A review of estimating population exposure to sea-level rise and the relevance for migration

Between 1902 and 2015, global mean sea-level rise (GMSLR) was 0.12–0.21 m according to the Intergovernmental Panel on Climate Change (IPCC 2019, p 334). Relative to 1986–2005, additional sea-level rise (SLR) of 0.43–0.84 m is projected by 2100 (0.29–1.10 m, likely range) (IPCC 2019, p 324), although it depends particularly on the rate of Greenland and Antarctic ice sheet melting and so it could be much higher (Thomas and Lin 2020). At the moment, climate change causes SLR predominantly through sea water expanding as it absorbs heat from the atmosphere (thermal expansion) and melting of land-based snow and ice (such as from glaciers). SLR interacts with other climatic factors, such as intensifying storms and wave action with consequences already observed including infrastructure damage, coastal erosion, salination of freshwater, and land habitat loss (IPCC 2018, 2019). Future SLR is projected to affect human health and wellbeing, cultural and natural heritage, freshwater, biodiversity, agriculture, and fisheries (WHO 2018, IPCC 2018, NRC 2019).

Consequently, various attempts have been made at the global scale to assess populations exposed to SLR (Muis et al 2017). While these assessments use different definitions, approaches and scenarios, they seek to estimate the number of people who might be directly affected by SLR and related impacts, which defines 'exposure'. The studies variously combine datasets on SLR, subsidence, coastal extreme weather, land elevation, population distribution, land-surface characteristics, adaptation options, and socio-economic change scenarios. One of the first global vulnerability assessments estimated that 200–250 million people per year (in 1990) were exposed to coastal flooding, defined as living below the 1-in-1000-year extreme sea level, and hence 1 m of GMSLR would increase exposure by 50% assuming no other changes (Hoozemans et al 1993). Recently, IPCC (2019) estimated that 680 million people currently live in the low-lying coastal zone and projected this number to reach more than one billion by 2050.

Despite the possibility for various forms of adaptation to SLR, with many people and populations already planning for adaptation in situ (e.g. Yamamoto and Esteban 2014), human mobility has been widely positioned as a deterministic certainty whereby climatic and environmental hazards such as SLR force people away from their coastal homes (c.f. Myers 2001). For example, Strauss et al (2015) state that future carbon emissions 'will determine which areas we can continue to occupy or may have to abandon' (Strauss et al 2015, p 13 508). The first IPCC assessment report (IPCC 1990) estimated that half a million people in archipelago and island countries might live in sites that were at risk of submergence or loss of land by 2100, contributing to increased numbers of so-called 'climate refugees' (Dronkers et al 1990, but see also Lewis 1990a &, 1990b). More recently, Nicholls et al (2011) estimated that if 2 m of GMSLR is realised by 2100, a risk of 'forced displacement' exists of up to 187 million people. Studies draw on estimates of exposure to SLR and other related hazards (e.g. living in a 1-in-100-year flood plain) as proxy indicators for population migration and relocation. In some sites, specifically low-lying coral atoll nations, population displacement is stated as being likely as even modest SLR is assumed to disrupt livelihoods and render land uninhabitable (Mcleman 2018).

Studies that estimate and forecast global and near-global exposure to, or presumed migration due to, SLR make choices about parameters, such as level of GMSLR, return periods of flooding, and time horizons. They rely on global datasets and draw conclusions on a global scale. While they provide important estimates of the scale of potential exposure, there are implicit assumptions that global trends are of primary significance for understanding the situation facing people and for planning for adaptation, so that the specificities of local sociocultural and environmental contexts are often obscured. Indeed, many analyses end at the point where local knock-on impacts and decision-making begin, irrespective of global trends. For example, the geographic distribution of coastal flooding is mapped against gridded global data sets, but the impact of this flooding on freshwater supplies for particular coastal sites is not clear and can sometimes be countered by local measures (Yang et al 2019).

To indicate the advantages and limitations of the studies available, and to better direct future work in this area, this paper reviews and discusses datasets and analytical methods for estimating global or near-global population exposure to SLR, with a specific focus on suggestions about SLR-related population mobility attributed to SLR and associated impacts.

Three databases—Web of Science, Scopus, and Google Scholar—were searched for publications published up until April 2020 that provide quantitative estimates of population exposure to SLR and associated hazards. The full-text search string used was English only: population* AND coast* AND flood* AND 'sea level*' AND (global OR international* OR worldwide) AND (model* OR indic*) followed by snowball sampling of citations in publications found. The selected publications were restricted to those published in English and with a focus on global or near-global estimates. Publications with a regional or smaller-scale focus (c.f. Ericson et al 2006, Dasgupta et al 2009, 2011, Hinkel et al 2011, Anderson et al 2018, Taherkhani et al 2020) were therefore excluded. Provided that the publication appeared in the database search or through the snowballing, it was considered irrespective of being peer-reviewed or not, but wider searches were not conducted to capture all the grey literature on that rationale that the wider material had not necessarily been validated through scientific investigation. No time limit was placed on the searches, but the selected publications range from 1993 to 2019. From the results, manual screening was completed based on the authors' expertise to seek comparability among the studies while being strict about the scoping of and definitions used by this review.

Thirty-three publications met the inclusion criteria: 30 peer reviewed journal articles, 1 book, 1 working paper and 1 report (table 1). The final study selection of 33 publications resulted in 11 publications estimating population exposure to specific levels of GMSLR, 13 publications estimating populations living in coastal floodplains, and 12 publications estimating populations living in LECZ or near-coastal zones. These publications were analysed by extracting the key information shown in the columns of table 1 permitting a synthesis of what is known about this review's research topic of estimating population exposure to SLR and the relevance for migration. To fulfil this review's mandate, the columns in table 1 focus on the publications' aims, sources of information (data sets) and analysis methods, time frames considered for forecasts, scenarios and numbers for population exposure, any assumptions or implications related to migration, and the data or analysis challenges mentioned, thereby also permitting discussion in the next sections of this review regarding what is not mentioned in these publications.

3.1. Overview

3.2. Population impacted by specified levels of SLR

3.3. Population living in coastal floodplains

3.4. Population living in low-elevation coastal zones (LECZs)

This review found numerous challenges in the literature when measuring population exposure to SLR and related impacts. The estimates are based on gridded datasets that include DEMs, flooding and extreme sea-levels, and population distribution. Hinkel et al (2014), Lichter et al (2011), and Mondal and Tatem (2012) have shown that estimates of land and population exposure to SLR and coastal flooding vary significantly according to which datasets are employed. Final estimates depend on the input data, and decisions about key parameters such as time horizons, warming scenarios, and ecological or socioeconomic processes and feedbacks including adaptation measures assumed. Four main challenges are discussed here based on our review and analysis.

First, estimates of populations exposed to SLR rely on elevation data to define zones of inundation or potential hazard parameters (Small and Nicholls 2003, Ericson et al 2006, Mcgranahan et al 2007, Lichter et al 2011). Global DEM datasets include GLOBE which combines six gridded DEMs and five cartographic sources; the US Geological Survey GTOPO30 which combines eight raster and vector sources of topographic information; and the Shuttle Radar Topography Mission (SRTM) elevation data with a vertical resolution of 1 m and spatial resolution of approximately 90 m at the equator (Mcgranahan et al 2007, Lichter et al 2011, Brown et al 2018). Any DEM has vertical and horizontal uncertainties (Wolff et al 2016). For example, while there are enhanced versions of SRTM data (see Mondal and Tatem 2012, Kulp and Strauss 2019), SRTM datasets have uncertainties in urban and forested areas where radar technologies capture infrastructure or tree elevation as opposed to ground elevation (Dasgupta et al 2011, Marzeion and Levermann 2014). Global mean error in SRTM's 1–20 m elevation band has been found to be 1.9 m (and 3.7 m in the US) (Kulp and Strauss 2019). Choice of DEM also has significant effects on estimates. Hinkel et al (2014) found the estimated number of people flooded according to the GLOBE elevation model to be double that calculated using the SRTM elevation model, and Kulp and Strauss (2019) found that using CoastalDEM instead of SRTM resulted in estimates of population exposure to extreme coastal water level that were three or more times higher. Improvements in elevation datasets are required to enable accurate estimates of land area exposure (Gesch 2009).

Second, estimating coastal floodplains and potential coastal flooding requires datasets on extreme sea levels. A significant limitation of flood analysis at all scales is limited availability of accurate datasets (Gesch 2009, Mondal and Tatem 2012, Neumann et al 2015). In their analysis of coastal flood exposure, Muis et al (2017) found that correcting vertical data of sea-level extremes and land elevation for two sea-level datasets—DINAS-COAST Extreme Sea Levels (DCESL) dataset and the GTSR dataset—resulted in an increase of 16% and 20% respectively in flood exposed land, and 39% and 60% respectively for exposed populations. Moreover, there are other drivers of flooding including storm intensity, and climate change affects frequencies, magnitudes, and tracks of storms thereby yielding low confidence in how storm surges and extreme sea levels may alter over time (Marzeion and Levermann 2014, Brown et al 2018, Knutson et al 2019). Hazard parameters must be set within models, but it is not always clear which hazard parameters to select, or whether to select extremes or means.

Third, estimates of population exposure to SLR require population distribution datasets. Key datasets are the Gridded Population of the World (GPW), the Global Rural Urban Mapping Project (GRUMP), and LandScan Global Population database, all of which are developed from census data. Census data are available by census accounting units, with uncertainty in the spatial distribution of populations within each unit. GPW was the first global and widely available dataset that transformed census data to a grid; it emphasises input data rather than modelling distributions (Nicholls et al 2008a). GRUMP combines population data with census units, allocating people into urban or rural areas to coincide with UN estimates and using an urban extent assessment derived mostly from the night-time lights dataset of the US National Oceanic and Atmospheric Administration (NOAA) (Mcgranahan et al 2007, CIESIN, IFPRI, World Bank and CIAT 2011, Mondal and Tatem 2012). LandScan disaggregates census data within administrative boundaries based on weightings derived from land cover data, proximity to roads, slope, and populated areas (Bhaduri et al 2007, Mondal and Tatem 2012). As always, all these datasets have limitations. Spatially detailed census data are often not available for low-income countries; some census data are over 10 years old; informal settlements and undocumented people might not be accounted for in census data; and datasets that use night-time lights as a proxy for population can miss smaller coastal settlements with limited development and where electricity supply is intermittent or unavailable (Dugoua et al 2017). Different datasets produce differing population distributions. An analysis of variation in estimates of populations in LECZ as derived from LandScan and GRUMP found that eight of the top ten locations with the largest differences in estimates were small low-lying island countries or territories, including for example Tuvalu (Mondal and Tatem 2012). Consequently, the limited spatial resolution of census data means there is uncertainty as to the location of populations relative to SLR and its related hazards (Small and Nicholls 2003, Foley 2018).

Finally, many studies set specified levels of future GMSLR based on different emission and warming pathways over the coming decades, centuries, and millennia (e.g. Pfeffer et al 2008, Brown et al 2016, Clark et al 2016, Mengel et al 2016). Debates regarding GMSLR estimates and forecasts relate to spatial variations, temporal uncertainties, rates of ice mass loss especially from Greenland and Antarctica, ocean dynamics, emission scenarios, and changes in gravity associated with water mass redistribution, leading to significant regional variations from the global mean (Clark et al 2016, Jevrejeva et al 2016, Mengel et al 2016, Geisler and Currens 2017). Subsidence and isostatic uplift further affect local sea level projections (Hinkel et al 2014, Erkens et al 2015, Brown and Nicholls 2015). There are temporal uncertainties in forecasting GMSLR associated with projected rates (e.g. Bindoff et al 2007, Solomon et al 2007); natural, multi-decadal oscillations (Sérazin et al 2016); and the pace of ice mass loss from the Greenland and Antarctica ice sheets (Tol et al 2006, Hansen 2007, Pfeffer et al 2008, Nicholls et al 2011, Jevrejeva et al 2016). Thus, future GMSLR is uncertain, something that many studies address by using higher and lower bounds of GMSLR in their analyses (IPCC 2019, c.f. Marzeion and Levermann 2014).

In summary, reliability of the estimates of both current and future population exposure to GMSLR and related hazards depend on the reliability of input datasets, with precision not always reflecting accuracy. Global quantitative estimates rely on global datasets, yet there are widely acknowledged challenges in estimating land elevation, extreme sea levels, population distribution, and GMSLR scenarios. These problems are amplified for studies seeking to estimate population exposure to GMSLR in the future. For example, the uncertainties in current population distribution estimates mean that future estimates also have large uncertainties.

While noting the challenges of estimating population exposure to SLR, most global estimates are in the order of tens or hundreds of millions of people exposed to coastal inundation and coastal flooding for different timeframes and scenarios. Given the current and expected impact of SLR, exposed populations must adapt (Murphy 2015). Adaptation to SLR and related hazards is widely defined under three main categories (Dronkers et al 1990, Brown et al 2016, IPCC 2019):(1) protection (measures reducing the parameters or likelihood of inundation in a specific location); (2) accommodation (policies or actions to live with the consequences of SLR); and (3) migration or relocation (planned movements of populations, sometimes termed 'managed retreat' or 'managed realignment'). Despite the possibility for different adaptive measures to SLR, migration or relocation—also variously referred to as retreat, displacement, resettlement, realignment, evacuation, and abandonment—has received significant attention in public and policy discussion.

First suggested in the 1970s (e.g. Brown 1976), the term 'environmental refugee' was the focus of much debate before being largely discredited on the basis of its legal definition, since the United Nations (UNHCR 1951/1967 ) does not include any environmental basis for claiming or granting refugee status. Additionally, the multi-causal nature of migration (Foresight: Migration and Global Environmental Change 2011, Felli and Castree 2012, Fiddian-Qasmiyeh et al 2016, Ayeb-Karlsson et al 2018) undermined concepts of environmental, climate, and climate change refugees, along with increasing resistance among residents of small island states to narratives of disappearing islands and population displacement as iconic examples of the threat of climate change (Farbotko 2010). Nonetheless, a recent ruling by the United Nations in response to a protection claim from a resident of Kiribati (OHCHR 2020) referred to 'climate change refugees' as people where 'the risk of serious harm arises from environmental factors indirectly caused by humans, rather than from violent acts' (OHCHR 2020, p 8). There are also emerging studies documenting early cases of human migration and relocation linked to the perceived impacts of SLR (c.f. Mcmichael and Katonivualiku 2020). Some countries—e.g. Micronesia, the Marshall Islands, the Solomon Islands, Fiji, and Kiribati—are developing policy and practice responses that include migration and planned resettlement of vulnerable populations in response to perceived climate impacts such as SLR (Hauer et al 2020). Despite the discursive, legal, and policy developments that link SLR to human migration, as well as emerging examples of low-lying places where SLR is considered a key climate change-related hazard which could affect migration decision-making, the question remains as to how, where, and to what extent population exposure to SLR might shape human mobility globally.

Twenty of the 33 publications reviewed in this article discuss connections between population migration and SLR. For example, Nicholls et al (2011) suggest that 2 m of SLR by 2100 yields the risk of 'forced displacement' of up to 187 million people. They suggest that, although displacement is avoidable through protection, the likelihood of protection being successfully implemented is lowest in small islands, Africa, and parts of Asia, hence these regions are likely to experience coastal abandonment. Brown et al (2018) note that although people may not want to move from their coastal sites, in some places this may be the most viable option, highlighting low-lying settlements in Fiji, the Maldives, and Panama. Kulp and Strauss (2019) suggest that coastal communities must prepare for difficult futures including large-scale migration away from coastlines, and they call for global modelling that accounts for the timing, locations, and scale of human migration in response to worsening coastal flooding. Nicholls et al (2008b) consider the scenario of West Antarctic Ice Sheet collapse (WAIS) producing 5 m of SLR, and consequent land loss that would lead to forced migration. Rowley et al (2007) suggest that inundation of coastal land due to SLR could displace millions of coastal residents, with up to 430 million people affected by SLR of 6 m.

Other studies that focus on wider SLR-related hazards such as coastal flooding have also suggested migration will be a key response. For example, Hinkel et al (2014) argue that where protection is not in place, populations will move in response to growing flood risk. Mcgranahan et al (2007) note that migration represents a key risk reduction strategy for populations in coastal lowlands, albeit a costly and disruptive form of adaptation. Given that exposure to SLR is not a reliable proxy indicator for migration, none of these studies rigorously or reliably quantifies the number of people who might be expected to move due to SLR or could do so. Beyond estimates of exposure, the connections between SLR and migration are complex and uncertain.

Consequently, population exposure to SLR and related impacts should not be conflated with inevitable migration. As has been widely noted, migration decisions are shaped by more than environmental factors, including economic, social, demographic, institutional, and political dimensions (Nicholls et al 2008a, Foresight: Migration and Global Environmental Change 2011, Felli and Castree 2012, Fiddian-Qasmiyeh et al 2016, Ayeb-Karlsson et al 2018, Hauer et al 2020). SLR could generate 'trapped' populations who have a desire to move but not the necessary resources, and there will be those who prefer not to move for social, cultural, and political reasons including place attachment (Farbotko and Mcmichael 2019).

Further, migration can be prevented or forestalled through other adaptive strategies. Nicholls and Cazenave (2010) emphasise that humans have historically adapted to coastal change, and many of the world's most populated coastlines and coastal cities are currently managed and engineered for flood risk (see also Hallegatte et al 2013). Over the 20th century, coasts have subsided by up to 5 m in Tokyo, 3 m in Shanghai, and 2 m in Bangkok; these cities depend on flood- and water management infrastructure to prevent further submergence. Some countries and cities with large coastal populations will try to adapt in situ (Davis et al 2018). Indeed, Lincke and Hinkel (2018) suggested that for 90% of exposed populations, who live on only 13% of the world's coasts, it will be cost-effective to protect. Nicholls et al (2008b) find that even in the scenario of the WAIS collapse, significant lengths of the world's populated coast could be defended, thereby significantly reducing migration.

Publications (including ten from this review) make the important point that coastal populations are growing due to migration into urban sites which increases population exposure to SLR (c.f. Small and Nicholls 2003, Nicholls 2004, Mcgranahan et al 2007, Hinkel et al 2014, Neumann et al 2015, Kummu et al 2016, Muis et al 2017). Jones and O'Neill (2016) suggest that models of future population distribution do not adequately account for the potential impacts of climate change. They argue that population distribution may be shaped by migration away from inland drought-affected regions and into coastal areas, and that population projections should include these parameters. These considerations highlight that populations are not static unless affected by SLR and related impacts. As the migration and mobilities literature has long explained (e.g. Fiddian-Qasmiyeh et al 2016), people continually move into and away from sites for numerous reasons, including coasts, so any estimates are subject to large uncertainties. These uncertainties do not decrease due to exposure to SLR.

Studies with a tighter geographic focus address some of these challenges, as they draw on datasets with higher spatial resolution and can include parameters that address the specificities of local social, economic, demographic, and political contexts (c.f. Hauer 2017, Chen and Mueller 2018). For example, Davis et al (2018) modify a diffusion-based model of human mobility in combination with population, elevation, and climatic data to forecast SLR-driven migration in Bangladesh in the years 2050 and 2100. They predict that 0.9 million people by 2050 and 2.1 million people by 2100 could be displaced by permanent inundation, predominantly within the southern half of Bangladesh. Hauer (2017) focuses on populations in the US, estimating both the number and destinations of potential SLR migrants by 2100. The analysis uses county-level projected population data, county-to-county migration flow data, and household economic status while assuming a 1.8 m SLR scenario, in order to simulate expected impacts. Hauer (2017) finds that by 2100, every US state and 56% of counties could be affected by SLR-related migration, of people moving either in or out.

Another advantage of more localised studies is that they can account for other environmental changes and coastal geomorphology that interact with SLR such as pre-existing subsidence, coastal erosion due to ecosystem destruction (e.g. wetlands, mangroves, and coral reefs), land use changes, or groundwater extraction (Mcleod et al 2010, Arkema et al 2013). Here, SLR and associated coastal change, might be a necessary but not sufficient driver of migration. For example, research in Louisiana illustrates that 5000 km² of coastal land has been lost since 1932—due to local, regional, and global factors driving relative sea level change—yet Louisiana's population has not moved landward in concert with observed shoreline encroachment (Hauer et al 2019). While this article is specifically examining global or near-global estimates for considering the implications of SLR and associated impacts, such as potential migration, more localised studies might be more appropriate.

Nonetheless, data are not readily available in many places identified as vulnerable to the impacts of SLR. For example, in 2014, residents of the small coastal village of Vunidogoloa, Fiji relocated from the foreshore to a higher site within their customary mataqali (i.e. clan) land. Over recent years and decades, villagers experienced high tides that regularly flooded their homes and village land, coastal erosion which led to the loss of some homes, storm surges, and saltwater intrusion. Earlier efforts to adapt included ad hoc relocation of houses and construction of two seawalls. The entire village relocated in January 2014, with the support of the Government of Fiji and donor organisations as well as significant community input and resources. Villagers consider their relocation to be driven by climate change, specifically SLR impacts (Mcmichael and Katonivualiku 2020). It is referred to and labelled by the Government of Fiji and institutions and donor organisations as a climate change-related relocation (c.f. The Fijian Government 2019).

Yet, many coastal sites are not well-represented in global gridded population and elevation datasets. An assessment of Pacific island countries and territories found that LandScan tended to overestimate numbers of Fijians in urban centres and GPWv4 tended to over-disperse the population, resulting in a smoothed population distribution rather than small dispersed settlements and urban centres (Andrew et al 2019). Moreover, SRTM elevation data could be unreliable in Fiji since SRTM imagery has large errors for sites vegetated with mangroves and evergreen tree cover (Hawker et al 2018). Furthermore, there is limited coastal monitoring in Fiji and other Pacific island countries to determine the extent to which coastal erosion and flooding are attributable to climate change related SLR, as opposed to other coastal geomorphology processes associated with subsidence, or human activity such as river dredging and removal of mangroves (Singh et al 2019). Vunidogoloa is an example of planned relocation in which emerging, local, coastal exposure aligns with the types of impacts expected with SLR, yet the resolution of global and regional SLR exposure estimates is too coarse to generate meaningful assessment of the causes of local coastal changes.

Coastal zones are some of the most densely populated areas in the world and include many of the world's large cities and many of the fastest growing urban areas (Smith 2011, Brown et al 2018). There are also coastal zones with small settlements and villages, such as those in outer islands of small island states or rural areas of Belize and Bangladesh. This review shows that the existing estimates of population exposure to GMSLR and related hazards largely depend on the reliability of input datasets and the methodologies adopted. The underlying uncertainties in the elevation datasets continue to result in significant differences in the estimates. Another limitation in the literature arises from the spatially gridded population datasets that are still unable to produce reliable high-resolution estimates in the lowest-lying coastal areas. The uncertainties in the current population distribution likely mean that future estimates will have large uncertainties. Low statistical confidence levels associated with extreme sea levels datasets are also common, especially regarding the vertical data of sea level extremes and land elevation. Finally, despite the fact that future estimates of GMSLR are uncertain, many existing studies do not provide confidence limits of their estimates of population exposure.

Irrespective of the uncertainties, the literature converges on the point that coming decades and centuries will see SLR radically redefine the world's coastlines (Taherkhani et al 2020). Global and near-global population estimates with their ranges highlight the large numbers of people exposed to SLR, both now and in the future. Even with climate change mitigation and more stable global temperatures, the land area and population exposed to SLR seem likely to continue increasing for centuries (Nicholls and Lowe 2004, Hauer et al 2020), so climate change adaptation is still required of which population mobility is frequently highlighted.

SLR has the potential to shape population mobility at all scales (Hauer et al 2020). There are significant and long-term implications of SLR-related hazards for coastal habitation (Brown et al 2018). Migration and relocation are ways for people to adapt to environmental and climatic changes (Black et al 2011, Murphy 2015), and migration has been a way of responding to shifting coastlines throughout human history (Fiddian-Qasmiyeh et al 2016). The main difference today is that climate and coastal changes contributing to human migration are increasingly human-induced, so global and national policy and practice debates are examining this topic.

Whether or not improved global estimates of population exposure to SLR are needed for policy and decision-making is debatable. Global estimates give a sense of the scale and distribution of exposure, but many other factors are needed for understanding and managing potential impacts and for planning for adaptation locally. Paramount are the moral, ethical, and legal debates regarding population stabilisation, land ownership, resources available, and decision-making for moving or staying. This review's value in these discussions is demonstrating that (i) despite the uncertainties and unknowns, it is clear that coastal changes and consequences for populations will be large and require significant action, (ii) there is time now to consider thoroughly the situations and options, so this should be done rather than waiting for more urgency, and (iii) many aspects of the topic need to be decoupled to avoid repeating myths or reproducing discursive narratives that uphold global power relations, especially those covering 'climate refugees' (Kelman 2019, Ayeb-Karlsson 2020).

For example, this review shows that the (imprecise) global estimates of population exposure to SLR are inaccurate proxies for estimates of SLR-related migration. Migration decisions are shaped by more than exposure to environmental changes (Fiddian-Qasmiyeh et al 2016); people may be trapped or choose to remain in sites of high exposure; people move into sites of exposure; and migration can be prevented or delayed through other adaptation and accommodation measures. It is necessary to connect SLR-related hazards to human migration at the temporal and spatial scale of lived experience and decision-making. More work is needed to better understand how to ensure that policies and decision-making focus on the research findings rather than the rhetoric, especially regarding how SLR interacts with other factors to shape migration decision-making.

A key area for further research could be to understand the ways in which adaptation measures (e.g. protection, accommodation, or migration) to SLR are shaped by wealth, including at the societal, household, and individual level. While some continue to highlight GDP (e.g. Kirezci et al 2020), this indicator is known to have such severe limitations that robust and meaningful results for policy and decision-making would be much better obtained by extracting information from micro surveys. Additionally, economic analyses suggest that managed migration might be cost-effective for the majority of the world's coastlines while protection might be pursued selectively in sites with dense populations and assets (Diaz 2016, Jevrejeva et al 2018). To illustrate, Hauer (2017) assumes that wealthier households in the US (earning more than US$100 000 per year) are more likely to adapt to SLR in situ, and thus unlikely to migrate. Economic analyses cannot fully account for intangible costs, such as health impacts, disrupted place attachments, and loss of culture and wellbeing, so further work should do so.

This work should also include focusing on improving terrain elevations in urban and coastal settings. Furthermore, although this review covers a global phenomenon, more localised studies are needed to better examine locally driven environmental stressors and coastal geomorphology. This will require developing coastal area elevation datasets by combining local terrain information with high-resolution satellite data. One other input into this work is considering scenarios of the Greenland and/or Antarctic ice sheets melting rapidly, because then, adapting in situ might not be an option for many. Understanding the drivers and any tipping points for SLR-related migration requires more detailed investigation, which then circles back to the decision-making for adaptation. As this review shows, without ice sheets melting, GMSLR might not exceed the maximum sea level since the last ice age (Dickinson 2009) until next century, in the absence of action related to climate change. Ice sheet melting provides substantially worse scenarios. Decisions need to be made over the next century considering these two widely divergent possibilities.

Thus, global and near-global estimates of SLR-related population exposure and their relevance for migration highlight (i) widespread impacts of human-caused climate change with the single phenomenon of GMSLR affecting a significant proportion of the human population and (ii) challenging circumstances for deciding how to act, considering the wide-ranging options from population stabilisation to engineering coastlines to moving away from the current shores. These decisions might be helped by spatial distribution and resolution of global datasets improving significantly to enable more reliable quantitative assessments of population exposure to SLR and related impacts, but decision-making should not wait for them, instead planning now for the various decision pathways. Meanwhile, people in LECZs are increasingly reporting environmental changes that are potentially attributable to SLR. At this local scale, more effort is needed to understand the complex interactions between localised SLR and related hazards, local social contexts and potential strategies regarding demographics, migration, and (im)mobility.

This paper was completed as part of the Lancet Countdown project (Tracking the connections between public health and climate change) at http://www.lancetcountdown.org which is supported by an unrestricted grant from the Wellcome Trust (200890/Z/16/Z).

No new data were created or analysed in this study.