Sea-level rise is one of the consequences of the current climate warming induced by anthropogenic greenhouse gas emissions [1
]. Specifically, sea-level rise is due to the following processes [1
]: First, the warming of the ocean triggers thermal expansion [2
] and increased melting of marine ice-sheets and glaciers, especially in Antarctica [5
]. Second, the warming of the atmosphere increases surface melting on glaciers and Greenland [8
]. Third, changes in precipitations modify the surface mass balance of the cryosphere [11
]. Fourth, cascading impacts resulting from the previous processes, such as the increased velocity of glaciers after removal of ice shelves in Antarctica, also contribute to more rapid sea-level rise [13
]. Finally, the extraction of groundwater is expected to become a net contribution to sea-level rise in the future, although it has been so far roughly counterbalanced by the construction of new dams [14
At the regional scale, ocean dynamics and the inverse barometer effect modulate thermal expansion [4
]. The mass contributions of mountain glaciers and ice sheets melting and groundwater extractions have regional patterns as well due to the response of the solid Earth to present-day mass redistributions [16
]. Furthermore, the Earth is still responding to the past deglaciation (Global isostatic adjustment, GIA), so that, for example, regions in Scandinavia and Canada are still uplifting and other areas located in the periphery of past ice sheets such as the Chesapeake Bay in the United States are subsiding [20
]. Finally, regional to local oceanic processes and vertical ground motions can alter regional relative water levels (with respect to the coast) and tides significantly [21
]. State of the art global and regional sea-level projections available today consider all the processes above except the latter regional to local oceanic processes and vertical land motions [17
Sea-level rise is a challenge for coastal adaptation, not only because it involves multidecadal and even multicentennial timescales, but also because it is associated with large uncertainties, even as soon as by the end of the 21st century [32
]. Many previous studies have delivered probabilistic sea-level projections to convey these uncertainties to coastal adaptation practitioners [30
]. Such projections usually assume a given climate scenario (e.g., RCP2.6 or 1.5° of global warming) and deliver a probability distribution to describe their uncertainties. However, computing probabilistic projections requires making choices, including: selecting a subset of climate models [4
]; choosing a particular ice-sheet modeling experiment (e.g., assuming Marine Ice Sheet Instabilities or Marine Ice Cliffs Instabilities [33
]); introducing an expert judgement on the top of modelling outcomes to overcome limitation of models; or assuming dependencies among future contributions (e.g., quantifying the correlation between the melting of glaciers and ocean thermal expansion) [35
]. These choices will result in different and potentially equally credible probabilistic sea-level projections [36
]. Consequently, coastal adaptation practitioners may have difficulties to select those most appropriate to their needs [37
In this paper, we deliver probabilistic sea-level projections by 2100 based on conservative assumptions (Available numerically as Supplementary Material
). We call these “low-end probabilistic sea-level projections”, because they are a credible—but extremely favorable—probabilistic representation of future sea-level rise. In other words, we call “low-end probabilistic sea-level projections” those delivering higher probability of exceedance than most credible probabilistic projections available in the literature. High-end sea-level scenarios and projections have received a lot of attention so far, because they are associated with high impacts and adaptation needs [40
]. Nonetheless, low-end projections are equally useful for practitioners applying robust decision making or assessing minimum adaptation needs.
To produce probabilistic projections relevant for low-end sea-level changes, we select the smallest published probabilistic projections of mountain glaciers [41
], of ice-sheet melting in Greenland [10
] and in Antarctica [7
]. Furthermore, we select a subset of models from the Coupled Model Intercomparison Project phase 5 (CMIP5) delivering moderate thermal expansion (Section 2
). We use state-of-the art approaches to compute global and regional sea-level projections, relying only on modeling outcomes, without adding any expert judgement on the top of them, and selecting dependencies schemes among contributors that favor low sea-level change values up to the 50th percentile of the distribution [35
] (Section 3
). We deliver not only global, but also regional sea-level projections because they are the only one applicable at regional scales and therefore for adaptation (Section 4
). Finally, we conclude by summarizing why our projections can be considered as low-end (Section 5
5. Discussion and Conclusions
In this paper, we provide global and regional low-end probabilistic sea-level projections (available as Supplementary Material
). Our projections are probabilistic because they are generated from probabilistic representations of all components of future sea-level rise, mostly obtained by fitting Gaussian distributions to global and regional modeling outcomes. They are low-end projections because they represent sea-level rise and its uncertainties under very optimistic sterodynamic and barystatic sea-level changes. Specifically, our probabilistic sea-level projections are very optimistic because:
We used recent studies delivering the smallest probabilistic contributions from Mountain Glaciers, Greenland, and Antarctica [7
For Antarctica, we relied on a study that probably underestimated the impact of ocean warming on the Antarctic marine ice-sheet melting, according to its own assessment [7
For the sterodynamic sea-level changes, we removed AOGCMs giving high thermal expansion values; in the regional AR5 assessment, these outliers increased the mean and uncertainties of the thermal expansion.
We relied on modelling outcomes only, and ignored the procedure consisting of multiplying the standard deviation of model outcomes by 1.64 applied in the AR5; we argue that this procedure artificially extends the lower tail of the distribution of future sea-level rise, whereas AOGCMs have been essentially criticized so far for minimizing future sea-level changes.
We assumed full dependency among the sterodynamic, Mountain Glaciers, and Greenland melting components, which slightly shifts the lower tail of the probability distribution to the right compared to partial dependency schemes;
We did not find physical arguments supporting probabilistic projections below our projections in the published literature.
Hence, our projections can be considered as a low-end probabilistic projection if not a probabilistic lower bound. The lower tails of some existing probabilistic sea-level projections are lower than ours, but these lower tails lack credibility, due to their being affected by modelling choices essentially made to deliver more realistic upper tails (see Section 4
). If new processes are identified in the future, slightly more conservative low-end probabilistic projections might become credible. However, we speculate that the resulting distribution should not be that much different from ours, as otherwise, this would mean an inexplicable decrease of the rates of sea-level rise despite climate warming.
Even with these optimistic assumptions, our results show that an acceleration of sea-level rise should be expected during the 21st century under business as usual greenhouse gas emissions (RCP8.5), which was not guaranteed by all previous probabilistic sea-level rise projections (see Section 4.1
). This result has significant implications for adaptation: for example, even under these optimistic projections, the regulatory sea-level scenario adopted in France has roughly one in two chance to be exceeded by 2100 under business as usual greenhouse gas emissions (RCP8.5). This result might be considered in future updates of such adaptation policy and engineering guidance, especially if the gap between climate mitigation targets and the real greenhouse gas emissions continues to grow.
There are physical arguments to argue that our projections are extremely conservative: for example, other trustful subsets of AOGCMs have delivered higher thermal expansion (e.g., in Kopp et al. (2014) [19
]). Furthermore, our study relies on CMIP5 data, which will be soon superseded by CMIP6 with potentially higher climate sensitivity and therefore thermal expansion. As a consequence, they should only be used cautiously, keeping in mind that they have good chances to be exceeded in the real world, so that nuisances and losses due to flooding, erosion, and salinization will be significantly larger [83