Sea Level Rise and Implications for Low-Lying Islands, Coasts and Communities

The number of tide gauges has increased over time from only a few in northern Europe in the 18th century to more than 2000 today along the world’s coastlines. Because of their location and limited number, tide gauges sample the ocean sparsely and non-uniformly with a bias towards the Northern Hemisphere. Most tide gauge records are short and have significant gaps. In addition, tide gauges are anchored on land and are affected by the vertical motion of Earth’s crust caused by both natural processes (e.g., GIA, tectonics and sediment compaction; Wöppelmann and Marcos, 2016¹¹⁷; Pfeffer et al., 2017¹¹⁸) and anthropogenic activities (e.g., groundwater depletion, dam building or settling of landfill in urban areas; Raucoules et al., 2013¹¹⁹; Pfeffer et al., 2017¹²⁰). When estimating the GMSL due to the ocean thermal expansion and land ice melt, tide gauges must be corrected for this VLM, where VLM = GIA + anthropogenic subsidence + (tectonics, natural subsidence). This is possible with stations of the Global Positioning System (GPS) network when they are co-located with tide gauges (Santamaría-Gómez et al., 2017¹²¹; Kleinherenbrink et al., 2018¹²²). However, this approach provides information on the VLM over the past two to three decades and has limited value over longer time scales for places where the VLM has varied significantly through the last century (Riva et al., 2017¹²³).

AR5 assessed the different strategies to estimate the 20th century GMSL changes. These strategies only accounted for the inhomogeneous space and time coverage of tide gauge data and for the VLM induced by GIA (Figure 4.5). Since AR5 two new approaches have been developed. The first one uses a Kalman smoother which combines tide gauge records with the spatial patterns associated with ocean dynamic change, change in land ice and GIA. It enables accounting for the inhomogeneous distribution of tide gauges and the VLM associated with both GIA and current land ice loss (Hay et al., 2015¹²⁴; Figure 4.5). The second approach uses ad hoc corrections to tide gauge records with an additional spatial pattern associated with changes in terrestrial water storage to account for the inhomogeneous distribution in tide gauges. It also accounts for the total VLM (Dangendorf et al., 2017¹²⁵; Figure 4.5). Both methods yield significantly lower GMSL changes over the period 1950–1970 than previous estimates, leading to long-term trends since 1900 that are smaller than previous estimates by 0.4 mm yr^–1 (Figure 4.5). Different arguments including biases in the tide gauge datasets (Hamlington and Thompson, 2015¹²⁶), biases in the averaging technique and absence of VLM correction (Dangendorf et al., 2017¹²⁷), or in the spatial patterns associated with the sea level contributions (Hamlington et al., 2018¹²⁸) have been proposed to explain these smaller GMSL rates. There is no agreement yet on which is the primary reason for the differences and it is not clear whether all the reasons invoked can actually explain all the differences across reconstructions. As there is no clear evidence to discard any reconstruction, this assessment considers the ensemble of AR5 sea level reconstructions augmented by the two recent reconstructions from Hay et al. (2015)¹²⁹ and Dangendorf et al. (2017)¹³⁰ to evaluate the GMSL changes over the 20th century. On this basis, it is estimated that it is very likely that the long-term trend in GMSL estimated from tide gauge records is 1.5 (1.1–1.9) mm yr^–1 between 1902 and 2010 for a total SLR of 0.16 (0.12–0.21) m (see also Table 4.1). This estimate is consistent with the AR5 assessment (but with an increased uncertainty range) and confirms that it is virtually certain that GMSL rates over the 20th century are several times as large as GMSL rates during the late Holocene (see 4.2.2.1). Over the 20th century the GMSL record also shows an acceleration (high confidence) as now four out of five reconstructions extending back to at least 1902 show a robust acceleration (Jevrejeva et al., 2008¹³¹; Church and White, 2011¹³²; Ray and Douglas, 2011¹³³; Haigh et al., 2014b¹³⁴; Hay et al., 2015¹³⁵; Watson, 2016¹³⁶; Dangendorf et al., 2017¹³⁷). The estimates of the acceleration ranges between -0.002–0.019 mm yr^–1 over 1902–2010 are consistent with AR5.

High precision satellite altimetry started in October 1992 with the launch of the TOPEX/Poseidon and Jason series of spacecraft. Since then, 11 satellite altimeters have been launched providing nearly global sea level measurements (up to ±82° latitude) over more than 25 years. Six groups (AVISO/CNES, SL_cci/ESA, University of Colorado, CSIRO, NASA/GSFC, NOAA; Nerem et al., 2010;¹³⁸ Henry et al., 2014¹³⁹; Leuliette, 2015¹⁴⁰; Watson et al., 2015¹⁴¹; Beckley et al., 2017¹⁴²; Legeais et al., 2018¹⁴³) provide altimetry-based GMSL time series. Since AR5, several studies using two independent approaches based on tide gauge records (Watson et al., 2015¹⁴⁴) and the sea level budget closure (Chen et al., 2017¹⁴⁵; Dieng et al., 2017¹⁴⁶) identified a drift of 1.5 (0.4–3.4) mm yr^–1 in TOPEX A from January 1993 to February 1999. Accounting for this drift leads to a revised GMSL rate from satellite altimetry of 3.16 (2.79–3.53) for 1993–2015 (WCRP Global Sea Level Budget Group, 2018¹⁴⁷; see Table 4.1) compared to 3.3 mm yr^–1 (2.7–3.9) for 1993–2010 in AR5. Compared to AR5, the revised satellite altimetry GMSL estimates now show with high confidence an acceleration of 0.084 (0.059–0.090) mm yr^–1 over 1993–2015 (5–95% uncertainty range; Watson et al., 2015¹⁴⁸; Nerem et al., 2018¹⁴⁹). This acceleration is due to an increase in Greenland mass loss since the 2000s (Chen et al., 2017¹⁵⁰; Dieng et al., 2017¹⁵¹) and a slight increase in all other contributions probably partly due to the recovery from the Pinatubo volcanic eruption in 1991 (Fasullo et al., 2016¹⁵²) and partly due to increased GHG concentrations e.g., (Slangen et al., 2016¹⁵³; high confidence). The current sea level rise is 3.6 ± 0.3 mm yr^–1 over 2006–2015 (90% confidence level). This is the highest rate measured by satellite altimetry (Ablain et al., 2019¹⁵⁴; medium confidence). Before the satellite altimetry era, the highest rate of sea level rise recorded was reached during the period 1935–1944. It amounted 2.5 ± 0.7 mm yr^–1 (estimate at the 90% confidence level from sea level reconstructions; Church and White, 2011¹⁵⁵; Ray and Douglas, 2011¹⁵⁶; Jevrejeva et al., 2008¹⁵⁷; Hay et al., 2015¹⁵⁸; Dangendorf et al., 2017¹⁵⁹). This is expected to be smaller than the current rate of sea level rise, making the current sea level rise the highest on instrumental record (medium confidence).

The ocean thermal expansion is caused by excess heat being absorbed by the ocean, as the climate warms. Thermal expansion is estimated from in situ ocean observations and ocean heat content reanalyses that rely on assimilation of data into numerical models (Storto et al., 2017¹⁶²; Sections 1.8.1.1 and 1.8.1.4; WCRP Global Sea Level Budget Group, 2018¹⁶³). Full-depth, high-quality and unbiased ocean temperature profile data with adequate metadata and spatio-temporal coverage are required to estimate thermal expansion and to understand drivers of variability and long-term change (Pfeffer et al., 2018¹⁶⁴; Section 5.2.2.2.2).

Historically, however, observational gaps exist and some ocean regions remain under-sampled to date (Sections 1.8.1.1 and 5.2.2.2.2; Figure 1.3; Appendix 1.A, Figure 1.1). Other factors also introduce uncertainty in estimates of thermal expansion like changes in instrumentation, systematic instrumental errors, changes in the quality control of the data and the mapping method used to produce regular grids (Section 5.2.2.2.2; Palmer et al., 2010¹⁶⁵). In the upper 700 m, the largest sources of uncertainty for estimates of global mean thermal expansion from 1970 to 2004 are the choice of mapping methods (Boyer et al., 2016¹⁶⁶), followed by the choice of bias correction for the bathythermographic observations (Cheng et al., 2016¹⁶⁷; Section 5.2.2.2.2). From 2006 onwards, the uncertainty is considerably reduced (Roemmich et al., 2015¹⁶⁸; von Schuckmann et al., 2016¹⁶⁹; Wijffels et al., 2016¹⁷⁰), because the Argo array reached its targeted near-global (up to ±60° latitude) coverage for the upper 2000 m in November 2007 (Riser et al., 2016¹⁷¹; Section 5.2.2.2.2).

Since AR5, in a community effort, the (WCRP Global Sea Level Budget Group, 2018¹⁷²) revisited the global mean thermal expansion estimates based on observations only. On the basis of a full-depth 13-member ensemble of global mean thermal expansion time series developed with the latest data and corrections available, they estimated that the global thermal expansion was 1.40 (1.08–1.72) mm yr^–1 for 2006–2015, 1.36 (0.96–1.76) mm yr^–1 for 1993–2015 (see Table 4.1). While the relative contribution of the upper 300 m did not change (~70%) between 2006–2015 and 1993–2015, the 700–2000 m contribution increased around 10% over the Argo decade (2006–2015), when observations for that depth interval soared (Figure 1.3; Appendix 1.A, Figure 1.1). This suggests that observed changes for 700–2000 m may have been underestimated for 1993–2005. Before 1993, estimates are based on a smaller ensemble of 4 datasets in which no thermal expansion is assumed below 2000 m because of lack of data (see Section 5.2.2.2.2 for more details). This ensemble shows a thermal expansion linear rate of 0.89 (0.84–0.94) mm yr^–1 for 1970–2015 (see Table 4.1).

Coupled climate models simulate the historical thermal expansion (see Table 4.1). However, for models that omit the volcanic forcing in their control experiment, the imposition of the historical volcanic forcing during the 20th century results in a spurious time mean negative forcing and a spurious persistent ocean cooling related to the control climate (Gregory, 2010¹⁷³; Gregory et al., 2013¹⁷⁴). Since AR5, the magnitude of this effect has been estimated from historical simulations forced by only natural radiative forcing. Then it has been used to correct the historical simulations forced with the full 20th century forcing (Slangen et al., 2016¹⁸⁷; Slangen et al., 2017b¹⁸⁸). The resulting ensemble mean of simulated thermal expansion provides a good fit to the observations within the uncertainty ranges of both models and observations (Slangen et al., 2017b¹⁸⁹; Cheng et al., 2019¹⁹⁰; Table 4.1). The spread, which is essentially due to uncertainty in radiative forcing and uncertainty in the modelled climate sensitivity and ocean heat uptake efficiency (Melet and Meyssignac, 2015¹⁹¹), is still larger than the observational uncertainties (Gleckler et al., 2016¹⁹²; Cheng et al., 2017¹⁹³; Table 4.1). Compared to AR5, the availability of improved observed and modelled estimates of thermal expansion and the good agreement between both confirm the high confidence level in the simulated thermal expansion using climate models and the high confidence level in their ability to project future thermal expansion.

The ocean mass changes correspond to the sum of land ice and terrestrial water storage changes. Since 2002, the GRACE and GRACE follow-on missions provide direct estimates of the ocean mass changes and thus they provide an independent estimate of the sum of land ice and terrestrial water storage contributions to sea level. Since AR5, GRACE-based estimates of the ocean mass rates are increasingly consistent (WCRP Global Sea Level Budget Group, 2018¹⁹⁴) because of the extended length of GRACE missions’ observations (over 15 years), the improved understanding of data and methods for addressing GRACE limitations (e.g., noise filtering, leakage correction and low-degree spherical harmonics estimates), and the improved knowledge of geophysical corrections applied to GRACE data (e.g., GIA). The most recent estimates (Dieng et al., 2015b¹⁹⁵; Reager et al., 2016¹⁹⁶; Rietbroek et al., 2016¹⁹⁷; Chambers et al., 2017¹⁹⁸; Blazquez et al., 2018¹⁹⁹; Uebbing et al., 2019²⁰⁰) report a global ocean mass increase of 1.7 (1.4–2.0) mm yr^–1 over 2003–2015 (see also Table 4.1). The uncertainty arises essentially from differences in the inversion method to compute the ocean mass (Chen et al., 2013²⁰¹; Jensen et al., 2013²⁰²; Johnson and Chambers, 2013²⁰⁴; Rietbroek et al., 2016²⁰⁵), uncertainties in the geocentre motion and uncertainty in the GIA correction (Blazquez et al., 2018²⁰⁵; Uebbing et al., 2019²⁰⁶). The consistency between estimates of the global mean ocean mass on a monthly time scale has also increased since AR5.

To assess the mass contribution of glaciers to sea level change, global estimates are required. Recent updates and temporal extensions of estimates obtained by different methods continue to provide very high confidence in continuing glacier mass loss on the global scale during the past decade (Bamber et al., 2018²⁰⁷; Wouters et al., 2019; Zemp et al., 2019²⁰⁸), see Section 2.2.3 and Appendix 2.A for a detailed discussion also on regional scales). Updates of the reconstructions of Cogley (2009)²¹⁰, Leclercq et al. (2011)²¹¹ and Marzeion et al. (2012)²¹², presented and compared in Marzeion et al. (2015)²¹², show increased agreement on rates of mass loss during the entire 20th century (Marzeion et al., 2015²¹³), compared to earlier estimates reported by AR5. The contribution of glaciers that may be missing in inventories or have already melted during the 20th century is hard to constrain (Parkes and Marzeion, 2018²¹⁵), and there is low confidence in their estimated contribution. These glaciers are thus neglected in the assessment of the sea level budget (Table 4.1).

While the agreement between the observational estimates of glacier mass changes and the modelled estimates from glacier models forced with climate model simulations has increased since AR5 (Slangen et al., 2017b²¹⁶), there is only medium confidence in the use of glacier models to reconstruct sea level change because of the limited number of well-observed glaciers available to evaluate models on long time scales, and because of the small number of model-based global glacier reconstructions.

Frequent observations of ice sheet mass changes have only been available since the advent of space observations (see Section 3.3.1). In the pre-satellite era, mass balance was geodetically reconstructed only for the GIS (Kjeldsen et al., 2015²¹⁷). These geodetic reconstructions empirically constrain the contribution of the GIS to SLR between 1900 and 1983 to 17.2 (10.7–23.2; Kjeldsen et al., 2015²¹⁸). During the satellite era, three approaches have been developed to estimate ice sheet mass balance: 1) Mass loss is estimated by direct measurements of ice sheet height changes with satellite laser or radar altimetry in combination with climatological/glaciological models for firn density and compaction, 2) the input–output method combines measurements of ice flow velocities estimated from satellite (synthetic aperture radar or optical imagery) across key outlets with estimates of net surface balance derived from ice thickness data, 3) space gravimetry data yields direct estimate of the mass changes by inversion of the anomalies in the gravity field (see Section 3.3.1 for more details). AR5 concluded that the three space-based methods give consistent results. They agree in showing that the rate of SLR due to the GIS and AIS’ contributions has increased since the early 1990s. Since AR5, up-to-date observations confirm this statement with increased confidence for both ice sheets (Rignot et al., 2019²¹⁹; see Section 3.3.1). The assessment of the literature since AR5 made in Section 3.3.1 shows that the contribution from Greenland to SLR over 2012–2016 (0.68 (0.64–0.72) mm yr^–1) was similar to the contribution over 2002–2011 (0.73 (0.67–0.79) mm yr^–1) and extremely likely greater than over 1992–2001 (0.02 (0.21–0.25) mm yr^–1). The contribution from Antarctica over 2012–2016 (0.55 (0.48–0.62) mm yr^–1) was extremely likely greater than over the 2002–2011 period (0.23 (0.16–0.30) mm yr^–1) and likely greater than over the period 1992–2001 (0.14 (0.12–0.16); see Section 3.3.1 for more details).

Here, the approach of Section 3.3.1 is followed, using the two multi-method assessments from Bamber et al. (2018)²²⁰ and the IMBIE team (2018) to evaluate the contribution of ice sheet mass loss to SLR over 1993–2015 and 2006–2015 (see Table 4.1). These two studies agree with results from the WCRP Global Sea Level Budget Group (2018). For the estimation of the AIS contribution, Bamber et al. (2018)²²¹ and the The IMBIE team (2018) use similar but not identical data sources and processing. Both studies find consistent results within uncertainties over both periods. In Table 4.1, the results of these two studies were averaged, and weighted the average on the basis of their uncertainties, because there is no apparent reason to discount either study. For the estimation of the GIS contribution only the Bamber et al. (2018)²²² estimate is used, as there is no other multi-method assessment available.

Water is stored on land not only in the form of ice but snow, surface water, soil moisture and groundwater. Temporal changes in land water storage, defined as all forms of water stored on land excluding land ice, contribute to observed changes in ocean mass and thus sea level on annual to centennial time scales (Döll et al., 2016²²⁵; Reager et al., 2016²²⁶; Hamlington et al., 2017²²⁷; Wada et al., 2017²²⁸). They are caused by both climate variability and direct human interventions, at the multi-decadal to centennial time scales. Over the past century, the main cause for land water storage changes are the groundwater depletion and impoundment of water behind dams in reservoirs (Döll et al., 2016²²⁹; Wada et al., 2016²³⁰). While the rate of groundwater depletion and thus its contribution to SLR increased during the 20th century and up to today (Wada et al., 2016²³¹), its effect on sea level was more than balanced by the increase in land water storage due to dam construction between 1950 and 2000 (Wada et al., 2016²³²). Since about 2000, based on hydrological models, the combined effect of both processes is a positive contribution to SLR (Wada et al., 2016²³³). Decreased water storage in lakes, wetlands and soils due to human activities are less important for ocean mass changes (Wada et al., 2016²³⁴). Overall, the integrated effects of the direct human intervention on land hydrology have reduced land water storage during the last decade, increasing the rate of SLR by 0.15–0.24 mm yr^–1 (Wada et al., 2016²³⁵; Wada et al., 2017²³⁶; Scanlon et al., 2018²³⁷; WCRP Global Sea Level Budget Group, 2018²³⁸). Over periods of a few decades, land water storage was affected significantly by climate variability (Dieng et al., 2015a²³⁹; Reager et al., 2016²⁴⁰; Dieng et al., 2017²⁴¹). Net land water storage change driven by both climate and direct human interventions can be determined based on GRACE observations and global hydrological modelling. They indicate different estimates of the rate of SLR. Over the period 2002–2014 GRACE-based estimates of the net land water storage (i.e., not including glaciers) show a negative contribution to sea level (e.g., Scanlon et al., 2018) resulting in the negative value after 2006 in Table 1while hydrological models determined a slightly positive one. The reasons for this difference between estimates are not elucidated. There is scientific consensus that uncertainties of both net land water storage contribution to sea level and its individual contributions remain high (WCRP Global Sea Level Budget Group, 2018²⁴²). The differences in estimates and the lack of multiple consistent studies give low confidence in the net land water storage contribution to current SLR.

Drawing on previous sections, the budget of GMSL rise (Table 4.1, Figure 4.5) is assessed with observations over 4 periods: 1901–1990 (which corresponds to the period in the 20th century that is prior to the increase in ice sheet contributions to GMSL rise), 1970–2015 (when ocean observations are sufficiently accurate to estimate the global ocean thermal expansion and when glacier mass balance reconstructions start), 1993–2015 (when precise satellite altimetry is available) and 2006–2015 (when GRACE data is available in addition to satellite altimetry and when the Argo network reaches a near-global coverage). The budget of GMSL rise is also assessed with sea level contributions simulated by climate models over the same periods (Table 4.1, Figure 4.5). The periods 1993–2015 and 2006–2015 are only 23 and 10 years long respectively, short enough so that they can be affected by internal climate variability. Therefore, it is not expected that observations over these periods will be precisely reproduced by climate model historical experiments. For the contribution from land water storage, the estimated effect of direct human intervention was used, neglecting climate-related variations until 2002 (Ngo‐Duc et al., 2005). From 2002 to 2015, total land water storage estimated with GRACE was used. In general, historical simulations of climate models end in 2005. Historical simulations were extended here to 2015 using the RCP8.5 scenario. This choice of RCP scenario is not critical for the simulated sea level, as the different scenarios only start to diverge significantly after the year 2030 (Church et al., 2013²⁴³).

For 1993–2015 and 2006–2015, the observed GMSL rise is consistent within uncertainties with the sum of the estimated observed contributions (Table 4.1). Over the period 1993–2015 the two largest terms are the ocean thermal expansion (accounting for 43% of the observed GMSL rise) and the glacier mass loss (accounting for a further 20%). Compared to AR5, the extended observations corrected for the TOPEX-A drift (see Section 4.2.2.1.2) allow us now to identify an acceleration in the observed SLR over 1993–2015 and to attribute this acceleration mainly to Greenland ice loss along with an acceleration in Antarctic ice loss (Velicogna et al., 2014²⁴⁴; Harig and Simons, 2015²⁴⁵; Chen et al., 2017²⁴⁶; Dieng et al., 2017²⁴⁷; Yi et al., 2017²⁴⁸; see also Sections 4.2.2.2.2, 4.2.2.3.4, 3.3.1). Since 2006, land ice, collectively from glaciers and the ice sheets has become the most important contributor to GMSL rise over the thermal expansion with mountain glaciers contributing 20% and ice sheets 33% (see Table 4.1). Over the periods 1993–2015, the sum of the observed sea level contributions is consistent with the total observed sea level within uncertainties at monthly-scales (not shown, e.g., Dieng et al., 2017). This is also true for the period 2006–2015, when uncertainties are significantly smaller. This agreement at monthly time scales represents a significant advance since the AR5 in physical understanding of the causes of past GMSL change. It provides an improved basis for the evaluation of models. Given these elements there is high confidence that the current observing system is capable of resolving decadal to multidecadal changes in GMSL and its components (with an uncertainty of <0.7 mm yr^–1 at decadal and longer time scales, see Table 4.1 and for example, WCRP Global Sea Level Budget Group, 2018). However, despite this advance since AR5 there are still no comprehensive observations of ocean thermal expansion below 2000 m, in regions covered by sea ice and in marginal seas. The understanding of glacier mass loss can be improved at regional scale and the understanding of the land water storage contribution is still limited. Thus, for smaller changes in sea level of the order of a few tenths of a mm yr^–1 at decadal time scales and shorter time scales there is medium confidence in the capability of the current observing system to resolve them (e.g., WCRP Global Sea Level Budget Group, 2018).

Before 1992, observations are not sufficient to confidently estimate the ice sheet mass balance and before 1970, the space and time sampling of ocean observations are not sufficient to estimate the global ocean thermal expansion. For these reasons, it is difficult to assess the closure of the GMSL rise budget over 1901–1990 and 1970–2015 (Church et al., 2013²⁴⁹; Gregory et al., 2013²⁵⁰; Jevrejeva et al., 2017²⁵¹; Meyssignac et al., 2017c²⁵²; Slangen et al., 2017b²⁵³; Parkes and Marzeion, 2018²⁵⁴). For the period 1970–2015, the thermal expansion of the ocean represents 43% of the observed GMSL rise while the glaciers’ contribution represents 22% (see Table 4.1). This result indicates a slightly smaller contribution from glaciers than reported by AR5. If the GIS contribution and the Antarctic SMB is added, then the sum of the contributors to sea level is in agreement with the low end observed SLR estimates over 1970–2015 (Frederikse et al., 2018²⁵⁵). This result suggests that the contribution of Antarctica ice sheet dynamics to SLR has been small, if any, before the 1990s.

Since AR5, extended simulations along with recent findings in observations and improved model estimates allow for a new more robust, consistent and comprehensive comparison between sea level estimates based on observations and climate model simulations (e.g., Meyssignac et al., 2017c; Slangen et al., 2017b²⁵⁶; Parkes and Marzeion, 2018²⁵⁷). Compared to AR5, the simulated thermal expansion from climate models has improved with a new correction for the volcanic activity (see Section 4.2.2.2.1). The glacier contribution from glacier models forced with inputs from climate models is updated with a new glacier inventory and improvements to the glacier mass balance model (Marzeion et al., 2015²⁵⁸). The simulated Greenland SMB is estimated with a new regional SMB-component downscaling technique, which accounts for the regional variations in components of the Greenland SMB (Noël et al., 2015²⁵⁹; Meyssignac et al., 2017a)²⁶⁰. In addition, an updated groundwater extraction contribution from Döll et al. (2014)²⁶¹ is now used for the land water storage contribution.

For the periods 1970–2015, 1993–2015 and 2006–2015 the simulated contributions from thermal expansion, glaciers mass loss and Greenland SMB explain respectively 84%, 81% and 77% of the observed GMSL (see Table 4.1). For all these periods the residual is consistent within uncertainty with the sum of the contribution from land water storage and ice discharge from Greenland and Antarctica. For each period the consistency is improved compared to AR5 (see Table 4.1) although the uncertainty on the residual is slightly larger because of a larger uncertainty in simulated Glaciers and Greenland SMB contributions.

For the period 1901–1990 the simulated contributions from thermal expansion, glaciers mass loss and Greenland SMB explain only 60% of the observed GMSL and the residual is too large to be explained by the sum of the contribution from land water storage and ice discharge from Greenland and Antarctica. The gap can be explained by a bias in the simulated Greenland SMB and glacier ice loss around Greenland in the early 20th century (Slangen et al., 2017b²⁶²). When the glacier model and the Greenland SMB downscaling technique are forced with observed climate from atmospheric reanalyses, rather than the simulated climate from coupled climate models, simulated SLR becomes consistent with the observed SLR (see the dashed blue line on Figure 4.5). This is because atmospheric reanalyses show an increase in air temperatures in and around Greenland over the period 1900–1940, which lead to increased melt in Greenland (Bjørk et al., 2012²⁶⁷; Fettweis et al., 2017²⁶⁸) and surrounding glaciers in the first half of the 20th century. This increase in air temperature over 1900–1940 is not reproduced by climate models (Slangen et al., 2017b²⁶⁹). It may be because this increase in air temperature was due to internal climate variability on temporal and spatial scales that cannot be precisely reproduced by climate models. It may also be due to a bias in atmospheric circulation in climate models (Fettweis et al., 2017²⁷⁰), or an issue with the spatial pattern of the historical aerosol forcing.

In summary, the agreement between climate model simulations and observations of the global thermal expansion, glacier mass loss and Greenland SMB has improved compared to AR5 for periods starting after 1970. However, for periods prior to 1970, significant discrepancies between climate models and observations arise from the inability of climate models to reproduce some observed regional changes in glacier and GIS SMB around the southern tip of Greenland. It is not clear whether this bias in climate models is due to the internal variability of the climate system or deficiencies in climate models. For this reason, there is still medium confidence in the ability of climate models to simulate past and future changes in glaciers mass loss and Greenland SMB.

Marcos and Amores (2014) found that during the period 1970–2005, 87% (95% confidence interval: 72–100%) of the observed thermosteric SLR in the upper 700 m of the ocean was anthropogenic. Slangen et al. (2014b) included the full ocean depth in their analysis. They concluded that a combination of anthropogenic and natural forcing is necessary to explain the temporal evolution of observed global mean thermosteric sea level change during the period 1957–2005. Anthropogenic forcing was responsible for the amplitude of observed thermosteric sea level change, while natural forcing caused the forced variability of observations. Observations could best be reproduced by scaling the patterns from ‘natural-only’ forcing experiments by using a factor of 0.70 ± 0.30 (2 standard deviations of the CMIP5 ensemble subset used), indicating a potential overestimation of forced variability in the CMIP5 ensemble. Patterns from the ‘anthropogenic-only’ forcing experiments needed to be scaled by a factor of 1.08 ± 0.13 (2 standard deviations of the CMIP5 ensemble subset used), indicating a realistic response of the CMIP5 ensemble to anthropogenic forcing.

For the glacier contribution to GMSL, Marzeion et al. (2014) concluded that while natural climate forcing and long-term adjustment of the glaciers to the end of the preceding Little Ice Age lead to continuous glacier mass loss throughout the simulation period of 1851–2010, the observed rates of glacier mass loss since 1990 can only be explained by including anthropogenic forcing. During the period 1851–2010, only 25 ± 35% of global glacier mass loss can be attributed to anthropogenic forcing, but 69 ± 24% during the period 1991–2010 (see Section 2.2.3 for a more detailed discussion of attribution of glacier mass change on regional scales).

There is medium confidence in evidence linking GIS mass loss to anthropogenic climate change, and low confidence in the evidence that AIS mass balance can be attributed to anthropogenic forcing (see Section 3.3.1.6 for a detailed discussion). The effects of groundwater depletion and reservoir impoundment on sea level change are anthropogenic by definition (e.g., Wada et al., 2012).

By estimating a probabilistic upper range of long-term persistent natural sea level variability, Dangendorf et al. (2015)³⁵⁶ detected a fraction of observed sea level change that is unexplained by natural variability and concluded by inference that it is virtually certain that at least 45% of the observed increase in GMSL since 1900 is attributable to anthropogenic forcing. Similarly, Becker et al. (2014) provided statistical evidence that the observed sea level trend, both in the global mean and at selected tide gauge locations, is not consistent with unforced, internal variability. They inferred that more than half of the observed GMSL trend during the 20th century is attributable to anthropogenic forcing.

Slangen et al. (2016)³⁵⁷ reconstructed GMSL from 1900 to 2005 based on CMIP5 model simulations separating individual components of radiative climate forcing and combining the contributions of thermosteric sea level change with glacier and ice sheet mass loss. They found that the naturally caused sea level change, including the long-term adjustment of sea level to climate change preceding 1900, caused 67 ± 23% of observed change from 1900 to 1950, but only 9 ± 18% between 1970 and 2005. Anthropogenic forcing was found to have caused 15 ± 55% of observed sea level change during 1900–1950, but 69 ± 31% during 1970–2005. The sum of all contributions explains only 74 ± 22% of observed GMSL change during the period 1900–2005 considering the mean of the reconstructions of Church and White (2011)³⁵⁸, Ray and Douglas (2011), Jevrejeva et al. (2014b) and Hay et al. (2015). However, the budget could be closed taking into contribution of glaciers that are missing from the global glacier inventory or have already melted (Parkes and Marzeion, 2018³⁶¹) which were not considered in Slangen et al. (2016)³⁶².

Based on these multiple lines of evidence, there is high confidence that anthropogenic forcing very likely is the dominant cause of observed GMSL rise since 1970.

The GIS is currently losing mass at roughly twice the pace of the AIS (see Chapter 3 and Table 4.1). About 60% of the mass loss between 1991 and 2015 has been attributed to increasingly negative SMB from surface melt and runoff on the lower elevations of the ice sheet margin. Ice dynamical changes and increased discharge of marine-terminating glaciers account for the remaining 40% of mass loss (Csatho et al., 2014³⁷⁶; Enderlin, 2014³⁷⁷; van den Broeke et al., 2016³⁷⁸). The ability of firn on Greenland to retain meltwater until it refreezes has diminished markedly since the late 1990s, especially in lower elevations and on peripheral ice caps (Noël et al., 2017³⁷⁹). Patterns of surface melt on Greenland are highly dependent on regional atmospheric patterns (Bevis et al., 2019³⁸⁰), adding uncertainty to future projections of SMB. Melt-albedo feedbacks associated with darkening of the ice surface from ponded water, changes in snow and firn properties, and accumulation of impurities are also important, because they can strongly enhance surface melt (Tedesco et al., 2016³⁸¹; Ryan et al., 2018³⁸²; Trusel et al., 2018³⁸³; Ryan et al., 2019³⁸⁴). These processes are not fully captured by most Greenland-scale models which is an important deficiency, because surface processes tend to dominate uncertainty in future GIS model projections (e.g., Edwards et al., 2014; Aschwanden et al., 2019³⁸⁵). Increases in meltwater and changes in the basal hydrologic regime, once thought to have a possible destabilising effect on the ice sheet (Zwally et al., 2002³⁸⁶), have been linked with recent reductions in ice velocity in western Greenland. On decadal time scales the effect of meltwater on ice dynamics are now assessed to be small (van de Wal et al., 2015³⁸⁷; Flowers, 2018³⁸⁸), which is supported by ice sheet model experiments (Shannon et al., 2013³⁸⁹). In sum, uncertain climate projections (Edwards et al., 2014³⁹⁰), albedo evolution, uncertainties around meltwater buffering by firn, complex processes linking surface, englacial and basal hydrology with ice dynamics (Goelzer et al., 2013³⁹¹; Stevens et al., 2016³⁹²; Noël et al., 2017³⁹³; Hempelmann et al., 2018³⁹⁴) and meltwater induced melting at marine-terminating ice fronts (Chauché et al., 2014³⁹⁵), and coarse spatial model resolution (Pattyn et al., 2018³⁹⁶), all continue to provide substantial challenges for ice sheet and SMB models.

Greenland-scale ice sheet modelling since AR5 (Edwards et al., 2014³⁹⁷; Fürst et al., 2015³⁹⁸; Vizcaino et al., 2015³⁹⁹; Calov et al., 2018⁴⁰⁰; Golledge et al., 2019⁴⁰¹; Aschwanden et al., 2019⁴⁰²) has built upon earlier work by coupling the ice models with regional climate models and using multiple climate and ice sheet models within single studies (Edwards et al., 2014⁴⁰³). Recent modelling studies use higher-order representations of ice flow (Fürst et al., 2015⁴⁰⁴), include more explicit representations of ice sheet processes including subglacial hydrology (Calov et al., 2018⁴⁰⁵), run the models at higher resolution and with updated boundary conditions (Aschwanden et al., 2019⁴⁰⁶), and account for two-way coupling between the ice sheet and the global ocean (Vizcaino et al., 2015⁴⁰⁷; Golledge et al., 2019⁴⁰⁸). Among these studies, Fürst et al. (2015), Vizcaino et al. (2015)⁴⁰⁹, and Aschwanden et al. (2019) provide projections following RCP2.6, RCP4.5, and RCP8.5 emissions scenarios. Calov et al. (2018)⁴¹² and Golledge et al. (2019)⁴¹³ did not consider RCP2.6. Edwards et al. (2014)⁴¹⁴ used the Special Report on Emissions Scenarios (SRES) A1B scenario which isn’t directly comparable to the other studies assessed here, but they do provide a rigorous analysis of uncertainty contributed by different climate forcings, varying simplifications of ice flow equations and height-SMB feedbacks. 

Fürst et al. (2015)⁴¹⁵ used ten different CMIP5 Atmosphere-Ocean General Circulation Model (AOGCM) simulations to provide offline SMB and ocean forcing for their Greenland-wide ice sheet model, accounting for influences of warming subsurface ocean temperatures and basal lubrication on ice dynamics. In their RCP8.5 ensemble, they found a GIS contribution to GMSL in 2100 of 10.15 cm ± 3.24 cm. Similarly, Calov et al. (2018)⁴¹⁷ found a range of GMSL contributions between 4.6–13 cm, depending on which CMIP5 GCM is used to force their regional climate model to produce SMB forcing. The wide range of RCP8.5 results in these studies highlights the substantial climate-driven uncertainty in 21st century projections of the GIS as emphasised by Edwards et al. (2014). It was found that central estimates and ranges for RCP8.5 simulated by Fürst et al. (2015), Calov et al. (2018), and Golledge et al. (2019) are in reasonable agreement with previous multi-model results (Bindschadler et al., 2013) and the assessment of AR5 (Church et al., 2013⁴¹⁹), which reported a likely RCP8.5 range of Greenland’s contribution to GMSL between 7–21 cm by 2100 (Table 4.2.). The GIS simulations provided by Vizcaino et al. (2015)⁴²⁰, using a relatively course-resolution ice model (10 km) with SMB forcing provided by a single GCM, estimate much less ice loss than other recent studies. Their GMSL projections (Vizcaino et al., 2015⁴²¹) also fall below the likely range of AR5 estimates. In contrast, the study by Aschwanden et al. (2019)⁴²² shows a significantly higher contribution to GMSL than the other studies, especially under RCP8.5 and beyond 2100 (see 4.2.3.5). This may be due to their SMB forcing, which is based on spatially uniform warming derived from future CMIP5 GCM climatologies averaged over the entire Greenland region. As noted by earlier work (e.g., Van de Wal and Wild, 2001; Gregory and Huybrechts, 2006⁴²³), this approach can overestimate melt rates in the ablation zone, which could account for their higher projected ice loss. It is noted that the process-based estimates of future GMSL rise from Greenland found in Aschwanden et al. (2019)⁴²⁴ are closest to those from an updated, structured judgement of glaciological and modelling experts (Bamber et al., 2019⁴²⁵). Calculations from the expert elicitation (Bamber et al., 2019⁴²⁶) result in higher estimates of Greenland ice loss than any of the process-based studies, with a mean and standard deviation of 33 ± 30 cm and a 17–83% range of 10–60 cm by 2100, following a climate scenario comparable to RCP8.5. The combination of the new process-based studies produces central estimates (Table 4.2) consistent with the likely ranges for Greenland’s contribution to GMSL in 2100 assessed by AR5.

Complimentary to the ice sheet scale simulations discussed above, Nick et al. (2013)⁴²⁷ used detailed flowline models of four Greenland outlet glaciers (Petermann, Kangerdlugssuaq, Jakobshavn Isbræ, and Helheim) to estimate a dynamical contribution to sea level in an RCP8.5 scenario of 11.3–17.5 mm by 2100, and 29–49 mm, by 2200. This demonstrates the limited potential of Greenland outlet glaciers alone to drive GMSL rise. Greenland-wide modelling studies (Table 4.2) consistently find a dominant role of runoff relative to dynamic discharge of ice loss, and a long-term reduction in the rate of dynamic ice discharge to the ocean as the ice sheet margin thins and the termini of outlet glaciers retreat from the coast (Goelzer et al., 2013⁴³⁸; Lipscomb et al., 2013⁴³⁹). Greenland’s bedrock geography and the limited, direct access of thick interior ice to the ocean ultimately limits the potential pace of GMSL rise from the GIS. Figure 4.7 illustrates a fundamental difference between Greenland and Antarctica. In Greenland, most of the bedrock at the ice sheet margin is above sea level (land terminating), with relatively narrow (generally <10 km wide) outlet glaciers reaching the ocean. In contrast, Antarctica has extensive areas with subglacial bedrock below sea level, and thick marine-terminating ice in direct contact with the open ocean. Recent subglacial mapping and mass conservation calculations since AR5 (Morlighem et al., 2014⁴⁴⁰; Morlighem et al., 2017⁴⁴¹) revise earlier bathymetric maps under and around the ice sheet, and reveal deeper and more extensive valley networks extending into the GIS interior than previously known. Accurate subglacial topography is important for modelling individual Greenland outlet glaciers (Aschwanden et al., 2016⁴⁴²; Morlighem et al., 2016⁴⁴³); however, the importance of these revised bedrock boundary conditions for the broader ice sheet has yet to be fully tested. Based on the limited cross sectional area of subglacial valleys and outlet glaciers on Greenland (Figure 4.7) and the results of Nick et al. (2013)⁴⁴⁴, the effects of uncertain bathymetric boundary conditions are assessed to be small relative to the uncertainties in future SMB forcing (medium confidence).

In summary, new modelling since AR5 is consistent with previous studies suggesting future Greenland ice loss over the 21st century will be dominated by surface processes, rather than dynamic ice discharge to the ocean, regardless of which emissions scenario is followed (high confidence). Based on these modelling studies, the GIS is not expected to contribute more than 20 cm of GMSL rise by 2100 in a RCP8.5 scenario, similar to the upper end of the likely range reported by AR5 (Church et al., 2013⁴⁴⁵). GIS simulations are most sensitive to uncertainties in the applied climate forcing, especially over this century (Edwards et al., 2014⁴⁴⁶), but updated climate projections since AR5 are not yet available. Because of the consistency of recent modelling with the assessment of Church et al. (2013⁴⁴⁷), Greenland’s contribution to future sea level reported in AR5 was used in our projections of GMSL.

Unlike Greenland, most of the AIS margin terminates in the ocean. The AIS also contains almost eight times more glacial ice above flotation than Greenland, and nearly half of this ice is marine-based, that is, grounded on bedrock hundreds of metres (or more) below sea level (Figure 4.7; Fretwell et al., 2013⁴⁴⁸). In places where the subglacial bedrock slopes downward away from the coast (reverse-sloped), the marine-based glacial ice is susceptible to dynamical instabilities (Weertman, 1974⁴⁴⁹; Schoof, 2007b⁴⁵⁰; Pollard et al., 2015⁴⁵¹) that can contribute rapid ice loss (Cross-Chapter Box 8 in Chapter 3). The instabilities can be triggered by the loss or thinning of ice shelves through changes in the surrounding ocean and increased sub-ice melt rates and changes in the overlying atmosphere affecting SMB and surface meltwater production. Much progress has been made since AR5 in the understanding of these processes, but their representation in continental-scale models continue to be heavily parameterised in most cases. Complex interactions between the ice sheet, ocean, atmosphere and underlying bedrock also remain difficult to simulate collectively. 

In contrast to Greenland, Antarctica’s recent contribution to SLR has been dominated by ice-dynamical processes rather than changes in SMB (Mouginot et al., 2014⁴⁵²; Rignot et al., 2014⁴⁵³; Scheuchl et al., 2016⁴⁵⁴; Shen et al., 2018⁴⁵⁵; The IMBIE team, 2018). Since AR5, it has become increasingly evident that this ice loss is being driven by sub-ice oceanic melt (thinning) of ice shelves (Paolo et al., 2015⁴⁵⁶; Wouters et al., 2015) and the resulting loss of back stress (buttressing) that impedes the seaward flow of grounded ice upstream. Elevated melt rates are generally associated with the increased presence of warm Circumpolar Deep Water (CDW) on the continental shelf (Khazendar et al., 2016⁴⁵⁷). Dynamic ice loss driven by ocean changes have also been observed on the East Antarctic margin (Li et al., 2016⁴⁵⁸; Shen et al., 2018⁴⁵⁹). This is an important development, because East Antarctica contains much more ice than West Antarctica, so even minor changes there could make major contributions to sea level in the future. 

Several of West Antarctica’s major outlet glaciers, including Pine Island Glacier, and Thwaites Glacier in the Amundsen Sea (Figure 4.8) have grounding lines currently retreating on retrograde bedrock (Rignot et al., 2014⁴⁶⁰). Thwaites Glacier is particularly important (Figure 4.8), because it extends into the interior of the WAIS, where the bed is >2000 m below sea level in places. By itself, the Thwaites drainage area contains the equivalent of ~0.4 m GMSL (Holt et al., 2006⁴⁶¹; Millan et al., 2017⁴⁶²), but loss of the glacier could have a destabilising impact on the entire WAIS (Feldmann and Levermann, 2015⁴⁶³). The WAIS contains enough ice to raise GMSL by ~3.4 m (Fretwell et al., 2013⁴⁶⁴). Since AR5, a number of ice sheet modelling studies have focussed on limited fractions of Antarctica and so are not included in estimating the SROCC Antarctic contribution to GMSL (see Section 4.2.3.2). However, these studies do allow an assessment of the potential for persistent and increasing ice loss, and the role of the marine ice sheet instability (MISI, see Cross-Chapter Box 8 in Chapter 3). 

Joughin et al. (2014)⁴⁶⁶ modelled the response of the Thwaites Glacier to a combination of elevated sub-ice melt rates and increased precipitation and found persistent future retreat, despite either the partial compensation of increased accumulation or a future reduction in melt. Sub-ice melt rates sustained at current levels were found to generate >1 mm yr–1 equivalent GMSL rise within a millennium. Higher melt rates and an assumed weak ice shelf triggered rapid retreat within a few centuries. Similarly, Waibel et al. (2018) used the BISICLES ice sheet model (Cornford et al., 2015⁴⁶⁵) to investigate the potential for self-sustained retreat of Thwaites Glacier, by incrementally increasing sub-ice melt rates until retreat is triggered, and then returning to pre-retreat melt rates. Consistent with Joughin et al. (2014)⁴⁶⁶, they found self-sustained retreat of Thwaites Glacier through MISI. Most uncertainty in their future WAIS simulations arises from uncertainties in the long-term response of Thwaites Glacier (Figure 4.8). Nias et al. (2016) demonstrated model sensitivity of Thwaites Glacier to poorly resolved bedrock boundary conditions (small scale topography), pointing to the need for better geophysical information to reduce model uncertainty (Schlegel et al., 2018⁴⁶⁸). Arthern and Williams (2017)⁴⁶⁹ used adaptive mesh techniques, but with a different formulation than Cornford et al., (2015)⁴⁷⁰, to simulate the future response of Amundsen Sea outlet glaciers. They demonstrate a sustained, but slow future retreat when sub-ice melt is maintained at current rates, and a direct relationship between the strength of ocean forcing and the pace of MISI-driven ice loss. Yu et al. (2018) simulate future Thwaites retreat using a range of model formulations with varying approximations of ice stress balance, different ocean melt schemes, and different basal friction laws. Like Arthern and Williams (2017) they find model-specific dependencies in the rate of ice loss, but all of their simulations demonstrate sustained ice loss and a bathymetrically controlled future acceleration.

Like Thwaites, the neighbouring Pine Island Glacier (PIG) has also been thinning and retreating at an accelerating rate in recent decades, in response to incursions of warm CDW in the waters underlying the glacier’s ice shelf. These incursions of CDW are controlled in part by sea floor bathymetry and climatic variability (Dutrieux et al., 2014⁴⁷¹). Favier et al. (2014) used three models with differing formulations to simulate PIG’s response to elevated sub-ice melt. Consistent with modelling of Thwaites Glacier (Joughin et al., 2014⁴⁷³), all three models demonstrate sustained future retreat at an increasing rate, as the glacier backs onto its retrograde bed. Only one of the three models used by Favier et al. (2014) demonstrates the possibility that the glacier can recover if sub-ice melt rates are reduced enough to allow the ice shelf to thicken and pin on bathymetric features to provide buttressing. These results highlight the long-term commitment to marine-based ice loss.

While limited to 50 year simulations, Seroussi et al. (2017)⁴⁷⁴ provide the first interactively coupled ice-ocean model simulations of Thwaites Glacier at a high spatial resolution. Their model demonstrates MISI-like grounding line retreat at a rate of ~1 km yr–1, comparable to observations between 1992 and 2011 (Rignot et al., 2014). The retreat is interrupted when the main trunk of the glacier stabilises on a bathymetric ridge, ~20 km upstream of the present-day grounding line (Figure 4.8), but due to the short duration of the simulation, the long-term potential for additional retreat into the interior of the ice sheet is not captured. 

Despite the use of independent model formulations, forcings, and different geographic settings, the overall agreement among these highly-resolved regional modelling studies and their ability to capture current rates of retreat, increases confidence since AR5 that observed retreat of Amundsen Sea outlet glaciers is driven by processes consistent with MISI theory (medium confidence), will continue (medium confidence), and could accelerate (medium confidence).

Observations of rapid bedrock uplift in the Amundsen Sea, low viscosity of the underlying mantle, and short GIA response times to glacial unloading suggest ice-Earth interactions could be important there (Barletta et al., 2018⁴⁷⁵). Bedrock uplift and reduced gravitational attraction between the ice sheet and ocean as an ice margin loses mass reduces RSL at the grounding line, promoting stability and providing a negative feedback on retreat (Adhikari et al., 2014⁴⁷⁶; Gomez et al., 2015⁴⁷⁷). Using a high-resolution ice sheet-Earth model, Larour et al. (2019) showed that long-term future retreat of Amundsen Sea grounding lines are slowed by these processes, but the effect is found to be minimal until after ~2250. This agrees with other recent modelling accounting for ice-Earth interactions, including the viscoelastic Earth response to changing ice loads and self-gravitation (Gomez et al., 2015⁴⁷⁸; Konrad et al., 2015⁴⁷⁹; Pollard et al., 2017⁴⁸⁰). These studies also showed a small negative feedback on future retreat over the next several centuries, particularly under strong climate forcing. However, the viscosity structure of the Earth under the AIS is not well resolved, and lateral variations in Earth structure could impact these results (Hay et al., 2017⁴⁸¹). Based on these consistent model results, and new observational evidence that PIG has been retreating on reverse-sloped bedrock for a half-century or more (Smith et al., 2017), ice-Earth interactions are not expected to substantially slow GMSL rise from marine-based ice in Antarctica over the 21st century (medium confidence). However, these processes could become important for GMSL rise on multi-century and longer time scales.

Atmospheric forcing is also becoming increasingly recognised to be an important factor for the future of the AIS. A sustained (15 days) melt event over the Ross Sea sector of the WAIS in 2016 illustrated both the connectivity of Antarctica to the tropics and El Niño, and the possibility that future meltwater production on ice shelf surfaces could change in the near future (Nicolas et al., 2017⁴⁸⁷). This was highlighted by Trusel et al. (2015)⁴⁸⁸, who evaluated the future expansion of surface meltwater using the snow component in the RACMO2 regional atmospheric model (Kuipers Munneke et al., 2012⁴⁸⁹) and output from CMIP5 GCMs. Under RCP8.5, they found a substantial expansion of surface meltwater production on ice shelves late in the 21st century that exceed melt rates observed before the 2002 collapse of the Larsen B Ice Shelf. Surface meltwater is important for both ice dynamics and SMB due to its potential to reduce albedo, saturate the firn layer, deepen surface crevasses, and to cause flexural stresses that can contribute to ice shelf break-up (hydrofracturing) (Banwell et al., 2013⁴⁹⁰; Kuipers Munneke et al., 2014⁴⁹¹). The presence of surface meltwater does not necessarily lead to immediate ice shelf collapse (Bell et al., 2017b⁴⁹²; Kingslake et al., 2017⁴⁹³), although surface meltwater was a precursor on ice shelves which have collapsed (Scambos et al., 2004⁴⁹⁴; Banwell et al., 2013⁴⁹⁵). This dichotomy illustrates the uncertain role of meltwater and the need for additional study. When and if melt rates will be sufficiently high in future warming scenarios to trigger widespread hydrofracturing is a key question, because the loss of ice shelves is associated with the onset of marine ice sheet instabilities (Cross-chapter Box 8 in Chapter 3). Based on the single modelling study by Trusel et al. (2015)⁴⁹⁶, it is not expected that widespread ice shelf loss will occur before the end of the 21st century, but due to limited observations and modelling to date, there is low confidence in this assessment.

Continental-scale ice sheet simulations are ultimately required to provide projections of future GMSL rise from Antarctica. At this spatial scale, most models rely on simplifying approximations of the equations representing three-dimensional ice flow, and in some cases they parameterise ice flow at the grounding line (Schoof, 2007b⁴⁹⁷) to improve computational efficiency. Such simplifications are necessary to allow long simulations that can be validated against geological information, in addition to modern observations (Briggs et al., 2013⁴⁹⁸; Pollard et al., 2016), however processes related to MISI are best represented at high spatial resolution and without simplifications of the underlying physics (Pattyn et al., 2013⁴⁹⁹; Reese et al., 2018c⁵⁰⁰).

Various ice sheet model formulations, including the choice of grounding line parameterisations and basal sliding schemes can strongly affect model response to a given forcing (Brondex et al., 2017⁵⁰¹; Pattyn, 2017⁵⁰²), although sophisticated statistical methodologies have been increasingly used since AR5 to quantitatively gauge model uncertainty (Bulthuis et al., 2019⁵⁰³; Edwards et al., 2019⁵⁰⁴). Accurate atmospheric forcing (SMB) and sub-ice melt are also prerequisite to resolving the time-evolving dynamics of the system, with sub-ice melt rates being particularly important (Schlegel et al., 2018⁵⁰⁵). An important ongoing deficiency is the lack of ice-ocean coupling in most continental-scale studies, which remains too computationally expensive to simulate the ocean at the spatial scales necessary to capture circulation in ice shelf cavities and time-evolving ice-ocean interactions (Donat-Magnin et al., 2017⁵⁰⁶; Hellmer et al., 2017⁵⁰⁷). Instead, melt rates are often parameterised as a depth dependent function of nearby ocean temperature derived from offline ocean models, but the lack of ice-ocean interaction can seriously overestimate melt rates in some settings (de Rydt et al., 2015; Seroussi et al., 2017⁵⁰⁸). Approaches that link offline ocean temperatures with efficient box models of the circulation in ice shelf cavities have been developed (Lazeroms et al., 2018⁵⁰⁹; Reese et al., 2018a⁵¹⁰) and used in long-term future simulations (Bulthuis et al., 2019⁵¹¹), although they still require uncoupled ocean models to provide time-evolving ocean conditions outside the cavities.

Ritz et al. (2015)⁵¹² used a hybrid physical-statistical modelling approach, whereby the timing of MISI onset is determined statistically rather than physically. They estimated probabilities of MISI onset in eleven different sectors around the ice sheet margin based on observations of continent-wide retreat and thinning over the last few decades, and expected future climate change following an IPCC SRES A1B emission scenario only. In places where MISI is projected to begin, the persistence and rate of grounding-line retreat is parameterised as a function of the local bedrock topography (slope), ice thickness at grounding lines following Schoof (2007b), and basal friction. This study represents a statistically rigorous approach in which model parameters are based on a synthesis of observations and projected surface and sub-shelf forcing, rather than coming directly from climate and ocean models. However, the model calibrations rely on recent observations, which may not provide adequate guidance under warmer future conditions.

Levermann et al. (2014)⁵¹³ use simplified emulations of temperature increase in order to estimate both SMB and sub-ice melt (including a parameterised delay for ocean warming) to determine the linearised response of five ice sheet models calibrated against recent rates of retreat. Substantial uncertainty arises from the different model treatments of grounding line dynamics and ice shelves. However, they conclude that the single greatest source of uncertainty stems from the external forcing. Golledge et al. (2015)⁵¹⁴ used PISM (Parallel Ice Sheet Model; Winkelmann et al., 2011⁵¹⁵) to simulate the future response of the AIS to RCP emission scenarios. PISM links grounded, streaming, and shelf flow, and has freely evolving grounding lines required to capture MISI. PISM’s parameterised treatment of sub-ice melt applies melt under partially grounded grid cells (Feldmann and Levermann, 2015⁵¹⁶), making the model sensitive to subsurface ocean warming, although the validity of this approach is contested (Arthern and Williams, 2017⁵¹⁷; Seroussi and Morlighem, 2018⁵¹⁸; Yu et al., 2018⁵¹⁹). While providing alternative outcomes with the two basal melt rate parameterisations, the model is not calibrated to observations and doesn’t provide a probability distribution. In a subsequent study Golledge et al. (2019)⁵²⁰ used PISM, but with updated RCP climate forcing based on CMIP5 GCMs, and with sub-ice ocean melt calibrated to observations. An offline, intermediate-complexity climate model was used to capture global ice-climate feedbacks ignored in most other studies, but the simulations only include RCP4.5 and RCP8.5 and do not extend beyond 2100. Accounting for the climatic effects of meltwater input from Greenland and Antarctica nearly doubled their estimates of Antarctic’s contribution to GMSL in 2100 from 2.4 cm to 4.6 cm in RCP4.5, and from 7.7 cm to 14 cm in RCP8.5. The increase is caused by a combination of SMB decrease over the WAIS, combined with subsurface ocean warming that increases sub-ice melt. However, the climate model used to diagnose the spatial patterns of the atmospheric and oceanic response to the meltwater input is simplistic. Bronselaer et al. (2018)⁵²¹ tested the global climatic response to future meltwater input from Antarctica using an ensemble of GCM simulations, but without an interactive ice sheet. They simulated an RCP8.5 scenario with and without a massive input of meltwater into the Southern Ocean and demonstrate that the addition of Antarctic meltwater expands sea ice in the Southern Ocean, delays the trajectory of global warming, and moderates atmospheric warming around the Antarctic coastline. Consistent with Golledge et al. (2019)⁵²², they found meltwater-induced stratification around Antarctica warms subsurface ocean temperatures, indicating the potential for a positive meltwater feedback on ice shelf melt. These studies reinforce the need for continental-scale studies to consider two-way ice-climate coupling, but with limited published studies to draw from and no simulations run beyond 2100, firm conclusions regarding the net importance of atmospheric versus ocean melt feedbacks on the long-term future of Antarctica can not be made.

Bulthuis et al. (2019)⁵²⁶ used a different continental-scale ice sheet model (Pattyn, 2017⁵²⁵) with the same simplified atmospheric and ocean forcing used by Golledge et al. (2015)⁵²⁸ to simulate RCP2.6, RCP4.5, and RCP8.5 scenarios. Simulations with varying model parameters were used to quantify uncertainties related to the atmospheric forcing, various ice-model physics, and bedrock response to changing ice loads. A key finding was that irrespective of model parametric uncertainty, the strongly mitigated RCP2.6 scenario prevents catastrophic WAIS collapse over the coming centuries. The probabilistic projections of Antarctic GMSL contributions (Bulthuis et al., 2019⁵³¹) represent a rigorous blending of physical ice sheet modelling and uncertainty quantification (UQ) techniques, albeit with a simplistic representation of future climate and using a relatively coarse-resolution ice sheet model. These results are well-supported by Schlegel et al. (2018), who blend UQ with a higher resolution ice sheet model than used by Bulthuis et al. (2019), but using an idealised climate forcing scheme not directly linked to time-evolving future climate trajectories. Their 800 simulations, run to 2100, provide not only probabilistic constraints on future GMSL-rise from Antarctica, but an assessment of key drivers of uncertainty, including uniform and regional dependencies on model physical parameters, climate forcing, and boundary conditions. Sub-ice shelf melt rates provide the greatest source of uncertainty in their projections, although the source region dominating the GMSL contribution is found to be dependent on the climate forcing applied, and different from those found by Golledge et al. (2015)⁵³⁰.

DeConto and Pollard (2016)⁵²⁹ used an ice sheet model with a formulation similar to that used by Golledge et al. (2015) and Bulthuis et al. (2019) but they include glaciological processes not accounted for in other continental-scale models: 1) surface melt and rain water influence on hydrofracturing of ice shelves; and 2) brittle failure of thick, marine-terminating ice fronts that have lost their buttressing ice shelves. Where the ice fronts are thick enough to form tall ice cliffs above the waterline, they can produce stresses exceeding the strength of the ice, causing calving (Bassis and Walker, 2012⁵³²). Once initiated, ice-cliff calving has been hypothesised to produce a self-sustaining Marine Ice Cliff Instability (MICI; Cross-chapter Box 8, Chapter 3). The validity of MICI remains unproven (Edwards et al., 2019⁵³³) and is considered to be characterised by ‘deep uncertainty’, but it has the potential to raise GMSL faster than MISI. DeConto and Pollard (2016) represent hydrofracturing and ice-cliff calving with simple parameterisations, but the glaciological processes themselves are supported by more detailed modelling and observations (Scambos et al., 2009⁵³⁴; Banwell et al., 2013⁵³⁵; Ma et al., 2017⁵³⁶; Wise et al., 2017⁵³⁷; Parizek et al., 2019⁵³⁸). DeConto and Pollard (2016)⁵³⁹ provide four ensembles for RCP2.6, RCP4.5, and RCP8.5 scenarios, representing two alternative ocean model treatments and two alternative palaeo sea level targets used to tune their model physical parameters. However, their ensembles do not explore the full range of model parameter space or provide a probabilistic assessment (Kopp et al., 2017⁵⁴⁰; Edwards et al., 2019⁵⁴¹). Under RCP2.6, DeConto and Pollard (2016) find very little GMSL rise from Antarctica by 2100 (0.02–0.16 m), consistent with the findings of Golledge et al. (2015)⁵⁴² and Bulthuis et al. (2019)⁵⁴³. In contrast, their four ensemble means range between 0.26–0.58 m for RCP4.5, and 0.64–1.14 m for RCP8.5. In RCP8.5, rates of GMSL rise from Antarctica exceed 5 cm yr-1 in the 22nd century and contribute as much as 15 m of GMSL rise by 2500, largely due to the ice cliff calving process. The climate forcing used by DeConto and Pollard (2016)⁵⁴⁴ simulates the appearance of extensive surface meltwater several decades earlier than indicated by other CMIP5 climate simulations (Trusel et al., 2015⁵⁴⁵). Because their model physics are sensitive to melt water through hydrofracturing, this makes the timing and magnitude of their simulated ice loss too uncertain to include in SROCC sea level projections. However, their results do demonstrate the potential for brittle ice sheet processes not considered by AR5 to exert a strong influence on future rates of GMSL rise and the possibility that GMSL beyond 2100 could be considerably higher than the likely range projected by models that do not include these processes.

Thermal expansion: Global mean thermal expansion projections rely on coupled climate models projections (Kopp et al., 2014⁶⁰⁸; Slangen et al., 2014a⁶⁰⁹; Jackson and Jevrejeva, 2016⁶¹⁰) or simple climate model projections (Perrette et al., 2013⁶¹¹; Nauels et al., 2017b⁶¹²; Wong et al., 2017⁶¹³), and are substantively unchanged since AR5. For those studies relying on the CMIP5 GCM ensemble, interpretations of the model output differ mainly with regard to how the range is understood. For example, Kopp et al. (2014)⁶¹⁴, interprets the 5–95 percentile of CMIP5 values as a likely range of thermal expansion. The differences among the studies yield discrepancies smaller than 10 cm, e.g., Slangen et al. (2014a)⁶¹⁵ use 20–36 cm in 2081–2100 with respect to 1986–2005, while (Kopp et al., 2014) project a likely range of 28–46 cm in 2081–2099 with respect to 1991–2009.

Glaciers: Projections of glacier mass change rely either on models of glacier SMB and geometry, forced by temperature and precipitation fields (Slangen and Van de Wal, 2011⁶¹⁵; Marzeion et al., 2012⁶¹⁶; Hirabayashi et al., 2013⁶¹⁷; Radić et al., 2014⁶¹⁸; Huss and Hock, 2015⁶¹⁹), or simple scaling relationships with global mean temperature (Perrette et al., 2013⁶²⁰; Bakker et al., 2017⁶²¹; Nauels et al., 2017a⁶²²). Glacier mass change projections published since AR5, based on newly developed glacier models, confirm the overall assessment of AR5 (see also Section 4.2.3.2).

Land water storage: Projections of the GMSL rise contributions due to dam impoundment and groundwater withdrawal are generally either calibrated to hydrological models (e.g., Wada et al., 2012) or neglected. Recent coupled climate-hydrological modelling suggests that a significant minority of pumped groundwater remains on land, which may reduce total GMSL rise relative to studies assuming full drainage to the ocean (Wada et al., 2016⁶²³). Kopp et al. (2014) estimated land water storage based on population projections. However, there are no substantive updates to projections of the future land-water storage contribution to GMSL rise since AR5.

Ice sheets: GMSL projections in previous IPCC assessments were based on results from physical models of varying degree of complexity interpreted using expert judgment of the assessment authors (Meehl et al., 2007⁶²⁵; Church et al., 2013⁶²⁶). AR5 (Church et al., 2013⁶²⁷) used this approach and is partly based on the assessment of statistical-physical modelling of the Antarctic contribution (Little et al., 2013⁶²⁸). As an alternative to the model-based approach, several studies have applied structured expert elicitation to the GMSL contribution of ice sheets. This approach is based on a more formal expert elicitation protocol (Cooke, 1991⁶²⁹; Bamber and Aspinall, 2013⁶³⁰; Bamber et al., 2019⁶³¹) instead of physically based models. Combining the Antarctic contribution from the expert elicitation with the non-Antarctic components from AR5 as done for Table 4.4 leads to an estimated SLR of 0.95 m (median) for the high scenario and an upper likely range of 1.32 m (Figure 4.2), which is slightly higher than the process-based results. Results by Bamber and Aspinall (2013)⁶³² were criticised because of their procedure for post-processing the expert data of individual ice sheets to a total sea level contribution from the ice sheets (de Vries and van de Wal, 2015; Bamber et al., 2016; de Vries and van de Wal, 2016). Bamber et al. (2019) avoids this issue by eliciting expert judgments about ice sheet dependence. Alternatively, Horton et al. (2014) used a simpler elicitation protocol focusing on the total SLR rather than the ice sheet contribution alone. Finally, several probabilistic studies (e.g., Bakker et al., 2017; Kopp et al., 2017⁶³³; Le Bars et al., 2017) used the results of a single ice sheet model study from DeConto and Pollard (2016)⁶³⁴ as the Antarctic contribution to GMSL.

Beside the total contribution of ice sheets several studies address the individual contribution of either Greenland or Antarctica (see Section 4.2.3.1.1 and 4.2.3.1.2) based on ice dynamical studies. Critical for GMSL projections is the low confidence in the dynamic contribution of the AIS beyond 2050 in previous assessments, as discussed in Section 4.2.3.1.2.

Differences between GMSL and RSL change are driven by three main factors: (1) changes in the ocean, for instance, the thermal expansion component and the circulation driven changes, (2) gravitational and rotational effects caused by redistribution of mass within cryosphere and hydrosphere, leading to spatial patterns, and (3) long term processes caused by GIA that lead to horizontal and VLM. Finally, the inverse barometer effect caused by changes in the atmospheric pressure, sometimes neglected in projections, can also make a small contribution, particularly on shorter time scales. For the 21st century as a whole, estimates of the latter are smaller than 5 cm at local scales (Church et al., 2013⁶³⁵; Carson et al., 2016⁶³⁶). 

Ocean Dynamic sea level: Projections of dynamic sea level change are necessarily derived through interpretations of coupled climate model projections. As with thermal expansion projections, interpretations of the CMIP5 ensemble differ with regard to how the model range is understood and the manner of drift correction, if any (Jackson and Jevrejeva, 2016⁶³⁷). However, relative to tide-gauge observations, coupled climate models tend to overestimate the memory in dynamic sea level; thus, they may underestimate the emergence of the externally forced signal of DSL change above scenario uncertainty (Becker et al., 2016⁶³⁸). ODSL from coupled climate models does not include the changes resulting from ice melt because ice melt is calculated off-line.

Gravitational-rotational and deformational effects (GRD; Gregory et al., 2019⁶³⁸): All projections of RSL change include spatial patterns in sea level for cryospheric changes, which however may differ in the details with which these are represented. Some studies also include a spatial pattern for land-water storage change (Slangen et al., 2014a⁶⁴⁰), anthropogenic subsidence is not included. Recent work indicates that, for some regions with low mantle viscosity, spatial patterns cannot be treated as fixed on multi-century time scales (Hay et al., 2017⁶⁴¹). This effect has not yet been incorporated into comprehensive RSL projections, but is probably only of relevance near ice sheets. For adaptation purposes, Larour et al. (2017) developed a mapping method to indicate which areas of ice mass loss are important for which major port city. There is high confidence in the patterns caused by GRD, as in AR5.

Vertical land motion (VLM): These processes can be an important driver of RSL change, particularly in the near- to intermediate-field of the large ice sheets of the LGM (e.g., North America and northern Europe). This process is incorporated either by physical modelling (Slangen et al., 2014a⁶⁴³) or by estimation of a long-term trend from tide-gauge data (e.g., Kopp et al., 2014), which is then spatially extrapolated. In the former case, only the long-term GIA process is included in the projections, but it excludes other important local factors contributing to VLM (e.g., tectonic uplift/subsidence and groundwater/hydrocarbon withdrawal); by using only tide gauge measurements, projections may assume that these other processes proceed at a steady rate and thus do not allow for management changes that affect groundwater extraction.

Semi-empirical models provide an alternative approach to process-based models aiming to close the budget between the observed SLR and the sum of the different components contributing to SLR. In general, motivated by a mechanistic understanding, semi-empirical models use statistical correlations from time series analysis of observations to generate projections (Rahmstorf, 2007⁶⁴⁴; Vermeer and Rahmstorf, 2009⁶⁴⁵; Grinsted et al., 2010⁶⁴⁶; Kemp et al., 2011⁶⁴⁷; Kopp et al., 2016⁶⁴⁸). They implicitly assume that the processes driving the observations and feedback mechanisms remain similar over the past and future. In the past, differences between semi-empirical projections and process-based models were significant but for more recent studies the differences are vanishingly small. Ongoing advances in closing the sea level budget and in the process understanding of the dynamics of ice have reduced the salience of estimates from semi-empirical models. Moreover, the results from semi-empirical models (Kopp et al., 2016⁶⁸²; Mengel et al., 2016⁶⁸³) are in general agreement with Church et al. (2013)⁶⁸⁴, except when those results reflect the combined hydrofracturing and ice cliff instability mechanism as presented by DeConto and Pollard (2016)⁶⁸⁵. At the same time, semi-empirical models based on past observations capture poorly or miss altogether the recent observed changes in Antarctica. MISI may lend a very different character to ice sheet evolution in the near future than in the recent past and hydrofracturing remains impossible to quantify from observational records only. For this reason, a new generation of semi-empirical models and emulators has been developed that estimate individual components of SLR, which the former models do not (Mengel et al., 2018⁶⁸⁶). These newer models aim to emulate the response of more complex models providing more detailed information for different climate scenarios or probability estimates than process-based models (Bakker et al., 2017⁶⁸⁷; Nauels et al., 2017a⁶⁸⁸; Wong et al., 2017⁶⁸⁹; Edwards et al., 2019⁶⁹⁰).

A wide range of probabilistic sea level projections exist, ranging from simple scaling relations to partly process-based components combined with scaling relations. Table 4.5 illustrates the overlap between many of the studies, a complete overview is presented by Garner et al. (2017), and differences between different classes of models are discussed in Horton et al. (2018)⁶⁹². Many studies rely on CMIP simulations for an important part of their sea level components. The largest difference can be found in the treatment of the ice dynamics, particularly for Antarctica, which are usually not CMIP5 based. Instead, each derives from one of several estimates of the Antarctic contribution. These results are useful for the purposes of elucidating sensitivities of process-based studies and effects of changing components to the total projection. This report relies on the Antarctic component from Section 4.2.3.2 for calculating the likely range of RSL. Hence the values in Table 4.5 are not used for the final assessment of RSL including the SROCC specific Antarctic contribution presented in Section 4.2.3.2. Comparing the probabilistic projections (Table 4.6) is difficult because of the subtle differences between their assumptions. Nevertheless, values range much more for 2100 than for 2050.

Changes in ESL are presented here, based on the projections as presented in 4.2.3.2 at the tide gauge locations in the GESLA2 database (Woodworth et al., 2016⁷¹²). Results include GIA effects, but anthropogenic subsidence is not prescribed. These calculations serve as a signal to guide adaption to SLR (Stephens et al., 2018⁷¹³). Return periods are calculated as a combination of regional RSL projections and a probabilistic characterisation of the variability in sea level as derived from the GESLA2 data set which contains a quasi-global set of tide gauges. By doing so, it is assumed that the variability in the tide gauge record does not change over time. Models are not accurate enough to address whether this is correct or not.

To quantify the average return period of ESL events, a peak-over-threshold method is applied following Arns et al. (2013)⁷¹⁴ and Wahl et al. (2017)⁷¹⁵. Tide gauge records are detrended by subtracting a running mean of one year. Peaks above the 99th percentile of hourly water levels are extracted and declustered by applying a minimum time between peaks of 72 hours. This threshold of 99% was recommended by Wahl et al. (2017)⁷¹⁶ for global applications. Using a maximum likelihood estimator, a Generalized Pareto Distribution (GPD) is fitted to these peaks, allowing for an extrapolation to return periods beyond the available period of observations. Changes in ESL events due to regional mean SLR are quantified following Hunter (2010)⁷¹⁷. Uncertainties in the GPD parameters and projections are propagated using a Monte Carlo approach, from which a best estimate is derived (see SM4.2). Only tide gauge records of 20 years of longer, which are at least 70% complete, are used. However, as can be seen for Guam (Fig 4.9), this does not ensure a good fit of the GPD to all peaks, as rare events may have been captured in this relatively short record. 

Projected changes in ESL events are shown for 12 selected tide gauges in Figure 4.11. The magnitude of these changes depends on the relation between ESL events and the associated return periods, as well as regional sea level projections, and the uncertainty therein (see inset Figure 4.11). The change in ESL events is commonly expressed in terms of the amplification factor and the allowance. The amplification factor denotes the amplification in the average occurrence frequency of a certain extreme event, often referenced to the water level with a 100-year return period during the historic period. The allowance denotes the increased height of the water level with a given return period. This allowance equals the regional projection of SLR with an additional height related to the uncertainty in the projection (Hunter, 2012⁷¹⁸).

Amplification factors are strongly determined by the local variability in ESL events. Locations where this variability is large due to large storm surges and astronomical tides (e.g., Cuxhaven, see Figure 4.9) will experience a relatively moderate amplification of the occurrence frequency of extremes. In comparison, locations with small variability in ESL events (e.g., Lautoka and Papeete) will experience large amplifications even for a moderate rise in mean sea level (Vitousek et al., 2017⁷¹⁹). Globally, this contrast between regions with large and small amplification factors becomes clear for projections by mid-century (Fig 4.11, left panels). Although regional differences in projected mean SLR are small for the coming centuries, regional contrasts in amplification factors are considerable. In particular, many coastal areas in the lower latitudes may expect amplification factors of 100 or larger by mid-century, regardless of the scenario as also shown in SR15 and Rasmussen et al. (2018). This indicates that, at these locations, water levels with return periods of 100 years during recent past will become annual or more frequent events by mid-century. By end-century and in particular under RCP8.5, such amplification factors are widespread along the global coastlines (Vousdoukas et al., 2018a⁷²⁰).

In summary, ESL events estimates as presented in this subsection, clearly show that as a consequence of SLR, events which are currently rare (e.g., with an average return period of 100 years), will occur annually or more frequently at most available locations for RCP8.5 by the end of the century (high confidence). For some locations, this change will occur as soon as mid-century for RCP8.5 and by 2100 for all emission scenarios. The affected locations are particularly located in low-latitude regions, away from the tropical cyclone (TC) tracks. In these locations, historical sea level variability due to tides and storm surges is small compared to projected mean SLR. Therefore, even limited changes in mean sea level will have a noticeable effect on ESLs, and for some locations, even RCP2.6 will lead to the annual occurrence of historically rare events by mid-century. Results should be treated with caution in regions where TCs are important as they are underrepresented in the observations (Haigh et al., 2014a⁷²¹).

Gender inequality came to prominence only recently in climate change studies (~15 years ago; see Pearse, 2017¹⁰⁹¹). In light of sea-related hazards and SLR specifically, the issue is still mainly investigated in the context of developing countries, although growing attention is paid to the issue in developed countries (e.g., Lee et al., 2015; Pearse, 2017). Recent studies in southern coastal Bangladesh, for example, show that women get less access than men to climate- and disaster-related information (both emergency information and training programmes), decision making processes at the household and community levels, economic resources including financial means such as micro-credit, land ownership, and mobility within and outside the villages (Rahman, 2013¹⁰⁹²; Alam and Rahman, 2014¹⁰⁹³; Garai, 2016¹⁰⁹⁴). Gender inequity may be inherent in unfavourable background conditions (higher illiteracy rates, deficiencies in food and calories intake and poorer health conditions) as a result of, among other things, traditions, social norms and patriarchy. Together, these barriers disadvantage women more than men in developing effective responses to anticipate gradual environmental changes such as persistent coastal erosion, flooding and soil salinisation (medium evidence, high agreement). Such conclusions are in line with the literature on gender inequality and climate change at large (Alston, 2013¹⁰⁹⁵; Pearse, 2017¹⁰⁹⁶), thus suggesting no major SLR-inherent specificities.

Despite the identification of this issue in AR4, its treatment in AR5 remained limited. Recent literature partly focussing on SLR reaffirms that indigenous knowledge and local knowledge (IK and LK; Cross-Chapter Box 4 in Chapter 1 and Glossary) are key to determining how people recognise and respond to environmental risk (Bridges and McClatchey, 2009¹⁰⁹⁷; Lefale, 2010¹⁰⁹⁸; Leonard et al., 2013¹⁰⁹⁹; Lazrus, 2015¹¹⁰⁰), and therefore to increasing adaptive capacity and reducing long-term vulnerability (Ignatowski and Rosales, 2013¹¹⁰¹; McMillen et al., 2014¹¹⁰²; Hesed and Paolisso, 2015¹¹⁰³; Janif et al., 2016¹¹⁰⁴; Morrison, 2017¹¹⁰⁵).

IK and LK contribute both as a foundation for and an outcome of customary resource management systems aimed at regulating resource use and securing critical ecosystem protection (examples in Indonesia; Hiwasaki et al., 2015¹¹⁰⁶), structuring the relationship between people and authorities, and framing and maintaining a strong sense of place in the community (examples in Timor Leste; Hiwasaki et al., 2015¹¹⁰⁷). In turn, this allows local communities to predict and prepare for both sudden shock events that have historical precedent and, when IK and LK are embedded in day-to-day rituals and decision making processes, to also anticipate the consequences of gradual changes, as in sea level (examples in Indonesia; Hiwasaki et al., 2015¹¹⁰⁸). Customary resource management systems based on IK and elders’ leadership—for instance, Rahui in French Polynesia (Gharasian, 2016¹¹⁰⁹), or Mo in the Marshall Islands (Bridges and McClatchey, 2009)—also allow communities to diversify access to marine and terrestrial resources using seasonal calendars, to ensure collective food and water security, and to maintain ecological integrity (McMillen et al., 2014¹¹¹¹). In rural Pacific atolls, traditional food preservation and storage (e.g., storing germinated coconuts or drying fish) still play a role in anticipating disruptions in natural resource availability (Campbell, 2015¹¹¹²; Lazrus, 2015¹¹¹³). Such practices have enabled the survival of isolated communities from the Arctic to tropical islands in constraining sea environments for centuries to millennia (McMillen et al., 2014¹¹¹⁴; Nunn et al., 2017a¹¹¹⁵). Morrison (2017) argues that IK and LK can also play a role in supporting internal migration in response to SLR, by avoiding social and cultural uprooting (Cross-Chapter Box 4 in Chapter 1).

In some specific contexts, climate change will also imply no-analogue changes, such as rapid ice-melt and changing conditions in the Arctic that have no precedent in the modern era, and could thus limit the relevance of IK and LK in efforts to address significantly different circumstances. Except in these specific situations, the literature suggests that the loss of IK and LK, and related social norms and mechanisms, will increase populations’ exposure and vulnerability to SLR impacts (Nakashima et al., 2012¹¹¹⁷). The literature notably points out that modern, externally-driven socioeconomic dynamics, such as the introduction of imported food (noodles, rice, canned meat and fish, etc.), diminish the cultural importance of IK-based practices and diets locally, together with introducing dependency on monetisation and external markets (Hay, 2013; Campbell, 2015¹¹¹⁸).

As a result, the loss of IK and LK may increase long-term vulnerability to SLR (medium evidence, high agreement). Given that IK and LK are largely based on observing and ‘making sense’ of the surrounding environment (moon, waves, winds, animal behaviours, topography, etc.), such a loss reflects a more general concern about the weakening of environmental connectedness in contemporary societies, which is not limited to remote, rural and developing communities (medium confidence). In developed contexts too, the loss of LK has played a critical role in recent coastal disasters (e.g., Katrina in 2005 in the USA, Kates et al., 2006) and increasing vulnerability to SLR (e.g., Newton and Weichselgartner, 2014; Wong et al., 2014¹¹¹⁹).

Coastal communities draw on social structures and capabilities that can reduce risk and increase adaptive capacity in the face of coastal hazards (Aldrich, 2017¹¹²⁰; Petzold, 2018¹¹²¹). Although the term is subject to debate (Meyer, 2018¹¹²²), social capital—that is, the level of cohesion between individuals, between groups of individuals, and between people and institutions, within and between communities—is considered to be a key enabler for collective action to reduce risk and build adaptive capacity (Adger, 2010¹¹²³; Aldrich and Meyer, 2015¹¹²⁴; Petzold and Ratter, 2015¹¹²⁵). Levels of social capital can be influenced by underlying social processes, such as socioeconomic (in)equalities, gender issues, health, social networks and social media. It applies to both developing and developed countries, for example in densely populated deltas (Jordan, 2015¹¹²⁶), European coasts (Jones and Clark, 2014¹¹²⁷; Petzold, 2016¹¹²⁸), Asian urban or semi-urban coastal areas (Lo et al., 2015¹¹²⁹; Triyanti et al., 2017¹¹³⁰) and Pacific islands (Neef et al., 2018¹¹³¹). Social capital framed as an enabler for reducing vulnerability has been studied in the context of extreme events (risk prevention mechanisms, emergency responses and post-crisis actions) and collective environmental management (e.g., replanting mangroves, beach cleaning, etc.). Social capital also enables adaptation prospects. For example, its role has been explored in public acceptability of long-term coastal adaptation policies in the UK (Jones and Clark, 2014¹¹³²; Jones et al., 2015¹¹³³). The role of social capital in building resilience to climate stress in coastal Bangladesh was explored by Jordan (2015), who found complex and even contradictory interactions between social capital and resilience to climate stress. Among others, Jordan (2015) also advises caution about uncritical importation of such Westernised concepts in seeking to understand and address coastal vulnerability in developing countries.

Risk perception, which is context-specific and varies from one individual to another, may influence communities’ exposure and vulnerability as it shapes authorities’ and people’s attitudes towards sudden and slow onset hazards, as shown by Terpstra (2011)¹¹³⁴, Lazrus (2015), Elrick-Barr et al. (2017)¹¹³⁵ and O’Neill et al. (2016) in the Netherlands, Tuvalu, Australia and Ireland, respectively. The progressive discounting of coastal hazard risks and subsequent loss of risk memory also played a role in coastal disasters such as Hurricane Katrina in 2005 in the USA (Burby, 2006¹¹³⁶; Kates et al., 2006¹¹³⁷) and Storm Xynthia in 2010 in France (Vinet et al., 2012¹¹³⁸; Genovese and Przyluski, 2013¹¹³⁹; Chadenas et al., 2014¹¹⁴⁰).

Risk perceptions stem from intertwined predictors such as ‘gender, political party identification, cause-knowledge, impact-knowledge, response-knowledge, holistic affect, personal experience with extreme weather events, [social norms] and biospheric value orientations’ (Kellens et al., 2011¹¹⁴¹; Carlton and Jacobson, 2013¹¹⁴²; Lujala et al., 2015¹¹⁴³; van der Linden, 2015, p. 112; Weber, 2016¹¹⁴⁴; Elrick-Barr et al., 2017¹¹⁴⁵; Goeldner-Gianella et al., 2019¹¹⁴⁶). In general, there is a lack of education, training and thus knowledge and literacy on recent and projected trends in sea level, which compromises ownership of science facts and projections at all levels, from individuals and institutions to society at large.

While some studies have begun to highlight the influence of the distance from the sea on risk perceptions (Milfont et al., 2014¹¹⁴⁷; Lujala et al., 2015¹¹⁴⁸; O’Neill et al., 2016), there is still little knowledge about how risk perceptions vary across different geographical and social contexts, and how this influences exposure and vulnerability to coastal hazards (e.g., Terpstra, 2011; van der Linden, 2015¹¹⁴⁹). There is a critical lack of studies specifically addressing SLR. Some recent works conducted in coastal Australia suggest that while people are confident about their ability to cope with an already experienced event, when it comes to SLR, the dominant narrative is articulated around the barriers related to the ‘uncertainty in the nature and scale of the impacts as well as the response options available’ (Elrick-Barr et al., 2017, p. 1147). Similar conclusions have been highlighted in the Caribbean islands of St. Vincent (Smith, 2018¹¹⁵⁰) and the Bahamas (Thomas and Benjamin, 2018¹¹⁵¹). SLR is rarely addressed separately from sea-related extreme events, which masks a crucial difference between already-observed and delayed impacts. Climate change is considered a “distant psychological risk” (Spence et al., 2012¹¹⁵²), making it and SLR per se ‘markedly different from the way that our ancestors have traditionally perceived threats in their local environment’ (Milfont et al., 2014¹¹⁵³; Lujala et al., 2015¹¹⁵⁴; van der Linden, 2015, p. 112; O’Neill et al., 2016).

SLR will affect agriculture mainly through land submergence, soil and fresh groundwater resources salinisation, and land loss due to permanent coastal erosion, with consequences on production, livelihood diversification and food security, especially in heavily coastal agriculture-dependent countries such as Bangladesh (Khanom, 2016¹³41). Recent literature confirms that salinisation is already a major problem for traditional agriculture in deltas (Wong et al., 2014¹³⁴²; Khai et al., 2018¹³⁴³) and low-lying island nations where some edible cultivated plants such as taro patches are threatened (Nunn et al., 2017b¹³⁴⁴). Taking the case of rice cultivation, recent works emphasise the prevailing role of combined surface elevation and soil salinity, such as in the Mekong delta (Vietnam; Smajgl et al., 2015¹³⁴⁵) and in the Ebro delta (Spain; Genua-Olmedo et al., 2016¹³⁴⁶), estimating for the latter a decrease in the rice production index from 61.2% in 2010 to 33.8% by 2100 in a 1.8 m SLR scenario. For seven wetland species occurring in coastal freshwater marshes in central Veracruz on the Gulf of Mexico, an increase in salinity was shown to affect the germination process under wetland salt intrusion (Sánchez-García et al., 2017¹³⁴⁷). In coastal Bangladesh, oilseed, sugarcane and jute cultivation was reported to be already discontinued due to challenges to cope with current salinity levels (Khanom, 2016¹³⁴⁸), and salinity is projected to have an unambiguously negative influence on all dry-season crops over the next 15–45 years (especially in the southwest; Clarke et al., 2018¹³⁴⁹; Kabir et al., 2018¹³⁵⁰). Salinity intrusion and salinisation can trigger land use changes towards brackish or saline aquaculture such as shrimp or rice-shrimp systems with impacts on environment, livelihoods and income stability (Renaud et al., 2015¹³⁵¹). However, increasing salinity is only one of the land use change drivers along with, for example, policy changes and market prices at the household level (Renaud et al., 2015¹³⁵²).

SLR may significantly affect tourism and recreation through impacts on landscapes (e.g., beaches), cultural features (e.g., Marzeion and Levermann, 2014; Fang et al., 2016¹³⁵³), and critical transportation infrastructures such as harbours and airports (Monioudi et al., 2018¹³⁵⁴). Coastal areas’ future tourism and recreation attractiveness will however also depend on changes in air temperature, seasonality and sea surface temperature (including induced effects such as invasive species, e.g., jellyfishes, and disease spreading; Burge et al., 2014¹³⁵⁵; Weatherdon et al., 2016¹³⁵⁶; Hoegh-Guldberg et al., 2018¹³⁵⁷; Section 5.4.2). Future changes in climatic conditions in tourists’ areas of origin will also play a role in reshaping tourism flows (Bujosa and Rosselló, 2013¹³⁵⁸; Amelung and Nicholls, 2014¹³⁵⁹), in addition to mitigation policies on air transportation, non-climatic features (e.g., accommodation and travel prices) and tourists’ and tourism developers’ perceptions of climate-related changes (Shakeela and Becken, 2015¹³⁶⁰). Since AR5, forecasting the consequences of climate change effects on global-to-local tourism flows has remained challenging (Rosselló-Nadal, 2014¹³⁶¹; Wong et al., 2014¹³⁶²; Hoegh-Guldberg et al., 2018¹³⁶³). There are also concerns about the effect of SLR on tourism facilities, for example hotels in Ghana (Sagoe-Addy and Addo, 2013¹³⁶⁴), in a context where tourism infrastructure often contributes to the degradation of natural buffering environments through, for example, coastal squeeze (e.g., Section 4.3.2.4) and human-driven coastal erosion. Again, forecasting is constrained by the lack of scientific studies on tourism stakeholders’ long-term strategies and adaptive capacity (Hoogendoorn and Fitchett, 2018¹³⁶⁵).

Recent studies support the AR5 conclusion that ocean warming and acidification are considered more influential drivers of change in fisheries and aquaculture than SLR (Larsen et al., 2014; Nurse et al., 2014¹³⁶⁶; Wong et al., 2014¹³⁶⁷). The negative effects of SLR on fisheries and aquaculture are indirect, through adverse impacts on habitats (e.g., coral reef degradation, reduced water quality in deltas and estuarine environments, soil salinisation, etc.), as well as on facilities (e.g., damage to small and large harbours). This makes future projections on SLR implications for coastal and marine fisheries and aquaculture an understudied field of research. Conclusions only state that future impacts will be highly context-specific due to local manifestations of SLR and local fishery-dependent communities’ ability to adapt to alterations in fish and aquaculture conditions and productivity (Hollowed et al., 2013¹³⁶⁸; Weatherdon et al., 2016¹³⁶⁹). Salinity intrusion also contributes to conversion of land or freshwater ponds to brackish or saline aquaculture in many low-lying coastal areas of Southeast Asia such as in the Mekong Delta in Vietnam (Renaud et al., 2015¹³⁷⁰).

Social values refer to what people consider of critical importance about the places in which they live, and range from material to immaterial things (assets, beliefs, etc.; Hurlimann et al., 2014¹³⁷¹; Rouse et al., 2017¹³⁷²). Consideration of social values offers an opportunity to address a wider perspective on impacts on human systems, for example, complementary to quantitative assessments of health impacts (e.g., loss of source of calories and food insecurity; Keim, 2010¹³⁷⁴). This also encompasses immaterial dimensions, such as threats to cultural heritage (Marzeion and Levermann, 2014¹³⁷⁴; Fatorić and Seekamp, 2017a¹³⁷⁵), socialising activities (Karlsson et al., 2015¹³⁷⁶), integration of marginalised groups (Maldonado, 2015¹³⁷⁷) and cultural ecosystem services (Fish et al., 2016), and provides an opportunity to better reflect context-specificities in valuing the physical/ecological/human/cultural impacts’ importance for and distribution within a given society (Fatorić and Seekamp, 2017b¹³⁷⁹). This field of research (no detailed mention found in AR5) is just emerging due to the transdisciplinary and qualitative nature of the topic. Graham et al. (2013)¹³⁸⁰ advance a 5-category framing of social values specifically at risk from SLR: health (i.e., the social determinants of survival such as environmental and housing quality and healthy lifestyles), feeling of safety (e.g., financial and job security), belongingness (i.e., attachment to places and people), self-esteem (e.g., social status or pride that can be affected by coastal retreat), and self-actualisation (i.e., people’s efforts to define their own identity). Another emerging issue relates to social values at risk due to land submergence in low-lying islands (Yamamoto and Esteban, 2014¹³⁸¹) and parts of countries and individual properties (Marino, 2012¹³⁸²; Maldonado et al., 2013¹³⁸³; Aerts, 2017¹³⁸⁴; Allgood and McNamara, 2017¹³⁸⁵). Recent studies also highlight the potential additional risks to social values in areas where displaced people relocate (Davis et al., 2018¹³⁸⁶).

The findings suggest that risks from SLR are already detectable for all of the geographies considered (Panel B in Figure 4.3), and that risk is expected to increase over this century in virtually all low-lying coastal areas whatever their context-specificities or nature (island/continental, developed/developing county) (Cross-Chapter Box 9). In the absence of high adaptation (bars (A)), risk is expected to significantly increase in urban atoll islands and the selected Arctic coastal communities even in a SROCC RCP2.6 scenario, and all geographies are expected to experience almost high to very high risks at the upper likely range of SROCC RCP8.5. These results allow refining AR5 conclusions by showing, first, that high risk can indeed occur before the 1m rise benchmark (Oppenheimer et al., 2014¹³⁹⁹; O’Neill et al., 2017) and, second, that risk as a function of SLR is highly variable from one geography to another. Some rationale is provided below for our assessment of illustrative geographies, summarising the more detailed description provided in SM4.3 (SM4.3.6 to SM4.3.8). Note however that the text below is not intended to be fully comprehensive and does not necessarily include all elements for which there is a substantive body of literature, nor does it necessarily include all elements which are of particular interest to decision makers.

Resource-rich coastal cities (SM4.3.8.1, Panel B in Figure 4.3) – Resource-rich coastal cities considered in this analysis are Shanghai, New York (see Box 4.1 for further details and references on Shanghai and NYC), and Rotterdam (Brinke et al., 2010¹⁴⁰⁰; Hinkel et al., 2018¹⁴⁰¹). High, and in many cases, growing population density and total population, and high exposure of people and infrastructure to GMSL rise and ESL events characterise coastal megacities (Hanson et al., 2011¹⁴⁰²). These are high concentrations of income and wealth in geographic terms but within relatively small area exhibit large distributional differences of both with important implications for emergency response and adaptation. Concentration translates into high exposure of monetary value to coastal hazards and the cities noted here have both historical and recent experience with damaging ESL events, such as Typhoon Winnie which struck Shanghai in 1997 (Xian et al., 2018¹⁴⁰³), Hurricane Sandy in New York in 2012 (Rosenzweig and Solecki, 2014¹⁴⁰⁴), and the North Sea storm of 1953 which impacted the Rotterdam area (Gerritsen, 2005¹⁴⁰⁵; Jonkman et al., 2008¹⁴⁰⁶). However, high density, limited space and high cost of land leads to development of below-ground space for transportation (e.g., subways, road tunnels; MTA, 2017) and storage, and even habitation, creating vulnerabilities not seen in low-density areas. Natural ecosystems within the megacity boundaries and nearby have been exploited for centuries and in some cases decimated or even extirpated (Hartig et al., 2002¹⁴⁰⁷). Accordingly, they provide limited benefits in terms of coastal protection for the densest part of these cities but can be critically important for protection of lower-density areas, for example, wetlands and sandy beaches in the Jamaica Bay/Rockaway sector of New York that protect nearby residential communities (Hartig et al., 2002¹⁴⁰⁸). Space limitations also constrain the potential benefits of EbA measures. Instead, resource-rich coastal cities depend largely on hard defences like sea walls and surge barriers for coastal protection (Section 4.4.2.2). Such defences are costly but generally cost effective due to the aforementioned concentration of population and value. However, barriers to planning and implementing adaptation include governance challenges (Section 4.4.2) such as limited control over finances and the intermittent nature of ESLs which inhibit focused attention over the long time scales needed to plan and implement hard defences (Section 4.4.2.2). As a result, coastal adaptation for resource-rich cities is uneven and the three presented here were selected with a view toward exhibiting a range of current and potential future effectiveness.

Urban atoll islands (SM4.3.8.2, Panel B in Figure 4.3) – The capital islands (or groups of islands) of three atoll nations in the Pacific and Indian Oceans are considered here: Fongafale (Funafuti Atoll, Tuvalu), the South Tarawa Urban District (Tarawa Atoll, Kiribati) and Male’ (North Kaafu Atoll, Maldives). Urban atoll islands have low elevation (<4 m above mean sea level; in South Tarawa, e.g., lagoon sides where settlement concentrates are <1.80 m in elevation) (Duvat, 2013) and are mainly composed of reef-derived unconsolidated material. Their future is of nation-wide importance as they concentrate populations, economic activities and critical infrastructure (airports, main harbours). They illustrate the prominence of anthropogenic-driven disturbances to marine and terrestrial ecosystems (e.g., mangrove clearing in South Tarawa or human-induced coral reef degradation through land reclamation in Male’; Duvat et al., 2013¹⁴⁰⁹; Naylor, 2015¹⁴¹¹) and therefore to services such as coastal protection delivered by the coral reef (i.e., wave energy attenuation that reduces flooding and erosion, and sediment provision that contributes to island persistence over time) (McLean and Kench, 2015¹⁴¹²; Quataert et al., 2015¹⁴¹³; Elliff and Silva, 2017¹⁴¹⁴; Storlazzi et al., 2018¹⁴¹⁵).

The controlling factors of urban atoll islands’ future habitability are the density of assets exposed to marine flooding and coastal erosion (SM4.3.8.2), future trends in these hazards, and ecosystem response to both ocean-climate related pressures and human activities. Urban atoll islands already experience coastal flooding, for example, in Male’ (Wadey et al., 2017¹⁴¹⁶) and Funafuti (Yamano et al., 2007¹⁴¹⁷; McCubbin et al., 2015¹⁴¹⁸). Coastal erosion is also a major concern along non-armoured shoreline in South Tarawa (Duvat et al., 2013¹⁴¹⁹) and Fongafale (Onaka et al., 2017¹⁴²⁰), but not in Male’ where surrounding fortifications have extended along almost the entire shoreline from several decades (Naylor, 2015). Salinisation already affects groundwater lenses, but its contribution to risk varies from one case to another, from low in Male’ (relying on desalinised seawater) to important for human consumption and agriculture in South Tarawa (Bailey et al., 2014¹⁴²²; Post et al., 2018¹⁴²³). 

Together, high population densities (from ~3,200 people per km2 in South Tarawa to ~65,700 people per km2 in Male’) (Government of the Maldives, 2014¹⁴²⁴; McIver et al., 2015¹⁴²⁵) and the concentration of critical infrastructure and settlements in naturally low-lying flood-prone areas already substantially contribute to coastal risk (Duvat et al., 2013¹⁴²⁶; Field et al., 2017¹⁴²⁷). Even stabilised densities in the future would translate into a substantial increase of risk under a 43cm GMSL rise. Risk will also be exacerbated by the negative effects of ocean warming and acidification, especially on coral reef and mangrove capacity to cope with SLR (Pendleton et al., 2016¹⁴²⁸; Van Hooidonk et al., 2016¹⁴²⁹; Perry and Morgan, 2017¹⁴³⁰; Perry et al., 2018¹⁴³¹) (Sections 4.3.3.5, 5.3). In addition, even small values of SLR will significantly increase risk to atoll islands’ aquifers (Bailey et al., 2016; Storlazzi et al., 2018). Finally, land scarcity in atoll environments will exacerbate the importance of SLR induced damages (on housing, agriculture and infrastructure especially) and cascading impacts (on livelihoods, for example, as a result of groundwater and soil salinisation). 

Large tropical agricultural deltas (SM4.3.8.3, Panel B in Figure 4.3) – River deltas considered in this analysis are the Mekong Delta and the Ganges-Brahmaputra-Meghna Delta. Both deltas are large, low-lying and dominated by agricultural production. The risk assessment to SLR considered the entire delta area (not only the coastal fringe; see SM4.3.6 for explanation). High population densities (1280 people per km2 and 433 people per km2 in the Ganges-Brahmaputra-Meghna and Mekong deltas, respectively) (Ericson et al., 2006¹⁴³³; Government of the Maldives, 2014¹⁴³⁴) and the removal of natural vegetation buffers contribute to high exposure rates to coastal flooding, erosion, and salinisation. Agricultural production contributes to GDP strongly (Smajgl et al., 2015¹⁴³⁵; Hossain et al., 2018¹⁴³⁶), making agricultural fields important assets. In both deltas, mangroves are partially degraded (Ghosh et al., 2018¹⁴³⁷; Veettil et al., 2018¹⁴³⁸) as well as other wetlands at the coast and further inland (Quan et al., 2018a¹⁴³⁹; Rahman et al., 2018¹⁴⁴⁰). Currently, riverine flooding dominates in both deltas (Auerbach et al., 2015¹⁴⁴¹; Rahman and Rahman, 2015¹⁴⁴²; Ngan et al., 2018¹⁴⁴³). However, high tides and cyclones can generate large coastal flooding events, especially in the Ganges-Brahmaputra-Meghna Delta (Auerbach et al., 2015¹⁴⁴⁴; Rahman and Rahman, 2015¹⁴⁴⁵). Human-induced subsidence increases the likelihood of flooding in both deltas (Brown et al., 2018b¹⁴⁴⁶). Coastal and river bank erosion is already a problem in both delta (Anthony et al., 2015¹⁴⁴⁷; Brown and Nicholls, 2015¹⁴⁴⁸; Li et al., 2017¹⁴⁴⁹) as well as salinity intrusion, which is impacting coastal aquifers, soils and surface waters (Anthony et al., 2015¹⁴⁵⁰; Brown and Nicholls, 2015¹⁴⁵¹; Li et al., 2017¹⁴⁵²). Salinisation of water and soil resources remains a coastal phenomenon (Smajgl et al., 2015¹⁴⁵³), but salinity intrusion can reach far inland in some extreme years and significantly contribute to risk at the delta scale (Section 4.3.3.4.2). Both deltas are partly protected with hard engineered defences such as dikes and sluice gates to prevent riverine flooding, and polders and dikes in some coastal stretches to prevent salinity intrusion and storm surges (Smajgl et al., 2015¹⁴⁵⁴; Rogers and Overeem, 2017¹⁴⁵⁵; Warner et al., 2018a¹⁴⁵⁶). Today, in both deltas, the measures implemented to restore natural buffers are still limited to mangroves ecosystems (Quan et al., 2018a¹⁴⁵⁷; Rahman et al., 2018¹⁴⁵⁸), and the measures aiming at reducing subsidence are underdeveloped (Schmidt, 2015¹⁴⁵⁹; Schmitt et al., 2017¹⁴⁶⁰). Assuming stable population densities in the future, coastal flooding will contribute increasingly to risk at the delta level (Brown and Nicholls, 2015¹⁴⁶¹; Brown et al., 2018a¹⁴⁶²; Dang et al., 2018¹⁴⁶³). Coastal erosion will increase (Anthony et al., 2015¹⁴⁶⁴; Liu et al., 2017a¹⁴⁶⁵; Uddin et al., 2019¹⁴⁶⁶) and salinisation of coastal waters and soils will be more significant (Tran Anh et al., 2018¹⁴⁶⁷; Vu et al., 2018¹⁴⁶⁸; Rakib et al., 2019¹⁴⁶⁹) and will strongly impact agriculture and water supply for the entire delta (Jiang et al., 2018¹⁴⁷⁰; Timsina et al., 2018¹⁴⁷¹; Nhung et al., 2019¹⁴⁷²). Without increased adaptation, coastal ecosystems will be largely destroyed at 110 cm of SLR (Schmitt et al., 2017¹⁴⁷³; Mehvar et al., 2019¹⁴⁷⁴; Mukul et al., 2019¹⁴⁷⁵). Given the size of these deltas, it is only under high emission scenarios, that flooding, erosion and salinisation lead to high risk at the entire delta scale.

Arctic communities (SM4.3.8.4, Panel B in Figure 4.3) – Five small indigenous settlements located on the Arctic Coastal Plain are considered in this analysis: Bykovsky (Lena Delta, Russian Federation), Shishmaref and Kivalina (Alaska, USA), and Shingle Point and Tuktoyaktuk (Mackenzie Delta, Canada). They lie on exposed coasts composed of unlithified ice-rich sediments in permafrost, in areas with seasonal sea ice and slow to moderate SLR. These communities have populations ranging from 380 to 900 (fewer and seasonal at Shingle Point) that are heavily dependent on marine subsistence resources (Forbes, 2011¹⁴⁷⁶; Ford et al., 2016a¹⁴⁷⁷). Shishmaref and Kivalina are located on low-lying barrier islands highly susceptible to rising sea level (Marino, 2012¹⁴⁷⁸; Bronen and Chapin, 2013¹⁴⁷⁹; Fang et al., 2018¹⁴⁸⁰; Rolph et al., 2018¹⁴⁸¹). Shingle Point is situated on an active gravel spit; Tuktoyaktuk is built on low ground with high concentrations of massive ice; and Bykovsky is mostly situated on an ice-rich eroding terrace about 20 m above sea level. All the selected communities are remote from regions of rapid positive GIA; many other areas in the Arctic experience rapid GIA uplift (James et al., 2015¹⁴⁸²; Forbes et al., 2018¹⁴⁸³) and have very low sensitivity to SLR, which may in fact help to reduce shoaling.

Especially in the Arctic, anthropogenic drivers in recent decades resulted in the induced settlement of indigenous peoples in marginalised climate-sensitive communities (Ford et al., 2016b) and the construction of infrastructure in nearshore areas, with the assumption of stable coastlines. This resulted in increased exposure to coastal hazards. Coastal erosion is already a major problem in all of the case studies, where space for building is usually limited. Accelerating permafrost thaw is promoting rapid erosion of ice-rich sediments, e.g., at Bykovsky (Myers, 2005¹⁴⁸⁵; Lantuit et al., 2011¹⁴⁸⁶; Vanderlinden et al., 2018¹⁴⁸⁷) and Tuktoyaktuk (Lamoureux et al., 2015¹⁴⁸⁸; Ford et al., 2016a¹⁴⁸⁹). Related to this, Kivalina, Shishmaref, Shingle Point, Tuktoyaktuk, and parts of the Lena delta (less so for Bykovsky) are already facing high risk of flooding. Shishmaref, for example, experienced 10 flooding events between 1973 and 2015 that resulted in emergency declarations (Bronen and Chapin, 2013¹⁴⁹⁰; Lamoureux et al., 2015¹⁴⁹¹; Irrgang et al., 2019¹⁴⁹²). There is however no evidence of salinisation in the selected communities, but brackish water flooding of the outer Mackenzie Delta caused by a 1999 storm surge (a rare event due to upwelling ahead of the storm) led to widespread die-off of vegetation with negative ecosystem impacts (Pisaric et al., 2011¹⁴⁹³; Kokelj et al., 2012¹⁴⁹⁴).

Permafrost thaw is already accelerating due to increasing ground temperatures that weaken the mechanical stability of frozen ground (Section 3.4.2.2). Arctic SLR and sea surface warming have the potential to substantially contribute to this thawing (Forbes, 2011¹⁴⁹⁵; Barnhart et al., 2014b¹⁴⁹⁶; Lamoureux et al., 2015¹⁴⁹⁷; Fritz et al., 2017¹⁴⁹⁸). An additional factor unique to the polar regions is the decrease in seasonal sea ice extent in the Arctic (Sections 3.2.1 and 3.2.2), which together with a lengthening open water season, provides less protection from storm impacts, particularly later in the year when storms are prevalent (Forbes, 2011¹⁴⁹⁹; Lantuit et al., 2011¹⁵⁰⁰; Barnhart et al., 2014a¹⁵⁰¹; Melvin et al., 2017¹⁵⁰²; Fang et al., 2018¹⁵⁰³; Forbes, 2019¹⁵⁰⁴) and therefore reduces the physical protection of the land (Section 6.3.1.3). 

Coastal protection through hard measures is widespread around the world, although it is difficult to provide estimates on how many people benefit from them. Currently, at least 20 million people living below normal high tides are protected by hard structures (and drainage) in countries such as Belgium, Canada, China, Germany, Italy, Japan, the Netherlands, Poland, Thailand, the UK, and the USA (Nicholls, 2010). Many more people living above high tides are also protected against ESL by hard structures in major cities around the world. There is a concentration of these measures in northwest Europe and East Asia, although extensive defences are also found in and around many coastal cities and deltas. For example, large scale coastal protection exists in Vancouver (Canada), Alexandria (Egypt) and Keta (Ghana; Nairn et al., 1999¹⁵⁴⁴) and 6000 km of polder dikes in coastal Bangladesh. Gittman et al. (2015) estimate that 14% of the total US coastline has been armoured, with New Orleans being an example of an area below sea level dependent on extensive engineered protection (Kates et al., 2006¹⁵⁴⁵; Rosenzweig and Solecki, 2014¹⁵⁴⁶; Cooper et al., 2016¹⁵⁴⁷). Defences built and raised for tsunami protection, such as post-2011 in Japan (Raby et al., 2015¹⁵⁴⁸), also provide protection against SLR.

The application of sediment-based protection measures also has a long history, offering multiple benefits in terms of enhancing safety, recreation and natural systems (JSCE, 2000¹⁵⁴⁹; Dean, 2002¹⁵⁵⁰; Hanson et al., 2002¹⁵⁵¹; Cooke et al., 2012¹⁵⁵²). About 24% of the world’s sandy beaches are currently eroding by rates faster than 0.5 m yr–1 (Luijendijk et al., 2018¹⁵⁵³). In the USA, Europe and Australia, these responses are often driven by the recreational value of beaches and the high economic benefits associated with beach tourism. More recently, sediment-based measures are implemented as effective and yet flexible measures to address SLR (Kabat et al., 2009¹⁵⁵⁴) and experiments are being conducted with innovative decadal scale application of sediments such as the sand engine in the Netherlands (Stive et al., 2013¹⁵⁵⁵).

There is high confidence that most major upgrades in defences happen after coastal disasters (Box 4.1). Dikes were raised and reienforced after the devastating coastal flood of 1953 in the Netherlands and the UK, and in 1962 in Germany. In New Orleans, investments in the order of 15 billion USD, including a major storm surge barrier, followed Hurricane Katrina in 2005 (Fischetti, 2015¹⁵⁵⁶), and in New York the Federal Government made available 16 billion USD for disaster recovery and adaptation after Superstorm Sandy in 2012 (NYC, 2015). Examples in which SLR has been considered proactively in the planning process include SLR safety margins in, for example, the UK, Germany and France, upgrading defences according to cost-benefit analysis in the Netherlands, and SLR guidance in the USA (USACE, 2011¹⁵⁵⁷).

There is high confidence that hard coastal protection will continue to be a widespread response to SLR in densely populated and urban areas during the 21st century, because this response is widely practised (Section 4.4.2.2.2), effective in reducing current (Section 4.4.2.2.2) and future flood risk (Section 4.3.3.2) and highly cost efficient in urban and densely populated areas (Section 4.4.2.7). There is, however, low agreement on the level of hard coastal protections to expect, with projections being based on different assumptions. A model assuming that coastal societies upgrade hard protection following scenario-based cost-benefit analysis finds that 22% of the global coastline will be protected under various SSPs and 1 m of 21st century global mean SLR (Nicholls et al., 2019¹⁵⁵⁸). Another model assuming that only areas for which benefit-cost ratios are above 1 under SLR scenarios up to 2 m, all SSPs and discount rates up to 6%, finds that this would lead to protecting 13% of the global coastline (Lincke and Hinkel, 2018¹⁵⁵⁹; Figure 4.14).

There is medium evidence and medium agreement on the costs of hard protection. Data on the costs of hard defences is only available for few countries and unit costs estimated from this data vary substantially depending on building/fill material used, labour cost, urban versus rural settings, hydraulic loads, etc. (Jonkman et al., 2013¹⁵⁸⁷; Lenk et al., 2017¹⁵⁸⁸; Aerts, 2018¹⁵⁸⁹; Nicholls et al., 2019¹⁵⁹⁰). In general, there has been limited systematic data collection across sites, although useful national guidance does exist in some cases (Environment Agency, 2015¹⁵⁹¹). Defences depend on good maintenance to remain effective. For some types of infrastructure such as surge barriers, maintenance costs are poorly described and hence more uncertain (Nicholls et al., 2007¹⁵⁹²). Protection-based adaptation to saltwater intrusion is more complex than adaptation to flooding and erosion, and there is less experience to draw upon. 

Based on these unit cost estimates, and different assumptions on future protection, global annual protection costs have been estimated to be 12–71 billion USD considering coastal dikes only (Hinkel et al., 2014¹⁵⁹³) and about 40–170 billion USD yr-1 considering coastal dikes, river dikes and storm surge barriers, under RCP2.6, and about 25–200 billion USD yr-1 considering coastal dikes only (Tamura et al. 2019¹⁵⁹⁴) under RCP8.5. If protection is widely practised through the 21st century, the bulk of the costs will be maintenance rather than capital costs (Nicholls et al., 2019¹⁵⁹⁵).

There is high confidence that well designed and maintained hard and sediment-based protection is very effective in reducing risk to the impacts of SLR and ESL (Horikawa, 1978¹⁵⁷²; USACE, 2002¹⁶⁷³; CIRIA, 2007¹⁶⁷⁴). This includes situations in which coastal megacities in river deltas have experienced, and adapted to, relative SLR of several metres caused by land subsidence during the 20th century (Kaneko and Toyota, 2011¹⁶⁷⁵; Esteban et al., 2019¹⁶⁷⁶; Box 4.1). In principle, there are no technological limits to protect the coast during the 21st century even under high-end SLR of 2 m (Section 4.3.3.2), but technological challenges can make protection very expensive and hence unaffordable in some areas (Hinkel et al., 2018¹⁶⁷⁷). Examples include southeast Florida, because protected areas can be flooded by rising groundwater through underlying porous limestone (Bloetscher et al., 2011¹⁶⁷⁸). Gradually rising water tables behind defences is also an issue, which can be managed by increasing pumping and drainage (Aerts, 2018¹⁶⁷⁹). Maintaining this effectiveness over time requires regular monitoring and maintenance, accounting for changing conditions such as SLR and widespread erosional trends in front of the defences. There will always be residual risks, which can be reduced, but never eliminated, by engineering protection infrastructure to very high standards, such as so-called ‘unbreakable dikes’ (de Bruijn et al., 2013). 

It is difficult to assess at what point in time and for which amount of SLR technical limits for coastal protection will be reached. Parts of Tokyo have been protected against five metres of relative SLR during the 21st century (Kaneko and Toyota, 2011¹⁶⁸⁰) and it has been argued that it is possible to preserve territorial integrity of the Netherlands even under 5 m SLR, using current engineering technology (Aerts et al., 2008¹⁶⁸¹; Olsthoorn et al., 2008¹⁶⁸²). This suggests that under RCP2.6, technical limits to adaptation will be rare even under longer-term SLR. Protecting against high-end SLR will be increasingly technically challenging as we move beyond the 21st century. This is not only due to the absolute amount of SLR, but also due to the very high rates of annual SLR (e.g., 10–20 mm yr–1 likely range under RCP8.5 in 2100), which challenge the planning and implementation of hard protection because major protection infrastructure requires decades to plan and implement (Gilbert et al., 1984¹⁶⁸³; Burcharth et al., 2014¹⁶⁸⁴). In summary, the higher and faster SLR, the more challenging coastal protection will be, but quantifying this is difficult. In any case, before technical limits are reached, economic and social limits will be reached because societies are neither economically able nor socially willing to invest in coastal protection (Sections 4.4.2.2 and 4.3.3.2; Hinkel et al., 2018; Esteban et al., 2019).

When space is limited (e.g., in an urban setting), co-benefits can be generated through multi-functional hard flood defences, which combine flood protection with other urban functions, such as car parks, buildings, roads or recreational spaces into one multifunctional structure (Stalenberg, 2013¹⁶⁰¹; van Loon-Steensma and Vellinga, 2014¹⁶²). An important co-benefit of sediment-based protection, such as beach nourishment and dune management, is that it preserves beach and associated environments, as well as tourism (Everard et al., 2010¹⁶⁰³; Hinkel et al., 2013a¹⁶⁰⁴; Stive et al., 2013¹⁶⁰⁵). 

Drawbacks of hard protection include the alteration of hydrodynamic and morphodynamic patterns, which in turn may export flooding and erosion problems downdrift (Masselink and Gehrels, 2015¹⁶⁰⁶; Nicholls et al., 2015¹⁶⁰⁷). For example, protection of existing shoreline in estuaries and tidal creeks may increase tidal amplification in the upper parts (Lee et al., 2017¹⁶⁰⁸). Hard protection also hinders or prohibits the onshore migration of geomorphic features and ecosystems (called coastal squeeze; Pontee, 2013¹⁶⁰⁹; Gittman et al., 2016¹⁶¹⁰), leading to both a loss of habitat as well as of the protection function of ecosystems (see Sections 4.3.2.4 and 4.4.2.2). Another drawback of raising hard structures, also emphasised in AR5, is the risk of lock-in to a development pathway in which development intensifies behind higher and higher defences, with escalating severe consequences in the event of protection failure (Wong et al., 2014¹⁶¹¹; Welch et al., 2017¹⁶¹²), as experienced in Hurricane Katrina impacted New Orleans (Burby, 2006¹⁶¹³; Freudenburg et al., 2009¹⁶¹⁴). This lock-in results from protection attracting further economic development in the flood zone within defenses, which then leads to further raising defences with SLR, and the growing value of exposed assets. 

Seabed dredging of sand and gravel can have negative impacts on marine ecosystems such as seagrass meadows and corals (Erftemeijer and Lewis III, 2006¹⁶¹⁵; Erftemeijer et al., 2012¹⁶¹⁶). Nourishment practices on sandy beaches have also been shown to have drawbacks for local ecosystems if local habitat factors are not taken into consideration when planning and implementing nourishment and maintenance (Speybroeck et al., 2006¹⁶¹⁷). A further emerging issue is beach material scarcity mainly driven by demand of sand and gravel for construction, but also for beach and shore nourishment (Peduzzi, 2014¹⁶¹⁸; Torres et al., 2017¹⁶¹⁹), which makes sourcing the increasing volumes of beach materials required to sustain beaches in the face of SLR more expensive and challenging (Roelvink, 2015¹⁶²⁰).

Reviews and comparative case studies confirm findings of AR5 that governance challenges are amongst the most common hindrance to implementing coastal measures (Ekstrom and Moser, 2014¹⁶²¹; Hinkel et al., 2018¹⁶²²). One main issue to resolve is conflicting stakeholder interests. This includes conflicts between those favouring protection and those being negatively affected by adaptation measures. In Catalonia, for example, the tourism sector welcomes beach nourishment because it provides direct benefits, whereas those dependent upon natural resources (e.g., fishermen) are increasingly in opposition because they fear that sand mining destroys coastal habitat and livelihood prospects (González-Correa et al., 2008¹⁶²³). 

There is also conflict related to the distribution of public money between communities receiving public support for adaptation and non-coastal communities who pay for this support through taxes (Elrick-Barr et al., 2015¹⁶²⁴). Generally, access to financial resources for adaptation, including from public sources, development and climate finance or capital markets, frequently constrain adaptation (Ekstrom and Moser, 2014¹⁶²⁵; Hinkel et al., 2018¹⁶²⁶). For example, homeowners are often not willing to pay taxes or levies for public protection or sediment-base measures even if they directly benefit, as found, for example in communities on the US east coast where beach nourishment is used to maintain recreational and tourism amenities (Mullin et al., 2019¹⁶²⁷). In many parts of the world, coastal adaptation governance is further complicated by existing conflicts over resources. For example, illegal coastal sand mining is currently a major driver of coastal erosion in many parts of the developing world (Peduzzi, 2014¹⁶²⁸). Examples of this can be found in Ghana (Addo, 2015¹⁶²⁹) and the Comoros (Betzold and Mohamed, 2017¹⁶³⁰).

An associated governance challenge is ensuring the effective maintenance of coastal protection. Ineffective maintenance has contributed to many coastal disasters in the past, such as in New Orleans (Andersen, 2007¹⁶³¹). AR5 highlighted that effective maintenance is challenging in a small island context due to a lack of adequate funds, policies and technical skills (Nurse et al., 2014¹⁶³²). In some countries in which coastal defence systems have a long history, effective governance arrangements for maintenance, such as the Water Boards in the Netherlands, have emerged. In Bangladesh, where Dutch-like polders were introduced in the 1960s, maintenance has been a challenge due to shifts in multi-level governance structures associated with independence, national policy priorities and donor involvement (Dewan et al., 2015¹⁶³³).

At global scales, new economic assessments of responses have mostly focused on the direct costs of hard protection and the benefits of reducing coastal extreme event flood risks. These studies confirm AR5 findings that the benefits of reducing coastal flood risk through hard protection exceed the costs of protection, on a global average, and for cities and densely populated areas, during the 21st century even under high-end SLR (medium evidence, high agreement; Hallegatte et al., 2013¹⁶³⁴; Wong et al., 2014¹⁶³⁵; Diaz, 2016¹⁶³⁶; Lincke and Hinkel, 2018¹⁶³⁷). For example, Lincke and Hinkel (2018) find that, during the 21st century, it is economically efficient to protect 13% of the global coastline, which corresponds to 90% of global floodplain population, under SLR scenarios from 0.3–2.0 m, five SSPs and discount rates up to 6% (Figure 4.14). While the above two studies have not considered the effects of hard protection in reducing the area of coastal wetlands, it is expected that coastal hard protection in densely populated areas and conserving wetlands in sparsely populated areas can go hand in hand. Protecting less than 42% of the global coastline would leave coastal wetlands sufficient accommodation space to even grow in areas under rising sea levels during the 21st century (Schuerch et al., 2018¹⁶³⁹). Diaz (2016), who includes the cost of wetland loss, using a simpler wetland model, finds that both protection and retreat reduce the global net present costs of SLR by a factor of seven as compared to no adaptation (applying a discount rate of 4%) under 21st century SLR of 0.3–1.3 m and SSP2. There is no global study that has considered social costs and benefits of responses (e.g., health, beach amenity, etc.) or looked at the economics of accommodate, retreat and advance responses.

At local scales, a large number of economic assessments of response options are available but mostly in the grey literature and again with a focus on hard and sediment-based protection. Similar to the global studies, hard protection is generally found to be economically efficient for urban and densely populated areas such as New York, USA (Aerts et al., 2014¹⁶⁴⁰) and Ho Chi Minh City, Vietnam (Scussolini et al., 2017¹⁶⁴¹). Both global and local studies show that sediment-based protection, such as beach nourishment is economically efficient in areas of intensive tourism development due to the large revenues generated within this sector (Rigall-I-Torrent et al., 2011¹⁶⁴²; Hinkel et al., 2013a¹⁶⁴³).

Relative to hard adaptation measures whose global distribution is not known in detail (Scussolini et al., 2015¹⁶⁴⁴), the current global distribution of coastal ecosystems is well-studied (e.g., for saltmarshes and mangroves, respectively; Giri et al., 2011¹⁶⁴⁵; Mcowen et al., 2017¹⁶⁴⁶). EbA, by definition, can only exist and function where the environmental conditions are appropriate for a given ecosystem. Mangroves, salt marshes and reefs occur along about 40–50% of the world’s coastlines (Wessel and Smith, 1996¹⁶⁴⁷; Burke, 2011¹⁶⁴⁸; Giri et al., 2011¹⁶⁴⁹; Mcowen et al., 2017¹⁶⁵⁰). However, there is no clear estimate on the global length of coastline covered by ecosystems relevant for EbA in the face of SLR in part because of a mismatch between the spatial resolutions of different estimates available. Mangroves occur on tropical and subtropical coasts, and cover 138,000–152,000 km² across about 120 countries (Spalding et al., 2010¹⁶⁵¹; Giri et al., 2011¹⁶⁵²). At least 150,000 km of coastline in over 100 countries benefit from the presence of coral reefs (Burke, 2011¹⁶⁵³) and these are estimated to protect over 100 million people from wave-induced flooding globally (Ferrario et al., 2014¹⁶⁵⁴). The extent of other coastal habitats is less well known: salt marshes are estimated to occur in 99 countries, especially in temperature to high latitude locations, with nearly 5,500,000 ha mapped across 43 countries (Mcowen et al., 2017¹⁶⁵⁵).

Since AR5 there has been growing recognition of the value of conserving existing coastal ecosystems, and where possible restoring them, for the flood protection and multiple other benefits they provide (Temmerman et al., 2013¹⁶⁵⁶; Arkema et al., 2015¹⁶⁵⁷). In parallel, EbA measures are increasingly being incorporated and required within national plans, strategies and targets (Lo, 2016), international adaptation funding mechanisms, such as the Adaptation Fund (AF; e.g., in Sri Lanka and India; Epple et al., 2016¹⁶⁵⁸), and national natural capital valuations (Beck and Lange, 2016¹⁶⁵⁹). Given their relative novelty, there is widespread interest in building and collecting knowledge of EbA implementation case-studies and examples (Table 4.7). Meanwhile, coastal communities around the globe are already implementing EbA responses at local scales, with emphasis on community participation and ownership and local priorities, needs and capacities (Reid, 2016¹⁶⁶⁰; see Section 4.4.4.4).

EbA has been used as an integral part of some retreat, advance and accommodation responses. For example, on coastlines where high-risk properties are relocated inland, space can be made for ecosystem restoration to enhance natural biodiversity and provide coastal protection (French, 2006¹⁶⁶¹; Coastal Protection and Restoration Authority of Louisiana, 2017¹⁶⁶²). There are also examples of ecosystem restoration to advance coastlines and build land elevation (Chung, 2006¹⁶⁶³). EbA can also be an element of accommodation responses by, for example, restoring or creating marshes to provide space for flood water (Temmerman et al., 2013¹⁶⁶⁴).

While there are projections available of ecosystem responses to climate change and SLR (Section 4.3.3), to date, there are no large-scale projections available on the future extent of EbA. However, several coastal nations, particularly Small Island Developing States (SIDS) explicitly advocate EbA measures as a means to address future coastal hazard and SLR concerns. Based on Nationally Determined Contributions (NDCs) submitted to the United Nations Framework Convention on Climate Change (UNFCCC), more than 30 SIDS cite EbA as a preferred SLR response, with mangrove planting being the most common measure (Wong, 2018¹⁶⁶⁵).

There is limited evidence and low agreement on the costs of ecosystem-based measures to make generally valid estimations of the unit costs across large spatial scales. The total cost of an ecosystem-based measure includes capital costs, maintenance costs, the cost of land and, in some situations, permitting costs (Bilkovic, 2017¹⁶⁶⁶). The costs of restoring and maintaining coastal habitats depend on coastal setting, habitat type and project conditions. In general, unit restoration costs are lowest for mangroves, higher for salt marshes and oyster reefs, and highest for seagrass beds and coral reefs (Table 4.8).

The conservation of coral reefs and other coastal habitats may also entail substantial opportunity costs because alternative uses of this land, such as through agricultural production, industry and settlements, are generally of high economic value (Stewart et al., 2003¹⁶⁶⁷; Balmford et al., 2004¹⁶⁶⁸; Adams et al., 2011¹⁶⁶⁹; Hunt, 2013¹⁶⁷⁰). The high value of these alternative uses are the reason why globally, coastal ecosystems are amongst the ecosystems that face the highest rates of anthropogenic destruction, with estimated annual losses of 1–3% of mangroves area, 2–5% seagrass area and 4–9% corals (Duarte et al., 2013¹⁶⁷¹). Conserving these areas means reversing these trends.

Under the right conditions, and to some extent, EbA measures are free of maintenance costs, because they respond and adapt to changes in their coastal environment. However, maintenance can become important in the aftermath of damage by storms or human action, for example, when wetlands and reefs can be damaged by high winds, waves and surges, or affected by dredging operations (Smith III et al., 2009¹⁶⁷²; Puotinen et al., 2016¹⁶⁷³). At present, there is limited evidence about the conditions under which EbA measures can self-adapt and when they require human intervention to recover.

While EbA has been able to reduce the impacts of sea level related hazards, there is still little agreement on the size of the effect (Gedan et al., 2011¹⁶⁷⁴; Doswald et al., 2012¹⁶⁷⁵; Lo, 2016; Renaud et al., 2016¹⁶⁷⁶). Dozens of independent field, experimental and numerical studies have observed and measured the wave attenuation and flood reduction benefits provided by natural habitats, such as marsh and mangrove wetlands (Barbier and Enchelmeyer, 2014¹⁶⁷⁷; Möller et al., 2014¹⁶⁷⁸; Rupprecht et al., 2017¹⁶⁷⁹), coral reefs (Ferrario et al., 2014¹⁶⁸⁰; Storlazzi et al., 2017¹⁶⁸¹), oyster reefs (Scyphers et al., 2011¹⁶⁸²) and submerged seagrass beds (Infantes et al., 2012¹⁶⁸³). Local and global numerical studies indicate that marshes and mangroves can reduce present-day surge-related flood damages by >15% annually, and the loss of a metre of living coral reef can double annual wave-related flood damages (Narayan et al., 2017¹⁶⁸⁴; Beck et al., 2018¹⁶⁸⁵). Artificial reef restoration along tens of metres of coastline using Reef Ball™ and other structures has been shown to reduce wave heights and stabilise beach widths (Reguero et al., 2018a¹⁶⁸⁶; Torres-Freyermuth et al., 2018¹⁶⁸⁷). The effectiveness of EbA measures, however, varies considerably depending on storm, wetland, reef and landscape parameters (Koch et al., 2009¹⁷⁰⁰; Loder et al., 2009¹⁷⁰¹; Wamsley et al., 2010¹⁷⁰²; Pinsky et al., 2013¹⁷⁰³; Quataert et al., 2015¹⁷⁰⁴), which makes it difficult to extrapolate the physical and economic benefits across geographies. Depending on these parameters, rates of surge attenuation can vary between 5–70 cm km^-1 (Krauss et al., 2009¹⁷⁰⁵; Vuik et al., 2015¹⁷⁰⁶).

Critical gaps remain in our understanding about those parameters that together affect the success of ecosystem-based measures including choice of species and restoration techniques, lead time, natural variability and residual risk, temperature, salinity, wave energy and tidal range (Smith, 2006¹⁷⁰⁷; Stiles Jr, 2006¹⁷⁰⁸). Among reasons commonly cited for the failure of mangrove restoration projects are poor choice of mangrove species, planting in the wrong tidal zones and in areas of excessive wave energy (Primavera and Esteban, 2008¹⁷⁰⁹; Bayraktarov et al., 2016¹⁷¹⁰; Kodikara et al., 2017¹⁷¹¹).

The effectiveness of ecosystem-based measures also exhibits high seasonal, annual and longer-term variability. For example, marsh and seagrass wetlands typically have lower densities in winter which reduces their coastal protection capacity (Möller and Spencer, 2002¹⁷¹²; Paul and Amos, 2011¹⁷¹³; Schoutens et al., 2019¹⁷¹⁴). In the long-term, there is limited evidence and low agreement on how changes in sea level, sediment inputs, ocean temperature and ocean acidity will influence the extent, distribution and health of marsh and mangrove wetlands, coral reefs and oyster reefs (Hoegh-Guldberg et al., 2007¹⁷¹⁵; Lovelock et al., 2015¹⁷¹⁶; Crosby et al., 2016¹⁷¹; Albert et al., 2017¹⁷¹⁸). EbA measures may have differential lead times before they are effective. For example, newly planted mangroves provide less wave attenuation until they mature (~3–5 years; Mazda et al., 1997¹⁷¹⁹). In contrast, a reef restoration project that uses submerged concrete structures performs as a breakwater as soon as the sub-structure is in place (Reguero et al., 2018a¹⁷²⁰).

There is high confidence that ecosystem-based measures provide multiple co-benefits such as sequestering carbon (Siikamäki et al., 2012¹⁷²¹; Hamilton and Friess, 2018¹⁷²²), income from tourism (Carr and Mendelsohn, 2003¹⁷²³; Spalding et al., 2017¹⁷²⁴), enhancing coastal fishery productivity (Carrasquilla-Henao and Juanes, 2017¹⁷²⁵; Taylor et al., 2018¹⁷²⁶), improving water quality (Coen et al., 2007¹⁷²⁷; Lamb et al., 2017¹⁷²⁸), providing raw material for food, medicine, fuel and construction (Hussain and Badola, 2010¹⁷²⁹; Uddin et al., 2013¹⁷³⁰), and a range of intangible and cultural benefits (Scyphers et al., 2015¹⁷³¹) that help improve the resilience of communities vulnerable to sea level hazards (Sutton-Grier et al., 2015¹⁷³²).

In comparison to hard structures like seawalls, EbA measures, particularly coastal wetlands, require more land (The Royal Society Science Policy Centre, 2014), and competition for land is often why the ecosystems have declined in the first place (4.4.2.3.1). On developed coasts, this land is often not available. In such cases, hybrid measures that either combine EbA measures with structural measures like mangrove forests in front of dikes (Dasgupta et al., 2019¹⁷³³), or build ecological enhancements into engineered structures can provide an effective solution. Like any other feature that interacts with coastal processes, natural wetlands and reefs can increase flooding in some instances, for example, due to the redistribution or acceleration of flows in channels within a wetland system (Marsooli et al., 2016¹⁷³⁴), or an increase in infragravity wave (i.e., surface gravity waves with frequencies lower than wind waves) energy behind a reef (Roeber and Bricker, 2015¹⁷³⁵).

The coastal protection benefits of natural ecosystems are increasingly being recognised within international discourse and national coastal adaptation, resilience and sustainable development plans and strategies (Section 4.4.2.3.1). In general, obtaining permits for EbA remains more difficult compared to established hard protection measures, in places like the USA (Bilkovic, 2017¹⁷³⁶). However, there are examples of instruments specifically tailored to retain the protective function of EbA (Borges et al., 2009¹⁷³⁷; Government of India, 2018¹⁷³⁸). The Living Shorelines Regulations of the state government of Maryland in the USA (Maryland DEP, 2013¹⁷³⁹), for instance, requires that private properties must include marsh creation or other non-structural measures when stabilising their shorelines, unless a waiver is obtained.

There are an increasing number of public and private financial mechanisms and policy instruments to encourage the use and implementation of EbA measures (Colgan et al., 2017¹⁷⁴⁰; Sutton-Grier et al., 2018¹⁷⁴¹). For example, a regulation by the Federal Emergency Management Agency (FEMA) of the USA, allows proponents of hazard mitigation projects, such as state, territorial and local governments, to take into account the co-benefits of EbA when assessing benefit-cost ratios of FEMA-funded recovery projects (FEMA, 2015¹⁷⁴²). International guidelines are being developed for designing and implementing EbA measures, with the intention to support wider implementation of these responses (Hardaway Jr and Duhring, 2010¹⁷⁴³; Van Slobbe et al., 2013; Van Wesenbeeck et al., 2017; Bridges et al., 2018¹⁷⁴⁴).

There is limited evidence regarding the economic efficiency of EbA, mainly due to the low agreement about EbA effectiveness (Section 4.4.2.3.2) and costs (Section 4.4.2.3.2). A study of coastal protection measures on the Gulf of Mexico coastline, USA, estimated that EbA measures have average benefit-cost ratios above 3.5 for 2030 flood risk conditions, assuming a discount rate of 2% (Reguero et al., 2018b¹⁷⁴⁵; see Section 4.4.2.3.2). This study also finds that EbA are nearly four times more cost-efficient along developed coastlines as compared to conservation-priority areas because protection benefits are higher in the former case due to the level of asset exposure.

There is limited evidence of migration occurring directly as a consequence of impacts associated with environmental change generally and SLR specifically. Research examining the linkages between migration and environmental change has been conducted in the Pacific (Connell, 2012; Janif et al., 2016; Perumal, 2018), South Asia (Szabo et al., 2016; Call et al., 2017; Stojanov et al., 2017), Latin America (Nawrotzki and DeWaard, 2016; Nawrotzki et al., 2017), Alaska, in North America (Marino and Lazrus, 2015; Hamilton et al., 2016) and Africa (Gray and Wise, 2016). While some limited evidence was found on population movement inland associated with shoreline encroachment in Louisiana, USA (Hauer et al., 2018), this research emphasises that the relationship between climate change impacts including SLR and migration is more nuanced than suggested by simplified cause-and-effect models (Adger et al., 2015). Migration is driven by a large number of individual, social, economic, political, demographic and environmental push and pull factors (Black et al., 2011; Koubi et al., 2016), interwoven with mega-trends such as urbanisation, land use change and globalisation, and is influenced by development and political practices and discourses (Bettini and Gioli, 2016; Cross-Chapter Box 7). For example, asset endowed individuals and households are more able to migrate out from flood-prone areas (Milan and Ruano, 2014; Logan et al., 2016), while the poorest households are significantly susceptible to material and human losses following an extreme event or disruptive environmental change (Call et al., 2017). Individual and social drivers include perceptions of environmental change (Koubi et al., 2016), formed by both direct experience of change and indirect information from social networks, mass media and governmental agencies. Environmental factors include the longer term impacts of climate variability and change, which can erode the capacity of ecosystems to provide essential services such as availability of freshwater, soil fertility and energy production acting as a threat multiplier for other drivers of migration(Hunter et al., 2015; McLeman, 2018).

There is robust evidence of disasters displacing people worldwide, but limited evidence that climate change or SLR is the direct cause. In 2017, 18.8 million people were displaced by disasters, of which 18 million were displaced by weather-related events including 8.6 million people displaced by floods and 7.5 million by storms, with hundreds of millions more at risk (IDMC, 2017; Islam and Khan, 2018). The majority of resultant population movements tend to occur within the borders of affected countries (Warner and Afifi, 2014; Hunter et al., 2015; Nawrotzki et al., 2017).

We find robust evidence of planned relocation taking place worldwide in low-lying zones exposed to the impacts of coastal hazards (Hino et al., 2017; Mortreux et al., 2018). While relocation plans are usually discussed after an extreme event occurs, they generally target the reduction of long-term environmental risks, including those of SLR (McAdam and Ferris, 2015; Hino et al., 2017; Morrison, 2017). For example, in the aftermath of Hurricane Katrina, the Louisiana Comprehensive Master Plan for a Sustainable Coast recommended the relocation of several communities in the next 50 years due to expected RSL rise, and relocation of inhabitants from Isle de Jean Charles is already taking place (Barbier, 2015; Coastal Protection and Restoration Authority of Louisiana, 2017). In Shismaref, an Iñupiat community in Alaska, increased shoreline erosion triggered government-led relocation (Bronen and Chapin, 2013; Maldonado et al., 2013). In the Pacific, current coastal risks aggravated by rising sea level are driving the government led relocation of the inhabitants of Taro, the provincial capital of Choiseul Province in the Solomon Islands (Albert et al., 2018). In 2014, the government of Kiribati purchased land on Vanua Levu, the second largest island of Fiji, with the purpose of economic development and food security, but many i-Kiribati associated the acquisition with future relocation to Fiji (Hermann and Kempf, 2017). In southeast Asia, the government of Vietnam assists and manages rural populations’ relocation from disaster prone areas exposed to coastal risks in the Mekong Delta to large industrial areas with high labour demand, such as Ho Chi Minh City and Can Tho City (Collins et al., 2017). Managed realignment carried out for the purposes of habitat creation, improved flood risk management and more affordable coastal protection, is increasingly popular in Europe, but usually involves small-scale projects and few people if any (Esteves, 2013). Most of the managed realignment projects in the UK and Germany have been carried out for habitat creation and to reduce spending on coastal defences (Hino et al., 2017)

There is high agreement that climate change has the potential to drastically alter the size and direction of migration flows (Connell, 2012; Gray and Wise, 2016; Janif et al., 2016; Nawrotzki and DeWaard, 2016; Szabo et al., 2016; Call et al., 2017; Nawrotzki et al., 2017), but there is low confidence in quantitative projections of migration in response to SLR and extremes of sea level. The number of modelling studies of migration in response to environmental drivers has increased rapidly over the past decade (Kumari et al. 2018), but only a small portion of these model studies address migration in response to SLR and sea level extremes. Amongst these, a variety of different modelling approaches have been applied, but no model currently accounts for all push and pull factors influencing migration decisions (see Section 4.4.2.6.1).

A model projecting future US county-level populations exposed to permanent inundation was combined with an empirical model of potential migration destinations to produce the first sea level/migration analysis of migrant destinations (Hauer, 2017). Assuming that households with incomes above 100,000 USD yr^-1 would have resources to stay and adapt, it was found that 1.8 m SLR by 2100 would displace over two million people in south Florida. Projected population gains due to SLR reach several hundred thousand for some inland urban areas. A gravity model modified to account for both distance to destinations and their attractiveness (deriving from such factors as economic opportunity and environmental amenities) projects a net migration into and out of the East African coastal zone, ranging from out-migration of 750,000 people between 2020 and 2050 to a small in-migration (Kumari et al., 2018). However, this range includes migration stimulated by freshwater availability as well as SLR and episodic flooding. A generalised radiation or diffusion model predicts 0.9 million people will migrate due to SLR in Bangladesh by 2050 and 2.1 million by 2100, largely internally, with substantial implications for nutrition, shelter and employment in destination areas (Davis et al., 2018).

A global dynamic general equilibrium framework (Desmet et al., 2018) provides a more comprehensive approach to accounting for economic factors including changes to trade, innovation, and agglomeration, and political factors, such as policy barriers to mobility, all of which influence the migration response to environmental change. Agent-based models attempt to simulate decisions by individuals who face a variety of socioeconomic and environmental changes (Kniveton et al., 2012). However, neither general equilibrium nor agent-based frameworks have been applied yet to migration responses to SLR. Econometric models, common in climate/migration studies (Millock, 2015), likewise have yet to be applied to the SLR context, except for a single case study where an econometric model was used to interpret the outcome of a discrete choice experiment (Buchanan et al., 2019). For example, an interesting distinction between migration responses to long term temperature and precipitation trends in contrast to extreme events like flooding has been noted (Bohra-Mishra et al., 2014; Mueller et al., 2014), but similar econometric studies have yet to be done comparing responses to gradual land loss versus flooding during ESL events.

We have limited evidence of estimates on the cost of retreat. There are few cost estimates in the literature and these are based on stylised assumptions as little empirical data is available.

The cost of managed relocation, including land acquisition, building of roads and infrastructure and other subsidies, was found to vary from 10,000–270,000 GBP per home in United Kingdom Coastal Change Pathfinder projects (Regeneris Consulting, 2011), and between 10,000 USD in Fiji and 100,000 USD per person in Alaska and in the Isle of Jean Charles in the USA (Hino et al., 2017). For people involved in planned relocation in Shaanxi Province, Northwest China, households receive subsidies ranging from 1200–5100 USD (Lei et al., 2017). The Louisiana’s National Disaster Resilience Competition, Phase II Application states that the proposed relocation of 40 households in the Isle de Jean Charles in Louisiana is estimated to cost 48,379,249 USD, including the cost for land acquisition, infrastructure and construction of new dwellings (State of Louisiana, 2015). Generally, maintenance costs do not arise if people are moved completely out of the hazard zone (Suppasri et al., 2015; Hino et al., 2017). In cases in which people are only moved so that short-term but not long-term risk is reduced, follow up costs for further responses will occur.

The individual costs associated with displacement after an environmental disaster are difficult to obtain. In the literature, there are limited estimates of the social costs to residents of Guadeloupe, Saint Croix, St. Thomas, Puerto Rico, and the southeast USAdisplaced after Hurricanes Hugo (1989) and Katrina (2005). A survey conducted across 18 parishes (i.e. counties) in Louisiana in 2006 revealed that non-displaced households had an average income of 36,000 USD compared to an average income of 30,000 USD recorded for displaced households (Hori and Schafer, 2010).

There is very high confidence that retreat is effective in reducing the risks and impacts of SLR as retreat directly reduces exposure of human settlements and activities (Gioli et al., 2016; Shayegh et al., 2016; Hauer, 2017; Morrison, 2017).

The other outcomes of retreat responses, beyond the one of effectively reducing SLR risks and impacts, are complex and affect both origin and destination. Generally, retreat impacts social networks, access to services and economic and social opportunities, and several well-being indicators (Jones and Clark, 2014; Adams, 2016; Herath et al., 2017; Kura et al., 2017; McNamara et al., 2018). The socioeconomic benefits of migration to individuals and households may include improved access to health and education services, as well as labour markets (Wrathall and Suckall, 2016). Destination areas may gain economically as populations and capital relocate and provide a new source of labour, capital and innovation to inland areas (see Section 4.4.2.6.2; de Haas, 2010). Income inequality may be reduced, but only through migration to areas with growing economies. Remittances can provide flexibility in livelihood options, supply capital for investment and spread risk (Scheffran et al., 2012).

Drawbacks of migration and displacement at the destination can be increased competition for resources and within labour markets, pressure on frontline services and on social cohesion as a result of heightened cultural or ethnic tension (Werz and Hoffman, 2015), as well as cultural, social and psychological losses related to disruptions to sense of place and identity, self-efficacy, and rights to ancestral land and culture (McNamara et al., 2018). The unplanned and unassisted voluntary relocation of the inhabitants of Nuatambu and Nusa Hope in the Solomon Islands to areas further from the coast poses a series of practical challenges with sanitation, access to drinking water and transport (Albert et al., 2018).

The success of planned relocation in terms of the balance of co-benefits and drawbacks varies across relocation schemes (Hino et al., 2017) and outcomes are highly uneven (Genovese and Przyluski, 2013; Ford et al., 2015; Nordstrom et al., 2015; Bukvic and Owen, 2017; Hino et al., 2017; Jamero et al., 2017). On the one hand, well designed and carefully implemented programmes, such as the ongoing resettlement of indigenous communities in Alaska, can improve housing standards and reduce vulnerability (Suppasri et al., 2015; Albert et al., 2018). On the other hand, relocated communities have often become further impoverished (Wilmsen and Webber, 2015), because they are removed from cultural and material resources on which they rely, compounded by poor implementation processes that may fail to ensure fairness, social and environmental justice and well-being (Herath et al., 2017; Mortreux et al., 2018; Nygren and Wayessa, 2018).

Environmentally driven migration and displacement gained major attention over the last decade in the international policy community (Goodwin-Gill and McAdam, 2017). Worldwide programmes, such as the Nansen Initiative, signed by 110 countries to address the serious legal gap around the protection of cross-border migrants impacted by natural disasters, have been implemented (Gemenne and Brücker, 2015). In 2016, the Platform on Disaster Displacement was established to follow up on the work conducted by the Nansen Initiative with the objective of implementing the recommendations of the Protection Agenda (McAdam and Ferris, 2015). Governments are further encouraged by civil society to relocate people at risk and displaced populations out of disaster-prone areas to avoid potential casualties (Lei et al., 2017; Mortreux et al., 2018). There have been discussions among Pacific Island countries and territories and other nations in the Pacific Rim around new policy mechanisms that would facilitate adaptive migration in the region in response to natural hazards including SLR (Burson and Bedford, 2015). There have been cases presented at the Immigration and Protection Tribunal of New Zealand testing refugee claims associated with climate change from Tuvaluan and i-Kiribati applicants, both citing environmental change on their home islands as grounds for remaining in New Zealand. One applicant was successful in the quest to remain in New Zealand on humanitarian grounds, but not on the grounds of refugee status (Farbotko et al., 2016).

The is high agreement that outcomes can be improved by upholding the principle of procedural justice and respecting the autonomy of individuals and their decisions about where and how they live (Warner et al., 2013; Schade et al., 2015; McNamara et al., 2018). However, there are cases where logistical and political stances constrain the application of such approach, such as when the government of Sri Lanka prohibited rebuilding along the coastline of the country after the 2004 tsunami (Hino et al., 2017). Proactive planning, including participation and consultation with those in peril, has the potential to improve outcomes (medium confidence; de Sherbinin et al., 2011; Gemenne and Blocher, 2017). Governments can assist migrants through policy reforms to enable relocation to fast growing economic regions in the country. An example of this approach was adopted in Vietnam by both the National Target Program to Respond to Climate Change and the National Strategy for Natural Disaster Prevention, Response and Mitigation targeted at locations within the Mekong Delta exposed to the impacts of SLR (Nguyen et al., 2015; Collins et al., 2017). Outcomes of retreat for both community of origin and destination can also be improved by building the human capital of migrants (skills, health and education), reducing costs of migration and remittance transfer, and provision of improved safety nets for migrants at their destinations (high agreement) (Gemenne and Blocher, 2017).

There is limited evidence on the efficiency of retreat responses in the scientific literature.

Decision analysis methods are formal methods that help to identify alternatives that perform best or well with regard to given objectives. An alternative (also called response option or, as a sequence of options over time: adaptation pathway) is a specific combination of SLR responses (See Section 4.4.3). Each alternative is characterised for each possible future state-of-the world (e.g., levels of SLR or socioeconomic development) by one or several attributes, which may measure any relevant social, ecological, or economic effect associated with choosing and implementing the alternative (Kleindorfer et al., 1993²¹³⁷). Attributes commonly used include cost of adaptation alternatives, monetary and non-monetary benefits of the SLR impacts avoided, or net present value (NPV), which is the difference between discounted monetised benefits over time and discounted costs over time. Formal decision analysis is one way to support social choices that is generally suggested for decision support if decisions are complex and involve large investments, as is frequently the case in coastal contexts in the face of SLR.

In order to be effective, decision analysis needs to be embedded in a governance process that accounts for societal needs and objectives (Sections 4.4.4.2 and 4.4.5). This is because decision analysis entails a number of normative choices about the objectives chosen, the criteria used, the specific methods and data applied, the set of alternatives considered, and the attributes used to characterise alternatives. These choices need to reflect the diversity of values, preference and goals of all stakeholders involved in and affected by a decision. Furthermore, decision analysis needs to consider all available knowledge, including all major uncertainties in both climate and non-climate factors, ambiguities in expert opinions, and differences in approaches, because a partial consideration of uncertainty and ambiguity could misguide the choice of adaptation alternatives (high confidence; Renn, 2008²¹³⁸; Jones et al., 2014²¹³⁹; Hinkel and Bisaro, 2016²¹⁴⁰).

Since AR5, the literature on coastal decision analysis has advanced significantly, specifically addressing the large uncertainty about post-2050 SLR through i) using robust decision approaches instead of expected utility, ii) iterating or adapting decisions over time, and iii) increasing flexibility of responses. Each advance is elaborated below. Furthermore, the coastal decision analysis literature also stresses the consideration of multiple criteria or attributes, because adaptation often involves stakeholders with differing objectives and ways of valuing alternatives (Oddo et al., 2017²¹⁴¹). Many decision making methods combine each of the three advances highlighted here (Marchau et al., 2019²¹⁴²). The suitability of each method depends strongly on the specific context, including available resources, technical capabilities, policy objectives, stakeholder preferences and available information.

A growing literature on decision analysis of coastal adaptation advocates the use of RDM approaches instead of maximising expected utility approaches (Hallegatte et al., 2012²¹⁴³; Haasnoot et al., 2013²¹⁴⁴; Lempert et al., 2013²¹⁴⁵; Wong et al., 2017²¹⁴⁶). The core criterion to be considered for choosing between the two types of approaches is whether one is confronted with a situation of shallow or deep uncertainty (high confidence) (Lempert and Schlesinger, 2001²¹⁴⁷; Kwakkel et al., 2010²¹⁴⁸; Kwakkel et al., 2016b²¹⁴⁹; Hinkel et al., 2019²¹⁵⁰). Uncertainty is shallow when a single unambiguous objective or subjective probability distribution can be attached to states-of-the-world. Uncertainty is deep, when this is not possible, either because there is no unambiguous method for deriving objective probabilities or the subjective probability judgements of parties involved differ (Cross-Chapter Box 4 in Chapter 1; Type 2).

Expected utility approaches can only be applied for identifying an optimal response in situations of shallow uncertainty. This is because these approaches require a probability distribution over states of the world in order to identify the optimal alternative which leads to the highest expected utility (i.e., the probability weighted sum of the utilities of all outcomes under a given alternative and all states-of-the-world; Simpson et al., 2016²¹⁵¹). A prominent example of this approach is cost-benefit analysis under risk, which assesses expected outcomes across states of the world in terms of NPV (the discounted stream of net benefits). Cost-benefit analysis has several well-known limitations, such as its sensitivity to discount rates and the difficulty to monetise ecological, cultural and other intangible benefits (Section 1.1.4) that have been widely discussed in the climate change literature (Chambwera et al., 2014²¹⁵²; Kunreuther et al., 2014²¹⁵³; Dennig, 2018²¹⁵⁴).

In the context of coastal adaptation, uncertainty is only shallow if projected SLR does not significantly differ between low end (e.g., RCP2.6) and high end (e.g., RCP8.5) scenarios (Hinkel et al., 2019²¹⁵⁵). The point in time when this is the case (i.e., time of scenario divergence) depends on what difference in expected utility matters to the particular stakeholders involved in a decision. The time of scenario divergence also differs across locations. In locations where the internal sea level variability is large as compared to relative SLR, it takes longer before the differences in sea levels under low end and high end scenarios become apparent. Figure 4.15 illustrates this effect for the ESL projections of this report (Sections 4.2.3.2 and 4.2.3.4), following the approach of Hinkel et al. (2019). Under the assumption that a 10% statistical distance between the distributions of RCP2.6 and RCP8.5 is decision relevant, scenario divergence occurs before 2050 for approximately two thirds of coastal sites with sufficient observational data, but for 7% of locations this occurs later than 2070.

In principle, a single unambiguous probability distribution on future sea levels could also be attained beyond the time of scenario divergence by attributing subjective probabilities to emission scenarios, but individuals may significantly disagree in their subjective probabilities, which again results in deep uncertainty (Lempert and Schlesinger, 2001²¹⁵⁶; Stirling, 2010²¹⁵⁷). For this reason, very few studies that assign subjective probabilities to emission scenarios are found in the literature (Woodward et al., 2014²¹⁵⁸; Abadie, 2018²¹⁵⁹). But even before the year of scenario divergence is reached, uncertainty about relative SLR can be deep, because of deep uncertainties about non-climatic contributors to relative sea level change, such as VLM during or after earthquakes and human-induced subsidence (Cross-Chapter 5 in Chapter 1; Section 4.2.2.4; Hinkel et al., 2019²¹⁶⁰).

Under situations of deep uncertainty, RDM approaches aim to identify alternatives that perform reasonably well (i.e., ‘are robust’) under a wide range of states-of-the-world or scenarios and hence do not require probability assessments. These approaches include minimax or minimax regret (Savage, 1951), info gap theory (Ben-Haim, 2006²¹⁶²), robust optimisation (Ben-Tal et al., 2009²¹⁶³) and exploratory modelling methods that create a large ensemble of plausible future scenarios for each alternative, and then use search and visualisation techniques to extract robust alternatives (Lempert and Schlesinger, 2000²¹⁶⁴). SLR examples of RDM include Brekelmans (2012) who minimise the average and maximum regret across a range of SLR scenarios for investments in dike rings in the Netherlands and Lempert et al. (2013)²¹⁶⁵ who apply RDM in Hoh-Chi-Minh City.

But even if SLR uncertainty is shallow, RDM are more suitable than expected utility approaches if parties involved or affected by a decision have a low uncertainty tolerance, because the goal of the uncertainty intolerant decision maker is to avoid major damages under most or all circumstances (Hinkel et al., 2019²¹⁶⁶).

An adaptation strategy developed based on the maximisation of expected utility may not meet this goal, because worst case damages occurring can exceed expected damages by orders of magnitude.

The uncertainty tolerance of stakeholders is also determining how large of a SLR range needs to be considered in RDM. Stakeholders (i.e., those deciding and those affected by a decision) that have a high uncertainty tolerance (e.g., those planning for investments that can be very easily adapted) can use the combined likely range of RCP2.6 and RCP8.5 (0.29–1.10 m by 2100) for long-term adaptation decisions. For stakeholders with a low uncertainty tolerance (e.g. those planning for coastal safety in cities and long term investment in critical infrastructure) it is meaningful to also consider SLR above this range, because a 17% chance of GMSL exceeding this range under RCP8.5 is too high to be tolerated from this point of view (Ranger et al., 2013²¹⁶⁷; Hinkel et al., 2015²¹⁶⁸; Hinkel et al., 2019²¹⁶⁹).

Independent of the debate about whether to apply expected utility or robust decision making approaches, there is an extensive literature that applies scenario-based cost-benefit analysis. For example, this approach has been applied for setting the safety standards of Dutch dike rings (Kind, 2014²¹⁷⁰; Eijgenraam et al., 2016²¹⁷¹), exploring future protection alternatives for New York (Aerts et al., 2014²¹⁷²), Ho Chi Minh City (Scussolini et al., 2017²¹⁷³), and for many other locations. Scenario-based cost-benefit analysis differs from cost-benefit analysis under risk discussed above in that scenario-based cost-benefit analysis is not applied to rank alternatives across scenarios, but a ‘separate’ cost-benefit analysis is applied within each emission or SLR scenario considered. While this identifies the optimal alternative under each scenario, it does not formally address the problem faced by a coastal decision maker, namely to decide across scenarios (Lincke and Hinkel, 2018²¹⁷⁴). Nevertheless, the results of scenario-based cost-benefit analysis (i.e., NPV of each alternative under each scenario) provide guidance for decision makers and can also be used as inputs (i.e., as attributes) to robust and flexible decision making approaches

Irrespective of whether expected utility or robustness criteria are applied, there is high confidence that an effective way of dealing with large uncertainties is adaptive decision making (also called iterative decision making, adaptive planning or adaptive management), which maintains that decision and decision analysis should be conducted within an iterative policy cycle. This approach includes monitoring of sea level variables and evaluation of alternatives in this light in order to learn from past decisions and collect information to inform future decisions (Haasnoot et al., 2013²¹⁷⁵; Barnett et al., 2014²¹⁷⁶; Burch et al., 2014²¹⁷⁷; Jones et al., 2014²¹⁷⁸; Wise et al., 2014²¹⁷⁹; Kelly, 2015²¹⁸⁰; Lawrence and Haasnoot, 2017²¹⁸¹). Such a staged approach is especially suitable for coastal adaptation due to the long lead and lifetimes of many coastal adaptation measures and the deep uncertainties in future sea levels (Hallegatte, 2009²¹⁸²; Kelly, 2015²¹⁸³). Prominent representatives of methods that entail this idea are Dynamic Adaptive Policy Pathways (Haasnoot et al., 2013²¹⁸⁴) and Dynamic Adaptation Planning (Walker et al., 2001²¹⁸⁵). An important prerequisite for any adaptive decision-making approach is a monitoring system that can detect sea level signals sufficiently early to enable the required responses (Hermans et al., 2017²¹⁸⁶; Haasnoot et al., 2018²¹⁸⁷; Stephens et al., 2018²¹⁸⁸).

In recent years, many different frameworks for adaptive decision making have been put forward, including Adaptive Policy Making (Walker et al., 2001²¹⁸⁹), Dynamic Adaptive Policy Pathways (Haasnoot et al., 2013²¹⁹⁰), Dynamic Adaptive Planning (Walker et al., 2013²¹⁹¹), Iterative risk management (Jones et al., 2014²¹⁹²) and Engineering Options Analysis (de Neufville and Smet, 2019). Each frameworks emphasises particular aspects of adaptive decision making and has merits in specific situations depending on the preferences, goals, uncertainties and information at stake (Marchau et al., 2019²¹⁹³). Nevertheless, all of these frameworks share the following generic and iterative steps

1. Set the stage: Identify current situation, objectives, options (alternatives) and uncertainties.
2. Develop a dynamic plan, which consists of a basic plan plus contingency actions to be carried out based on observed triggers.
3. Implement basic plan and monitor system for triggers.
4. Monitor and act upon triggers.

An idea closely related to adaptive decision making is to keep future alternatives open by favouring flexible alternatives over non-flexible ones. An alternative is said to be ‘flexible’ if it allows switching to other alternatives once the implemented alternative is no longer effective. For example, a flexible protection approach would be to build small dikes on foundations designed for higher dikes, in order to be able to raise dikes in the future should SLR necessitate this.

A prominent and straightforward method that addresses the objective of flexibility is adaptation pathways analysis (Haasnoot et al., 2011²¹⁹⁴; Haasnoot et al., 2012²¹⁹⁵), which is one component of Dynamic Adaptive Policy Pathways. The method graphically represents alternative combinations of measures over time together with information on the conditions under which alternatives cease to be effective in meeting agreed objectives, as well as possible alternatives that will then be available. As time and SLR progress, monitoring may trigger a decision to switch to another alternative. Adaptation pathway analysis has been widely applied both in the scientific literature as well as in practical cases. Applications after AR5 include Indonesia (Butler et al., 2014²¹⁹⁶), New York City (Rosenzweig and Solecki, 2014²¹⁹⁷), Singapore (Buurman and Babovic, 2016²¹⁹⁸) and Australia (Lin and Shullman, 2017²¹⁹⁹). In New Zealand, the method has been included in national guidance for coastal hazard and climate change decision making (Lawrence et al., 2018²²⁰⁰). There is high confidence that the method is useful in interaction with decision makers and other stakeholders, helping to identify possible alternative sequences of measures over time, avoiding lock-in, and showing decision makers that there are several possible pathways leading to the same desired future (Haasnoot et al., 2012²²⁰¹; Haasnoot et al., 2013²²⁰²; Brown et al., 2014²²⁰³; Werners et al., 2015²²⁰⁴).

Alternatives can also be characterised through multiple attributes such as costs, effectiveness, co-benefits, social acceptability, etc., which in turn can be used in multi-attribute decision making methods (Haasnoot et al., 2013²²⁰⁵). An important attribute is transfer cost, which is the cost of course correction (switching from one alternative to another), reflecting the potential for path dependency (Haasnoot et al., 2019²²⁰⁶). Delaying decisions and opting for flexible measures introduces extra costs, such as transfer costs. Also, flexible measures are often more expensive than inflexible ones, and damages may occur whilst delaying the decision. An important question therefore is whether it is cheaper to implement a flexible measure now or to wait and implement a less flexible (i.e., cheaper) measure later in time when more information is at hand.

Technically more demanding methods such as real-options analysis (Dixit et al., 1994²²⁰⁷), and decision tree analysis (Conrad, 1980²²⁰⁸), can also find pathways that are economically efficient in terms of flexibility and timing of adaptation. There is little application of these approaches in the SLR literature. For example, Woodward et al. (2014) applied real-options analysis to determine flood defences around the Thames Estuary, London, England; Buurman and Babovic (2016) for climate-proofing drainage networks in Singapore; Dawson et al. (2018) for coastal rail infrastructure in southern England; and Kim et al. (2018) for assessing flood defences in southern England. A requirement for applying real-options analysis and decision tree analysis is to quantify today how much will have been learned at a given point in time in the future. The few applications of these methods to SLR-related decisions in the literature have generally used ad-hoc assumptions. For example, Woodward et al. (2011) assumed either perfect learning (i.e., in 2040, which SLR trajectory is occurring will be known) or no learning (i.e., uncertainty ranges and confidence in these remains as today). Others have derived learning rates from comparing past progress in SLR projections and then applied these to the future. An example is given by Dawson et al. (2018) who derive learning rates from the 2002 and 2009 SLR projections of the UK Climate Impacts Programme and apply these in real-options analysis.

Four general gaps can be identified in the literature. First, the generation of SLR information is insufficiently coupled to the use of this information in decision analysis. This constitutes a limitation, as different coastal decision contexts require different decision analysis methods, which in turn require different SLR information. Specifically, applications of decision analysis methods generally convert existing sea level information to fit their method, often misinterpreting the information, making arbitrary assumptions or losing essential information in the process (Hinkel et al., 2015²²¹⁰; Bakker et al., 2017²²¹¹; Van der Pol and Hinkel, 2018). Second, with the exception of adaptation pathway analysis, methods of robust and flexible decision making are under-represented in the literature despite their suitability (Van der Pol and Hinkel, 2018). Third, research is necessary to compare the various methods, to identity which methods are most suitable in which context and to develop consistent categorisations of methods (Hallegatte et al., 2012²²¹²; Haasnoot et al., 2013²²¹³; Hinkel et al., 2015²²¹⁴; Watkiss et al., 2015²²¹⁵; Dittrich et al., 2016²²¹⁶; Suckall et al., 2018²²¹⁷). Fourth, future research needs to address how to embed decision analysis better in real world planning and decision making processes, recognising that adaptation to SLR is a multi-stakeholder process often characterised by conflicting interests and interdependence between stakeholders (Section 4.4.3). Addressing these gaps requires closer cooperation between SLR sciences, decision science, and planning and governance scholars. An underlying challenge is to design and integrate relevant formal decision making approaches into the heterogeneous reality of local planning and decision making cultures, institutions, processes and practices, often with community-specific needs and requirements (see Box 4.4).