The rising of the global mean sea-level (GMSL) is observed since the early 20th century and is projected to continue and further accelerate over the 21st century [1
], hence posing a major challenge for coastal regions worldwide [3
]. Since the 1970s, the ocean thermal expansion and melting of land glaciers, largely caused by the anthropogenic global warming [4
], are the main contributors to the GMSL rise. While the thermal expansion is expected to continue increasing over the 21st century, the total contribution of ice mass loss by ice-sheets is projected to become more substantial, and it is the first driver of the GMSL rise acceleration since 1993 [5
]. The range of projections of the GMSL by the end of the 21st century is very large, however, namely due to high uncertainties in the understanding of physical processes that drive components of the GMSL and uncertainties in future greenhouse gas emissions. Since the release of the IPCC AR5 (Intergovernmental Panel on Climate Change Fifth Assessment Report) in 2013, the debate on long-term projections of GMSL has strongly focused on the potentially very large contribution of the Antarctica ice-sheet [2
], which constitutes a deep source of uncertainty.
Sea-level changes at the regional scale can substantially differ from GMSL change. Thermal expansion is modulated regionally by changes in ocean circulation, density, and atmospheric pressure [12
]. In addition, water mass transfer from land to the ocean—due, e.g., to mountain glaciers and ice-sheets melting or groundwater extraction—induces regional changes by altering the Earth’s gravity field, Earth rotation, and solid-Earth deformation [14
]. Ongoing changes in the solid Earth are also still caused by the viscous adjustment of the mantle to the important mass redistribution that followed the Last Glacial Maximum (named Glacial Isostatic Adjustment; GIA) [15
]. At the European scale, the influence of GIA is particularly prominent in the Scandinavian area. Finally, at a local scale, high-resolution oceanic processes [16
] and vertical ground motion [17
] can affect further the relative sea-level.
Projections of future regional sea-level are crucial to support adaptation planning. So far, coastal adaptation practitioners have often relied on IPCC sea-level projections, which are provided in the form of a likely range (probability larger than 66%), and do not reflect the whole range of uncertainties of sea-level projections [19
]. However, as coastal climate services are being developed [24
], it is increasingly recognized that different types of sea-level projections are required depending on the degree of uncertainty tolerance of decision-makers [26
]. If the uncertainty tolerance is medium to high, one can use probabilistic projections that are particularly suited to identify the adaptation alternative that has the best-expected outcome [26
]. In contrast, in the case of high-risk aversion, the uncertainty tolerance is low and, therefore, robust decision-making could be required [26
]. Probabilistic projections cannot be used within robust decision-making approaches because ensuring robustness implies testing adaptation options against any plausible scenarios, whereas the tail of sea-level projections is highly uncertain [22
]. Instead, one should consider high-end projections or scenarios, which explore plausible—although unlikely—high impact sea-level scenarios beyond the likely range [28
]. In particular, high-end sea-level scenarios are particularly useful to plan the full range of coastal adaptation responses [26
Cascading effects from sea-level rise to coastal impacts are generally quantified with the aim of computing the most likely impacts [32
]. In fact, previous studies that have addressed future high-end sea-level changes at the regional and local scale [34
], generally did not assess subsequent coastal impacts on erosion consistently, except in some locations. For instance, Jiménez et al. [37
] projected the sea-level rise-induced erosion on the Catalan coast and quantified the impacts on beach functions. Multiple challenges pertain to the evaluation of high-end or likely shoreline changes. First, there is no unique approach to quantify future sea-level rise impacts on shoreline retreats, as shown for example for the coast of Asturias [38
], Balearic Islands [39
], or Black Sea beaches [40
]. Furthermore, in addition to sea-level rise, multiple natural and anthropogenic processes are acting at different space and time scales, and contribute to the observed shoreline changes [41
]. Finally, the validation of prospective shoreline change modeling frameworks remains challenging because of the observed variability of shoreline changes. For example, it has been estimated that half of the world’s sandy beaches are stable, one quarter is accreting, and a remaining quarter is eroding [43
], in particular, due to human interventions [44
]. Nevertheless, in the climate change context, sandy shorelines will be altered by waves, storm surges, tides, and river flows [38
], whose changes have either not fully quantified effects at European scale or have fewer impacts than projected sea-level rise [49
]. As a consequence, sea-level rise is the most common coastal impact of climate change considered in European regulations, engineering designs, and adaptation plans [50
]. Shoreline retreat projections induced by a high-end sea-level rise at the European scale and for each individual European country has, however, not been quantified yet.
The present study provides a first estimate of the contribution of sea-level rise for shoreline change of European sandy coasts for likely and high-end sea-level rise scenarios by 2100 under business-as-usual greenhouse gas emissions. Our study focuses on beaches, therefore not considering other vulnerable coasts such as non-consolidated cliffs, artificialized coasts and wetlands, and their potential permanent inundation. To do so, we design high-end sea-level scenarios at the European scale that we combined with a database of European coastal settings (EUROSION, 2004) and a recently released nearshore slope dataset [52
]. Our sea-level projections account for all relevant regional sea-level component (i.e., sterodynamic, land-water mass transfer, etc.), except local contributions such as local vertical ground motion or high-resolution ocean changes, which needs to be assessed locally [36
]. We build two high-end sea-level scenarios, informed by the most recent literature, following two distinct approaches. The first relies on the upper bound of the likely range, while the second follows a “worst-model” approach. Finally, we deliver separate shoreline change projections and beach area loss for each European coastal country.
3. Approach for the Assessment of High-End Coastal Impacts
High-ends are defined as plausible—although unlikely—high-impact sea-level scenarios [28
]. There is still no unique approach in the sea-level literature on how to quantify high-ends today. In fact, different lines of evidence can be used to define potential future contributions and to combine them [28
]. High-end scenarios are often designed in a probabilistic frame based on: (i) Representative Concentration Pathway (RCP) scenarios of the IPCC, (ii) assumptions with regard to physical processes to be considered (e.g., drivers of Antarctic ice-sheet melting) and (iii) particular subsets of models [72
]. High-end projections are also informed based on expert elicitations [73
], whether within a probabilistic approach or not [75
Here, since we focused on high impact scenarios of high greenhouse gas emissions, our sea-level projections rely on the RCP8.5 emission scenario and its likely range. We deliver two high-end scenarios; the high-end A which is less pessimistic, defined based on the upper limit of the RCP8.5 likely range, while the high-end B follows a “worst-model” approach, that is, not necessarily the upper limit to sea-level rise, which may exceed current modeling outcomes. Details on the high-end A and B scenarios design are given in the following two subsections.
3.1. Sterodynamic Component: CMIP5 Model Selection and High-End Definition
shows the sea-level change RCP8.5 projections of CMIP5 models, as provided by the ICDC data center, at the end of the 21st century and calculated for the seven ocean basins surrounding the the European coastline (see Appendix A
). The model spread is particularly pronounced in the Atlantic sector, with sterodynamic changes projection ranging from 0.3 to 0.7 m. Despite the overall consistency in the magnitude of model projections at the European scale (e.g., MIROC5 and ACCESS1-0 climate models are found in the upper tail of all distributions), we note that there are significant changes in model ordering between basins. This reflects the regional influence of ocean dynamics and circulation changes on sea level.
Among the 21 models, MIROC-ESM and MIROC-ESM-CHEM project anomalously large sea-level rise in the Atlantic and North Sea areas. If these two models are discarded, the distribution obtained by the 19 remaining CMIP5 models in these areas is no longer significantly different from a Gaussian distribution according to the Shapiro–Wilk normality test. Furthermore, by 2100, the global-mean thermosteric sea-level rise of these two models (0.5 m for the RCP8.5 scenario) exceeds the median global-mean thermosteric sea-level rise of all other models (0.3 m) beyond 5 sigma [76
]. Finally, the CMIP5 historical MIROC-ESM and MIROC-ESM-CHEM simulations revealed unrealistic sea-surface height values of -15 m in the Mediterranean area that may suggest important biases in the regional sea-level calculations in these two models [77
]. MIROC-ESM and MIROC-ESM-CHEM were, therefore, removed from our model selection.
also shows that semi-enclosed seas are not fully covered by all models—among the 21 models, only 14 and 12 cover the Baltic and Mediterranean basins, respectively. These differences between model spatial coverage result in inconsistencies when computing multi-model ensemble statistics, which in turn could significantly affect the spatial homogeneity of regional sea-level rise projections and hence alter their credibility. Furthermore, the rather coarse resolution of AOGCMs prevents an accurate representation of small-scale processes, for example, the water exchange at Gibraltar, which in turn affects regional sea-level estimates in marginal seas [78
]. Hence, to remove these potential sources of errors, the Mediterranean sterodynamic sea-level projections are constrained with those of the Atlantic area near Gibraltar, which is the Mediterranean Sea entry point. In a similar way, the Baltic Sea projections are calculated using those of the North Sea. This procedure leads to an increase in the consistency of multi-model statistics compared to those relying on the CMIP5 models within the semi-enclosed basins [54
] and is also more realistic with regard to the processes governing multi-decadal sea-level changes in these marginal seas [81
]. Finally, our selection choice leads to an increase in the Mediterranean projections compared to previous assessments based on AR5 data.
Hereafter, we follow the IPCC AR5 method to define the “likely-range” and therefore compute it as the standard deviation interval around the multi-model mean multiplied by 1.64. For the two high-end scenarios, sterodynamic contributions for each grid cell are defined using the upper bound of the multi-model likely-range (High-end A) and the multi-model outcome maximum value (High-end B, worst model estimate).
3.2. Barystatic-GRD Components
Barystatic-GRD induced contributions to European sea-level change projections are defined based on sea-level barystatic-fingerprints computed from data of the ICDC data center, which we multiply by the global sea-level equivalent for each component [54
]. The global sea-level equivalent contribution corresponds to the change in the sea-level induced by the transfer of water mass from land water storage to the ocean. As shown in Table 2
, the high-end A scenario—corresponding to a “moderate” scenario—is prescribed based on the barystatic component projections of the upper limit of the RCP8.5 global warming scenario “likely-range” [1
], except for Antarctica ice-sheet dynamics contribution which experienced several updates since 2013 as summarized in the recent IPCC Special Report on Ocean and Cryosphere in a Changing Climate (SROCC) [2
]. For the high-end B—a more pessimistic scenario—the “worst model approach” was followed.
In the AR5, projections of sea-level equivalent for glaciers were estimated based on outcomes of global glacier models forced by temperature and precipitation projections from RCP scenarios simulations by climate models [85
]. Revisions of existing projections (e.g., Marzeion et al., 2012 [86
]) and new modeling estimates [88
] resulted in slightly lower glacier mass losses [79
]. Nonetheless, given the limited evidence of substantial changes in glacier mass loss projections since the AR5 [2
], glacier contributions for the high-end A scenario are defined as the upper limit of the likely range, which corresponds to a sea-level equivalent of 0.26 m. Following the “worst-model approach” glacier sea-level equivalent (SLE) for the high-end B is set to 0.29 m, which is the maximum estimate obtained by Marzeion et al. (2012) [86
] when forcing their glacier model with CMIP5 simulation outputs of the Hadley Global Environment Model 2—Earth System (HADGEM-ES).
AR5 projections of land water account for groundwater depletion [89
], which contributes to sea-level rise, and a negative contribution through increasing land water storage due to dams over the 21st century [90
]. Although Wada et al. (2016) [91
] showed recently that previous studies might have inflated groundwater depletion contribution to sea-level rise since full groundwater drainage to the ocean was assumed without accounting for pumped water remaining on land (~20%), the AR5 projections have not yet been re-assessed substantially. Therefore, in the absence of worst-case model estimates in the literature (to our knowledge), both high-end scenarios thus rely on the upper limit of the likely range provided in the AR5 and SROCC (0.11 m).
Greenland ice-sheet contribution to sea-level change is driven by changes in surface mass balance (SMB) and dynamic effects (DYN). IPCC AR5 [1
] (and unchanged in SROCC [2
]) estimated that, by 2100, the upper limit of the likely range of Greenland’s SLE contribution would be 0.28 m under the RCP8.5 scenario, dominated by SMB by two thirds. This value defines our high-end A scenario. Recently, Fürst et al. (2015) [92
] forced an ice-sheet model with 10 AOGCMs from the CMIP5 dataset and found a slightly smaller median contribution of Greenland to global sea-level rise by 2100 than in the AR5 (0.10 m versus 0.15 m). Their highest Greenland SLE projection (0.17 m by 2100) was found when the ice-sheet model was forced by canESM2 RCP8.5 simulation results. However, using also a subset of CMIP5 models (including canESM2) and comparing with reanalysis, Delhasse et al. (2018) [93
] have shown that some changes observed over the last two decades in atmospheric circulation (i.e., increase in blocking high frequencies in summer) could not be reproduced by the CMIP5 models. They further concluded that, if the current summer atmospheric circulation pattern over Greenland happens to persist, projected Greenland SMB contribution to sea-level rise could be amplified by a factor of two for a similar temperature increase. Therefore, we build our high-end B scenario by considering that the largest model projection found in Fürst et al. (2015) [92
] could be amplified by a factor two following the arguments of Delhasse et al. (2018) [93
]; this would lead to a Greenland SLE of 0.34 m by 2100. Note that the most recent expert elicitations on ice-sheet contribution suggested an upper bound of the likely range at 60 cm [73
], which gives us confidence that our high-end B Greenland component projection remains credible.
Since the release of the AR5, the Antarctica ice-sheet contribution has been highly debated [2
]. It could be one of the most important contributions to future sea-level rise, in particular under a high global warming scenario. The uncertainties on this contribution are, however, very large and strongly depend on the understanding of Antarctica ice-sheet dynamic processes and their evolution under a warming climate. Lately, two mechanical processes that may trigger important dynamic mass loss of the ice sheet have been intensively discussed. First, the marine ice sheet instability (MISI), which is probably observed already in West-Antarctica [94
], and second, the marine ice cliffs instabilities (MICI), more hypothetical than MISI (and never observed over the instrumental period), which involves a rapid retreat of ice shelves through hydrofracturing and subsequent collapse of ice cliffs formed at the ice sheet margins [8
]. Although the effectiveness and magnitude of these two mechanisms remain to be determined, considering their potential contribution to SLR strongly inflates the likely range provided in the AR5. Therefore, the median contribution of the Antarctica ice-sheet dynamics has been recently re-assessed to 0.16 m in 2100 under the RCP8.5 scenario in the SROCC [2
] (against 0.08 m in the AR5). Note that among the studies (five in total) considered in the SROCC to retrieve this new estimate, some relied on simulations that included MISI (but not MICI). The large spread in the projections of these five studies further led to extend the upper bound of the RCP8.5 likely range substantially from 0.19 m (AR5) to 0.37 m (SROCC) by 2100. The latter value is hence used for the high-end A scenario. For the high-end B, we consider a mean projection assuming MICI (and not a worst-case model outcome) because the confidence in MICI projection is still debated, and it is unsure that it will be initiated over the 21st century. In their former paper, DeConto and Pollard (2016) [8
] estimated that MICI could contribute to global sea-level rise to more than 1 m by 2100. Very recently, Edwards et al. (2019) [9
] revisited the latter results by considering the full range of uncertainties of the ice-sheet model parameters used by DeConto and Pollard (2016) [8
]. This more robust statistical treatment by Edwards et al. (2019) [9
] led to revised downward the DeConto and Pollard (2016) [8
] projection to 0.8 m. The latter value 0.8 m is hence used for the high-end B.
In contrast with Greenland, the Antarctic ice-sheet SMB is projected to contribute to a drop of global sea-level, namely due to an increase of snow accumulation under future atmospheric warming [96
]. Given the absence of new estimates since the AR5 (particularly because no new CMIP simulations are available), the high-end A scenario was designed based on the upper bound of the AR5 likely range. The high-end B assumes no increase of precipitations over the Antarctic.
5. Discussion and Conclusions
In this study, we provided first pass high-end estimates of the sea-level rise contribution to European sandy coastline retreat by the end of the 21st century. High-end scenarios are high-impact and unlikely—but possible—sea-level scenarios that are particularly suited in robust decision-making contexts [26
]. Three sea-level scenarios were considered, all built upon greenhouse gas emissions that follow an RCP8.5 trajectory. The first scenario is simply defined by summing the different sea-level contributions and considering the corresponding likely range, as originally published in the AR5 and updated recently in the SROCC. The second scenario (High-end A) relies on the upper bound of the AR5/SROCC likely range. Finally, the third scenario (High-end B) follows a worst-model approach; i.e., we selected the outcome of the model showing the highest sea-level projection for every component based on the most recent literature. The shoreline changes induced by the sea-level rise were estimated from the Bruun rule applied to the EUROSION database sandy beach segments, and for which, we considered either the new nearshore slope dataset provided by Athanasiou et al. (2019) [52
] or a uniform slope of 1%, a common approach used for continental-to-global scale shoreline retreat assessments [61
The RCP8.5, high-end A and high-end B scenarios induce, by the end of the 21st century, a relative sea-level rise off European coasts larger than 0.7 m, 1.2 m and 1.7 m, respectively. Nevertheless, these projections feature substantial regional variations, which in turn strongly influence the spatial distribution of SLR-induced shoreline changes in Europe. Our main findings are:
The magnitude and regional distribution of SLR-induced shoreline change projections by 2100 utterly depend on the nearshore slope, the regional distribution of sea-level changes (i.e., hence the various regional contributions) and the trajectory of the future scenario.
Ignoring the variability of nearshore slopes and assuming a 1% constant uniform nearshore slope instead may lead to a substantial underestimation of SLR-induced shoreline retreat and beach area removal (reduction by 50%) in Europe. In the absence of any coastal adaptation measure, and assuming an infinite erosion potential of each EUROSION segment, we found that Europe is projected to accumulate a land loss area of 4040 km2, 8950 km2, and 13,470 km2 for the RCP8.5 median, High-end A and High-end B scenarios, respectively.
The sequencing of countries with respect to their exposure to future shoreline retreat varies very importantly with scenarios. In particular, in Northern Europe, the impacts of high-end sea-level scenarios are disproportionately high compared to those of likely scenarios and to those of the Mediterranean area. This is because the softer the nearshore slope, the more sensitive to sea-level changes in the shoreline. Subsequently, large uncertainties in future regional sea-level changes affect primarily coastal regions with gentle nearshore slopes. This results in substantial changes in the ranking of coastal impacts and adaptation needs in Europe, which may be relevant to consider in adaptation finance mechanisms.
In a previous assessment [71
], the European Union land loss rate without adaptation measure was estimated to be of 3.4 km2
/year, 6.7 km2
/year, 9.9 km2
/year and 16.4 km2
/year in 2010, 2030, 2050, and 2100, respectively, considering a climate change scenario leading to a global-mean sea-level rise of 45 cm by 2100 (scenario A2). This corresponds to a European cumulated land loss area of ~1000 km2
over the 21st century. To a zeroth-order approximation, assuming proportionality between land loss and sea-level, a global-mean sea-level rise of 84 cm by 2100 (RCP8.5) would lead to a cumulated land loss of 1870 km2
, which, at first glance, appears to be consistent with our results based on the 1% uniform nearshore slope (we found 2060 km2
loss). However, the DIVA model [60
], upon which these estimates rely, strongly differs from our modeling framework. First, the percentage of erodible sandy shoreline at the global scale in DIVA is about 11% [33
], while in EUROSION and other independent global databases [43
], this percentage is generally larger than 20%. Second, the DIVA model not only accounts for the direct shoreline change predicted by the Bruun rule but also considers the indirect erosion in the tidal basin using a simplified version of the ASMITA model [61
]. More importantly, they found that indirect erosion accounts for more than 50% of the total land loss [61
]. In our study, indirect erosion in the tidal basin is not accounted for because the open tidal basins are either not included in the EUROSION database or because they are dominated by muddy coasts, which are not identified as sandy coasts (Table 1
). Overall, this indicates that by assuming a 1% uniform nearshore slope, both our and DIVA assessments potentially underestimate the land loss area (in the absence of adaptation measures) and, therefore, likely minimize the subsequent socio-economic impact. Furthermore, as shown in our study, ignoring the local nearshore slope characteristics—at least for Europe—could lead to a large underestimation of the land loss area. This stresses the need to promote further coastal impact models intercomparison exercises (such as COASTMIP) in order to quantify coastal impacts in face of sea-level rise better.
Besides, the above-discussed sources of uncertainty, our shoreline change modeling approach has limitations that may significantly affect our shoreline change estimates. First, the limitations inherent to the Bruun rule are mainly its non-applicability in coasts where processes other than sea-level rise-induced cross-shore sediment transport prevail, and the lack of its comprehensive validation, since sea-level rise-induced shoreline change can be significantly masked by shoreline changes driven by other processes (e.g., sand losses during storms, aeolian transport, alongshore gradients in longshore transport) [67
]. Recently, alternative approaches to the Bruun rule have been proposed. They assume that the effects of sea-level rise differ depending on waves and storm climates in each areas and compute sediment losses at dunes toes as sea-level rises [38
]. So far, these approaches have delivered lower rates of shoreline retreat than the Bruun rule [102
]. Neither the Bruun rule nor the alternative modeling approaches have been convincingly validated yet. However, the fact that the Bruun rule predicts shoreline retreats larger than those of other models suggests it can be used for a high-end estimation of shoreline changes induced by sea-level rise.
Additional limitations of our analysis include: (i) the fact that local vertical ground motion (beside GIA) are not accounted for, (ii) the non-consideration of nearshore slope uncertainties and potential beach slope changes in the future (e.g., through notably to changes in the wave climate [103
]), (iii) the consideration of beaches only, while other systems (e.g., wetlands) are also vulnerable to sea-level rise, (iv) the fact that we do not assess shoreline changes induced by the permanent inundation of low-lying areas (which however will in practice very much depend on adaptation practices), or (v) the assumption that erosion can continue indefinitely rather than be limited by geological constraints. Regarding the latter point, nonetheless, a qualitative estimate of the geological constraint on the erodibility is provided in the coastline geology layer of the EUROSION database. This layer lists more than 30 types of lithology that we classified with respect to three degrees of erodibility: weak (e.g., granite, basalt), moderate (e.g., limestone rock, sandstone) and high (e.g., sand, loess). The shoreline distribution per country (expressed in %) of the degree of erodibility is shown in Figure 6
(bottom). This qualitative analysis can be viewed as a confidence index of the land loss projections. For instance, Ireland, Greece, and the United Kingdom present a substantial part of the shoreline (more than 50%) that has a weak to moderate erodibility degree. Hence, it is very likely that our high-end land loss projections are overestimated for these countries. In contrast, Germany or Denmark sandy coasts are found to be more prone to erosion as revealed by their large erodibility potential; in this case, our projections are less affected. Finally, as our estimate of shoreline changes only consider erosion of sandy beaches induced by sea-level rise, the actual shoreline retreat induced by sea-level rise could be larger in many low-lying coastal environments. For example, Mediterranean lagoon-type coasts bounded by sandy spits will not only be affected by the erosion of the sandy coast, but also by the permanent inundation of parts of the low-lying coastal plain lying behind the sand spit.
The high-end approach explored in our study is particularly adapted for decision-making applications and adaptation planning, especially for stakeholders with low tolerance to uncertainty [26
]. Shoreline retreat projections are, however, deeply uncertain, owing to the high diversity of uncertainty sources and their large magnitude. Hence, to reduce uncertainties and improve the reliability of shoreline change projections, it is crucial to improve coastal impact models and regional sea-level projections. In addition, to account for the full range of uncertainties and quantify their relative weight on future projections, the extra-probabilistic framework appears well suited. This will be the purpose of near-future work.