned relocations – small island-to-small island (e.g., Kandholhudhoo), small island-to-mainland/large island (e.g., Gardi Sugdub), and mainland/large island-to-mainland/large island (e.g., Taholah, Vunidogoloa) with consistent differences in exposure reduction. In this analysis, a small island is less than 2 square kilometers. On average, small island-to-small island relocation had the highest end-of-century projected inundation at origin (low emissions: 78% to high emissions: 87%) and destinations (low emissions: 36% to high emissions: 45%) (Fig. 4), the latter implying lower exposure reduction potential for small island-to-small island relocations, especially under high emissions scenarios.

Fig. 4: Exposure to SLR and floods at origin and destination sites across emissions scenarios, by spatial pattern.
figure 4

Exposure to SLR and a once-a-year flooding event at origin and destination sites averaged by topological type with variation across the low, mid, and high emissions scenarios at the end of the century (2100). Each pie chart represents the percentage of the destination or origin site area that is likely to be inundated (blue) or not inundated (orange) under each emissions scenario.

Full size image

In 2100, small island-to-mainland/large island planned relocations had high levels of projected inundation at origins (low emissions: 83% to high emissions: 90%), close to small island-to-small island cases, but considerably lower inundation at destination (low emissions: 13% to high emissions: 14%) (Fig. 4). Hence, this type of planned relocation has, on average, the most exposure reduction potential, regardless of emissions scenario. Mainland/large island-to-mainland/large island planned relocation, usually a move from a coastal location to an inland location at higher elevation, had more modest but still substantial 2100 projected inundation at origin (low emissions: 39% to high emissions: 46%) but less inundation at destination (low emissions and high emissions: 5%).

Discussion

Across 17 cases, planned relocation reduced coastal hazard exposure from SLR and flooding. While near term exposure was generally reduced, this did not always extend through the 21st century, especially under a high emissions scenario. Consequently, the effectiveness of planned relocation is dynamic, changing over time, and dependent on emissions pathways32,33.

Geographic context helps explain variation in exposure reduction across cases. The three classes of cases described above highlight the importance of differences in land availability for planned relocation, with some countries having land at safer higher elevation, whereas others have no locations available with zero SLR and flood exposure. Many small island developing states (SIDs), especially atoll island countries such as the Maldives, where 80% of land area is less than 1-m above mean sea level34, have no places that are completely safe from coastal hazards. This may explain why exposure reduction is more modest in small island-to-small island moves. By contrast, mainland-to-mainland moves often involve both origins and destinations at higher elevation, meaning less exposure reduction overall. That said, SIDs are not homogenous, with differences in geography, institutional context, and historical and political status that have critical implications for climate adaptation35, yet many face severe constraints on land availability. While the limits to adaptation in SIDs and specifically atoll island contexts are well known36,37, our study raises questions about the specific limits of planned relocation as a sustainable climate change adaptation strategy over the long term and under higher emissions scenarios.

Exposure in some destination sites could be also a consequence of the absence of prior knowledge through systematic assessment of risk. Strikingly, for 15 of the 17 cases, we could not find assessments of long-term site suitability conducted prior to relocation. Based on a thorough search (Supplementary Table 5), we found only two assessments38,39 conducted by stakeholders involved in the 17 planned relocations. Both were conducted in 2016 and considered current coastal hazard exposure and risk. Neither considered future sea level rise projections, let alone multiple future time horizons, emissions scenarios, and hazard types. Both included many criteria for site selection beyond environmental considerations, including, for example, economic livelihood opportunities and political constraints of land availability. Anecdotal evidence suggests that assessments were not always key to decision making, even if conducted: “the community’s decision to relocate was based on the experience of the matriarch (grandmother) … not on the study”40. For future relocations, assessments of future scenarios tailored to decision-making context have the potential to function as decision support tools that offer communities and governments a way to more clearly understand risks, benefits, and uncertainties41.

While natural hazard(s) were present at each of the 17 cases, the relative influence of the hazard drivers likely varied. Available information about these cases suggests a range of motivations for relocation site selection, often centering on land availability, “political interests”42, or other social, economic, and demographic factors alongside environmental reasons23. Gardi Sugdub was initiated because of SLR and also demographic concerns of overcrowding on a small island43; Jakarta was initiated due to concerns about coastal hazards, subsidence, and long-standing political considerations44. This confirms findings of other studies: Tacloban destination sites selected post typhoon were bolstered by political enablers (i.e., already government owned, unoccupied, land with a clean title) alongside exposure reduction8. These specific examples point to the diversity of motivations and the potential for multicausal drivers for any relocation. The pentagon conceptual framework popularized by the Foresight Report45,46 captures the concept of multicausal drivers, including environmental, political, economic, demographic and social factors, for environmental migration at the individual level. Extending this conceptualization to the community scale and adding cultural factors, this framework can provide a starting point for exploring the role of future SLR projections in influencing decisions about whether and where to relocate entire communities. All these factors may be important determinants of whether and where a community (or other decision maker such as government or non-governmental organization) will plan a relocation.

For this study, we made several simplifications that might be addressed in future work47. First, even for a given SSP, a range of sea level rise projections are possible. Second, estimates of extreme high water based on a model calibrated on data from the past may be biased low, as the intensity and frequency of coastal storms and associated storm surge and flooding are projected to increase in many regions31. Third, as sea levels rise and coastal storms potentially change, the two factors may interact in complex ways that lead to inundation patterns that increasingly depart from the linear ‘bathtub’ approach described here48, and future analyses could consider hydrologic connectivity49. Fourth, despite advances in reducing errors in the vertical accuracy of the Coastal Digital Elevation Model (CoastalDEM)29 and the global reanalysis of storm surges and extreme sea levels (GTSR) dataset50, uncertainties in vertical resolution and challenges with downscaling large-scale models remain and may be reduced in future analyses with improved datasets and models. Fifth, a range of measured or modeled flood thresholds (e.g., daily to once per century) and flood height distributions could facilitate alignment with a range of risk-reduction objectives; future research should examine and compare a range of flood metrics. Sixth, an analysis that includes risks from hazards other than coastal flooding such as storms and landslides (which occur at the destination site for Denimanu community in Fiji18) would allow for a more complete picture of exposure reduction; future research should examine compounding, multi-hazard contexts. Seventh, accounting for other confounding factors (e.g., extant climate adaptation infrastructure at origin and destination sites, social capital, economic opportunity, political dynamics) would allow for better assessment of exposure reduction due to relocation, as other factors may help explain changes in risk exposure. Eighth, including cases with multiple origin and destination sites would expand the data available and help explore exposure reduction in these contexts8,28. Finally, complementary analyses of socioeconomic vulnerability51, sustainable livelihoods5, and equity are essential to consider in metrics of relocation effectiveness, since the same exposure does not result in the same outcome for all52, and relocating people perceive risk and habitability based on their unique identities53. Acknowledging all these compounding uncertainties, our findings should be interpreted cautiously and paired with in depth community-based assessments for local decision-making.

The decision to relocate an entire community is often fraught21, and many may not want to move54. Yet planned relocation may be deemed the best adaptation option by some communities most exposed to rising seas and flood risks. When planned relocation is considered, systematic prior assessment can play a key role in the characterization of future risk that is essential for effective policy and practice. There needs to be anticipatory forethought about what, where, how, why, and when planned relocation takes place, and how circumstances may evolve over time. Understanding the level of exposure reduction delivered in a move is a critical starting point. Assessments informing planned relocation processes should also consider socioeconomic and cultural dimensions of risk5, leverage community-engaged participatory co-produced approaches55, and be inclusive of traditional and local knowledge56. The methodological approach and empirical evidence presented here of what planned relocation does – or, critically, does not57 – achieve throughout the 21st century supports communities and governments as they consider planned relocation to reduce human suffering and improve livelihood outcomes. Especially for countries that lack safe destination sites, this is an issue of climate injustice: the international community may need to consider burden sharing and multilateral support for exposed communities that lack internal planned relocation alternatives.

Methods

In this paper we ask: does planned relocation effectively reduce exposure of communities to coastal hazards (SLR, flooding) and how does this change over time? To answer this question, we undertook: 1) selection of relevant cases; 2) geolocation of origin and destination sites; 3) projection of SLR and floods across multiple scenarios; 4) calculation of exposure reduction; and 5) comparison of assessment approaches.

Phase 1: Case selection

We identified cases from our team’s new database of over 400 planned relocations across the globe28. Cases were selected if they met the following criteria: a) initiated in relation to one or more coastal hazards (e.g., SLR, coastal floods, coastal erosion, tsunamis), b) origin site is coastal58, here defined as within 1 kilometers of a coastline, c) followed a common spatial pattern involving a move of a whole community from a single origin to a single destination site, d) had available information required to geolocate the precise polygons of the origin and destination sites, and e) were located within the geographic scope of Climate Central’s CoastalDEM v2.1 (Phase 3). For criterion a), we included cases where coastal erosion or tsunami flooding contributed to relocation because these hazards often co-occur with and can be expected to worsen with varying degrees of SLR. Based on criterion e), we excluded Alaska cases given that they were located beyond the coverage of CoastalDEM. We identified origin and destination sites through review of documents published by a range of stakeholders (e.g., researchers, governments, communities, and the media) that contained aerial or taken-from-the-ground photographs, maps, and specific narrative description that could be cross-referenced with satellite imagery available via Google Earth. Seventeen cases met these five criteria. See Supplementary Table 1 for origin and destination site sources.

Phase 2: Identifying origin and destination sites

For each of these cases, we used QGIS version 3.22.1. to create shapefiles of the origin and destination sites. To do this, we first uploaded the Google Earth Satellite Imagery hybrid base map. Our approach then differed depending on whether we had access to a site-specific map with defined perimeters of origin and destination site polygons. If a map with predefined site polygons was available, we traced the boundaries of the site and georeferenced specific points to ensure alignment with the site source information. For example, this approach was used for tracing origin and destination sites of Saney Village, Indonesia59. If a map with predefined site polygons was not available, we used photographs and narrative description of the case to trace the perimeter of inhabited areas on Google Earth. For consistency, we included houses and community centers, but excluded lands used for agricultural or other livelihoods purposes, as well as airstrips. For example, this approach was used for Gardi Sugdub, Panama60.

Phase 3: SLR and flood projections across multiple scenarios

Conventional approaches to study sea-level rise projections at a global scale are too low-resolution for understanding the effectiveness of hyper-local adaptation such as small-scale community-wide planned relocations. Therefore, we created inundation surface raster files derived from high-quality digital elevation models, sea level rise projections, and coastal flood risk statistics for each origin and destination site. We employed Climate Central’s CoastalDEM v2.1 for elevation.

In this study, we considered three scenarios from IPCC Sixth Assessment Report61 that span different socio-economic pathways (SSPs): SSP1-2.6 represents a future with dramatic emissions reductions and stabilization of temperatures at around 1.8 °C warming by the end of the century; SSP2-4.5 represents a middle of the road future where emissions stay around current levels and fall mid-century with around 2.7 °C warming by 2100; and finally, SSP5-8.5 represents a future where current emissions double by mid-century and the mean temperature is expected to rise by around 4.4 °C by end of the century. For each of these scenarios, we considered SLR alone plus the added height of a local annual flood, meaning a water level that is reached once per year on average, based on estimates from the GTSR dataset50. We also considered projections at different years, including a near-term (2030), mid-term (2050), and long-term (2100) planning horizon. For each of these scenarios, we considered the median (50th percentile) of the projection’s distribution and created an inundation layer which represents all areas where the elevation is less than or equal to a given threshold of SLR projection plus the once per year flood.

Phase 4: Calculation of exposure reduction

To evaluate the extent to which a planned relocation case has reduced exposure for the community, we calculated the intersection between each scenario’s inundation layer and the area of the origin site and the destination site. Our metric for a communities’ exposure to a coastal hazard is the percentage of the full origin or destination site area that intersects with the inundation layer. We quantified the effectiveness (narrowly defined) of a relocation project as the difference in percentage of inundated area between the origin and destination sites. Thus, we calculated this difference in percentage for each of the 17 cases, in each of the scenarios described above. As noted in our discussion of future research to address simplifications in this analysis, other socioeconomic metrics, beyond exposure reduction, are needed to inform more comprehensive evaluations of relocation effectiveness22.

Phase 5: Assessment comparison

To complement this spatial analysis, we also identified assessment documents used to make planned relocation decisions through systematic searches for key words: “Assesment” and “Case Name”. We also contacted authors of the primary sources used for origin/destination site to identify additional assessment documents. For each case with an identifiable assessment document, we conducted a content analysis of the document according to a short codebook. We identified characteristics of the document including consideration of future time horizons, emissions scenarios, options for relocation alternatives and destination sites, and rationale for why destination sites were selected including site selection criteria.