Editing IPCC:AR6/SROCC/Chapter-4 (section)

=== 4.2.3 Projections of Sea Level Change ===

<div id="section-4-2-3projections-of-sea-level-change-block-1"></div>

As a consequence of climate change, the global and regional mean sea level will change. Coupled climate models are used to make projections of the climate changes and the associated SLR. Results from the CMIP5 model archive used for AR5 provide information on expected changes in the oceans and on the evolution of climate, glaciers and ice sheets. New estimates from CMIP6 are not yet available and will be discussed in the IPCC 6th Assessment Report (AR6), hence only a partly updated projection can be presented here.

Coupled climate models can be applied on century time scales, to provide estimates of the steric (temperature and salinity effects on sea water density) and ocean dynamical (ocean circulation) components of sea level change, both globally and regionally. However, the glacier and ice sheet component are calculated off-line based on temperature and precipitation changes. In the AR5 report, changes in the SMB of glaciers and ice sheet were calculated from the global surface air temperature. In addition, GCMs also resolve climate variability related to changes in precipitation and evaporation. These changes are used to calculate short duration sea level changes (Cazenave and Cozannet, 2014 <sup>[[#fn:r363|363]]</sup> ; Hamlington et al., 2017 <sup>[[#fn:r364|364]]</sup> ). With various degrees of success those models capture ENSO, PDO and other modes of variability (e.g., Yin et al., 2009; Zhang and Church, 2012 <sup>[[#fn:r365|365]]</sup> ), which affect sea level through redistributions of energy and salt in the ocean on slightly longer time scales. Off-line temperature and precipitation fields can be dynamically or statistically downscaled to match the high spatial resolution required for ice sheets and glaciers, but serious limitations remain. This deficiency limits adequate representation of potentially important feedbacks between changes in ice sheet geometry and climate, for example through fresh water and iceberg production that impact on ocean circulation and sea ice, which can have global consequences (Lenaerts et al., 2016 <sup>[[#fn:r366|366]]</sup> ; Donat-Magnin et al., 2017 <sup>[[#fn:r367|367]]</sup> ). Another limitation is the lack of coupling with the solid Earth which controls the ice sheet evolution (Whitehouse et al., 2019 <sup>[[#fn:r368|368]]</sup> ). Dynamics of the interaction of ice streams with bedrock and till at the ice base remain difficult to model due to lack of direct observations. Nevertheless, several new ice sheet models have been generated over the last few years, particularly for Antarctica (Section 4.2.3.1) focusing on the dynamic contribution of the ice sheet to sea level change, which remains the key uncertainty in future projections (Church et al., 2013), particularly beyond 2050 (Kopp et al., 2014 <sup>[[#fn:r370|370]]</sup> ; Nauels et al., 2017b <sup>[[#fn:r371|371]]</sup> ; Slangen et al., 2017a <sup>[[#fn:r372|372]]</sup> ; Horton et al., 2018 <sup>[[#fn:r373|373]]</sup> ).

Information beyond that provided by climate models is needed to describe local and RSL changes. Geodynamic models are used to calculate RSL changes due to changes in ice mass in the past and future. This includes solid Earth deformation, gravitational and rotational changes, as ice and water are redistributed around the globe. Input for those models is provided by the mass changes following from the off-line land ice models, time series of terrestrial water mass changes which typically require climate input, and reconstruction of past ice sheet changes over the last glacial cycle provided by coupled ice-Earth models (de Boer et al., 2017). Combining these different models leads to projections of RSL (Section 4.2.3.2).

At the local spatial scales of specific cities, islands and stretches of coastlines, hydrodynamical models (Section 4.2.3.3) and knowledge about anthropogenic subsidence are necessary to analyse the impacts of highly variable processes leading to ESL, such as tropical cyclone-driven storm surges. These hydrodynamical models are capable of providing statistics on the variability or the change in variability of the water level required for flood risk calculations at specific locations and at spatial scales of less than 1 km. The models also rely on input from climate models, like temperature, precipitation, wind regime, and storm tracks (Colbert et al., 2013 <sup>[[#fn:r374|374]]</sup> ; Garner et al., 2017 <sup>[[#fn:r375|375]]</sup> ).

In summary, climate models play an important role at the various stages of projections in providing, together with emission scenarios, geodynamic, ice-dynamic and hydrodynamic models, the required information for hazard estimation for coasts and low-lying islands. This report relies on results of the CMIP5 model runs.

<div id="section-4-2-3-1contribution-of-ice-sheets-to-gmsl"></div>

<span id="contribution-of-ice-sheets-to-gmsl"></span>
==== 4.2.3.1 Contribution of Ice Sheets to GMSL ====

<div id="section-4-2-3-1contribution-of-ice-sheets-to-gmsl-block-1"></div>

<span id="greenland"></span>
===== 4.2.3.1.1 Greenland =====

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., 2014 <sup>[[#fn:r376|376]]</sup> ; Enderlin, 2014 <sup>[[#fn:r377|377]]</sup> ; van den Broeke et al., 2016 <sup>[[#fn:r378|378]]</sup> ). 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., 2017 <sup>[[#fn:r379|379]]</sup> ). Patterns of surface melt on Greenland are highly dependent on regional atmospheric patterns (Bevis et al., 2019 <sup>[[#fn:r380|380]]</sup> ), 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., 2016 <sup>[[#fn:r381|381]]</sup> ; Ryan et al., 2018 <sup>[[#fn:r382|382]]</sup> ; Trusel et al., 2018 <sup>[[#fn:r383|383]]</sup> ; Ryan et al., 2019 <sup>[[#fn:r384|384]]</sup> ). 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., 2019 <sup>[[#fn:r385|385]]</sup> ). 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., 2002 <sup>[[#fn:r386|386]]</sup> ), 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., 2015 <sup>[[#fn:r387|387]]</sup> ; Flowers, 2018 <sup>[[#fn:r388|388]]</sup> ), which is supported by ice sheet model experiments (Shannon et al., 2013 <sup>[[#fn:r389|389]]</sup> ). In sum, uncertain climate projections (Edwards et al., 2014 <sup>[[#fn:r390|390]]</sup> ), albedo evolution, uncertainties around meltwater buffering by firn, complex processes linking surface, englacial and basal hydrology with ice dynamics (Goelzer et al., 2013 <sup>[[#fn:r391|391]]</sup> ; Stevens et al., 2016 <sup>[[#fn:r392|392]]</sup> ; Noël et al., 2017 <sup>[[#fn:r393|393]]</sup> ; Hempelmann et al., 2018 <sup>[[#fn:r394|394]]</sup> ) and meltwater induced melting at marine-terminating ice fronts (Chauché et al., 2014 <sup>[[#fn:r395|395]]</sup> ), and coarse spatial model resolution (Pattyn et al., 2018 <sup>[[#fn:r396|396]]</sup> ), all continue to provide substantial challenges for ice sheet and SMB models.

Greenland-scale ice sheet modelling since AR5 (Edwards et al., 2014 <sup>[[#fn:r397|397]]</sup> ; Fürst et al., 2015 <sup>[[#fn:r398|398]]</sup> ; Vizcaino et al., 2015 <sup>[[#fn:r399|399]]</sup> ; Calov et al., 2018 <sup>[[#fn:r400|400]]</sup> ; Golledge et al., 2019 <sup>[[#fn:r401|401]]</sup> ; Aschwanden et al., 2019 <sup>[[#fn:r402|402]]</sup> ) 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., 2014 <sup>[[#fn:r403|403]]</sup> ). Recent modelling studies use higher-order representations of ice flow (Fürst et al., 2015 <sup>[[#fn:r404|404]]</sup> ), include more explicit representations of ice sheet processes including subglacial hydrology (Calov et al., 2018 <sup>[[#fn:r405|405]]</sup> ), run the models at higher resolution and with updated boundary conditions (Aschwanden et al., 2019 <sup>[[#fn:r406|406]]</sup> ), and account for two-way coupling between the ice sheet and the global ocean (Vizcaino et al., 2015 <sup>[[#fn:r407|407]]</sup> ; Golledge et al., 2019 <sup>[[#fn:r408|408]]</sup> ). Among these studies, Fürst et al. (2015), Vizcaino et al. (2015) <sup>[[#fn:r409|409]]</sup> , and Aschwanden et al. (2019) provide projections following RCP2.6, RCP4.5, and RCP8.5 emissions scenarios. Calov et al. (2018) <sup>[[#fn:r412|412]]</sup> and Golledge et al. (2019) <sup>[[#fn:r413|413]]</sup> did not consider RCP2.6. Edwards et al. (2014) <sup>[[#fn:r414|414]]</sup> 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) <sup>[[#fn:r415|415]]</sup> 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) <sup>[[#fn:r417|417]]</sup> 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., 2013 <sup>[[#fn:r419|419]]</sup> ), 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) <sup>[[#fn:r420|420]]</sup> , 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., 2015 <sup>[[#fn:r421|421]]</sup> ) also fall below the ''likely'' range of AR5 estimates. In contrast, the study by Aschwanden et al. (2019) <sup>[[#fn:r422|422]]</sup> 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, 2006 <sup>[[#fn:r423|423]]</sup> ), 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) <sup>[[#fn:r424|424]]</sup> are closest to those from an updated, structured judgement of glaciological and modelling experts (Bamber et al., 2019 <sup>[[#fn:r425|425]]</sup> ). Calculations from the expert elicitation (Bamber et al., 2019 <sup>[[#fn:r426|426]]</sup> ) 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.

<span id="table-4.2"></span>
<!-- START IMG -->
<!-- TABLE IMG -->
<!-- IMG TITLE -->
'''Table 4.2'''
<!-- IMG CAPTION -->
Estimates of the Greenland Ice Sheet (GIS) contribution to Global Mean Sea Level (GMSL; cm) in 2100 reported by process-based modelling studies including the effects of both surface mass balance (SMB) and ice dynamics published since the IPCC 5th Assessment Report (AR5). Only model results including elevation-SMB feedback are shown. All values are reported as the contribution to GMSL in 2100 relative to 2000, with the exception of Aschwanden et al. (2019) <sup>[[#fn:r428|428]]</sup> who report values relative to 2008. The median estimate for comparison with AR5 is based on the average of the three simulations in Calov et al. (2018) <sup>[[#fn:r429|429]]</sup> using different General Climate Models (GCMs), combined with the central estimates from the other studies. RMSD (Fürst et al., 2015 <sup>[[#fn:r430|430]]</sup> ) is the Root Mean Squared Deviation from their ensemble median. The range reported by Aschwanden et al. (2019) <sup>[[#fn:r431|431]]</sup> refers to the 16–84% interval of a 500 member ensemble with varying model physical parameters. RCP is Representative Concentration Pathway.
<!-- IMG FILE -->
[[File:0867c405058758b3a0714c022b257a62 table4.2.png]]

Complimentary to the ice sheet scale simulations discussed above, Nick et al. (2013) <sup>[[#fn:r427|427]]</sup> 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., 2013 <sup>[[#fn:r438|438]]</sup> ; Lipscomb et al., 2013 <sup>[[#fn:r439|439]]</sup> ). 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 &lt;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., 2014 <sup>[[#fn:r440|440]]</sup> ; Morlighem et al., 2017 <sup>[[#fn:r441|441]]</sup> ) 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., 2016 <sup>[[#fn:r442|442]]</sup> ; Morlighem et al., 2016 <sup>[[#fn:r443|443]]</sup> ); 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) <sup>[[#fn:r444|444]]</sup> , the effects of uncertain bathymetric boundary conditions are assessed to be small relative to the uncertainties in future SMB forcing ( ''medium confidence'' ).

<!-- END IMG -->
<span id="figure-4.7"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.7'''

<span id="figure-4.7-bedrock-topography-below-the-existing-ice-sheets-in-greenland-morlighem-et-al.-2017-and-antarctica-right-fretwell-et-al.-2013.-horizontal-scales-are-not-the-same-in-both-panels.-note-the-deep-subglacial-basins-in-west-antarctica-and-the-east-antarctic-margin.-the-ice-above-floatation-in-these-areas-is-equivalent-to"></span>
<!-- IMG CAPTION -->
'''Figure 4.7 | Bedrock topography below the existing ice sheets in Greenland (Morlighem et al., 2017) and Antarctica (right) (Fretwell et al., 2013). Horizontal scales are not the same in both panels. Note the deep subglacial basins in West Antarctica and the East Antarctic margin. The ice above floatation in these areas is equivalent to […]'''
<!-- IMG FILE -->
[[File:7c5a735653ef82fd11e92c7c390ed2d9 IPCC-SROCC-CH_4_7-3000x1355.jpg]]

Figure 4.7 | Bedrock topography below the existing ice sheets in Greenland (Morlighem et al., 2017) and Antarctica (right) (Fretwell et al., 2013). Horizontal scales are not the same in both panels. Note the deep subglacial basins in West Antarctica and the East Antarctic margin. The ice above floatation in these areas is equivalent to &gt;20 m of Global Mean Sea Level (GMSL).

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., 2013 <sup>[[#fn:r445|445]]</sup> ). GIS simulations are most sensitive to uncertainties in the applied climate forcing, especially over this century (Edwards et al., 2014 <sup>[[#fn:r446|446]]</sup> ), but updated climate projections since AR5 are not yet available. Because of the consistency of recent modelling with the assessment of Church et al. (2013 <sup>[[#fn:r447|447]]</sup> ), Greenland’s contribution to future sea level reported in AR5 was used in our projections of GMSL.

<!-- END IMG -->
<div id="section-4-2-3-1contribution-of-ice-sheets-to-gmsl-block-2"></div>

<span id="antarctica"></span>
===== 4.2.3.1.2 Antarctica =====

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., 2013 <sup>[[#fn:r448|448]]</sup> ). 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, 1974 <sup>[[#fn:r449|449]]</sup> ; Schoof, 2007b <sup>[[#fn:r450|450]]</sup> ; Pollard et al., 2015 <sup>[[#fn:r451|451]]</sup> ) 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., 2014 <sup>[[#fn:r452|452]]</sup> ; Rignot et al., 2014 <sup>[[#fn:r453|453]]</sup> ; Scheuchl et al., 2016 <sup>[[#fn:r454|454]]</sup> ; Shen et al., 2018 <sup>[[#fn:r455|455]]</sup> ; 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., 2015 <sup>[[#fn:r456|456]]</sup> ; 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., 2016 <sup>[[#fn:r457|457]]</sup> ). Dynamic ice loss driven by ocean changes have also been observed on the East Antarctic margin (Li et al., 2016 <sup>[[#fn:r458|458]]</sup> ; Shen et al., 2018 <sup>[[#fn:r459|459]]</sup> ). 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., 2014 <sup>[[#fn:r460|460]]</sup> ). Thwaites Glacier is particularly important (Figure 4.8), because it extends into the interior of the WAIS, where the bed is &gt;2000 m below sea level in places. By itself, the Thwaites drainage area contains the equivalent of ~0.4 m GMSL (Holt et al., 2006 <sup>[[#fn:r461|461]]</sup> ; Millan et al., 2017 <sup>[[#fn:r462|462]]</sup> ), but loss of the glacier could have a destabilising impact on the entire WAIS (Feldmann and Levermann, 2015 <sup>[[#fn:r463|463]]</sup> ). The WAIS contains enough ice to raise GMSL by ~3.4 m (Fretwell et al., 2013 <sup>[[#fn:r464|464]]</sup> ). 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) <sup>[[#fn:r466|466]]</sup> 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 &gt;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., 2015 <sup>[[#fn:r465|465]]</sup> ) 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) <sup>[[#fn:r466|466]]</sup> , 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., 2018 <sup>[[#fn:r468|468]]</sup> ). Arthern and Williams (2017) <sup>[[#fn:r469|469]]</sup> used adaptive mesh techniques, but with a different formulation than Cornford et al., (2015) <sup>[[#fn:r470|470]]</sup> , 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., 2014 <sup>[[#fn:r471|471]]</sup> ). 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., 2014 <sup>[[#fn:r473|473]]</sup> ), 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) <sup>[[#fn:r474|474]]</sup> 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., 2018 <sup>[[#fn:r475|475]]</sup> ). 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., 2014 <sup>[[#fn:r476|476]]</sup> ; Gomez et al., 2015 <sup>[[#fn:r477|477]]</sup> ). 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., 2015 <sup>[[#fn:r478|478]]</sup> ; Konrad et al., 2015 <sup>[[#fn:r479|479]]</sup> ; Pollard et al., 2017 <sup>[[#fn:r480|480]]</sup> ). 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., 2017 <sup>[[#fn:r481|481]]</sup> ). 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.

<span id="figure-4.8"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.8'''

<span id="figure-4.8-processes-affecting-the-thwaites-glacier-in-the-amundsen-sea-sector-of-antarctica-adapted-from-scambos-et-al.-2017.-the-grounding-line-is-currently-retreating-on-reverse-sloped-bedrock-at-a-water-depth-of-600-m-joughin-et-al.-2014-mouginot-et-al.-2014.-the-glacier-terminus-is-120-km-wide-widens-upstream-and"></span>
<!-- IMG CAPTION -->
'''Figure 4.8 | Processes affecting the Thwaites Glacier in the Amundsen Sea sector of Antarctica (adapted from Scambos et al., 2017). The grounding line is currently retreating on reverse-sloped bedrock at a water depth of ~600 m (Joughin et al., 2014; Mouginot et al., 2014). The glacier terminus is ~120 km wide, widens upstream, and […]'''
<!-- IMG FILE -->
[[File:ce6dc95edf37fa50cc08aa9551b90b2d IPCC-SROCC-CH_4_8-3000x1028.jpg]]

Figure 4.8 | Processes affecting the Thwaites Glacier in the Amundsen Sea sector of Antarctica (adapted from Scambos et al., 2017). The grounding line is currently retreating on reverse-sloped bedrock at a water depth of ~600 m (Joughin et al., 2014 <sup>[[#fn:r483|483]]</sup> ; Mouginot et al., 2014 <sup>[[#fn:r484|484]]</sup> ). The glacier terminus is ~120 km wide, widens upstream, and is minimally buttressed by a laterally discontinuous ~40 km long ice shelf. The remaining shelf is thinning in response to warm, sub-shelf incursions of circumpolar deep water (CDW), with melt rates up 200 m yr–1 near the groundling line in some places (Milillo et al., 2019 <sup>[[#fn:r485|485]]</sup> ). The bathymetry upstream of the grounding zone is complex, but it generally slopes downward into a deep basin, up to 2000 m below sea level under the centre of the West Antarctic Ice Sheet (WAIS) (far left), making the glacier vulnerable to marine ice sheet instabilities (Cross-Chapter Box 8 in Chapter 3).

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., 2017 <sup>[[#fn:r487|487]]</sup> ). This was highlighted by Trusel et al. (2015) <sup>[[#fn:r488|488]]</sup> , who evaluated the future expansion of surface meltwater using the snow component in the RACMO2 regional atmospheric model (Kuipers Munneke et al., 2012 <sup>[[#fn:r489|489]]</sup> ) 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., 2013 <sup>[[#fn:r490|490]]</sup> ; Kuipers Munneke et al., 2014 <sup>[[#fn:r491|491]]</sup> ). The presence of surface meltwater does not necessarily lead to immediate ice shelf collapse (Bell et al., 2017b <sup>[[#fn:r492|492]]</sup> ; Kingslake et al., 2017 <sup>[[#fn:r493|493]]</sup> ), although surface meltwater was a precursor on ice shelves which have collapsed (Scambos et al., 2004 <sup>[[#fn:r494|494]]</sup> ; Banwell et al., 2013 <sup>[[#fn:r495|495]]</sup> ). 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) <sup>[[#fn:r496|496]]</sup> , 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, 2007b <sup>[[#fn:r497|497]]</sup> ) 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., 2013 <sup>[[#fn:r498|498]]</sup> ; 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., 2013 <sup>[[#fn:r499|499]]</sup> ; Reese et al., 2018c <sup>[[#fn:r500|500]]</sup> ).

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., 2017 <sup>[[#fn:r501|501]]</sup> ; Pattyn, 2017 <sup>[[#fn:r502|502]]</sup> ), although sophisticated statistical methodologies have been increasingly used since AR5 to quantitatively gauge model uncertainty (Bulthuis et al., 2019 <sup>[[#fn:r503|503]]</sup> ; Edwards et al., 2019 <sup>[[#fn:r504|504]]</sup> ). 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., 2018 <sup>[[#fn:r505|505]]</sup> ). 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., 2017 <sup>[[#fn:r506|506]]</sup> ; Hellmer et al., 2017 <sup>[[#fn:r507|507]]</sup> ). 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., 2017 <sup>[[#fn:r508|508]]</sup> ). Approaches that link offline ocean temperatures with efficient box models of the circulation in ice shelf cavities have been developed (Lazeroms et al., 2018 <sup>[[#fn:r509|509]]</sup> ; Reese et al., 2018a <sup>[[#fn:r510|510]]</sup> ) and used in long-term future simulations (Bulthuis et al., 2019 <sup>[[#fn:r511|511]]</sup> ), although they still require uncoupled ocean models to provide time-evolving ocean conditions outside the cavities.

Ritz et al. (2015) <sup>[[#fn:r512|512]]</sup> 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) <sup>[[#fn:r513|513]]</sup> 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) <sup>[[#fn:r514|514]]</sup> used PISM (Parallel Ice Sheet Model; Winkelmann et al., 2011 <sup>[[#fn:r515|515]]</sup> ) 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, 2015 <sup>[[#fn:r516|516]]</sup> ), making the model sensitive to subsurface ocean warming, although the validity of this approach is contested (Arthern and Williams, 2017 <sup>[[#fn:r517|517]]</sup> ; Seroussi and Morlighem, 2018 <sup>[[#fn:r518|518]]</sup> ; Yu et al., 2018 <sup>[[#fn:r519|519]]</sup> ). 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) <sup>[[#fn:r520|520]]</sup> 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) <sup>[[#fn:r521|521]]</sup> 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) <sup>[[#fn:r522|522]]</sup> , 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) <sup>[[#fn:r526|526]]</sup> used a different continental-scale ice sheet model (Pattyn, 2017 <sup>[[#fn:r525|525]]</sup> ) with the same simplified atmospheric and ocean forcing used by Golledge et al. (2015) <sup>[[#fn:r528|528]]</sup> 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., 2019 <sup>[[#fn:r531|531]]</sup> ) 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) <sup>[[#fn:r530|530]]</sup> .

DeConto and Pollard (2016) <sup>[[#fn:r529|529]]</sup> 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, 2012 <sup>[[#fn:r532|532]]</sup> ). 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., 2019 <sup>[[#fn:r533|533]]</sup> ) 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., 2009 <sup>[[#fn:r534|534]]</sup> ; Banwell et al., 2013 <sup>[[#fn:r535|535]]</sup> ; Ma et al., 2017 <sup>[[#fn:r536|536]]</sup> ; Wise et al., 2017 <sup>[[#fn:r537|537]]</sup> ; Parizek et al., 2019 <sup>[[#fn:r538|538]]</sup> ). DeConto and Pollard (2016) <sup>[[#fn:r539|539]]</sup> 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., 2017 <sup>[[#fn:r540|540]]</sup> ; Edwards et al., 2019 <sup>[[#fn:r541|541]]</sup> ). 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) <sup>[[#fn:r542|542]]</sup> and Bulthuis et al. (2019) <sup>[[#fn:r543|543]]</sup> . 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) <sup>[[#fn:r544|544]]</sup> simulates the appearance of extensive surface meltwater several decades earlier than indicated by other CMIP5 climate simulations (Trusel et al., 2015 <sup>[[#fn:r545|545]]</sup> ). 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.

<!-- END IMG -->
<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise"></div>

<span id="global-and-regional-projections-of-sea-level-rise"></span>
==== 4.2.3.2 Global and Regional Projections of Sea Level Rise ====

<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-1"></div>

In addition to the model including MICI from DeConto and Pollard (2016) , only a subset of studies ( Levermann et al., 2014; Golledge et al., 2015; Ritz et al., 2015; Bulthuis et al., 2019; Golledge et al., 2019) , and statistical emulation of DeConto and Pollard (2016) by Edwards et al. (2019) provide continental-scale estimates of future Antarctic ice loss, under a range of GHG emissions scenarios. They all provide probabilistic information, but vary considerably, both in their physical approaches and their resulting projections of Antarctica’s future contribution to GMSL. Such variations facilitate the first quantitative uncertainty assessment of the full dynamical contribution of Antarctica, which could not be made by Church et al. (2013) in AR5. The assessment by Church et al. (2013) , based on a single statistical-physical model, reported median values (and ''likely'' ranges) of 0.05 m (-0.04–0.13) and 0.04 m (-0.06–0.12), for RCP4.5 and RCP8.5, respectively, for the total Antarctic contribution in 2081–2100 relative to 1986–2005, and added the following: ‘Based on current understanding, only the collapse of marine-based sectors of the AIS, if initiated, could cause GMSL to rise substantially above the ''likely'' range during the 21st century. This potential additional contribution cannot be precisely quantified but there is ''medium confidence'' that it would not exceed several tenths of a metre of SLR during the 21st century (Church et al., 2013) . Given the above-mentioned publications after AR5, Antarctica’s contribution to sea level change was reassessed and now include the possibility of MISI allowing for a more complete assessment of the ''likely'' range of the projections for three RCP scenarios. Our assessment is based on process-based numerical models of the AIS, driven by diverse climate scenarios. Results are discussed in the context of an expert elicitation study (Bamber et al., 2019 <sup>[[#fn:r546|546]]</sup> ) , probabilistic studies (Perrette et al., 2013 <sup>[[#fn:r547|547]]</sup> ; Slangen et al., 2014a <sup>[[#fn:r548|548]]</sup> ; Grinsted et al., 2015 <sup>[[#fn:r549|549]]</sup> ; Jackson and Jevrejeva, 2016 <sup>[[#fn:r550|550]]</sup> ) and a sensitivity study (Schlegel et al., 2018 <sup>[[#fn:r551|551]]</sup> ) assessing the uncertainty in snow accumulation, ocean-induced melting, ice viscosity, basal friction, bedrock elevation and the effect of ice shelves on ice mass loss in 2100, Figure 4.4.

Ritz et al. (2015) is difficult to contextualise as they only provided estimates for the A1B scenario and not for the RCP scenarios. Despite this limitation their results, which are close to the other studies, are included as if they represent RCP8.5 and as such supports the assessment. The results by DeConto and Pollard (2016) <sup>[[#fn:r552|552]]</sup> indicate significantly higher mass loss even for RCP4.5, potentially related to their high surface melt rates on the ice shelves as contested by Trusel et al. (2015) <sup>[[#fn:r553|553]]</sup> . This early onset of high surface melt rates in DeConto and Pollard (2016) <sup>[[#fn:r554|554]]</sup> leads to extensive hydrofracturing of ice shelves before the end of the 21st century and therefore to rapid ice mass loss. For this reason, their results and probabilistic (e.g., Kopp et al., 2017; Le Bars et al., 2017) and statistical emulation estimates that build on them (Edwards et al., 2019 <sup>[[#fn:r555|555]]</sup> ) , are not used in SROCC sea level projections. Consequently, the process-based studies by Golledge et al. (2015) <sup>[[#fn:r556|556]]</sup> , Ritz et al. (2015) , Levermann et al. (2014) <sup>[[#fn:r558|558]]</sup> , Golledge et al. (2019) <sup>[[#fn:r559|559]]</sup> , and Bulthuis et al. (2019) are used to assess the Antarctic contribution for the different RCP scenarios. The study by Schlegel et al. (2018) does not provide RCP based scenarios, but is considered as an extensive sensitivity estimate providing a high-end estimate based on physical process understanding of the Antarctic contribution.

Each study expresses an uncertainty in the Antarctic contribution to GMSL rise which is, in part, dependent on a common driver, namely regional warming. The uncertainties were therefore interpreted as being dependent and propagate the total uncertainty accordingly. As a result, the total uncertainty exceeds that of the individual studies, which reflects that the individual studies only sample a fraction of the total uncertainty. The uncertainty estimates of Levermann et al. (2014) <sup>[[#fn:r561|561]]</sup> concentrate on the oceanic basal melt rates including a time delay between atmosphere and ocean temperature, but do not consider other sources of uncertainty. Ritz et al. (2015) <sup>[[#fn:r562|562]]</sup> is constrained by observations and provides an asymmetric distribution of the rate of mass loss. The ice sheet simulations by Golledge et al. (2015) and Golledge et al. (2019) only provide two alternative subgrid parameterisations for sub-ice melt, rather than a statistical estimate of the uncertainty. The more sensitive of these two parameterisations which induces more ice loss is challenged by Seroussi and Morlighem (2018) <sup>[[#fn:r565|565]]</sup> . In order to assess a realistic uncertainty for the total Antarctic contribution, it was first assumed that Golledge et al. (2015) <sup>[[#fn:r566|566]]</sup> and Golledge et al. (2019) <sup>[[#fn:r567|567]]</sup> are dependent, because they use similar parameterisations. For each study, a probabilistic distribution is used, assuming a normal distribution with a ''likely'' range bounded by the high and low estimate from those studies. Levermann et al. (2014) <sup>[[#fn:r568|568]]</sup> also provides two alternatives, one with and one without a time delay between oceanic temperatures below the Antarctic ice shelves and global mean atmospheric temperature. As it is unclear which version best matches the updated record of ice loss presented by The IMBIE team, (2018) , results are combined assuming full probabalistic dependence as for the two Golledge studies. Bulthuis et al. (2019) <sup>[[#fn:r572|572]]</sup> uses a simplified ice sheet model to study the uncertainty caused by the atmospheric forcing, ice dynamics, ice and bed rheology, calving and sub-shelf melting. Finally, the studies by Ritz et al. (2015) <sup>[[#fn:r571|571]]</sup> , Bulthuis et al. (2019) and the averages for Golledge and Levermann are combined to identify a best estimate for the Antarctic contribution under RCP8.5. This results in a median contribution of 16 cm in 2100 under RCP8.5. A Monte Carlo technique is used to combine the uncertainties in the aforementioned studies, assuming mutual dependence. The resulting 5–95 percentile range, 2–37 cm in 2100 under RCP8.5, is assessed as the ''likely'' range. This assessment is used in order to reflect ongoing limited understanding of the physics and the fact that the individual studies only reflect part of the total uncertainty. The distribution is slightly skewed to higher values, because of an underlying skewness in the studies of Levermann et al. (2014) <sup>[[#fn:r569|569]]</sup> and Ritz et al. (2015) <sup>[[#fn:r571|571]]</sup> . This skewed distribution is supported by an expert elicitation study (Bamber et al., 2009) . The expert elicitation approach (Bamber et al., 2018) , which applied elicitation to both ice sheets, suggests considerably higher values for total SLR for RCP2.6, RCP4.5 and RCP8.5 than provided in Table 4.3.

As the importance of MISI and MICI is difficult to assess on longer time scales, there remains deep uncertainty for the Antarctic contribution to GMSL after 2100 (Cross-Chapter Box 4 in Chapter 1). Results on these long-time scales are discussed in 4.2.3.5.

<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-2"></div>

<span id="table-4.3"></span>
<!-- START IMG -->
<!-- TABLE IMG -->
<!-- IMG TITLE -->
'''Table 4.3:'''
<!-- IMG CAPTION -->
An overview of different studies estimating the future Antarctic contribution to sea level rise (SLR), listed here are median values. Estimates from Golledge et al. (2015) are based on the average contribution to Global Mean Sea Level (GMSL) over the full 21st century, based on two alternative ensembles using different sub-ice melt schemes. This average is not explicitly reported in the original paper where the individual values of 0.1 and 0.39 m are reported. SMB is the surface mass balance, BMB the basal melt balance, LIG is Last Interglacial, MICI is marine ice cliff instability, GCM is General Circulation Model, PDD is positive-degree day.

<!-- IMG FILE -->
[[File:4bc1e6307f36dfeee2ae0eafcf4efea3 table4.3.png]]

<!-- END IMG -->
<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-3"></div>

There is limited evidence for major changes since AR5 in the non-Antarctic components. Recent projections of the glacier contribution are nearly identical to AR5 results used here (see Cross-Chapter Box 6 in Chapter 2). Greenland, thermal expansion and land water storage are also not updated, mainly due to a lack of updated CMIP simulations. Hence, our revised projections replace only the AR5 estimate for Antarctica by a new assessment as outlined in the previous paragraph based on post-AR5 literature and maintaining identical contributions for the non-Antarctic components. As no general dependence between the Antarctic contribution and the non-Antarctic components can be derived from the four studies, independent uncertainties are assumed, which is close to the uncertainty propagation by Church et al. (2013) <sup>[[#fn:r576|576]]</sup> .

Time series for the different RCP scenarios are shown in Figure 4.9 indicating a divergence in median and upper ''likely'' range for RCP8.5 during the second half of the century between this report and the AR5 projections (Church et al., 2013 <sup>[[#fn:r577|577]]</sup> ) . The value of the Antarctic contribution in 2081–2100 under RCP8.5 is the individual component with the largest uncertainty. As a consequence, the uncertainty in the GMSL projections is slightly increased compared to Church et al. (2013) . Nevertheless, results can also be considered to be consistent with Church et al. (2013) <sup>[[#fn:r578|578]]</sup> . In AR5, the potential additional contribution by ice dynamics, was estimated to be not more than several tenths of a metre but excluded from projections; here this value was assessed to be 16 cm (5–95 percentile; 2–37 cm) and include it in the projections. As the projections build on the CMIP5 work presented in AR5, and also given the limited exploration of uncertainty in estimates from each individual study, the results of the 5–95 percentile are interpreted to represent the ''likely'' range, that is, the 17–83 percentile, as assessed by Church et al. (2013) and as assessed in AR5 for other CMIP5-derived results.

<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-4"></div>

<span id="table-4.4"></span>
<!-- START IMG -->
<!-- TABLE IMG -->
<!-- IMG TITLE -->
'''Table 4.4'''
<!-- IMG CAPTION -->
Median values and likely ranges for projections of global mean sea level (GMSL) rise in metres in 2081–2100 relative to 1986–2005 for three scenarios. In addition, values of GMSL rise are given for 2046-2065 and 2100, and the rate of GMSL rise is given for 2100. Values between parentheses reflect the likely range. SMB is surface mass balance, DYN is dynamical contribution, LWS is land water storage. Total AR5 minus Antarctica AR5 is the GMSL rise contribution in Church et al. (2013) without the Antarctic contribution of Church et al. (2013). The newly derived Antarctic contribution is added to this to arrive at the GMSL rise.

<!-- IMG FILE -->
[[File:aa63ba81f162ddaf5b939f462585d229 table4.4.png]]

Notes:

\*The uncertainty in this value is calculated as in Church et al. (2013).

<!-- END IMG -->
<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-5"></div>

<span id="figure-4.9"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.9'''

<span id="figure-4.9-time-series-of-global-mean-sea-level-gmsl-for-representative-concentration-pathway-rcp2.6-rcp4.5-and-rcp8.5-as-used-in-this-report-and-for-reference-the-ipcc-5th-assessment-report-ar5-results-church-et-al.-2013.-results-are-based-on-ar5-results-for-all-components-except-the-antarctic-contribution.-results-for-the-antarctic"></span>
<!-- IMG CAPTION -->
'''Figure 4.9 | Time series of Global Mean Sea Level (GMSL) for Representative Concentration Pathway (RCP)2.6, RCP4.5 and RCP8.5 as used in this report and, for reference the IPCC 5th Assessment Report (AR5) results (Church et al., 2013). Results are based on AR5 results for all components except the Antarctic contribution. Results for the Antarctic […]'''
<!-- IMG FILE -->
[[File:77840ce1305f31bf8c3a27055b8fff52 IPCC-SROCC-CH_4_9-3000x895.jpg]]

Figure 4.9 | Time series of Global Mean Sea Level (GMSL) for Representative Concentration Pathway (RCP)2.6, RCP4.5 and RCP8.5 as used in this report and, for reference the IPCC 5th Assessment Report (AR5) results (Church et al., 2013). Results are based on AR5 results for all components except the Antarctic contribution. Results for the Antarctic contribution in 2081–2100 are provided in Table 4.4. The shaded region is considered to be the likely range.

<!-- END IMG -->
<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-6"></div>

Projections as presented in Table 4.4 are used to calculate the regional RSL projections as outlined in AR5 by including gravitational and rotational patterns as shown in Figure 4.10 and subsequently used in 4.2.3.4 to calculate ESL projections. Including the updated results in terms of magnitude and uncertainty for the Antarctic component also changes the regional patterns in sea level projections. Results of the regional patterns in Figure 4.10 show an increased SLR with respect to the results presented in AR5 nearly everywhere for RCP8.5 because of the increased Antarctic contribution.

<div id="section-4-2-3-2global-and-regional-projections-of-sea-level-rise-block-7"></div>

<span id="figure-4.10"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.10'''

<span id="figure-4.10-regional-sea-level-change-for-rcp2.6-rcp4.5-and-rcp8.5-in-metres-as-used-in-this-report-for-extreme-sea-level-esl-events.-results-are-median-values-based-on-the-values-in-table-4.4-for-antarctica-including-gia-and-the-gravitational-and-rotational-effects-and-results-by-church-et-al.-2013-for-glaciers"></span>
<!-- IMG CAPTION -->
'''Figure 4.10 | Regional sea level change for RCP2.6, RCP4.5 and RCP8.5 in metres as used in this report for extreme sea level (ESL) events. Results are median values based on the values in Table 4.4 for Antarctica including GIA and the gravitational and rotational effects, and results by Church et al. (2013) for glaciers, […]'''
<!-- IMG FILE -->
[[File:551a1ce25a7fff5251607a55b7ae6dc3 IPCC-SROCC-CH_4_10-3000x2591.jpg]]

Figure 4.10 | Regional sea level change for RCP2.6, RCP4.5 and RCP8.5 in metres as used in this report for extreme sea level (ESL) events. Results are median values based on the values in Table 4.4 for Antarctica including GIA and the gravitational and rotational effects, and results by Church et al. (2013) for glaciers, land water storage (LWS) and Greenland. The left column is for the time slice 2046–2065 and the right column for 2081–2100.

<!-- END IMG -->
<div id="section-4-2-3-3probabilistic-sea-level-projections"></div>

<span id="probabilistic-sea-level-projections"></span>
==== 4.2.3.3 Probabilistic Sea Level Projections ====

<div id="section-4-2-3-3probabilistic-sea-level-projections-block-1"></div>

Since AR5, several studies have produced SLR projections in coherent frameworks that link together global-mean and RSL rise projections. The approaches are generally similar to those adopted by AR5 for its global-mean sea level projections: a bottom-up accounting of different contributing processes (e.g., land-ice mass loss, thermal expansion, dynamic sea level), of which many are ‘probabilistic’, in that they attempt to describe more comprehensive probability distributions of sea level change than the ''likely'' ranges presented by Church et al. (2013) <sup>[[#fn:r600|600]]</sup> . An overview of probabilistic approaches is presented in Garner et al. (2017) <sup>[[#fn:r601|601]]</sup> , indicating higher values for post AR5 studies mainly reflecting increased uncertainty based on a single contested study for the Antarctic contribution (DeConto and Pollard, 2016 <sup>[[#fn:r602|602]]</sup> ) . As such many of these probabilistic studies present full probability density function conditional not only on an RCP scenario, but with additional and equally important a priori assumptions concerning for instance the Antarctic contribution over which a consensus has yet to solidify. An example is the study by Le Bars et al. (2017) who expand the projection by Church et al. (2013) <sup>[[#fn:r603|603]]</sup> in a probabilistic way with the Antarctic projections by DeConto and Pollard (2016) <sup>[[#fn:r604|604]]</sup> to obtain a full probability density function for SLR for RCP8.5. Other probabilistic approaches are provided by Kopp et al. (2014) <sup>[[#fn:r605|605]]</sup> and Jackson and Jevrejeva (2016) <sup>[[#fn:r606|606]]</sup> using different ice sheet representations drawing on expert elicitation (Bamber and Aspinall, 2013 <sup>[[#fn:r607|607]]</sup> ) . Probabilistic estimates are useful for a quantitative risk management perspective (see Section 4.3.3). An even more general approach than the probabilistic estimates has been taken by Le Cozannet et al. (2017) who frame a ‘possibilistic’ framework of SLR including existing probabilistic estimates and combining them.

This section first briefly reviews key sources of information for probabilistic projections (Section 4.2.3.3.1), with a focus on new results since AR5, then summarises the different global and regional projections (Section 4.2.3.3.2). Eventually, bottom-up projections were distinguished which explicitly describe the different components of SLR (Section 4.2.3.3.3) from semi-empirical projections (Section 4.2.3.3.4).

<div id="section-4-2-3-3probabilistic-sea-level-projections-block-2"></div>

<span id="components-of-probabilistic-global-mean-sea-level-projections"></span>
===== 4.2.3.3.1 Components of probabilistic global mean sea level projections =====

Thermal expansion: Global mean thermal expansion projections rely on coupled climate models projections (Kopp et al., 2014 <sup>[[#fn:r608|608]]</sup> ; Slangen et al., 2014a <sup>[[#fn:r609|609]]</sup> ; Jackson and Jevrejeva, 2016 <sup>[[#fn:r610|610]]</sup> ) or simple climate model projections (Perrette et al., 2013 <sup>[[#fn:r611|611]]</sup> ; Nauels et al., 2017b <sup>[[#fn:r612|612]]</sup> ; Wong et al., 2017 <sup>[[#fn:r613|613]]</sup> ) , 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) <sup>[[#fn:r614|614]]</sup> , 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) <sup>[[#fn:r615|615]]</sup> 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, 2011 <sup>[[#fn:r615|615]]</sup> ; Marzeion et al., 2012 <sup>[[#fn:r616|616]]</sup> ; Hirabayashi et al., 2013 <sup>[[#fn:r617|617]]</sup> ; Radić et al., 2014 <sup>[[#fn:r618|618]]</sup> ; Huss and Hock, 2015 <sup>[[#fn:r619|619]]</sup> ) , or simple scaling relationships with global mean temperature (Perrette et al., 2013 <sup>[[#fn:r620|620]]</sup> ; Bakker et al., 2017 <sup>[[#fn:r621|621]]</sup> ; Nauels et al., 2017a <sup>[[#fn:r622|622]]</sup> ) . 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., 2016 <sup>[[#fn:r623|623]]</sup> ) . 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., 2007 <sup>[[#fn:r625|625]]</sup> ; Church et al., 2013 <sup>[[#fn:r626|626]]</sup> ) . AR5 (Church et al., 2013 <sup>[[#fn:r627|627]]</sup> ) used this approach and is partly based on the assessment of statistical-physical modelling of the Antarctic contribution (Little et al., 2013 <sup>[[#fn:r628|628]]</sup> ) . 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, 1991 <sup>[[#fn:r629|629]]</sup> ; Bamber and Aspinall, 2013 <sup>[[#fn:r630|630]]</sup> ; Bamber et al., 2019 <sup>[[#fn:r631|631]]</sup> ) 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) <sup>[[#fn:r632|632]]</sup> 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., 2017 <sup>[[#fn:r633|633]]</sup> ; Le Bars et al., 2017) used the results of a single ice sheet model study from DeConto and Pollard (2016) <sup>[[#fn:r634|634]]</sup> 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.

<div id="section-4-2-3-3probabilistic-sea-level-projections-block-3"></div>

<span id="from-probabilistic-global-mean-sea-level-projections-to-regional-relative-sea-level-change"></span>
===== 4.2.3.3.2 From probabilistic global mean sea level projections to regional relative sea level change =====

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., 2013 <sup>[[#fn:r635|635]]</sup> ; Carson et al., 2016 <sup>[[#fn:r636|636]]</sup> ) .

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, 2016 <sup>[[#fn:r637|637]]</sup> ) . 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., 2016 <sup>[[#fn:r638|638]]</sup> ) . 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., 2019 <sup>[[#fn:r638|638]]</sup> ) : 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., 2014a <sup>[[#fn:r640|640]]</sup> ) , 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., 2017 <sup>[[#fn:r641|641]]</sup> ) . 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., 2014a <sup>[[#fn:r643|643]]</sup> ) 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.

<div id="section-4-2-3-3probabilistic-sea-level-projections-block-4"></div>

<span id="semi-empirical-projections"></span>
===== 4.2.3.3.3 Semi-empirical projections =====

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, 2007 <sup>[[#fn:r644|644]]</sup> ; Vermeer and Rahmstorf, 2009 <sup>[[#fn:r645|645]]</sup> ; Grinsted et al., 2010 <sup>[[#fn:r646|646]]</sup> ; Kemp et al., 2011 <sup>[[#fn:r647|647]]</sup> ; Kopp et al., 2016 <sup>[[#fn:r648|648]]</sup> ) . 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., 2016 <sup>[[#fn:r682|682]]</sup> ; Mengel et al., 2016 <sup>[[#fn:r683|683]]</sup> ) are in general agreement with Church et al. (2013) <sup>[[#fn:r684|684]]</sup> , except when those results reflect the combined hydrofracturing and ice cliff instability mechanism as presented by DeConto and Pollard (2016) <sup>[[#fn:r685|685]]</sup> . 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., 2018 <sup>[[#fn:r686|686]]</sup> ) . 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., 2017 <sup>[[#fn:r687|687]]</sup> ; Nauels et al., 2017a <sup>[[#fn:r688|688]]</sup> ; Wong et al., 2017 <sup>[[#fn:r689|689]]</sup> ; Edwards et al., 2019 <sup>[[#fn:r690|690]]</sup> ) .

<div id="section-4-2-3-3probabilistic-sea-level-projections-block-5"></div>

<span id="recent-probabilistic-and-semi-empirical-projections"></span>
===== 4.2.3.3.4 Recent probabilistic and semi-empirical projections =====

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) <sup>[[#fn:r692|692]]</sup> . 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.

<span id="table-4.5"></span>
<!-- START TABLE -->
'''Table 4.5'''

'''Table 4.5:''' Sources of Information Underlying Probabilistic Projections of Sea level Rise (SLR) Projections.

CMIP5 is Coupled Model Intercomparison Project Phase 5, GRD is gravitational, rotational and deformation effects, SMB is surface mass balance, AR4 is IPCC 4th Assessment Report, VLM is vertical land motion, GIA is glacio-isostatic adjustment.

<!-- TABLE -->
{| class="wikitable"
|-
| Study
| Thermal expansion
| Glaciers
| Land water storage
| Ice Sheets
| Dynamic sea level
| GRD
| VLM
|-
| Perrette et al. (2013)
| CMIP5
| Global SMB sensitivity and exponent from AR4; total glacier volume from Radić and Hock (2010)
| Not included
| Greenland’s SMB from AR4; semi-empirical model using historical observations.
| CMIP5
| Bamber et al. (2009)
| Not included
|-
| Grinsted et al. (2015)
| CMIP5
| Church et al. (2013)
| Wada et al. (2012)
| Church et al. (2013); Expert elicitation from Bamber and Aspinall (2013)
| CMIP5
| Bamber et al. (2009)
| GIA projections from Hill et al. (2010) using observations
|-
| Slangen et al. (2014a)
| CMIP5
| CMIP5; glacier area inventory Radić and Hock (2010) in a glacier mass loss model
| Wada et al. (2012)
| SMB Meehl et al. (2007), ice dynamics Meehl et al. (2007) and Katsman et al. (2011)
| CMIP5
| Slangen et al. (2014a)
| GIA resulting of ice sheet melt from glacier mass loss model
|-
| Kopp et al. (2014)
| CMIP5
| CMIP5; Marzeion et al. (2012)
| Chambers et al. (2017); Konikow (2011)
| Church et al. (2013); Expert elicitation from Bamber and Aspinall (2013)
| CMIP5
| Mitrovica et al. (2011)
| GIA, tectonics, and subsidence from Kopp et al. (2013)
|-
| Kopp et al. (2017)
| CMIP5
| CMIP5; Marzeion et al. (2012)
| Chambers et al. (2017); Konikow (2011)
| DeConto and Pollard (2016)
| CMIP5
| Mitrovica et al. (2011)
| GIA, tectonics, and subsidence from Kopp et al. (2013)
|-
| Le Bars et al. (2017)
| CMIP5
| Four glacier models:
Giesen and Oerlemans (2013)

Marzeion et al. (2012),

Radić et al. (2014)

Slangen and Van de Wal (2011)

| Wada et al. (2012)
| DeConto and Pollard (2016);
Fettweis et al. (2013)

Church et al. (2013)

| CMIP5
| –
|-
| Jackson and Jevrejeva (2016)
| CMIP5
| Marzeion et al. (2012)
| Wada et al. (2012)
| Church et al. (2013); Expert elicitation from Bamber and Aspinall (2013)
| CMIP5
| Bamber et al. (2009)
| GIA resulting of ice sheet melt from glacier mass loss model Peltier et al. (2015)
|-
| de Winter et al. (2017)
| CMIP5
| CMIP5; glacier area inventory Radić and Hock (2010) in a glacier mass loss model
| Wada et al. (2012)
| Church et al. (2013); Expert elicitation de Vries and van de Wal (2015); Ritz et al. (2015)
| CMIP5
| Mitrovica et al. (2001)
| GIA resulting of ice sheet melt from glacier mass loss model
|}
<!-- END TABLE -->

<span id="table-4.6"></span>
<!-- START TABLE -->
'''Table 4.6:'''

'''Table 4.6:''' Median and ''likely'' Global Mean Sea Level (GMSL) rise projections (m). Values between brackets are ''likely'' range, if no values are given the ''likely'' range is not available. The table shows result from the probabilistic and semi-empirical results. A is 2000 as base line year up to 2100; B is the average of 1986–2005 as base line for the projection up to 2081–2100, C 1980–1999 as baseline up to 2090–2099.

<!-- TABLE -->
{| class="wikitable"
|-
|
| colspan="3"| 2050
| colspan="3"| 2100
|-
|
| Period
| RCP2.6
| RCP4.5
| RCP8.5
| RCP2.6
| RCP4.5
| RCP8.5
|-
| Perrette et al. (2013)
| C
|
| 0.28
(0.23–0.32)

| 0.28
(0.23–0.34)

|
| 0.86
(0.66–1.11)

| 1.06
(0.78–1.43)

|-
| Grinsted et al. (2015)
| A
|
| 0.8
(0.58–1.20)

|-
| Slangen et al. (2014a) ''' '''
| B AB B
|
| 0.54
(0.35–0.73)

| 0.71
(0.43–0.99)

|-
| Kopp et al. (2014)
| A
| 0.25
(0.21–0.29)

| 0.26
(0.21–0.31)

| 0.29
(0.24–0.34)

| 0.50
(0.37–0.65)

| 0.59
(0.45–0.77)

| 0.79
(0.62–1.00)

|-
| Kopp et al. (2017)
| A
| 0.23
(0.16–0.33)

| 0.26
(0.18–0.36)

| 0.31
(0.22–0.40)

| 0.56
(0.37–0.78)

| 0.91
(0.66–1.25)

| 1.46
(1.09–2.09)

|-
| de Winter et al. (2017)
| B
|
| 0.68/0.86
|-
| Jackson and Jevrejeva (2016)
| B
|
| 0.54
(0.36–0.72)

| 0.75
(0.54–0.98)

|-
| Le Bars et al. (2017)
| B
|
| 1.06
(0.65-1.47)

| 1.84
(1.24-2.46)

|-
| Nauels et al. (2017b)
| B
| 0.24
(0.19–0.30)

| 0.25
(0.21–0.30)

| 0.27
(0.23–0.33)

| 0.45
(0.35–0.56)

| 0.55
(0.45–0.67)

| 0.79
(0.65–0.97)

|-
| Bakker et al. (2017)
| A
| 0.20
| 0.23
| 0.25
| 0.53
| 0.72
| 1.16
|-
| Wong et al. (2017)
| A
| 0.26
| 0.28
| 0.30
| 0.55
| 0.77
| 1.50
|-
| Jevrejeva et al. (2014a)
| A
|
| 0.80
(0.6-1.2)

|-
| Schaeffer et al. (2012)
| A
|
| 0.90
| 1.02
|-
| Mengel et al. (2016)
| B
| 0.18
| 0.21
| 0.39
| 0.53
| 0.85
|}
<!-- END TABLE -->

<div id="section-4-2-3-4changes-in-extreme-sea-level-events"></div>

<span id="changes-in-extreme-sea-level-events"></span>
==== 4.2.3.4 Changes in Extreme Sea Level events ====

<div id="section-4-2-3-4changes-in-extreme-sea-level-events-block-1"></div>

ESL events are water level heights that consist of contributions from mean sea level, storm surges and tides. Compound effects of surges and tides are drivers of the ESL events. Section 4.2.3.4.1 discusses the combination of mean sea level change with a characterisation of the ESL events derived from tide gauges over the historical period and the sections 4.2.3.4.2 and 4.2.3.4.3 evaluate possible changes in these characteristics caused by cyclones and waves. This section discusses the importance of ESL and different modelling strategies to improve our understanding of ESL projections.

Even a small increase in mean sea level can significantly augment the frequency and intensity of flooding. This is because SLR elevates the platform for storm surges, tides, and waves, and because there is a log-linear relationship between a flood’s height and its occurrence interval. Changes are most pronounced in shelf seas. Roughly 1.3% of the global population is exposed to a 1 in 100-year <sup>-1</sup> flood (Muis et al., 2016 <sup>[[#fn:r699|699]]</sup> ) . This exposure to ESL and resulting damage could increase significantly with SLR, potentially amounting to 10% of the global GDP by the end of the century in the absence of adaptation (Hinkel et al., 2014 <sup>[[#fn:r700|700]]</sup> ) .

The frequency and intensity of ESL events can be estimated with statistical models or hydrodynamical models constrained by observations. Hydrodynamic models simulate a series of ESL events over time, which can then be fitted by extreme value distributions to estimate the frequency and intensity (e.g. the return level of an event occurring with a period of 100 years or frequency of 0.01 yr <sup>-1</sup> , also called the 100-year event). A tide model is sometimes included and sometimes added offline to estimate the ESL events. Statistical models fit tide gauge observations to extreme value distributions to directly estimate ESL events or combine probabilistic RSL scenarios with storm surge modelling. This can be done on global scale or local scale. For example, Lin et al. (2016) and Garner et al. (2017) <sup>[[#fn:r701|701]]</sup> estimate the increase in flood frequency along the US east coast. Both of these modelling approaches can account for projections of SLR. Rasmussen et al. (2018) used a combination of a global network of tide gauges and a probabilistic localised SLR to estimate expected ESL events showing inundation reductions for different temperature stabilisation targets as shown in the SR15 report.

An advantage of the use of hydrodynamic models is that they can quantify interactions between the different components of ESL (Arns et al., 2013 <sup>[[#fn:r702|702]]</sup> ) . Hydrodynamical models can be executed over the entire ocean with flexible grids at a high resolution (up to 1/20° or ~5 km) where necessary, appropriate for local estimates (Kernkamp et al., 2011 <sup>[[#fn:r703|703]]</sup> ) . Input for these models are wind speed and direction, and atmospheric pressure. Results of those models show that the Root Mean Squared Error between modelled and observed sea level is less than 0.2 m for 80% of a data set of 472 stations covering the global coastline (Muis et al., 2016 <sup>[[#fn:r704|704]]</sup> ) at 10-minute temporal resolution over a reference period from 1980–2011. This implies that for most locations it can be used to describe the variability in ESL. However, the areas where ESL is dominated by tropical storms are problematic for hydrodynamical models. Another difficulty arises when these models are forced with climate models: they inherit the limitations (resolution, precision and accuracy) of wind and pressure in climate projections, which is often insufficient to describe the role of waves.

Statistical models have shown that the estimation of ESL is highly sensitive to the characterisation of SLR and flood frequency distributions (Buchanan et al., 2017) . This is confirmed by Wahl et al. (2017) who estimate that the 5–95 percentile uncertainty range, attained through the application of different statistical extreme value methods and record lengths, of the current 100-year event is on average 40 cm, whereas the corresponding range in projected GMSL of AR5 under RCP8.5 is 37 cm. For ESL events with a higher return period, differences will be larger. Capturing changes in the ESL return periods in the future is even more complicated because both the changing variability over time and the uncertainty in the mean projection must be combined. A statistical framework to combine RSL and ESL, based on historical tide gauge data was applied to the US coastlines (Buchanan et al., 2016 <sup>[[#fn:r705|705]]</sup> ) . Hunter (2012) <sup>[[#fn:r708|708]]</sup>  and the AR5 (Church et al., 2013 <sup>[[#fn:r709|709]]</sup> ) projected changes in flood frequency worldwide; however, these analyses used the Gumbel distribution for high water return periods, which implies that the frequency of all ESLs (e.g., whether the 1in 10-year or 1 in 500-year) will change by the same magnitude for a given RSL, an approximation that can underestimate or overestimate ESL (Buchanan et al., 2017 <sup>[[#fn:r710|710]]</sup> ) . Hence, the amplification factors of future storm return frequency in AR5 WGI Figure 13.25 may underestimate flood hazards in some areas, while overestimating them in others. By using the Gumbel distribution, Muis et al. (2016) <sup>[[#fn:r711|711]]</sup> may also inadequately estimate flood frequencies.

<div id="section-4-2-3-4changes-in-extreme-sea-level-events-block-2"></div>

<span id="relative-sea-level-and-extreme-sea-level-events-based-on-tide-gauge-records"></span>
===== 4.2.3.4.1 Relative sea level and extreme sea level events based on tide gauge records =====

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., 2016 <sup>[[#fn:r712|712]]</sup> ). Results include GIA effects, but anthropogenic subsidence is not prescribed. These calculations serve as a signal to guide adaption to SLR (Stephens et al., 2018 <sup>[[#fn:r713|713]]</sup> ). 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) <sup>[[#fn:r714|714]]</sup> and Wahl et al. (2017) <sup>[[#fn:r715|715]]</sup> . 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) <sup>[[#fn:r716|716]]</sup> 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) <sup>[[#fn:r717|717]]</sup> . 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, 2012 <sup>[[#fn:r718|718]]</sup> ).

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., 2017 <sup>[[#fn:r719|719]]</sup> ). 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., 2018a <sup>[[#fn:r720|720]]</sup> ).

<span id="figure-4.11"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.11'''

<span id="figure-4.11-the-relation-between-expected-extreme-sea-level-esl-events-and-return-period-at-a-set-of-characteristic-tide-gauge-locations-see-upper-left-for-their-location-referenced-to-recent-past-mean-sea-level-based-on-observations-in-the-gesla-2-data-base-grey-lines-and-20812100-conditions-for-three-different-rcp-scenarios-as"></span>
<!-- IMG CAPTION -->
'''Figure 4.11 | The relation between expected extreme sea level (ESL) events and return period at a set of characteristic tide gauge locations (see upper left for their location), referenced to recent past mean sea level, based on observations in the GESLA-2 data base (grey lines) and 2081–2100 conditions for three different RCP scenarios as […]'''
<!-- IMG FILE -->
[[File:2c3ee38945aaa3d2678c0a9389d75ccb IPCC-SROCC-CH_4_11-2560x3000.jpg]]

Figure 4.11 | The relation between expected extreme sea level (ESL) events and return period at a set of characteristic tide gauge locations (see upper left for their location), referenced to recent past mean sea level, based on observations in the GESLA-2 data base (grey lines) and 2081–2100 conditions for three different RCP scenarios as presented in Section 4.2.3.2. The grey bands represent the 5–95% uncertainty range in the fit of the extreme value distribution to observations. The upper right hand panel provides an example illustrating the relationship between ESL events and return period for historical and future conditions; the blue line in this panel shows the best estimate ESL event above the 1986–2005 reference mean sea level. The coloured lines for the different locations show this expected ESL events for different RCP scenarios. The horizontal line denoting the amplification factor expresses the increase in frequency of events which historically have a return period of once every 100 years. In the example, a water level of 2.5 m above mean sea level, recurring in the recent climate approximately every 100 years in recent past climate, will occur every 2 to 3 years under future climate conditions. The allowance expresses the increase in ESL for events that historically have a return period of 100 years.

<!-- END IMG -->
<span id="figure-4.12"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 4.12'''

<span id="figure-4.12-the-colours-of-the-dots-express-the-factor-by-which-the-frequency-of-extreme-sea-level-esl-events-increase-in-the-future-for-events-which-historically-have-a-return-period-of-100-years.-hence-a-value-of-50-means-that-what-is-currently-1-in-100-year-event-will-happen-every-2-years-due"></span>
<!-- IMG CAPTION -->
'''Figure 4.12 | The colours of the dots express the factor by which the frequency of extreme sea level (ESL) events increase in the future for events which historically have a return period of 100 years. Hence a value of 50 means that what is currently 1-in-100 year event will happen every 2 years due […]'''
<!-- IMG FILE -->
[[File:07ae878d8b95d32d616dfd462e96e0c1 IPCC-SROCC-CH_4_12-3000x2581.jpg]]

Figure 4.12 | The colours of the dots express the factor by which the frequency of extreme sea level (ESL) events increase in the future for events which historically have a return period of 100 years. Hence a value of 50 means that what is currently 1-in-100 year event will happen every 2 years due to a rise in mean sea level. Results are shown for three RCP scenarios and two future time slices as median values. Results are shown for tide gauges in the GESLA2 database. The accompanying confidence interval can be found in SM4.2 as well as a list of all locations. The data underlying the graph are identical to those presented in Figure 4.11. The amplification factor is schematically explained in the upper right panel of figure 4.11. Storm climatology is constant in these projections.

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., 2014a <sup>[[#fn:r721|721]]</sup> ).

<!-- END IMG -->
<div id="section-4-2-3-4changes-in-extreme-sea-level-events-block-3"></div>

<span id="waves"></span>
===== 4.2.3.4.2 Waves =====

A warming climate is expected to affect wind patterns and storm characteristics, which in turn will impact wind waves that contribute to high coastal water levels. Wind-wave projections are commonly based on dynamical and statistical wave models forced by projected surface winds from GCMs, notably those participating in the CMIP. In the framework of the Coordinated Ocean Wave Climate Project (COWCLIP), an ensemble of Coupled Model Intercomparison Project Phase 3 (CMIP3)-based global wave projections (Hemer et al., 2013) was produced and the results were summarised in the AR5 (WGI Chapter 13). Casas-Prat et al. (2018) expanded the geographic domain to include the Arctic Ocean, highlighting the vulnerability of high-latitude coastlines to wave action as ice retreats. Reduced sea ice allows larger waves and stronger cyclones in the Arctic Ocean, which can further disrupt and break up sea ice (e.g., Thomson and Rogers, 2014; Day and Hodges, 2018). A review and consensus-based analysis of regional and global scale wave projections, including CMIP5-based projections, has been provided by Morim et al. (2018) as part of COWCLIP. Projections of annual and seasonal mean significant wave height changes agree on an increase in the Southern Ocean, tropical eastern Pacific and Baltic Sea; and on a decrease over the North Atlantic, northwestern Pacific and Mediterranean Sea. Projections of mean significant wave height lack consensus over the eastern North Pacific and southern Indian and Atlantic Oceans. Projections of future extreme significant wave height are consistent in projecting an increase over the Southern Ocean and a decrease over the northeastern Atlantic and Mediterranean Sea. Regional projections of wind-waves have mostly been applied to Europe so far, while highly vulnerable regions have been largely overlooked. This is the case for low-lying islands where impacts of SLR and wave-induced flooding are expected to be severe and adaptive capacity is reduced (Hoeke et al., 2013; Albert et al., 2016).

A number of studies have included waves, in addition to tides and sea level anomalies, to assess coastal vulnerability to SLR using dynamical and statistical approaches. CoSMoS (Barnard et al., 2014) includes a series of embedded wave models to estimate high resolution projections of total water levels along the Southern California coast for different extreme scenarios (O’Neill et al., 2017). Arns et al. (2017) find that an increase in sea level may reduce the depth-limitation of waves, thereby resulting in waves with greater energy approaching the coast. Including wave effects is crucial for coastal adaptation and planning (e.g., Isobe, 2013). For example Arns et al. (2017) report that coastal protection design heights need to be increased by 48–56% in the German Bight region relative to a design height based on the effect of SLR on ESL only. Combining SLR with extreme value theory applied to past observations of tides, storm surges and waves, Vitousek et al. (2017) found that a 10–20 cm SLR could result in a doubling of coastal flooding frequency in the tropics. For the southern North Sea region, Weisse et al. (2012) argue that increasing storm activity also increases hazards from ESL events. Global-scale projections of ESL event changes including wave setup indicate a very likely increase of the global average 100-year ESL of 58–172 cm under RCP8.5 (Vousdoukas et al., 2018c). Changes in storm surges and waves enhance the effects of relative SLR along the majority of northern European coasts, with contributions up to 40% in the North Sea (Vousdoukas et al., 2017).

A stationarity of the wave climate is often assumed for projections of ESL events (Vitousek et al., 2017). Yet, wave contributions to coastal sea level changes (setup and swash) depend on several factors that can vary in response to internal climate variability and climate change, including deep-water wave field, water-depth, and geomorphology. Melet et al. (2018) reported that over recent decades, wave setup and swash interannual-to-decadal changes induced by deep-water wave height and period changes alone were sizeable compared to steric and land-ice mass loss coastal sea level changes.

Comprehensive broad-scale projections of sea level at the coast including regional sea level changes, tides, waves, storm surges, interactions between these processes and accounting for changes in period and height of waves and frequency and intensity of storm surges are yet to be performed.

<div id="section-4-2-3-4changes-in-extreme-sea-level-events-block-4"></div>

<span id="effects-of-cyclones"></span>
===== 4.2.3.4.3 Effects of cyclones =====

Tropical and extratropical cyclones (TCs and ETCs) tend to determine ESL events, such as coastal storm surges, high water events, coastal floods, and their associated impacts on coastal communities around the world. The projected potential future changes in TCs and ETCs frequency, track and intensity is therefore of great importance. After AR5, it was realised that the modelled global frequency of TCs is underestimated and that the geographical pattern is poorly resolved in the case of TC tracks, very intense TCs (i.e., Category 4/5) and TC formation by using low resolution climate models (Camargo, 2013). Over recent years, multiple methods including downscaling

CMIP5 climate models (Knutson et al., 2015; Yamada et al., 2017), high-resolution simulations (Camargo, 2013; Yamada et al., 2017), TC–ocean interaction (Knutson et al., 2015; Yamada et al., 2017), statistical models (Ellingwood and Lee, 2016) and statistical-deterministic models (Emanuel et al., 2008) have been developed, and the ability to simulate TCs has been substantially improved. Most models still project a decrease or constant global frequency of TCs, but a robust increase in the lifetimes, precipitation, landfalls and ratio of intense TCs under global warming. This is consistent with IPCC AR5 and many additional studies (Emanuel et al., 2008; Holland et al., 2008; Knutson et al., 2015; Kanada et al., 2017; Nakamura et al., 2017; Scoccimarro et al., 2017; Zheng et al., 2017). It is expected that these projected increases are intensified by favourable marine environmental conditions, expansion of the tropical belt, or ocean warming in the northwest Pacific and north Atlantic, and increasing water vapour in the atmosphere (Kossin et al., 2014; Moon et al., 2015; Cai et al., 2016; Mei and Xie, 2016; Cai et al., 2017; Kossin, 2017; Scoccimarro et al., 2017; Kossin, 2018). However, it is noted that, in contrast to most models, some models do predict an increase in global TC frequency during the 21st century (Emanuel, 2013; Bhatia et al., 2018).

Previous extensive studies indicated the important role of warming oceans in the TC activity (Emanuel, 2005; Mann and Emanuel, 2006; Trenberth and Fasullo, 2007; Trenberth and Fasullo, 2008; Villarini and Vecchi, 2011; Trenberth et al., 2018) and also revealed TCs stir the ocean and mix the subsurface cold water to the surface (Shay et al., 1992; Lin et al., 2009). The resulting increased thermal stratification of the upper ocean under global warming will reduce the projected intensification of TCs (Emanuel, 2015; Huang et al., 2015; Tuleya et al., 2016). A recent study suggests a strengthening effect of ocean freshening in TC intensification, opposing the thermal effect (Balaguru et al., 2016). It is concluded that it is likely that the intensity of severe TCs will increase in a warmer climate, but there is still low confidence in the frequency change of TCs in the future.

Recent projection studies indicate that trends in regional ETCs vary from region to region, for example, a projected increase in the frequency of ETCs in the South and the northeast North Atlantic, the South Indian Ocean, and the Pacific (Colle et al., 2013; Zappa et al., 2013; Cheng et al., 2017; Michaelis et al., 2017) and a decrease in the numbers of ETCs in the North Atlantic basin and the Mediterranean (Zappa et al., 2013; Michaelis et al., 2017). Note that the projected frequency in ETCs still remains uncertain due to different definitions of cyclone, model biases or climate variability (Chang, 2014; Cheng et al., 2016). Considering these processes implies that changes in TC and ETC characteristics will vary locally and therefore there is low confidence in the regional storm changes, which is in agreement with AR5 WGI Chapter 14 (Christensen et al., 2013).

Observed damages from ETCs/TCs to coastal regions has increased over the past 30 years and will continue in the future (Ranson et al., 2014). The global population exposed to ETCs/TCs hazards is expected to continue to increase in a warming climate (Peduzzi et al., 2012; Blöschl et al., 2017; Emanuel, 2017a; Michaelis et al., 2017). The probabilities of sea level extreme events induced by TC storm surge are very likely to increase significantly over the 21st century. Risk from TCs increases in highly vulnerable coastal regions (Hallegatte et al., 2013), e.g., on coasts of China (Feng and Tsimplis, 2014), west Florida, north of Queensland, the Persian Gulf, and even in well protected area such as the Greater Tokyo area (Tebaldi et al., 2012; Lin and Emanuel, 2015; Ellingwood and Lee, 2016; Hoshino et al., 2016; Dinan, 2017; Emanuel, 2017b; Lin and Shullman, 2017). The ESL return period has greatly decreased over recent decades and is also expected to decrease greatly in the near future, for example, in NYC (by 2030–2045; Garner et al., 2017). It is very likely that the ESL return period in low-lying areas such as coastal megacities decreased over the 20th century and frequencies of still unusual ESL events are expected to increase in frequency in the future. In addition, the compound effects of SLR, storm surge and waves on ESL events and the associated flood hazard are assessed in Chapter 6 (Section 6.3.3.3 and 6.3.4).

<div id="section-4-2-3-5long-term-scenarios-beyond-2100"></div>

<span id="long-term-scenarios-beyond-2100"></span>
==== 4.2.3.5 Long-Term Scenarios, Beyond 2100 ====

<div id="section-4-2-3-5long-term-scenarios-beyond-2100-block-1"></div>

Sea level at the end of the century will be higher than present day and continuing to rise in all cases even if the Paris Agreement is followed (Nicholls, 2018 <sup>[[#fn:r794|794]]</sup> ). The reasons for this are mainly related to the slow response of glacier melt, thermal expansion and ice sheet mass loss (Solomon et al., 2009 <sup>[[#fn:r794|794]]</sup> ). These processes operate on long time scales, implying that even if the rise in global temperature slows or the trend reverses, sea level will continue to rise (SR1.5 report, AR5). A study by Levermann et al. (2013) <sup>[[#fn:r799|799]]</sup> based on palaeo-evidence and physical models formed the basis of the assessment by Church et al. (2013) <sup>[[#fn:r803|803]]</sup> indicating that committed SLR is approximately 2.3 m per degree warming for the next 2000 years with respect to pre-industrial temperatures. This rate is based on a relation between ocean warming and basal melt as used by Levermann et al. (2013), without accounting for surface melt, hydrofracturing of ice shelves and subsequent ice cliff failure, suggested to be a dominant long term mechanism for ice mass loss (DeConto and Pollard, 2016 <sup>[[#fn:r808|808]]</sup> ). Deep uncertainty (Cross-Chapter Box 5 in Chapter 1) remains on the ice dynamical contribution from Antarctica after 2100.

Beyond the 21st century, the relative importance of the long-term contributions of the various components of SLR changes markedly. For glaciers, the long-term is of limited importance, because the sea level equivalent of all glaciers is restricted to 0.32 ± 0.08 m when taking account of ice mass above present day sea level (Farinotti et al., 2019 <sup>[[#fn:r796|796]]</sup> ). Hence, there is ''high confidence'' that the contribution of glaciers to SLR expressed as a rate will decrease over the 22nd century under RCP8.5 (Marzeion et al., 2012 <sup>[[#fn:r797|797]]</sup> ). For thermal expansion the gradual rate of heat absorption in the ocean will lead to a further SLR for several centuries (Zickfeld et al., 2017 <sup>[[#fn:r798|798]]</sup> ). By far, the most important uncertainty on long time scales arises from the contribution of the major ice sheets. The time scale of response of ice sheets is thousands of years. Hence, if ice sheets contribute significantly to sea level in 2100, they will necessarily also contribute to sea level in the centuries to follow. Only for low emission scenarios, like RCP2.6, can substantial ice loss be prevented, according to ice dynamical models (Levermann et al., 2014 <sup>[[#fn:r799|799]]</sup> ; Golledge et al., 2015 <sup>[[#fn:r800|800]]</sup> ; DeConto and Pollard, 2016 <sup>[[#fn:r801|801]]</sup> ; Bulthuis et al., 2019 <sup>[[#fn:r802|802]]</sup> ). For Greenland, surface warming may lead to ablation becoming larger than accumulation, and the associated surface lowering increases ablation further (positive feedback). As a consequence, the ice sheet will significantly retreat. Church et al. (2013) <sup>[[#fn:r803|803]]</sup> concluded that the threshold for perpetual negative mass balance based on modelling studies lies between 1ºC (Robinson et al., 2012 <sup>[[#fn:r804|804]]</sup> ; ''low confidence'' ) and 4ºC ( ''medium confidence'' ) above pre-industrial temperatures. Pattyn et al. (2018) <sup>[[#fn:r805|805]]</sup> demonstrated that with more than 2.0ºC of summer warming, it becomes ''more likely than not'' that the GIS crosses a tipping point, and the ice sheet will enter a long-term state of decline with the potential loss of most or all of the ice sheet over thousands of years. If the warming is sustained, ice loss could become irreversible due to the initiation of positive feedbacks associated with elevation-SMB feedback (reinforced surface melt as the ice sheet surface lowers into warmer elevations), and albedo-melt feedback associated with darkening of the ice surface due to the presence of liquid water, loss of snow, changes in firn and biological processes (Tedesco et al., 2016 <sup>[[#fn:r806|806]]</sup> ; Ryan et al., 2018 <sup>[[#fn:r807|807]]</sup> ). The precise temperature threshold and duration of warming required to trigger such irreversible retreat remains very uncertain, and more research is still needed.

The mechanisms for decay of the AIS are related to ice shelf melt by the ocean, followed by accelerated loss of grounded ice and MISI, possibly exacerbated by hydrofracturing of the ice shelves and ice cliff failure (Cross-Chapter Box 8 in Chapter 3). The latter processes have the potential to drive faster rates of ice mass loss than the SMB processes that are ''likely'' to dominate the future loss of ice on Greenland. Furthermore, the loss of marine-based Antarctic ice represents a long-term (millennial) commitment to elevated SLR, due to the long thermal memory of the ocean. Once marine based Antarctic ice is lost, local ocean temperatures will have to cool sufficiently for buttressing ice shelves to reform, allowing retreated grounding lines to re-advance (DeConto and Pollard, 2016 <sup>[[#fn:r808|808]]</sup> ). A minimum time scale, whereby the majority of West Antarctica decays, was derived from a schematic experiment with an ice flow model by Golledge et al. (2017) <sup>[[#fn:r809|809]]</sup> , where ice shelves were removed instantaneously and prohibited from re-growing. Results of this experiment indicate that most of West Antarctica’s ice is lost in about a century.

Gradual melt of ice shelves accompanied by partial retreat of East Antarctic ice would yield greater ice melt but on a time scale of millennial or longer (Cross-Chapter Box 8 in Chapter 3). Prescribing a uniform warming of 2°C–3°C in the Southern Ocean triggers an accelerated decay of West Antarctica in a coarse resolution model with a temperature-driven basal melt formulation yielding 1–2 m SLR by the year 3000 and up to 4 m by the year 5000 (Sutter et al., 2016 <sup>[[#fn:r810|810]]</sup> ). Formulating an ice sheet model with Coulomb friction in the grounding line zone yields a SLR of 2 m after 500 year for a sub-ice shelf melt of 20 m a <sup>–1</sup> (Pattyn, 2017 <sup>[[#fn:r811|811]]</sup> ). On decadal to millennial time scales the interaction between ice and the solid Earth indicates the possibility of a negative feedback slowing retreat by viscoelastic uplift and gravitational effects that reduce the water depth at the grounding line (Gomez et al., 2010 <sup>[[#fn:r812|812]]</sup> ; de Boer et al., 2014; Gomez et al., 2015 <sup>[[#fn:r813|813]]</sup> ; Konrad et al., 2015 <sup>[[#fn:r814|814]]</sup> ; Pollard et al., 2017 <sup>[[#fn:r815|815]]</sup> ; Barletta et al., 2018 <sup>[[#fn:r816|816]]</sup> ; Section 4.2.3.1.2).

A blended statistical and physical model, calibrated by observed recent ice loss in a few basins (Ritz et al., 2015 <sup>[[#fn:r817|817]]</sup> ) projects an Antarctic contribution to sea level of 30 cm by 2100 and 72 cm by 2200, following the SRES A1B scenario, roughly comparable to RCP6.0. The projected contribution of WAIS was found to be limited to 48 cm in 2200 following the A1B scenario. The key uncertainty in these calculations comes from the dependency on the relation between the sliding velocity and the friction at the ice-bedrock interface. Several parameterisations are used to describe this process. Golledge et al. (2015) <sup>[[#fn:r820|820]]</sup> present values between 0.6–3 m by 2300 for the RCP8.5 scenario. In contrast to the previous studies, Cornford et al. (2015) used an adaptive grid model, which can describe more accurately grounding line migration (Cross-Chapter Box 8 in Chapter 3). Due to the computational complexity of their model, simulations are limited to West Antarctica. Starting from present-day observations, they find that the results are critically dependent on initial conditions, sub ice shelf melt rates, and grid resolution. The glacier with the most uncertain vulnerability is the 120 km-wide Thwaites Glacier, in the Amundsen Sea sector of West Antarctica. Thwaites Glacier is currently retreating in a reverse-sloped trough extending into the central WAIS (Figure 4.8), where the bed is up to 2 km below sea level. In addition to Thwaites, several smaller outlet glaciers and ice streams may contribute to sea level on long time scales, but in the study by Cornford et al. (2015), a full West Antarctic retreat does not occur with limited oceanic heating under the two major ice shelves (Filchner-Ronne and Ross) keeping ice streams flowing into the Ross and Weddell Seas in place. However, the representation of these processes remains simplistic at the continental ice sheet scale (Cross-Chapter Box 8 in Chapter 3).

Nonetheless, recent studies using independently developed Antarctic ice dynamical models (Golledge et al., 2015; DeConto and Pollard, 2016 <sup>[[#fn:r821|821]]</sup> ; Bulthuis et al., 2019 <sup>[[#fn:r822|822]]</sup> ) agree that low emission scenarios, are required to prevent substantial future ice loss ( ''medium confidence'' ). However, observations (Rignot et al., 2014 <sup>[[#fn:r823|823]]</sup> ) and modelling of the Thwaites Glacier in West Antarctica (Joughin et al., 2014 <sup>[[#fn:r824|824]]</sup> ), suggest grounding line retreat on the glacier’s reverse sloped bedrock is already underway and possibly capable of driving major WAIS retreat on century time scales. Whether the retreat is driven by ocean changes driven by climate change or by climate variability (Jenkins et al., 2018 <sup>[[#fn:r825|825]]</sup> ) is still under debate. Hence it is not possible to determine whether a low emission scenario would prevent substantial future ice loss ( ''medium confidence'' ). This is a further elaboration on the SR15 assertion that the chance for passing a threshold is larger for 2°C warming than for 1.5°C warming.

A study by Clark et al. (2016) addresses the evolution of the ice sheets over the next 10,000 years and concludes that given a climate model with an equilibrium climate sensitivity of 3.5°C, the estimated combined loss of Greenland and Antarctica ranges from 25–52 m of equivalent sea level, depending on the emission scenario considered, with rates of GMSL as high as 2–4 m per century. A worst-case scenario was explored with an intermediate complexity climate model coupled to a dynamical ice model (Winkelmann et al., 2015 <sup>[[#fn:r826|826]]</sup> ), in which all readily available fossil fuels are combusted at present-day rates until they are exhausted. The associated climate warming leads to the disappearance of the entire AIS with rates of SLR up to around 3 m per century. A follow up study by Clark et al. (2018) <sup>[[#fn:r827|827]]</sup> addressing the long-term commitment of SLR based on cumulative carbon dioxide emissions points to SLR as an additional measure for setting emission targets. It shows that a 2ºC scenario would result in 0.9 m in 2300 and around 7.4 m in the year 9000 CE.

Similar to the strategy for the 21st century, the long-term projections of sea level were assessed. Since no new CMIP runs are available there are no major new insights in the thermal expansion and glacier component which deviate from the AR5 assessment for the long-term contribution of these components. Some studies updated the contribution of the GIS on long time scales. Vizcaino et al. (2015) <sup>[[#fn:r828|828]]</sup> used a GCM coupled to an ice sheet model to calculate the Greenland contribution which is within the range of estimates presented by Church et al. (2013) <sup>[[#fn:r829|829]]</sup> . This is also true for the ice sheet simulations by Calov et al. (2018) <sup>[[#fn:r830|830]]</sup> which are based on off line simulations with a regional climate model forced by RCP4.5 and RCP8.5 scenarios of three different CMIP5 models. On the other hand, Aschwanden et al. (2019) <sup>[[#fn:r831|831]]</sup> used temperatures to calculate SMB which was used to force an ice sheet model to arrive at much higher values for SLR. However, they used a spatially uniform temperature forcing, which is in conflict with earlier work and overestimated temperatures in the ablation zone (e.g., Van de Wal and Wild, 2001; Gregory and Huybrechts, 2006 <sup>[[#fn:r832|832]]</sup> ). Given this limited and contrasting evidence for Greenland, the assessed values presented in Table 13.8 of Church et al. (2013) <sup>[[#fn:r836|836]]</sup> were also used, but again replacing the Antarctic component by the assessed value from the process and climate scenario-based studies published after 2013. The low scenario in Table 13.8 of Church et al. (2013) without the Antarctic contribution was combined with the RCP2.6 estimates for Antarctica simulated by Golledge et al. (2015) <sup>[[#fn:r833|833]]</sup> , the mean of the RCP2.6 simulations with and without time delay between global mean atmosphere and ocean temperature around Antarctica of Levermann et al. (2014) <sup>[[#fn:r834|834]]</sup> , and the model results of Bulthuis et al. (2019) <sup>[[#fn:r835|835]]</sup> . The medium scenario from Church et al. (2013) <sup>[[#fn:r836|836]]</sup> is combined with RCP4.5 results and the high scenario with RCP8.5. Results are shown in Figure 4.2, Section 4.1 and show a strong divergence of RSL rise over time, whereby the estimates in 2300 range from about 1–2 m under RCP2.6 up to 2–5.5 m for RCP8.5.

The specific trajectories that will be followed may depend critically on if and when certain tipping points are reached. Most critical in that respect are presumably the tipping points corresponding (1) to the threshold where the ablation in Greenland becomes larger than the accumulation, causing an irreversible and nearly full retreat of the ice sheet; and (2) the thresholds for ice shelf stability in West Antarctica, which depend on surface melt and sub-ice melt, combined with uncertainties surrounding MISI and/or MICI. There is deep uncertainty about whether and when a tipping point will be passed. For RCP8.5, the chance of passing a tipping-point are considered to be substantially higher than for RCP2.6.

In summary, there is ''high confidence'' in continued thermal expansion and the loss of ice from both the GIS and AIS sheets beyond 2100. A complete loss of Greenland ice contributing about 7 m to sea level over a millennium or more would occur for sustained GMST between 1°C ( ''low confidence'' ) and 4°C ( ''medium confidence'' ) above pre-industrial levels. Due to deep uncertainties regarding the dominant processes that could trigger a major retreat, there is ''low confidence'' in the estimates of the contribution of the AIS beyond 2100, but our estimates (2.3–5.4 m in 2300) for RCP8.5 are considerably higher than presented in AR5. High-emission scenarios or exhaustion of fossil fuels over a multi-century period lead to rates of SLR as high as several metres per century in the long term ( ''low confidence'' ). Low-emission scenarios lead to a limited contribution over multi-century time scales ( ''high confidence'' ). Discriminating between 1.5°C and 2°C scenarios in terms of long-term sea level change is not possible with the limited evidence. Hence, it is concluded that the SLR on millennial time scales is strongly dependent on the emission scenario followed. This, combined with the lack in predictability of the tipping points, indicates the importance of emissions mitigation for minimising the risk to low-lying coastlines and islands ( ''high confidence'' ).

<div id="section-4-2-3-5long-term-scenarios-beyond-2100-block-2" class="box"></div>

<span id="box-4.1-case-studies-of-coastal-hazard-and-response"></span>